Historical forest structure and fire in Sierran mixed-conifer forests reconstructed from General Land Office survey data

Dry forests of the western United States (ponderosa pine, dry mixed conifer) are often considered at risk of uncharacteristic severe fires, but recent research has found historically extensive severe fire. This has left divergent perspectives about how to restore dry forests, protect people and infrastructure from fire, and interpret the ecological effects of large fires, such as the 2013 Rim fire, a human-set 104,000 ha fire in the western Sierra Nevada Mountains. To help resolve this uncertainty, I used new methods to reconstruct historical forest structure and fire and test 11 hypotheses about them, using A.D. 1865–1885 General Land Office surveys, across 330,000 ha of Sierran mixed-conifer forests. The reconstructions show these historical forests were open and park-like in places, but generally dense, averaging 293 trees/ha; shade-tolerant trees and large trees were abundant, but smaller (,60 cm diameter) pines and oaks numerically dominated. These smaller trees, along with common understory seedlings and saplings and almost pervasive shrubs, created abundant ladder fuels. It is not surprising, given these conditions, that just 13–26% of historical Sierran mixed-conifer forests had only low-severity fire, with mixed-severity fire over 43–48%, and high-severity fire over 31–39% of the land area. The high-severity fire rotation was 281 years in the northern and 354 years in the southern Sierra, short enough to contribute to high levels of heterogeneity, including abundant areas and large patches (up to 9400 ha) of earlysuccessional forest and montane chaparral, but long enough to allow recovery of old-growth forest over large land areas. Proposals to reduce fuels and fire severity would actually reduce, not restore, historical forest heterogeneity important to wildlife and resiliency. Sierran mixed-conifer forests are inherently dangerous places to live, which cannot be changed without creating artificial forests over large land areas. However, people can adapt to fires by channeling development to safer areas and modifying ignition zones near houses and communities to survive fire.


INTRODUCTION
''...the departure of views begins with the relative certainty of fire frequency and spatial intensity in presettlement times. There is too little compelling evidence and incomplete rangewide research to conclude a precise pattern of fire frequency or severity in presettlement times. There were very probably areas that burned frequently (less than tenyear intervals), but some areas within the same vegetation type probably escaped burning for much longer periods and built up sufficient fuel loads to burn with high intensity...forest conditions were not largely 'open or parklike,' in the words of John Muir; rather, there was a mix of dark, dense, or thick forests in unknown comparative quantities...'' (Alternative View: Sierra Nevada Ecosystem Project 1996 Volume 1:63) Dry forests of the western United States could face increased drought, fire, and insect outbreaks in an altered condition because of past logging, livestock grazing, and fire exclusion, yet how to restore them is clouded by competing evidence and uncertainty about historical forests and wildfires, as is evident in the quote above. Evidence shows that some dry forests, which include ponderosa pine (Pinus ponderosa) forests and dry mixed-conifer forests with more firs (Abies, Pseudotsuga), were historically maintained in a relatively open, often park-like condition by low-severity fires (e.g., Covington and Moore 1994, Fulé et al. 2003. These fires periodically limited fuel buildup and large high-severity fires. However, paleoecological studies have revealed past high-severity fire in these forests (e.g., Pierce et al. 2004). Early scientific reports (Shinneman andBaker 1997, Baker et al. 2007) and aerial photography (Hessburg et al. 2007), tree-ring reconstructions (Ehle and Baker 2003), reconstructions from the General Land Office surveys in the late-1800s (Williams and Baker 2012a, b), and spatially-extensive age-structure analysis ) present substantial evidence high-severity fire and dense forests were a significant part of historical forests. In the Sierra Nevada Mountains, the subject of this study, similar competing evidence led to both a main and alternative view of the role of fire and structure of historical forests after an extensive multi-author scientific study (Sierra Nevada Ecosystem Project 1996). The main view is above, and the alternative is in the initial quote.
Uncertainty about the structure of historical forests and role of high-severity fire has scientific and policy implications. Are sensitive species (e.g., spotted owls), endangered by uncharacteristic high-severity fires (Weatherspoon et al. 1992, Spies et al. 2006 or have these fires long provided early-successional vegetation favored for foraging (e.g., Bond et al. 2009)? Are highseverity fires increasing to unnatural levels (e.g., Adams 2013), threatening natural ecosystems as well as houses, or are these fires burning episodically at rates similar to historical rates (Baker 2012)? Is it restoration if these forests are extensively thinned to prevent high-severity fires, or will this reduce historically important high-severity fires and add to adverse effects of fire exclusion?
In a series of recent studies (Baker 2012, Baker 2012a, b, 2013), we used the General Land Office (GLO) surveys to help answer these questions. The surveys, mostly done in the late-1800s in the western U.S., laid out the public land-survey system as a grid of 1.6 km 3 1.6 km section lines intersecting at section corners. Surveyors collected systematic data on tree attributes at corners and listed dominant trees and shrubs in order of abundance along section lines. These geographically precise data can be used, with new methods (Williams and Baker 2011), to accurately reconstruct forest structure (e.g., tree density) and fire severity across large land areas. Earlier use of the GLO data in the Sierra Nevada included analysis of tree sizes and forest composition (Fites-Kaufman 1997, Manley et al. 2000, Hyde 2002), analysis of changes in a burned chaparral area (Wilken 1967), and reconstruction of tree density (Maxwell et al. 2014).
I used GLO survey data to analyze 11 hypotheses (Table 1) about historical forest structure and fire in Sierran mixed-conifer (SMC) forests of the western Sierra Nevada, California. I supplemented survey data with 208 quotes from early scientific and agency reports about fire and forest structure (Appendix A). Hypothesis H 1 , that historical SMC forests were somewhat open, is supported by studies that suggest historical SMC forests had low tree densities, and by early observations (Appendix A: Q112-Q134). North et al. (2009:9) indicate that ''All reconstruction studies, old forest survey data sets, and 19th-century photographs (Gruell 2001, McKelvey andJohnston 1992) suggest that frequently burned forests had very low tree densities. '' McKelvey and Johnston (1992:237) said: ''The stand structure at the turn of the century [A.D. 1900] was often quite open'' and was ''...one dominated by large, old, widely spaced trees...' ' (McKelvey and Johnston 1992:241). Early photographs of unlogged SMC forests support a generally open forest structure, except in the northern Sierra on cooler, moister sites (Gruell 2001). Scholl and Taylor (2010:375) v www.esajournals.org also say: ''Our reconstruction supports written descriptions of mixed-conifer stands as being low in density...'' In contrast, Sudworth's (1900) early data suggest historical SMC forests averaged 229-235 trees/ha in the northern (Stephens 2000) and 278 trees/ha in the southern Sierra (Stephens and Elliott-Fisk 1998) for trees .30.5 cm. Early observations also support the idea that SMC forests were dense (Appendix A: Q135-Q147). Thus, evidence supports both the main view and the alternative in the initial quote.
Hypothesis H 2 suggests historical SMC composition had similar amounts of shade-tolerant and intolerant trees. McKelvey and Johnston (1992:235), said ''Pines did not dominate the forests, either in numbers or in volume.'' In contrast, Scholl and Taylor (2010:375) found ''a large proportion of large-diameter shade-intolerant and fire-tolerant pines and oak.'' Regarding the basal area hypothesis, H 3 , reported dominance by large trees suggests historical basal area and quadratic mean diameter were also moderately large. Basal area (BA) is the total cross-sectional area of tree stems, and quadratic mean diameter (QMD) is the diameter of the tree of mean basal area. Safford (2013) suggested a historical mean BA of 33.2 m 2 /ha based on 13 reference values, and I thus use that as the hypothesis here. No specific estimate of historical quadratic mean diameter is available, thus a hypothesis is not posed.
Hypotheses H 4 and H 5 are that historical SMC forests were dominated by large trees, and small trees were lacking or rare. Based on data in Sudworth (1900), McKelvey and Johnston (1992:234) suggested that ''...most stems exceeded 25 inches in d.b.h.'' and trees ,28 cm (11 inches) ''...were uncommon, though patches of very small regeneration appear to have been present.'' Sudworth's data also suggested to these authors that ''sugar pine, Douglas-fir, and white fir occurred only as very large trees.'' Similarly, Scholl and Taylor (2010) found large-diameter trees were historically dominant. Gruell (2001:106) said that early photographs showed that SMC forests generally had ''large trees either widely spaced or close together...'' Early observations report old-growth with dominant large trees (Appendix A: Q112-Q117) and relatively few small trees, at least where fire was common (Appendix A: Q164, Q167-Q173).
Hypothesis H 6 , that historical SMC forests had high shrub cover, is based on the observation that shrub cover declined after EuroAmerican settlement due to shading by increased conifer cover (Gruell 2001, North et al. 2009), intense early sheep grazing (Sudworth 1900, Leiberg 1902, Vankat and Major 1978, McKelvey and Johnston 1992, overbrowsing by deer, and a decline in fire-stimulated shrubs due to fire exclusion (Vankat and Major 1978). Early photographs (Gruell 2001:106-107) often suggested only ''scattered understory trees or shrubs...'' or ''a patchy chaparral understory with numerous openings, '' or grassy or oak-dominated understories. Chang (1996) also suggested patchy and variable understory shrubs in historical SMC forests. Some early observations report loss of shrubs to livestock grazing (Appendix A: Q198-Q199) and a sparse understory by about 1900 Historical SMC forests were dominated by large trees (i.e., .50% of trees were .60 cm in diameter). H 5 Historical SMC forests had relatively low abundance of small trees (i.e., ,10% of area with understory trees). H 6 Historical SMC forests had generally high shrub cover (i.e., .75% of area). H 7 Low-severity fire historically characterized !85% of SMC landscapes, with ,15% of area having other fire severities. H 8 High-severity fire in historical SMC forests occurred with long fire rotations (i.e., .500 years). H 9 Patches of contiguous high-severity fire area did not exceed 250 ha. H 10 Historical northern Sierran forests differed in forest structure and fire from southern Sierran forests, based on the attributes tested in H 1 -H 9 . H 11 Phases of Sierran mixed-conifer forests differed in forest structure and fire, based on the attributes tested in H 1 -H 9 .
(Appendix A: Q200-Q203). Hypotheses H 7 -H 9 about historical fire are based on the finding that SMC forests historically had a fire regime with ''frequent, low-severity fires'' and ''a low incidence of high-severity, or stand-replacing fire'' but with some uncertainty about the latter (Collins and Stephens 2012:7). This is supported by Chang (1996). Scholl and Taylor (2010) suggested fires were mainly low severity but patchy, allowing some shade-tolerant trees to reach the canopy. Upper-elevation SMC forests may have had perhaps 15% of total burned area in high-severity fire, but as ''many small stand-replacing patches (,4 ha) and few large patches (.60 ha)'' (Collins and Stephens 2010:937). Stephens et al. (2007) estimated about 5% high severity in historical SMC forests. These ideas are supported by early observations in some cases (Appendix A: Q14-Q27).
The last two hypotheses suggest SMC forests varied in structure and fire regime between north and south (H 10 ) and among compositional phases (H 11 ). Gruell (2001), for example, found that historical SMC forests were denser on cooler, moister sites in the northern Sierra.

Study area
The study focuses on Sierran mixed-conifer (SMC) forests in the lower/middle montane zone (Barbour and Minnich 2000) on the western side of the Sierra Nevada Mountains from south of Quincy and Blairsden (Fig. 1a) to near Miracle Hot Springs, California (Fig. 1d). This forest extends from about 300-1800 m elevation in the northern Sierra and from about 1200 to 2300 m elevation in the southern Sierra Nevada (Fites-Kaufman et al. 2007).
Major trees include ponderosa pine (Pinus ponderosa Lawson and C. Lawson), sugar pine (Pinus lambertiana Douglas), incense cedar (Calocedrus decurrens (Torr.) Florin), Sierra white fir (Abies concolor (Gord. & Glend.) Lindl. Ex Hildebr. var. lowiana (Gord. & Glend.) Lemmon), Douglas-fir (Pseudotsuga menziesii (Mirb.) Franco), and California black oak (Quercus kelloggii Newberry). About 25 other trees and large shrubs were used by surveyors as bearing trees (see Appendix B for taxonomic authorities). Ponderosa and Jeffrey pine (Pinus jeffreyi Balf.) can co-occur, especially at higher elevations and in the southern Sierra, but were poorly distinguished by surveyors, thus are both included in ponderosa. Although giant sequoias (Sequoia gigantea (Lamb. Ex D. Don) Endl.) also occur in small groves, GLO survey data are often too coarse to provide useful data about them, and they are thus not studied. Major shrubs include Ceanothus integerrimus (and C. parvifolius in the southern Sierra) at lower elevations and C. cordulatus at higher elevations, often both with Arctostaphylos patula or with A. viscida at lower elevations. Other common shrubs include Prunus emarginata, Quercus vacciniifolia, Cornus nuttallii, Chrysolepis sempervirens, Ribes roezlii, Corylus cornuta var. californica, and Chamaebatia foliolosa. Surveyors used many common names for shrubs (Appendix B).
I selected study areas in the northern and southern Sierra Nevada Mountains ( Fig. 1) to address hypothesis H 10 : that the two regions differed. Entering GLO survey data is laborious, and in each study area I completed about 25 townships or 230,000 ha. A township is a publicland survey-system unit about 9.6 km 3 9.6 km containing 36 sections each about 1.6 km 3 1.6 km. Townships were chosen that had early (pre-1890) surveys, high-quality surveyors (Appendix C), and to span the elevational and latitudinal range of SMC forests. Selection favored areas relatively undisturbed at the time of the surveys to provide reference information about historical forests. Thus, I included current national parks, wilderness areas, and other protected areas. This undisturbed condition was rarer in the north than the south.
To test hypothesis H 11 , I divided study areas into compositional and roughly elevational phases (Barbour and Minnich 2000): (1) ponderosa pine-Douglas-fir, (2) Sierran mixed-conifer, and (3) white fir (Table 2). These phases best correspond with Society of American Foresters cover types mapped in CALVEG, a satellitebased mapping system created by the Pacific Southwest Region of the U.S. Forest Service (www.fs.fed.us/r5/rsl/projects/mapping). CAL-VEG has been shown to have reasonable accuracy (Franklin et al. 2000). Associated vegetation (e.g., California black oak, canyon live oak, chaparral, grasslands) was not assigned to these phases in CALVEG, but can occur in any phase.
v www.esajournals.org The GLO surveys and their use in reconstructing historical forest structure General Land Office (GLO) surveyors were required to record species, diameter, distance, and azimuth of two bearing trees on opposite sides of section-lines at quarter corners at the 0.8km mark and four bearing trees, one per section, at section corners at the 1.6-km mark along section lines (USDI General Land Office 1881, 1894. Surveyors also recorded where they left forests and entered openings (e.g., grasslands, chaparral patches) and vice versa, and often also recorded entry and exit locations for areas of scattered trees or other conditions. Most surveyors used qualitative descriptors of forest density and tree size or the quality of timber (e.g., heavily timbered, dense forest, good timber). At the end of each section line, they were required to list dominant overstory trees or shrubs in order of abundance and do the same for understory trees and shrubs, also using density terms (e.g., dense, scattered).
Bearing-tree data at section corners and the section-line data constitute the GLO data used in the analysis. Bearing trees, typically .10 cm diameter, were measured accurately and selected with little bias, based on resampling at relocated section corners, and thus provide a valid statistical sample of dry forests (Williams and Baker 2010). Section-line data provide a statistically valid line-intercept estimate of percent cover (Butler and McDonald 1983).
Bearing-tree data can be used, with new methods we developed (Williams and Baker 2011), to reconstruct tree density, composition, basal area, quadratic mean diameter, and diameter distributions. Tree data are pooled to produce sufficient sample size. I pooled six contiguous corners (518 ha) for tree density, nine corners (777 ha) for composition, basal area, and quadratic mean diameter, and 12 corners (1036 ha) for reconstructing diameter distributions. In an extensive modern plot-based accuracy trial (Williams and Baker 2011), these levels of pooling led to the lowest relative mean absolute errors (RMAE) for tree density, varying from Table 2. Area in three phases of the Sierran mixed-conifer forest and the Society of American Foresters (SAF) cover types that occur in each phase. Note that these are before removal of human-affected areas.  (207,215,217,218,221,235,238,248,250,255,256) 14.4% to 23.0% among three states, and for basal area, varying from 21.0% to 25.4%. RMAE is 100 3 (jGLO Survey estimate-plot estimatej)/plot estimate. Composition was 89-94% accurate and diameter distributions 87-88% accurate. GLO reconstructions thus approach the accuracy of tree-ring reconstructions of historical forest structure, which also have error (Scholl and Taylor 2010).
The new methods (Williams and Baker 2011) require field data to develop equations to estimate the Voronoi area of each tree and its crown radius from tree diameter (Williams and Baker 2011). A field assistant and I revisited 56 section corners in the study areas, spanning phases and a wide cross-section of stand structures. At each corner, we collected needed data for 2-4 of the nearest trees, aiming for 25-30 trees of each major species in the study area. For each tree, we measured diameter (at about 30 cm height, which is where surveyors likely measured diameter) using a caliper and crown radius using a densitometer and laser distance meter. The Voronoi area for the tree is the area of ground nearer to the subject tree than any other tree. We estimated this area by mapping, using the laser distance meter and a sighting compass, the location of each of at least six neighboring trees, with at least one per 908 of azimuth (Delincé 1986). In ArcGIS, I recreated the Voronoi polygon and measured its area. I then fit crown radius and Voronoi equations using regression (Minitab) for each major species, species groups (e.g., pines), and a pool of minor species (Appendix D). Bearing-tree data were then used with the equations to do reconstructions (Williams and Baker 2011). We also revisited about 100 section lines, where common names used by surveyors for trees and shrubs were crosschecked.
GLO survey data have some limitations. The fine-scale historical structure of SMC forests is often described as having highly clustered groups of trees separated by openings (North et al. 2009). The GLO survey data cannot discern this structure, as they are limited to scales exceeding a minimum three-corner reconstruction polygon of about 259 ha (Williams and Baker 2011). Surveyors did not all follow the instructions (e.g., USDI General Land Office 1881). Section-line descriptions or bearing trees can be missing, with no explanation, and some surveyors did not use density terms consistently. To offset this, I selected areas with high-quality surveys, analyzed and rated surveyor quality (Appendix C), then used the highest-quality records for specific analyses.
Some critiques of our GLO methods have appeared. Fulé et al. (2014) suggested some of our GLO methods were invalid. In response, we explained that our methods were extensively tested, validated, and shown to be accurate, and we added new corroboration . Hagmann et al. (2013) suggested our GLO methods overestimated tree density in one area, but their inventory data, collected decades after surveys, omitted small trees and were collected in areas that had been logged . Maxwell et al. (2014) suggested surveyors were biased, but appeared unaware of Williams and Baker (2010) who found very low surveyor bias and error. Maxwell et al. also said GLO data are aggregated small point samples that could lead to misguided restoration, then used an aggregate of small point samples (185 small plots/transects in ,0.1 ha) totaling ,20 ha in a watershed of about 80,000 ha (a 0.025% sample) to guide restoration. Maxwell et al. had data for 1509 trees .100 years old in their 80,000 ha study area, thus one tree per 53 ha. GLO data provide 66% more tree data, with one tree per 32 ha. Maxwell et al. (2014:2) said ''GLO estimates of forest structure and species composition lack sufficient detail to guide forest restoration management,'' but their sample of 1509 trees in 185 plots averages only 8 trees per plot, which seems to provide little detail to guide forest restoration management. Authors of some tree-ring studies thus appear to have overstated the merits of their methods relative to GLO methods, which often use similar tree sample sizes to produce comparable reconstructions with similar accuracy Baker 2012a, Maxwell et al. 2014). Tree-ring methods could provide finer detail and higher accuracy if sample size was larger, but this would then be infeasible to replicate across large landscapes. Even the roughly 12,500 direct measurements of trees across 400,000 ha of GLO data would be challenging to replicate with tree-ring methods, and the evidence needed to use tree-ring methods has often disappeared, because of disturv www.esajournals.org bance from land uses (Maxwell et al. 2014).

Separating EuroAmerican effects
The extended period of Indian influence on fire and forest structure in the study area had complex and heterogeneous effects, but ended abruptly in the middle 1800s (Parker 2002). The western Sierra experienced a gold rush beginning in 1848 and also early livestock grazing and logging (Beesley 2004). Mining impacted forests directly and also required logs for the mines, houses, and water systems. Since the study's focus is the pre-EuroAmerican historical forest, I used buffer analysis in GIS, with the survey data, to identify and exclude parts of the study areas where EuroAmerican land uses may have directly altered forests by the time of the surveys. As surveyors walked section lines, they recorded, in links from a section corner, where observed landuses occurred, including point locations for small features (e.g., a building) and entry and exit locations for larger features (e.g., mining, farming). These features were mapped accurately, likely to within a few tens of meters. Early roads and mines often correspond closely with modern roads and abandoned mines visible on topographic maps. In the GIS, I assigned 650 links (20 m total length) as an extent for point features, so they would show up on maps and also as a rough estimate of extent.
A total of 2326 records of human uses in the northern and 389 in the southern Sierra was digitized. The number of sawmills identified by surveyors in the northern Sierra was 37. Beesley (2004) estimated 150 sawmills were in operation in the counties of the northern Sierra Nevada north of Sacramento between 1850 and 1900, thus it seems reasonable that about a fourth of those would have been operating in my study area. Features that represent the transportation system (i.e., roads, trails) or water system (i.e., ditches, reservoirs) may have been associated with specific mining or logging operations or other activities, but surveyors did not consistently record associations. These features also may have had multiple uses in many cases. Thus, I put them only in broad categories (i.e., transportation, water system).
I buffered all features to estimate the potential width of an ''effect zone'' (e.g., Forman and Deblinger 2000) in nearby forests, a zone adjoining the land use where trees may have been removed by mining, logging or other activities. Methods for estimating the effect zones for each land use and using buffers to spatially model the effect zones are given in Appendix E, along with detailed findings. I merged all estimated effect zones, then used the resulting map to erase affected areas, leaving a complementary area comparatively unaffected by Euro-American land uses. The merged map of buffered human effects covers 102,323 ha (39.5%) of the 235,805-ha northern Sierra Nevada area, leaving a 133,482-ha unaffected area (Fig.  1a), and in the southern Sierra covers 38,856 ha, leaving an unaffected area of 196,461 ha (Fig. 1bd).
Reconstructing historical fire severity and fire rotation Fire severity was reconstructed using three approaches: (1) structure-based evidence from a combination of tree sizes and tree density from survey tree data, (2) evidence from section-line descriptions of patches of chaparral and scattered trees, and (3) evidence from combined tree data and section-line data. In the first approach, reconstruction of fire severity is based on forest structure from combined small trees, large trees, and tree density, as used previously Baker 2012a, 2013). I first intersected reconstructed tree-density (6-corner pools) with diameter distributions (12-corner pools) and used the 6-corner intersection. Fire-severity is based on calibration using 64 tree-ring reconstructions, where authors reconstructed forest structure and identified fire severity at the same locations (Williams and Baker 2012a). Low fire severity identified by authors corresponds with 6-corner polygons in which small trees were 46.9% of total trees, large trees were .29.2% of all trees, and tree density was ,178 trees/ha. High fire severity occurs where small trees were .50.0% of total trees and large trees were ,20.0% of total trees. Mixed severity corresponds with polygons between low and high.
The second approach uses evidence of highseverity fire shown by chaparral. Surveyors were required to record where they left forest and entered chaparral patches and vice versa. Chaparral patches are identified in section-line data by dominance by montane chaparral shrubs, priv www.esajournals.org marily Ceanothus integerrimus, C. cordulatus, Arctostaphylos patula or A. viscida. Surveyors described 57% of chaparral section-line length as ''dense.'' Montane chaparral dominates after high-severity fire in SMC forests (Cronemiller 1959, Nagel andTaylor 2005), which is strongly supported by early observations (Appendix A: Q57-Q78). Montane chaparral is favored after high-severity fire because the two Ceanothus and A. patula have fire-stimulated seed, persistent seedbanks, and can resprout after fire (Cronemiller 1959, Knapp et al. 2012. Arctostaphylos viscida is an obligate seeder, but its seed germination is not stimulated by fire (Kauffman and Martin 1991). Ceanothus integerrimus, C. cordulatus, and A. patula have refractory seed, but heat shock breaks dormancy imposed by a hard seed coat in Ceanothus and a chemical in charred wood or smoke does the same in A. patula (Keeley 1991). Thus, these shrubs are postfire seed recruiters (Keeley 1991). Severe fires in SMC forests usually greatly increase montane chaparral shrubs. Mean shrub cover, for example, increased from 8.4% in a low-severity area to 17.2% in medium severity to 53.0% on a highseverity area of one fire (Crotteau et al. 2013). In five high-severity fires, shrub cover in 60% of plots was .60% (Collins and Roller 2013).
Not all chaparral originated after forest fires or is successional to forests, at least on century timescales (Show 1924, Nagel andTaylor 2005). I estimated ;80% of chaparral area was successional to forests after these fires (Appendix F), which is similar to early estimates of two thirds to three quarters (Appendix A: Q95, Q98). The 20% not successional to forest either experienced another fire that maintained the chaparral (e.g., Appendix A: Q100-Q103) or was in an environmental setting unfavorable to forest (Appendix F). To approximately account for the 20% not successional to forest, I reduced chaparral area that indicates high-severity fire by 20%.
Unfortunately, surveyors did not record chaparral entry and exit locations for 27.7% of the area in the northern and 55.0% in the southern Sierra Nevada. These surveyors did not follow instructions (USDI General Land Office 1881, 1894), and instead described lines as having forest ''and chaparral,'' providing no entry/exit locations or even the length of chaparral along section-lines. Thus, we know chaparral occurred on these lines, but not how much. To estimate missing chaparral, I divided documented chaparral area by the fraction of total unaffected area with chaparral entry/exit data. The added chaparral area occurred in areas mapped as low-or mixedseverity. As an approximation, I removed the hectares of added chaparral area from hectares of low-and mixed-severity areas in a manner that maintains their relative proportions.
Section-lines explicitly recorded as having scattered trees also likely represent high-severity fires. These lines typically also had one or more of the four main chaparral shrubs dominant in the understory, which was described as dense on about one-third of line-length. These lines also had many missing bearing trees, about 5% of line-length had understory seedling/sapling pines and 17% had understory oaks. These lines likely represent high-severity fires with more survivors and/or more post-fire trees than in chaparral patches. This pattern was also recorded many times in early observations (Appendix A: Q79-Q94, Q97). Slightly reduced dense shrubs with scattered small trees were reported 2-6 decades after high-severity fires in SMC forests (Merriam 1899, Show 1924, Cronemiller 1959, Wilken 1967, Conard and Radosevich 1982. This interpretation is also supported by comparison with maps by Leiberg (1902), described in the next section. In the northern Sierra Nevada, 100.0% of chaparral patches identified by surveyors were mapped later by Leiberg (1902) in his 75-100% burned category (high-severity fire). Similarly, 79.9% of patches with scattered trees were mapped later by Leiberg as high-severity fire.
In the third approach, combined tree-and section-line data were used to classify fire severity for forest openings (Fig. 1). An opening is a section corner lacking recorded bearing trees, and a corner with scattered trees had ,50% of expected bearing trees. These are closely intermixed where they occur. Most surveyors in the southern (Appendix C), but not northern Sierra, used combinations of density terms for the overstory and understory in descriptions of section-lines containing these corners. A null hypothesis of random usage is rejected for both types of corners (Table 3). Indeed, 95% of corners with openings and 99% of corners with scattered trees had understories described as dense where v www.esajournals.org overstory density was not described, thus was neither dense nor scattered, but still forested (Table 3). The dense/thick understories were almost entirely dominated by fire-stimulated Ceanothus integerrimus and Arctostaphylos patula, and nearly all also had understory trees, which were much less common in SMC forests in general.
I interpret this combination as a forest regenerated after high-severity fire that still retains the dense montane chaparral understory. The chaparral and scattered trees that identify highseverity fire in the second approach are similar but more fully forested. Trees regenerating in post-fire chaparral start overtopping shrubs within 8-10 years in the southern Sierra, and shrubs may reach the end of their lifespan after 35 years (Cronemiller 1959). Second-growth forests 30 years old may have 1000 or more trees/ha, averaging about 10-15 cm diameter (Dunning and Reineke 1933), also evident from early observations (Appendix A: Q96, Q99). Dense/thick understories seldom occurred in mature forests described as mature timber, but I used this as an indicator of mixed-severity fire. I left the other combinations as uncertain forest corners.
Fire rotation, the expected time to burn once across an area equal to a study area of interest, was calculated for high-severity fire as the period of observation divided by the fraction of the study area burned at high severity during that period (Baker 2009). The period of observation is based on the time needed to reach the size of a large tree. Tree growth is faster in the western Sierra Nevada, and I used ,40 cm to define small trees and !50 cm to define large trees, in contrast to ,30 cm and !40 cm, respectively, used in Arizona and Oregon (Williams and Baker 2012a). A 40-cm tree averaged about 120 years old in Arizona and 140 years old in the Blue Mountains of Oregon (Williams and Baker 2012a). In contrast, in the Sierra, a 50-cm tree was about 105-120 years old for sugar pine (Hodge 1906, Hall 1909, about 100-125 years for ponderosa pine (Hodge 1906, Hall 1909, Moore 1913, and about 100-155 years for white fir (Hodge 1906, Hall 1909, Moore 1913. I use 110 years as the average for a 50-cm tree, which is then used to reconstruct fire severity and estimate fire rotation for the 110-year period preceding the surveys, beginning about 1755-1775 and ending about 1865-1885 (Fig. 2).
Individual fires cannot be reconstructed from GLO data, but the area of contiguous patches that burned at high severity during the reconstruction period can be estimated from the final map of fire severity. I merged contiguous (touching) polygons of reconstructed high-severity fire and measured the area of each merged polygon. To avoid slivers and small polygons created by GIS operations, I omitted polygons ,50 ha in area. I then constructed a binned histogram of patch sizes, and compared the

Cross-validation
I compiled data from locations within or near the study area for which there are tree-ring reconstructions or early scientific inventories that report tree density or basal area, then used the GLO surveys to reconstruct tree-density and basal area for these same locations (Appendix G). Comparing these allows cross-validation of accuracy (Williams and Baker 2011). As a quantitative measure of the accuracy of the GLO estimates, I used RMAE, defined earlier.
Corroboration: Comparing the reconstructions to the Leiberg maps A government scientist, John B. Leiberg, studied part of the northern Sierra Nevada about A.D. 1900, before national forests were established, to inventory and map forest structure (composition, timber volume, etc.) and threats to the forest (logging, fires, grazing, other human activities). I obtained digital versions of Leiberg's maps from the U.S. Forest Service's Pacific Southwest Research Station (www.fs.fed.us/ psw/topics/ecosystem_processes/sierra/gis). Leiberg's study area overlaps 208,481 ha of my northern-Sierra study area, facilitating detailed comparison. However, after removing humanaffected areas, the overlap area was 108,787 ha.
The spatial error in the Leiberg maps is unknown. In the GIS maps, the total area of forest in the cover-type map is only 1.03 times (3% error) the area reported by Leiberg, and total chaparral area, foothill woodland area, and logged area have errors of only 2.3-6.7%. However, Leiberg's burn map may have one larger error. He reported in two places (Leiberg 1902:41, 186) that there were 715,440 acres that had !50% burned. In the GIS map of burns in Leiberg's whole study area, there are only 447,302 acres in the sum of the 50-75% and 75-100% burned categories. Much of the 50-75% category appears missing from the map. This is not a digitizing error; the printed map also has little area in this category. Thus, I focus on 5-25% and 75-100% burned categories.
Testing hypotheses I used an initial alpha level of 0.05 for statistical tests, Bonferroni adjusted for multiple tests. I  Leiberg (1902) mapping relative to dates of surveys.
v www.esajournals.org tested H 1 and H 3 using one-sample t-tests and H 2 using ANOVA. I tested H 4 to H 9 using null hypotheses and chi-square tests of counts. For H 4 , I converted expected and actual percentages to counts (number of trees). For H 5 and H 6 , which use section-line data, I converted expected and actual percentages to counts of the number of 1-km segments of section-line length. For H 7 , which uses area data, I converted expected and actual percentages to counts of the number of 1000-ha areas. H 8 will be rejected if the actual value does not exceed the expected value of 500 years. H 9 was tested using a chi-square test with Yates' correction. I tested the null hypotheses of no difference between north and south (H 10 ) and among phases (H 11 ) using two-way ANOVAs with followup one-way ANOVAs and Tukey's multiple comparison procedure. H 10 and H 11 are tested along the way while testing corresponding hypotheses. I tested median tree densities using Kruskal-Wallis tests followed by a Bonferronicorrected Mann-Whitney as a follow-up test. I compared diameter distributions using chisquare tests of counts in each 10-cm diameter class. Clipping with maps of phases produced some slivers, as pools of corners were not constructed to match boundaries of phases, thus I first omitted polygons ,100 ha in area for 6corner pools, ,200 ha for 9-corner pools, and ,300 ha for 12-corner pools, which still led to a large sample in each phase. Statistical testing used Minitab (Minitab, State College, Pennsylvania, USA).

Historical forest density (H 1 )
Hypothesis H 1 , that historical SMC forests had mean tree densities that were low (i.e., ,150 trees/ha), was rejected for all phases in both regions, except white fir in the north (Table 4), and rejected overall in the north and south, as well as pooled across regions and phases. I used an initial a ¼ 0.05, Bonferroni-corrected to a ¼ 0.0055, given nine planned tests, one per phase, one overall per region, and one for pooled data. Only 23% of area in the north and 33% in the south had a somewhat open, low-density condition with ,150 trees/ha (Fig. 1).
Mean forest densities (Table 4) did not differ significantly between the northern and southern Sierra Nevada (F ¼ 1.23, df ¼ 1, 810, p ¼ 0.268) or among the three phases (F ¼ 0.07, df ¼ 2, 810, p ¼ 0.935), based on a two-way ANOVA, thus hypotheses H 10 and H 11 are rejected for this measure. Standard deviations did not differ significantly between north and south (Levene's test statistic ¼ 0.21, p ¼ 0.645) or among phases (Levene's test statistic ¼ 0.13, p ¼ 0.874). Reconstructed mean forest density across the study area thus can be pooled across the phases and regions, and was 293 trees/ha with a standard deviation of 477 trees/ha (Table 4).
Median forest densities, which ranged across phases and regions from 179 to 239 trees/ha (Table 4) are also useful, particularly because the distribution of tree densities was right skewed, with a long tail. About 16% of forest area exceeded 400 trees/ha and 3% had 1,000-9,000 trees ha. Indeed, 65% of the northern and 46% of the southern Sierra were dense, with .200 trees/ ha, and 34% of the northern and 21% of the southern Sierra were very dense, with .300 trees/ha.
An initial a ¼ 0.05 was Bonferroni-corrected to a ¼ 0.025, given planned tests of medians, one for region, one for phase. Median forest density overall was significantly higher at 229 trees/ha in the north than the 191 trees/ha in the south (H ¼ 21.67, df ¼ 1, p , 0.001). Sample medians were 186 trees/ha in white fir, 208 trees/ha in ponderosa pine-Douglas-fir and 209 trees/ha in mixedconifer, pooled across north and south. These were quite close, but not significantly different among the 3 phases (H ¼ 7.29, df ¼ 2, p ¼ 0.026). If that test had instead been barely significant, the follow-up tests would have confirmed lack of significance. Thus, for medians, H 10 , that regions differed, is not rejected, but H 11 , that phases differed, is rejected.
Openings, which are section corners with no recorded bearing trees, and patches of scattered trees, which are corners with ,50% of expected bearing trees, covered about 22% of each area (Table 4, Fig. 1). These corners are used in the fire-severity reconstructions.
Forest density was heterogeneous at a fine scale in both north and south (Fig. 1). Contiguous blocks of low-density forest (i.e., ,150 trees/ha) seldom were .1,000 ha, although there was one larger area west of Johnsondale (Fig. 1d). Somewhat denser forests (150-293 trees/ha) appear to cover the largest contiguous areas, perhaps up to about 2,500 ha, and often were peppered with 250-500 ha patches of very dense forests (.293 trees/ha). However, the northern half of the northern Sierra had a large area of contiguous very-dense forest, albeit interrupted by human-affected areas (Fig. 1a). At the township scale (9,328 ha), SMC landscapes nearly always had diverse tree densities, along with openings and scattered trees.
Cross-validation (Appendix G) showed the GLO reconstructions had low RMAE and were quite accurate. Estimates were rare and unusable for specific cross-validation for the ponderosa Historical composition of pines, shade-tolerant trees, and oaks (H 2 ) Regarding H 2 , that historical SMC forests had about half shade-intolerant and half shadetolerant trees, compositional trends are apparent in the reconstruction between the two regions and three phases in the percentage of trees that were shade-intolerant pines, shade-tolerant firs and incense cedar, and oaks (Table 4). Thus, the outcome for H 2 is complex.
Pines (primarily ponderosa pine and sugar pine) varied, among the six phases in the two regions, from about 30-33% of total trees in the north to about 46-49% of total trees in the south (Table 4). Two-way ANOVA showed that the mean percentage of trees that were pines did not differ significantly among the phases, when pooled across regions (F ¼ 0.97, df ¼ 2, 498, p ¼ 0.382), but did differ significantly between north and south (F ¼ 107.74, df ¼ 1, 498, p , 0.001), with an average of 30.8% pines in the north and 47.8% pines in the south, pooled across phases.
The mean percentage of trees that were oaks (primarily California black oak and canyon live oak) varied from 9.3 to 42.1% among the six phases in the two regions (Table 4). Two-way ANOVA showed that the mean percentage of oaks, 29.1% in the north, was significantly higher than the mean of 22.4% in the south (F ¼ 7.52, df ¼ 1, 498, p ¼ 0.006), and differed significantly among phases when pooled across regions (F ¼ 63.67, df ¼ 2, 498, p , 0.001). Follow-up tests showed mean percentages differed among phases in both the north (F ¼ 25.81, df ¼ 2, 203, p , 0.001) and south (F ¼ 39.98, df ¼ 2, 294, p , 0.001). The percentage of oaks was highest in the ponderosa pine-Douglas-fir phase in the north, where oaks were 42.1% of total trees, and was lowest in the white fir phase in the south, where oaks were 9.3% of total trees (Table 4).
Shade-tolerant trees (white fir, Douglas-fir, incense cedar) varied among the six phases in the two regions (Table 4). Two-way ANOVA showed that mean percentage of trees that were shade-tolerant, 38.4% in the north, was significantly higher than the mean of 28.3% in the south (F ¼ 51.92, df ¼ 1, 498, p , 0.001), and also differed significantly among phases (F ¼ 70.04, df ¼ 2, 498, p , 0.001). Followup tests showed mean percentages differed among phases in both the north (F ¼ 26.19, df ¼ 2, 203, p , 0.001) and south (F ¼ 48.50, df ¼ 2, 294, p , 0.001). The mean percentage of shade-tolerant trees was highest in the white fir phase in the north, at 56.0% of total trees, and lowest in the ponderosa pine-Douglas-fir phase in the south, at 16.7% of total trees (Table 4). Thus, hypothesis H 2 is rejected as a general pattern in historical SMC forests. Hypothesis H 10 , that regions differed, and hypothesis H 11 , that phases differed, are not rejected.

Historical basal area (H 3 ) and quadratic mean diameter
Reconstructed mean basal area varied from 27.9 to 40.5 m 2 /ha among the six phases in the two regions (Table 5). An initial a ¼ 0.05 was Bonferroni-corrected to a ¼ 0.00625, given eight planned t-tests, one per phase and one overall in the two regions. Hypothesis H 3 , that historical mean basal area equaled 33.2 m 2 /ha, could not be rejected for any phase or for overall values in either region (Table 5). Two-way ANOVA showed that mean basal area was not significantly different between north and south (F ¼ 0.70, df ¼ 2, 497, p ¼ 0.404), but was significantly different among phases (F ¼ 3.48, df ¼ 2, 497, p ¼ 0.032). Followup tests showed that mean basal area did not differ among phases in the south (F ¼ 0.84, df ¼ 2, 294, p ¼ 0.433), but did in the north (F ¼ 4.34, df ¼ 2, 202, p ¼ 0.014). Basal area differed in the ponderosa pine-Douglas-fir phase and the white fir phase, but mixed-conifer did not differ from either. Hypothesis H 10 , that historical SMC forests differed between north v www.esajournals.org and south, is rejected for basal area, and H 11 , that historical SMC forests differed between phases, is rejected in the north for basal area.
Cross-validation with tree-ring reconstructions and early-inventory estimates (Appendix G) was hampered by few estimates and potential problems with estimates. No tree-ring/inventory estimates are available for the ponderosa pine-Douglas-fir phase. In the mixed-conifer phase, specific comparisons are possible with five estimates, but three are from Sudworth (1900), whose estimates are atypical of historical forests (Bouldin 1999). Sudworth's estimates, which vary from 221-387 m 2 /ha, are in the top 0.5% of reconstructed GLO estimates. I also suspect something is wrong with Sudworth's data. If Sudworth's data are left out, specific comparisons are not possible for the white fir phase, and the only specific comparisons are for the two in mixed-conifer, where RMAE is high, at 55.0%. However, the mean of four available general reconstructions and inventories is 33 m 2 /ha for mixed-conifer (omitting Sudworth), compared to a pooled mixed-conifer mean of 36.2 m 2 /ha, a 9.7% RMAE. Thus, cross-validation for basal area is limited by few reconstructions, and poor agreement with the two available ones, but has some general support for the overall study area. Basal area had only 21-25% RMAE in an extensive accuracy trial (Williams and Baker 2011).
Reconstructed quadratic mean diameter (QMD) ranged from 47.6 to 63.1 cm across the six phases in the north and south (Table 5). Twoway ANOVA showed that QMD was significantly different between north and south, when pooled across phases (F ¼ 34.92, df ¼ 2, 497, p , 0.001), and also differed among phases, when pooled across regions (F ¼ 6.06, df ¼ 2, 497, p ¼ 0.003). A follow-up test identified significantly different groups: (1) white fir, mixed-conifer, and ponderosa pine-Douglas-fir phases in the south, all with the highest mean QMDs of about 58-63 cm and (2) ponderosa pine-Douglas-fir in the north, with the lowest mean QMD of 48 cm. A third group, containing the remaining two phases, white fir and mixed-conifer in the north, had intermediate QMD from 51 to 59, and its phases differed from groups 1 and 2, respectively.
Cross-validation with tree-ring reconstructions and early-inventory estimates (Appendix G) was also hampered by few estimates, after omitting Sudworth's data. No tree-ring/inventory estimates are available for specific comparisons for white fir or ponderosa pine-Douglas-fir phases. In the mixed-conifer phase, specific comparisons are possible with two estimates, where RMAE is

Historical diameter distributions (H 4 , H 5 )
Regarding hypothesis H 4 , the null hypothesis, that the number of trees 60 cm diameter equals the number .60 cm, is rejected for all trees in both north and south, except ponderosa pine in the south (Fig. 3e). In all cases, except sugar pine (Fig. 3f ) ,50% of trees were .60 cm diameter; overall, only 21% of trees in the north and 33% in the south were .60 cm diameter (Fig. 3). Percentages of trees .60 cm were higher in the south than north, across all species and in total. Hypothesis H 5 , that historical SMC forests had low abundance of small trees, is also not supported. Trees 40 cm were 30.7% of all trees in the south and 41.1% of all trees in the north. Trees 20 cm were 11.2% of all trees in the south and 20.2% of all trees in the north.
Regarding hypothesis H 10 , the northern and southern Sierra Nevada show some similarities and differences in diameter distributions. White fir, incense cedar, Douglas-fir, ponderosa pine, and sugar pine all show peaks in the 40-50 cm size class, relative to adjoining size classes; the peaks are more pronounced in the north. Sample sizes are so large that all species differed significantly ( p , 0.001) in distributions between north and south, using chi-square. Size-classes contributing most to differences appear anecdotal in most cases (e.g., higher proportion of 60-80 cm incense cedar in the south than north; Fig. 3). However, there is a higher proportion of trees in Fig. 3. Reconstructed diameter distributions for the northern (blue) and southern (red) Sierra Nevada. Note that y-axes are not always the same in the two areas. Omitted are trees only identified as ''pine'' or ''fir'' and species with ,100 total trees in an area. Diameters were likely estimated at stump height (about 30 cm), thus are likely larger than at breast height (about 1.4 m). Also shown is the percentage of total trees that exceeded 60 cm diameter, and the result of the chi-square test of the null hypothesis that the number of trees .60 cm diameter equals the number of trees 60 cm diameter.
v www.esajournals.org the smallest size class (10-20 cm) in the north than the south, particularly for white fir, incense cedar, California black oak, ponderosa pine, and all trees, but canyon live oak is an exception (Fig.  3). Incense cedar, Douglas-fir, ponderosa pine, and sugar pine all appear to have few trees in the smallest size class (10-20 cm). Both sugar pine and ponderosa pine had quite a few trees exceeding 120 cm, but sugar pine stands out for its large trees (Fig. 3).

Section-line data and understory trees and shrubs (H 5 and H 6 )
Hypothesis H 5 , that historical SMC forests had relatively low abundance (i.e., ,10%) of small trees, is not supported by section-line data in either the northern (v 2 (1, N ¼ 1122) ¼ 877.2, p , 0.001) or southern Sierra (v 2 (1, N ¼ 1521) ¼ 5132.4, p , 0.001). Instead, understory trees were abundant on 36.5% of area in the north and 65% in the south (Table 6) and 35-37% of these areas had dense understory trees. Among phases, the percentage with trees of any species was highest in ponderosa pine-Douglas-fir in the north, and white fir in the south.
Hypothesis H 6 , that historical SMC forests had high shrub abundance is supported, as the null hypothesis that the percentage of shrub cover was not greater than 75% is rejected in the northern (v 2 (1, N ¼ 1122) ¼ 343.2, p , 0.001) and in the southern Sierra (v 2 (1, N ¼ 1521) ¼ 1748.9, p , 0.001). Overall, 91% of section-line length in the north and 96% in the south had shrubs ( Table  6); 41% of these occurrences in the north in the white fir phase and 41-46% in the south in the three phases were described as dense shrubs (Table 6). Only 16% of the other two phases in the north had dense shrubs. Ceanothus integerrimus was most abundant in both north and south, followed by Arctostaphylos patula/viscida, then Ceanothus cordulatus (Table 6). Surveyors were required to record understory trees and shrubs in order of abundance along the section line. à Firs include Abies concolor and Pseudotsuga menziesii.
§ Pines include Pinus ponderosa, P. jeffreyi, and P. lambertiana. } Oaks include primarily Quercus kelloggii and Q. chrysolepis. # Section-line lengths in samples differ depending on the whether surveyors recorded the particular data (see Appendix C).
jj If a surveyor did not record information for an attribute, that could mean the attribute was truly lacking, which is how the percentages were calculated. However, this provides a low estimate, if the surveyor just neglected to record the information, in which case the data are missing. In this latter case, the correct percentage is obtained by applying the multiplier.
v www.esajournals.org Hypothesis H 7 , that low-severity fire characterized !85% of historical landscapes is rejected for both the northern (v 2 (1, N ¼ 116) ¼ 555.0, p , 0.001) and southern Sierra (v 2 (1, N ¼ 187) ¼ 329.1, p , 0.001). Exclusive low-severity area covered only about 13% of the northern and 26% of the southern Sierra Nevada (Table 7, Fig. 4). Hypothesis H 8 , that the high-severity fire rotation in historical SMC forests was .500 years is not supported for either the northern (281 years) or southern Sierra Nevada (354 years; Table 7). North and south differed substantially in their historical fire regime (Table 7) thus hypothesis H 10 is supported for fire.
Within each region, percentages varied among phases (Fig. 5), but are inconsistent, thus the outcome for hypothesis H 11 is complex. Somewhat more low severity was found in the white fir phase and slightly more mixed severity in the mixed-conifer phase in each region. In the north, the greatest high severity was in white fir, but in the south, it was in the ponderosa pine-Douglasfir phase. Corresponding fire rotations varied among the six phases in two regions from 223 years in white fir in the north to 542 years in white fir in the south (Fig. 5).

Contiguous fire areas (H 9 )
There is some spatial pattern to reconstructed fire severities (Fig. 4). Mixed-severity fire covered a little less than half of both the north and south (Table 7) and formed the matrix within which were found smaller areas of high-and lowseverity fire ( Fig. 4). Contiguous areas of lowseverity fire were rare in the northern Sierra Nevada, and usually only a few hundred hectares in extent, except along its southern border (Fig. 4a). In the southern Sierra Nevada, several contiguous areas of low-severity fire of 3,000 to 6,000 ha occurred .
Large patches of contiguous high-severity fire occurred historically in the north (Fig. 4a) and south ( Fig. 4b-d). Patch-size distributions were similar between north and south ( Fig. 6), with most patches ,1000 ha. Ten in the north and eight in the south were .1000 ha, two each in north and south were .4000 ha. The largest were 8050 ha in the north and 9400 ha in the south. Thirty-six of 75 patches in the north and 25 of 70 patches in the south were .250 ha. The hypothesis (H 9 ), that all patches of contiguous high-severity fire were ,250 ha, was rejected for the north (v 2 (1, N ¼ 75) ¼ 36.6, p , 0.001) and south (v 2 (1, N ¼ 70) ¼ 29.7, p , 0.001). High-severity components include all the areas reconstructed as high-severity fire. These include: (1) the structure-based model with survey tree data to reconstruct high-severity forest area, (2) chaparral patches, which are high-severity areas reconstructed from section-line data, (3) forest lines-scattered trees, which are also reconstructed from section-line data, (4) forest corners-openings, which are forest corners lacking bearing trees that also have dense shrubs in the understory, and (5) forest corners-scattered trees, which are forest corners missing .50% of bearing trees. See text for details on these reconstructions.
à Calculated as 110 years/(percentage/100) using the percentage of the area within which fire components were reconstructed.

Corroboration and new findings from comparison with Leiberg's (1902) maps
Leiberg's burn categories appear reliable, as they are consistent with expected effects of different fire severities, as can be seen by comparing cover types inside versus outside burn areas across his maps (Fig. 7). Percentages in all categories are, as expected, reduced inside the 75-100% burned category relative to outside the burns, except chaparral, which is greatly increased (Fig. 7a). Also as expected, the area inside the lower severity 5-25% burned category has only slightly elevated area in chaparral and ,2,000 board-feet/acre (Fig. 7b). The 25-50% category covers insufficient area, and the 50-75% category does not occur.
The areas mapped by Leiberg in the 75-100% burned category on about 32,360 ha of the overlap area in 1900 are also consistent with high-severity fires. First, the areas in the 75-100% burned category were mapped in 1885-1890 as 42.4% chaparral (13,707 ha), 19% forest (6,306 ha) having ,2,000 board-feet/acre, 25% forest (8,020 ha) having 2,000-5,000 board-feet/acre, and only 13% forest (4,177 ha) having .5,000 board-feet/ acre (Fig. 7a). These areas thus mostly had no timber or low timber volume, as mature forests had .10,000 board-feet/acre. Some or all this area had likely already burned at high severity by 1885-1890. Second, the chaparral area in this 75-100% burned category (13,707 ha) was 90.4% forested (12,391 ha) at the time of the surveys 1-25 years prior to Leiberg's mapping (Fig. 2); only 9.5% of it was chaparral that reburned. Third, areas of chaparral are a strong indicator of highseverity fire in forests. Leiberg says: ''There can not be the slightest doubt that every acre of chaparral represents so much ground once v www.esajournals.org forested, denuded by fire, then overgrown with brush'' (Leiberg 1902:43). The chaparral in the 75-100% category was also 87% of total chaparral Leiberg mapped in the overlap area, thus chaparral was concentrated in this high-severity category.
When the Leiberg 75-100% burned category from 1900 is compared to the survey section-line data from 1865-1890, it is clear that high-severity fire burned 18,769 ha of mature forest after the surveys (Fig. 8). Surveyors described 58% (201 km) of the 346-km section-line length in the 75-100% burned category as ''heavily timbered,'' ''good timber,'' or ''excellent timber,'' thus mature forest. The 58% is a line-intercept estimate of the area (0.58 3 32,360 ha ¼ 18,769 ha) of mature forest that burned. This occurred 8.8% (1650 ha) in the ponderosa pine phase, 63.0% (11,827 ha) in mixed-conifer, and 28.2% (5,294 ha) in white fir. Section lines listed the first tree or shrub (the dominant) as: 58% pine, 20% white fir, 10% chaparral, 9% oak, and 3% Douglas-fir, thus high-severity fire was favored in pine-dominated parts of all phases.
The remaining 145 km of the 346-km sectionline length (13,500 ha) in the 75-100% burned category is also consistent with high-severity fire, but in younger forests. This area was described by surveyors as having: (1) scattered and/or scrubby trees, (2) poor, fair, or medium timber, or Fig. 6. Size distribution of contiguous patches of high-severity fire in the (a) northern Sierra Nevada and (b) southern Sierra Nevada. The distribution was truncated at the low end at 50 ha, to avoid small polygons created by clipping and other GIS operations.
Leiberg did not provide sufficient mapping detail to be able to fully determine patch size, but the 18,769-ha high-severity area that burned mature forest is concentrated in two contiguous, roughly 8,000 ha areas (Fig. 8). This underestimates patch size for the high-severity area as a whole, as it is only for the mature forest, but adds evidence that hypothesis H 9 , that patches of contiguous high-severity fire area did not exceed 250 ha, is rejected.

Historical forest structure
Historical SMC forests were on average denser than other dry forests in the western United States. The mean density of 293 trees/ha exceeds mean densities reconstructed from GLO data Baker 2012a, 2013) for northern Arizona and the Blue Mountains, Oregon (142-167 trees/ha), and the Colorado Front Range (217 trees/ha), but is similar to the 275 trees/ha mean for dry mixed conifer forests in Oregon's eastern Cascades (Baker 2012). Cross-validation shows that mean tree density is reconstructed with low error, averaging ,10% RMAE in nine comparisons, better than the 11.4% mean RMAE in five comparisons in northern Arizona (Williams and Baker 2011). The cross-validation also validates tree-ring reconstructions and early inventories (Appendix G). In addition, dense forests were often described by early observers (Appendix A: Q135-Q147). Thus, tree-ring reconstructions, early scientific reports and the GLO reconstructions concur that historical SMC forests were dense to very dense on average, not on average open and park-like as in the main view in the introduction.
Somewhat open, park-like forests with ,150 trees/ha did occur, but only on 23% of the northern and 33% of the southern Sierra Nevada. These open forests were often described by early observers (Appendix A: Q118-Q134). Dry forests in the eastern Oregon Cascades were similar to the northern Sierra Nevada, with 25% of the landscape having ,143 trees/ha (Baker 2012), but other dry forests had a larger percentage of lowdensity forests Baker 2012a, 2013). These open, park-like forests are a striking, ecologically important component of historical SMC landscapes that warrants protection and restoration. However, reconstructions from these areas (e.g., Scholl and Taylor 2010), which my reconstruction also validates, are atypical of most historical SMC forest landscapes, which averaged almost twice as dense.
Relative to other dry-forest landscapes, historical SMC landscapes generally had much higher proportions of dense forest. The 65% of the northern and 46% of the southern Sierra that was dense (.200 trees/ha) contrasts with 45% in the Colorado Front Range, 29% in Oregon's Blue Mountains, and 15-17% in northern Arizona Baker 2012a, 2013). Very dense forest (.300 trees/ha) was 34% of the northern and 21% of the southern Sierra Nevada, similar to the Eastern Oregon Cascades, with !25% Fig. 7. Percentage of land area, by cover type and timber volume, as mapped by Leiberg in 1885-1890 inside and outside burns mapped by Leiberg (1902) about 1900, for two severities of fire: (a) 75-100% burned, (b) 5-25% burned.
Historical SMC forests were more heteroge-neous in tree density than other dry-forests reconstructed with GLO data. The coefficientof-variation of tree density was 162.8%, about three times as large as in dry-forest landscapes of Fig. 8. Areas of mature Sierran mixed-conifer forest burned at high severity after the surveys and before Leiberg's mapping. The Leiberg 75-100% burned category from 1900 was overlain on the survey section-line data from 1865-1890. Section lines shown in red were described by surveyors as ''heavily timbered,'' ''good timber,'' or ''excellent timber,'' thus mature forest, in 1865-1890 before Leiberg mapped these areas in 1900 as severely burned.
v www.esajournals.org northern Arizona (Williams and Baker 2013). At a township scale, patches of low-or very-dense forest were peppered across broader expanses of generally dense forest (Fig. 1). Variability is from variation in fire severity and associated post-fire succession, and variation in environment, from open rocky slopes with scattered trees (Appendix A: Q148-Q159) to north-facing or higher-elevation moister slopes with denser forests (Appendix A: Q143-Q147). This historical variability supports its recent focus in ecological restoration programs (van Wagtendonk and Lutz 2007, North et al. 2009, Collins et al. 2011. Historical basal area appears to have been similar across SMC forests and comparatively large for dry forests, with a mean between 28-41 m 2 /ha, averaging 33-36 m 2 /ha in the two regions (Table 5). Reconstructed means do not differ significantly from the hypothesized mean of 33.2 m 2 /ha from a sample of reference data (Safford 2013). This is about three times the mean historical basal area on the Coconino Plateau in northern Arizona (Williams and Baker 2013). The relative consistency in this attribute of forest structure in SMC forests likely reflects a combination of environmental constraints and limitation by wildfires and other disturbances. Overall means for basal area have reasonable accuracy relative to means of tree-ring reconstructions and inventories, with 7.0% RMAE in mixed-conifer.
Historical quadratic mean diameter varied over a limited range (means of 48-63 cm) among phases and regions (Table 5), with a low coefficient-of-variation (24-32%) compared to other attributes. QMD was higher than the 40.1 cm reconstructed for ponderosa pine forests on the Coconino Plateau, Arizona (Williams and Baker 2013). QMD had RMAE of 30.7% at two sites with specific comparisons, but general comparisons show a 16.0% RMAE in the study area overall, comparable to the 12-16% RMAE in the accuracy trial (Williams and Baker 2011).
The results show that historical SMC forests were not dominated by trees .60 cm diameter (dbh), as often suggested (McKelvey and Johnston 1992, Gruell 2001, Scholl and Taylor 2010. Northern Sierran forests overall had only 21% and southern Sierran forests only 33.3% of trees .60 cm (Fig. 3). These included oaks, but conifers .60 cm were only 29% of 9532 total trees. Two early observations suggested SMC forests were multi-aged (Appendix A: Q113-Q114). Large trees, although not numerically dominant, were a key feature of historical SMC forests.
The results show that historical SMC forests were instead numerically dominated by smaller trees and also had abundant seedlings and saplings beneath these small trees. Trees 10-50 cm in diameter were 61% of 9532 total trees (Fig.  3). Since an average 50-cm diameter tree was roughly 110 years old, SMC forests were numerically dominated by relatively smaller and younger trees. Section-line data show that additional smaller understory tree-regeneration (,10 cm diameter) was likely present across most forests but abundant on 37% of the northern and 65% of the southern Sierra and more than a third of this regeneration was dense. Early reports indicate that fires could reduce or eliminate seedlings and saplings (Appendix A: Q160, Q168-Q174), but also stimulated abundant post-fire recruitment (Appendix A: Q163, Q166). Some areas lacked understory trees, but most contained abundant or even very dense understory trees, varying among species and with environment and fire (Appendix A: Q160-Q197). SMC forests had understory tree abundance similar to dry forests in eastern Oregon (Baker 2012, Williams and Baker 2012a), but much more than in the Colorado Front Range or northern Arizona, which had only 1-10% of area with understory trees Baker 2012a, 2013).
The results show that historical SMC forests had nearly ubiquitous and often abundant shrubs, more than in other dry western forests. Shrubs were present on 91% of forest area in the north and 96% in the south, and were dense on 41-46% of shrub area in the south and the white fir phase in the north (Table 6). By about A.D. 1900, overgrazing had substantially reduced shrub cover (Vankat and Major 1978; Appendix A: Q198-Q199), thus sparse understories at this time could reflect overgrazing (Appendix A: Q200-Q203). However, denser forests reportedly had fewer shrubs (Appendix A: Q205-Q206). Both fire and canopy openings favored denser shrubs (Appendix A: Q205, Q208). The eastern Cascades, Oregon, had shrubs on 71% of forest area (Baker 2012). The Blue Mountains, Colorado Front Range, and northern Arizona had shrubs v www.esajournals.org on only 0.3-18.0% of forest area (Williams and Baker 2012a).
I hypothesize that the peak in diameter distributions in the 40-50 cm class for several trees in both the northern and southern Sierra (Fig. 3) reflects elevated recruitment after regional drought and/or moderate and high-severity fires in the late-1700s to early-1800s. Extreme elevated temperatures (Scuderi 1993) occurred in 1779 and 1786-1805, which also had large areas burned in both Yosemite (Scholl and Taylor 2010) and Sequoia National Parks (Swetnam et al. 2009) SMC forests. Elevated ponderosa pine and sugar pine recruitment is evident in age structures in Yosemite after this period (Scholl and Taylor 2010;Fig. 6). The peak is not likely surveyor bias, as that would require similar bias across most of 35 surveyors. Relocated bearing trees also document little bias in bearing-tree selection (Williams and Baker 2010).

Historical fire severity
Other droughts are documented in the western Sierra during the period of the fire-severity reconstruction, which begins about 1755-1775 and ends about 1865-1885. Major periods of drought and high temperature occurred during 1764-1794 and 1806-1861 (Graumlich 1993) as well as 1856-1865 and 1870-1877 (Herweijer et al. 2006). These warm, dry periods increased after the preceding Little Ice Age, but continued into the 20th century (Herweijer et al. 2006). Thus, the reconstruction period and modern period may both have had climate favoring fire.
The hypothesis that low-severity fire nearly exclusively maintained dry-forest landscapes is rejected for historical SMC forests, as only 13-26% of these landscapes had only low-severity fire over the 110 years preceding the surveys. However, early reports suggest that high-severity patches did occur, associated with low-severity fires in these areas, and often were small (Appendix A: Q18-Q26, Q71, Q72, Q74, Q75). Show and Kotok (1924) reported that 15 early low-severity fires in the pine region (including the western Sierra) had an average of about 15% high-severity fire, but in small patches. Early reports that suggested low-severity fires were mostly the only fires (Appendix A: Q14-Q16, Q41-Q43) likely reflect the limited data they had available. The low percentage of exclusive low severity (13-26% of study areas) is shared with mixed-conifer forest in Oregon's eastern Cascades (Baker 2012). Low-severity fire was more common in the Blue Mountains (Williams and Baker 2012a) and northern Arizona Baker 2012a, 2013). The hypothesis that lowseverity fire exclusively maintained entire dryforest landscapes has been rejected for all areas with spatially-extensive reconstructions (Baker 2012, Baker 2012a, 2013). Dry-forest landscapes in the western US were instead most strongly influenced by mixed-and high-severity fire, as also shown by .
Mixed-severity fire was the dominant fire severity in SMC forests, found on 48% of the northern and 43% of the southern Sierra. Leiberg (1902) first documented the extent of mixedseverity fires in SMC forests (Appendix A: Q29, Q32-Q40). He described in detail the diversity of forest structures left behind and created by a mixture of fire severities: (1) chaparral patches and ''lanes'' often with surviving individual trees and tree groups or larger patches of surviving trees, (2) severely-thinned forests often with heavy chaparral understories, (3) scattered young trees regenerating in the chaparral, if observed 5-20 years after the fire, and (4) remnant denser unburned or lightly burned patches of forest often directly adjacent to the chaparral (Appendix A: Q35, Q39, Q94). Show and Kotok (1924) described this same suite (Appendix A: Q106), but did not recognize it as mixed-severity fire. Mixed-severity fire in SMC forests is most similar to dry forests in the eastern Cascades of Oregon (Baker 2012) and Washington (Hessburg et al. 2007), and Oregon's Blue Mountains (Williams and Baker 2012a), which had 43-59% mixed-severity.
In the Sierra, high-severity fire was somewhat more prominent in the north, found across 39% of these landscapes and 31% of the southern Sierra, lower than in the Colorado Front Range and on northern Arizona's Black Mesa (Williams and Baker 2012a), similar to the 30% in dry mixed-conifer in the eastern Cascades of Washington (Hessburg et al. 2007), but higher than in other areas. The high-severity rotation of 281 years in the north is similar to the 271-year rotation in dry forests in the Colorado Front Range and the 278-year rotation in the central region in Oregon's Eastern Cascades (Baker 2012). The 354-year rotation in the southern Sierra is a little longer, but not as long as the 435-year rotation overall in Oregon's eastern Cascades or the 828-year rotation in Oregon's Blue Mountains (Williams and Baker 2012a). High-severity fire likely was often a component of mixed-severity fires rather than independent, but extended at times as contiguous patches over large areas (Fig. 6).
Dry western forests were considered resistant to high-severity fire because understory fuels were kept low by frequent fires (Covington and Moore 1994). If so, fires could not have burned at high severity in the Sierra Nevada during a period of intense overgrazing by livestock in the late-1800s, when understory fuels were reduced. Comparison of survey data and Leiberg's data show that high-severity fire burned about 18,770 ha of mature forest after the surveys (Fig. 8). Moreover, the idea that low-severity fire kept understories free of fuels is not applicable to most SMC forests, which had pervasive ladder fuels in understory shrubs and small trees, that were often dense over large areas. Also, historical tree densities averaged 293 trees/ha, not including smaller understory trees. Calibration with fire severities from tree-ring reconstructions shows that forests this dense did not have a low-to moderate-severity fire regime (Williams and Baker 2012a), also shown by simulation analysis (Johnson et al. 2011).
The idea that low-severity fire kept fuel loads low and prevented high-severity fires is also behind the modern notion that historical highseverity fires did not produce patches exceeding a few hundred hectares (e.g., Collins and Stephens 2010). However, the reconstructions show that contiguous areas of historical highseverity fire commonly exceeded 250 ha and reached as high as 9400 ha. Show and Kotok (1924) also reported chaparral areas produced by fire over contiguous areas .2000 ha (Appendix A: Q95). In the Colorado Front Range, historical high-severity patch sizes had a geometric mean of 171 ha for patches .20 ha (Williams and Baker 2013). I found means only a little higher (Fig. 6) for patches .50 ha. Maximum historical patch size in Colorado was 8331 ha, similar to the maxima of 8050 ha in the northern and 9400 ha in southern Sierra, as well as the 8000-ha patches in Leiberg's maps (Fig. 8).
Extensive historical mixed-and high-severity fire and associated diverse forest structures are now well-established as characterizing much of the historical dry forest across the western US (Baker 2009, Williams and Baker 2012a, including in the western Sierra Nevada. This finding from spatially extensive GLO studies and spatially extensive analysis of forest age structures ) is also well corroborated by early scientific accounts, early primary observations and photographs, paleoecological studies, and other age-structures (synopses in Baker 2009. The alternative view of SMC forests in the 1996 Sierra Nevada Ecosystem Project Final Report to Congress, excerpted in the introductory quote, was supported by Leiberg's detailed study in 1902, and is again now that spatially extensive data are available from GLO survey data. Historical SMC forests were not largely open or park-like, but instead were mostly dense or very dense, high-severity fire was common, and mixed-severity fires and topography fostered very heterogeneous forest structure.
Contrasting historical forest structure and fire: northern and southern Sierra Nevada The northern and southern Sierra and the phases had similarities and differences in historical fire and forest structure. North and south did not differ in mean tree density or basal area, had similar amounts of mixed-severity fire, similarly high understory shrubs, and heterogeneous landscapes with a matrix of dense forests interrupted by patches of low-density forest and very dense forest. Northern forests had less open, low density forest, about 50% more dense and very dense forest, and higher median tree density than southern forests. They were slightly dominated by shade-tolerant trees with equal accompanying oaks and pines, whereas southern forests were almost half pines, nearly a third shade-tolerant trees, and less than a fourth oaks. Southern forests had more trees .60 cm, and a higher quadratic mean diameter, fewer 10-20 cm trees, but almost twice as much coverage by understory trees and dense shrubs. I suggest this may reflect more fire in the southern Sierra Nevada nearer the time of the surveys. The northern Sierra had a quarter more high-severity v www.esajournals.org fire and a high-severity fire rotation about one quarter shorter than in the southern Sierra, which had about twice as much area of exclusive lowseverity fire. Larger amounts of dense forest, more shade-tolerant trees and oaks, and smaller trees are congruent with more high-severity fire and less low-severity fire in the northern Sierra. In both regions, oaks declined by almost three quarters from the ponderosa pine-Douglas-fir phase to the white fir phase, whereas shadetolerant trees roughly doubled. Neither mean nor median tree densities differed among phases.

Limitations
Although I explicitly spatially controlled for EuroAmerican effects in the reconstruction of the historical fire regime in the northern Sierra Nevada (Appendix E), it remains possible that high-severity fire was elevated somewhat by EuroAmericans in this area. The spatial control is for fixed locations (e.g., sawmills), but people moving through these landscapes could set fires that spread over larger areas. Another limitation is the relatively short period for the fire-severity reconstructions, 110 years, which is less than the estimated rotations. Thus, it is likely that the estimates are imprecise. However, modern data for comparison (e.g., Hanson and Odion 2014) also are limited, typically to ,30 years. It is an unfortunate reality that analysis and comparison of fire severity is limited by short periods of record. Another limitation is that the amount of high-severity fire and the fire rotation for highseverity fire do not include the high-severity parts of mixed-severity or low-severity fires, as they cannot be separated and measured. Thus, the high-severity fire rotation was likely shorter than my estimates. All reconstructions of historical forests that provide reference data have limitations, but their limitations increase with the passing of time since EuroAmerican settlement. The GLO-based reconstructions are nearly all for 1865-1884, with a median of 1873 (Fig. 2). This is not ideal, as EuroAmerican land uses expanded rapidly after 1848 (Beesley 2004), 17-36 years before the surveys (median 25 years). However, the GLO reconstructions provide the earliest spatially extensive reconstructions.
Other methods of reconstructing historical forests have limitations as well. First, detailed tree-ring reconstructions of forest structure (Ap-pendix G), are unfortunately few for SMC forests (n ¼ 5) relative, for example, to northern Arizona where . 100 reconstructions are available across large land areas (e.g., Abella and Denton 2009). Second, tree-ring reconstructions are typically limited to current old forests in protected areas, where evidence of historical forests is relatively undisturbed and best preserved. Forests that may have originated after mixed-and highseverity fires in the early to middle-1800s, which may be denser and ,150 years old today are often not studied, leading to a sampling bias against younger, denser historical forests and mixed-and high-severity fire. This explains why Mallek et al.'s (2013) estimates of historical mixed-and high-severity fire are too low. Finally, later inventories (e.g., Collins et al. 2011) and the VTM plots from the 1930s (Keeley 2004) provide significantly diminished evidence about historical forests, since they took place 60-90 years after the 1848 EuroAmerican expansion.

MANAGEMENT IMPLICATIONS
Fuel-reduction programs will not restore historical fire, forest structure, or resiliency The reconstructions show that the historical fire regime in SMC landscapes included low-to moderate-severity fire, likely at modest intervals, combined with mixed-severity fires, at longer intervals, which included substantial high-severity fire. The low-to moderate-severity component of the fire regime included many small highseverity patches. These fires did not keep fuels at low levels, as forests were dominated numerically by smaller trees (,50 cm diameter) and abundant shrubs commonly considered ladder fuels. SMC forests thus were not generally resistant to the mixed-severity fires, but instead: (1) burned completely and became chaparral across contiguous patches or lanes, or (2) were severely thinned by them, leaving scattered surviving trees or tree groups, often with dense chaparral understories, or (3) were thinned less severely and developed an open, park-like structure with large, old trees.
This old-growth structure may have conferred some resistance to higher fire severity, but these stands, too, were at best incompletely resistant (Hessburg et al. 2007), as demonstrated by the large area of mature forest that burned after the v www.esajournals.org surveys. Given that high-severity fire rotations were about 280-350 years, which is the mean expected time between stand-replacing fires at any point in SMC landscapes, there was ample time for full recovery of old-growth forests across large areas. Moreover, mixed-and high-severity fires often left surviving trees and tree groups that became older emergent trees in recovering forests. Although these forests were likely not very resistant to mixed-and high-severity fires, tree and shrub regeneration after these fires was abundant (Appendix A) and forest resilience was thus historically very high.
Episodic recovering early-successional forests from mixed-severity fire had many ecological benefits (DellaSala et al. 2014). As an example, I analyzed whether SMC oaks, thought to be declining recently due to fire exclusion, were damaged or favored by these fires. I found oak concentrations favored in areas burned by mixed-severity fires before the surveys (Appendix H), which is also supported by early observations (Appendix A: Q64, Q76-78, Q84, Q90).
A current agency focus on lowering fuel loads so that only low-severity fires occur is not supported by the findings of this study. Agency proposals (North et al. 2009, North 2012 seek to lower fuel loads, remove most small trees and shrubs, and create and maintain low-density forests with large fire-resistant trees, low fuel loads, and nearly homogeneous low-severity fire, which this study shows were atypical of historical SMC forests. For example: ''Mixed-conifer resilience might be best ensured by (1) reducing fuels such that if the forest burned, the fire would most likely be a low-severity surface fire...'' (North et al. 2009:v). However, this study shows that a mix of fire severities historically created and maintained SMC forests.
Moreover, lowering fuel loads to eliminate all but low-severity fires will not restore the high levels of heterogeneity that characterized historical SMC forests. Multiple authors agree that more intense fires are needed (Schmidt et al. 2006, van Wagtendonk and Lutz 2007, Collins et al. 2011). However, agency proposals appear conflicted. For example, North et al. (2009:20;Fig. 9) use the 2007 Moonlight Fire to illustrate desirable landscape heterogeneity they suggest should be created. However, just 31% of Moon-light's burned area was from low-severity fire, 25% was from moderate-severity, and 43% was from high-severity fire (http://www.mtbs.gov). The desirable heterogeneity from the Moonlight fire cannot be created with the mostly lowseverity fire that North et al. (2009) also recommend, or with mechanical thinning which does not mimic habitat structures (e.g., snags, down logs, chaparral patches) created by moderate-and high-severity fire. Reducing fuels to eliminate moderate-and high-severity fires would, if successful, reduce the historical landscape-level heterogeneity that provided wildlife habitat and conferred resiliency to drought, insect outbreaks, and fires (Millar et al. 2007).
Working with nature to restore historical fire, forest structure, and resilience What is needed to restore SMC fire regimes so that these landscapes remain as resilient as they were historically? First, the higher-severity component of SMC fire regimes may still be functioning, but its rate (fire rotation) is deficient. Historical high-severity fire rotations of 281 years in the northern and 354 years in the southern Sierra (Table 7) are both shorter than estimated high-severity rotations for 1984-2010 of 461 years for the lower montane and 893 years for the midupper montane, using data from Monitoring Trends in Burn Severity (http://www.mtbs.gov; Hanson and Odion 2014). This suggests a deficit in high-severity fire in recent relative to historical landscapes. Lack of significant trend from 1984-2010 in high-severity fire proportion or annual area of high-severity fire  indicates the deficit was not being reduced through 2010 by increased high-severity fire. The percentage of total burn area that burned at high severity between 1984-2010 varied from year to year (Hanson and Odion 2014), but likely averaged close to, or only a little less than the 31-39% reconstructed for historical forests (Table  7), so it did not appear to be in deficit or surplus through 2010.
Some are concerned that recent high-severity patch sizes are uncharacteristically large and damaging in dry western forests (Stephens et al. 2013, Fulé et al. 2014). However, these articles surprisingly presented no patch-size data (Williams and Baker 2014). Hanson and Odion (2014) found that maximum annual patch sizes across v www.esajournals.org 27 years  in Sierran montane forests included one year of about 8,000 ha, a few years with 3,000-7,000 ha and many years with 1,000ha maximum patch sizes. These are the sizes Fule et al. mention, but they are very similar to reconstructed historical patch-sizes (Fig. 6). Also, Williams and Baker (2012b) found that recent patch-size distributions for high-severity fire in the Colorado Front Range did not differ from historical distributions, except for a recent deficit in the largest sizes. Thus, the data suggest that the only restoration need regarding high-severity fire through 2010 was to remedy a deficit in the rate (fire rotation) of high-severity fire. The 2013 Rim fire added several thousand hectares that will offset some of the deficit in high-severity burned area, help restore landscapes, and maintain their resilience.
The low-to moderate-severity part of the historical fire regime may need restoration and maintenance in two ways. First, the rate (fire rotation) at which these fires burn is likely not matching the historical rate, but it is still unresolved what the historical rate was. Past estimates derived from the widespread composite-fire-interval method (e.g., North et al. 2012) suggest much more low-severity fire than actually occurred, due to methodological flaws in this method (Baker and Ehle 2001, Dugan and. Available direct estimates of overall fire rotation compiled by Mallek et al. (2013) are better, but have a sampling bias toward lowdensity mature forests that makes them unreliable for the whole SMC landscape, which was denser. Mallek et al.'s conclusion that lowerseverity fires are burning at lower rates than historically may be valid, but valid evidence is very limited. New landscape (Farris et al. 2010) and plot-based methods (Dugan and Baker 2014) can produce valid and accurate estimates, but many more are needed.
Second, small high-severity patches from lowto moderate-severity fires provide tree-regeneration sites, abundant shrub cover, dead snags, and important wildlife habitat that likely are deficient relative to historical forests. The reconstructions from section-line data (Table 6) show that firestimulated understory shrubs (Ceanothus, Arctostaphylos) and understory trees that provided ladder fuels essential to maintain the highseverity component of the low-severity fire regime were historically abundant. These fuels were substantially reduced by overgrazing by domestic livestock in the late-1800s and further reduced by fire exclusion that removed the fire stimulus that maintains these shrubs (Vankat and Major 1978). Mis-directed fuel-reduction programs are removing more of these fuels, which likely need to be increased, not reduced.
Fire is the logical choice for restoring historical fuels. However, previous work in Yosemite National Park showed that prescribed fires were typically ignited in shoulder seasons, and were lower in fireline intensity and fire severity than wildfires or wildland fire-use fires (van Wagtendonk and Lutz 2007). Low-intensity prescribed fires will not restore a substantial component of small high-severity patches, as they seldom increase fire-stimulated shrubs (e.g., Collins et al. 2009). In contrast, wildfires from 1974-2005 in Yosemite averaged about 37% low severity, 44% moderate severity, and 19% high severity, excluding unchanged areas, which is much more similar to the historical distribution of 26% low-, 43% mixed-, and 31% high severity fire (Table 7) than are fire-severity distributions for prescribed or wildland fire-use fires (van Wagtendonk and Lutz 2007). Wildfires thus were most restorative. Wildland fire-use, or multi-objective fires managed for specific goals, may help, but also may be insufficient unless allowed to create more highseverity patches, something that managers likely can accomplish.

Protecting people and infrastructure
Sierran mixed-conifer forests have likely long been subject to severe fires, and the plants and animals that live in them have remarkable capabilities to thrive both after these fires and in the interludes between them. It is us who have perhaps not had sufficient time to adapt to the rather fiery wild nature that characterized historical Sierran mixed-conifer forests. Extensive property damage, loss of human life, and ignitions by people are symptoms of this lack of adaptation. Perhaps adaptation has not occurred because the public has heard that the problem largely lies in the forest and can be fixed because it is an artifact of past mis-directed forest management.
However, the reconstructions and early scientific reports both show that these forests are v www.esajournals.org inherently dangerous places to live, and will remain so if restored. Even low-intensity fires often blew up into small high-severity patches, and episodically fires became very severe and unstoppable over thousands of hectares. Even if society did not want to restore Sierran mixedconifer forests and instead just wanted to prevent severe fires, this study shows fuel-reduction programs in wildlands are unlikely to work well. The understory fuels targeted in contemporary fuel-reduction programs were very extensively reduced in the late-1800s by overgrazing, yet high-severity fire still burned thousands of hectares of mature forest at high intensity.
The focus instead can be where it is essential and effective, which is to make our homes firesafe through fuel reduction in home-ignition zones (Calkin et al. 2014), and reduce ignitions by people. Communities can also create growth boundaries, and rearrange their land uses to place ball fields, parks, wetlands, canals, irrigated crops and other low, open, less-flammable land uses on the outskirts (Baker 2009). In public forests, Smokey the Bear is still needed. About half the burned area in the 20 largest fires in California was from ignitions by people (www. fire.ca.gov). Closure of public forests during severe droughts is sensible. We can also reduce accidental ignitions by people in forests by redirecting development away from forests (Syphard et al. 2007) and by creating passive firesafe features in places where people recreate, camp, stop, and drive, as well as near infrastructure (e.g., powerlines). People and wildfire can coexist in dangerous dry western forests if we accept that historically dominant and ecologically essential mixed-severity fires inherently have overwhelming physical power that requires us to adapt (Calkin et al. 2014).

ACKNOWLEDGMENTS
Audrey Harvey entered more than half the data into GIS, a key contribution. I appreciate field assistance and helpful insights by Ted Bartlett. Carl Skinner, Kevin McKelvey, and Chad Hanson found important early reports. Chad Hanson provided helpful comments. Funding is from Environment Now, but antecedent work behind the study was funded by the National Science Foundation and the Cooperative State Research, Education and Extension Service, U.S. Department of Agriculture. Funding sources had no role in the study design, its execution, or its publication. I appreciate helpful peer-review comments. Thanks to Yosemite and Sequoia National Parks for permits that allowed field research. This article is dedicated to John Leiberg, whose remarkably detailed observations, mapping, and insights more than a century ago revealed the power of wildfires in shaping Sierran mixed-conifer forests.    Logging caused fires Show and Kotok (1924:5-6) Whole Sierra Q10: ''The written historical records of the period, though extraordinarily meager on this question, indicate that the early miner was the cause of many forest fires . . . there is evidence that these early prospectors found themselves hampered by brush and young growth, and adopted the practice of setting fire to the woods in order to facilitate their search for gold-bearing outcrops.''

LITERATURE CITED
Mining caused fires Leiberg (1902:63-64) North Fork of the Feather River Q11: ''The burned region west of the river corresponds exactly to the extent of the auriferous areas where mining has been carried on since 1850. The tracts more severely burned east of the river are not situated in a very rich mineral region, but connect directly with burned tracts adjoining the placer grounds east of Spanish Peak. On its face the evidence would seem to warrant the conclusion that the fires which have ravaged the basin most extensively followed in the steps of miners and prospectors of the early days. The correctness of this conclusion is further strengthened by the fact that the big burns throughout the country examined lie contiguous or very close to much of the richest mineral ground.''

Mining caused fires
Flintham (1904:38) Southern Sierra Q12: ''Mining. . . Generally carried on on the lower foothill slopes below the lower line of the forest, the prospecting in various regions has occasioned practically no modification to the forest, except where locally the stand in the pine belt has been culled for mine timbers.'' Mining did not cause fires Leiberg (1902:41) Northern Sierra Q13: ''The only older burns which give any clues to their age are those which stretch in a line from northwest to southeast through the central district of the region. They are marked by the occurrence of large tracts covered with chaparral. Most of these areas are situated contiguous to placer camps, worked from the earliest times, and might be regarded as having been burned over by fires spreading from such camps. In some instances this most likely happened, but a large proportion of the chaparral tracts was denuded of forest so long ago that nearly all the stumps have decayed. Hence the fires which overran them probably date back to the early part of the last century.''

Mining did not cause older fires
Low-to moderate-severity fire Sudworth (1900:557) Southern Sierra Q14: ''The fires of the present time are peculiarly of a surface nature, and with rare exception there is no reason to believe that any other type of fire has occurred here. . . The tree roots are for the most part buried deep in the crevices of bare rock, in gravel, sand, or shale, over which surface fires run annually without the slightest direct injury to the roots. Barring the debris left from timber-cutting, the only food for these fires is the scanty fall of pine and fir needles, irregular patches of low conifer seedlings and chaparral. In general, these materials limit the fires to surface burning.'' where the timber has not been wholly destroyed, has more than quadrupled in density, while on some of the southern slopes, like the ridges between Long Canyon and Rubicon River, soil aridity has followed to such an extent that the chaparral is scarcely able to obtain a foothold. All the areas around French Meadows tell the same tale and show the same picture of scattered broken stands of timber set in dense undergrowth, or separated by lanes of chaparral. All the way down the main canyon of the Middle Fork of the American River there is a succession of these fire glades, alternating with heavy stands which serve to indicate the former density of the forest. All the slopes of Duncan Canyon from its head down show the same marks of fire-dead timber, dense undergrowth, stretches of chaparral, thin lines of trees or small groups rising out of the brush, and heavy blocks of forest surrounded by chaparral. North of Duncan Peak and connecting with the burns on the northern slopes of North Fork of American River Canyon the forest has been burned out in narrow lanes and patches. In some places brush has replaced the timber, in other localities the ground has been too rocky, soil aridity has set in, and low shrubs or coarse weeds thinly cover the ground.'' Fires were mixed severity with about 20-30% high severity in lanes and patches Leiberg (1902:130) Leiberg (1902:41) Northern Sierra Q40: ''It is estimated that the areas badly burned-that is, those on which 50 per cent or more has been destroyed-comprise 715,440 acres [this includes moderate-and high-severity fire], and of this amount there are 213,730 acres, in tracts larger than 80 acres, on which the destruction has been total [this is the highseverity part]. If the many small lots of badly burned forest which are scattered throughout the still growing stands were taken into the account, the figures given above would be considerably increased.'' Fires of moderate and high severity covered more than 290,000 ha High-severity fire Sudworth (1900:558) Southern Sierra Q41: ''The instances in this region where large timber has been killed outright by surface fires are comparatively rare. Two cases only were found, and are shown on the accompanying map. . . One of these burns involved less than an acre, and the other included several hundred acres. They are exceptional cases, and the killing of the trees is accounted for by the fact that long protection from fire and from all but cattle grazing had resulted in the accumulation of much fallen timber, considerable humus in depressions and on benches, and a dense undergrowth of brush and seedlings. The fires burned deep enough to badly injure the surface roots, which resulted in subsequent death of the timber.'' High-severity fire was rare, occasionally up to 100-200 ha Show and Kotok (1924:31) Whole Sierra Q42: ''Extensive crown fires, though common in the forests of the western white pine region, are almost unknown in the California pine region. Local crown fires may extend over a few hundred acres, but the stands in general are so uneven-aged and broken and have such a varied cover type that a continuous crown fire is practically impossible.'' High-severity fire was rare, occasionally up to 100-200 ha Kotok (1933:4020) Eldorado County Q43: ''. . .complete wiping-out of the original coniferous forests representing the climax types, never extended over large areas during the Indian period. In contrast, with the advent of white man, lumbering, grazing, and an enormous increase in fires (in a 75-year period) brought rapid retrogression.'' High-severity fire historically rare, became common after EuroAmerican settlement Show and Kotok (1924:13) Whole Sierra Q44: ''A dense, closed stand of timber, on the other hand, will more readily develop a true crown fire. Such fires are the rule in dense, even-aged second-growth stands where there is an uninterrupted tree canopy.'' High-severity fire likely in dense second growth Show and Kotok (1924:27) Whole Sierra Q45: ''With summer fires, these dense groups of reproduction, even in sapling and pole stages, are peculiarly susceptible to crown fires, just as the larger second-growth stands are.'' High-severity fire likely in dense second growth Show and Kotok (1924 Hall (1909:29) Plumas National Forest Q49: ''The most destructive fire in 1908 occurred on the western edge of the Forest, mostly outside. It was started in the foothill type outside of the Forest and ran up a ways into the Forest burning over in all outside and inside some 10,000 acres . . . in the timber destroying everything in its central path, even large timber.'' High-severity fire in 1908 over a large area Hall (1909:29) Plumas National Forest Q50: ''On private land near Quincy, a very bad fire burned over some sixty acres of mostly thick stands of pure yellow pine almost completely killing every one and killing nearly all of the large mature trees.'' High-severity fire of about 60 ha in dense ponderosa pine Hodge (1906:39) Whole Sierra Q51: ''The leaves also are very inflammable, and crown fires in this species [white fir] are not uncommon.'' High-severity fire common in white fir Leiberg (1902:117) Middle Fork of the Yuba River Q52; ''South of the Middle Fork Canyon the region of the yellow-pine forest is fire marked and damaged very much as in the corresponding areas north of the river, to a point just beyond Bloomfield. From here on, eastward to Shands, the timber is composed largely of white fir, probably to the extent of 60 per cent, and long swaths have been burned in all directions through these stands of low fire-resisting capacity.'' High-severity fire in white fir in long swaths Leiberg (1902: Leiberg (1902:43) Northern Sierra Q58: '' An increase in density and extent of brush growth below the 7,300-foot level is here an unfailing consequence of fires. In the yellow-pine type of forest and in the woodland it grows to larger proportions, and here and there, where the timber has been totally destroyed, it forms patches of pure growth. . .there can not be the slightest doubt that every acre of chaparral represents so much ground once forested, denuded by fire, then overgrown with brush. At the head of Slate Creek, and near Howland Flat, in the western portion of Sierra County, chaparral identical with these growths elsewhere in the region is now in the process of formation on tracts which were covered with forest to within twenty years ago.'' High-severity fire led to chaparral Leiberg (1902:65) North Fork of the Feather River Q59: ''Elsewhere [outside the woodland belt below SMC forests] a brush growth close enough to be called chaparral is invariably a sequel to the total destruction of the forest on any area below the highest subalpine elevations.'' High-severity fire led to chaparral Hodge (1906:14-15) Whole Sierra Q60: ''Patches of pure chaparral without trees occur throughout the Sierras, particularly at the higher elevations. They almost invariably occupy situations that are capable of producing timber, and their origin lies in forest fires.'' High-severity fire led to chaparral Cooper (1906:30) Whole Sierra Q61: ''In many portions of the State where fires have been exceptionally bad this process has gone so far that the tree growth has been entirely replaced by chaparral.'' High-severity fire led to chaparral Hodge (1906:39) Whole Sierra Q62: ''The white fir forest is usually extremely dense, containing little undergrowth. When it is destroyed by fire, however, it is usually seeded up promptly to chaparral.'' High-severity fire in white fir led to chaparral Leiberg (1902:84) Middle Fork of the Feather River Q63: ''Bush-covered tracts occur everywhere in the basin, but chaparral proper is found only where the forest has been destroyed by fire. . . In yellow-pine forests the brush is chiefly a thin undergrowth scattered among the growing trees, except on ground where the forest has been wholly or partially destroyed by fire, when it forms true chaparral.'' High-severity fire led to chaparral Leiberg (1902:118-119) Middle Fork of the Yuba River Q64; ''Where fires have swept through the lower areas of the yellow-pine forest and consumed the timber, a heavy growth of manzanita . . . has followed as a sequel. . . In the upper areas of the yellow-pine forest, especially where white fir constituted the principal species, heavy masses of brush growth composed of species of ceanothus, scrub oak, and manzanita . . . for a chaparral cover on the denuded or partially denuded tracts.'' High-severity fire led to chaparral Leiberg (1902:156) North Fork of the American River Q65: ''From the western limits of the forested regions to Emigrant Gap the traces of fire are more obvious. Partly or wholly dead timber seared by fires and the brush growths following in their wake exist in every canyon and on every ridge.'' High-severity fire led to chaparral Leiberg (1902:171)  Southern Sierra Q71: ''The most noticeable effect of the fire in the fir forests is the denudation of the cover-the opening of areas in the forest. . . There result within the forest small open spots, in which dead stubs of the fire-killed timber stand . . . and frequently on the upper slopes and crests of ridges great areas of openings, which fire or death following upon it have completely denuded of the cover formerly occupying the site as evidence from charred logs and occasional stubs or trees left standing. The areas opened in the fir stand after the fire, more especially the denuded crests of the ridges and the divides, have generally been overgrown by a specially dense and heavy chaparral of species occurring as undergrowth under the stand at this altitude. . . Fire and chaparral extension have stood to each other throughout the Sierras in the relation of cause to effect. The encroachment of the chaparral has uniformly followed fires, and has been uniformly inimical to the forest cover. . .'' High-severity fire led to chaparral Flintham (1904:166) Southern Sierra Q72: ''However, denudation of the forest cover has been confined to relatively small areas, and has led to no serious deterioration of any section of the mountains from excessive erosion or other injurious agency, because the removal of the forest cover does not imply complete opening of an area but generally its recovering with a dense chaparral growth. . .
High-severity fire led to chaparral Leiberg (1902:95) South Fork of the Feather River Q73: ''The young growth is in thickset stands with little underbrush. Here and there it is broken . . . by hillsides formerly nearly deforested by fires and now covered with a close growth of manzanita.'' High-severity fire led to chaparral with manzanita Leiberg (1902:132) South Fork of the Yuba River Q74: ''In the yellow-pine forest, especially in the restocking where patches have been burned clean, a chaparral springs up 3 to 5 feet high, composed almost exclusively of manzanita.'' High-severity fire led to chaparral with manzanita Leiberg (1902:145) Bear River Basin Q75: ''Small brush-covered patches of ground occur here and there in the eastern portion, having followed as a sequel to fires in the yellow-pine forest. The brush consists chiefly of manzanita.'' High-severity fire led to chaparral with manzanita Leiberg (1902:157) North Fork of the American River Q76: ''All along the Blue Canyon and Canyon Creek drainage a large proportion of the reforestation consists of scrubby oak instead of the coniferous species of trees which formerly constituted the timber in these localities. The extensive and heavy stands of brush which have here come as a sequel to fires show no sign of being replaced with tree growth. . . All the higher slopes of Monumental Hill are covered with dense chaparral, which will not be replaced by forest for a century or more.'' High-severity fire led to oak and chaparral v www.esajournals.org Source Location Quote Interpretation Kotok (1933:4019-4020) Eldorado County Q77: ''In reconstructing the ecological changes in this county due to fire the following major retrogressions have been noted: The Pinus ponderosa with its usual understory of Quercus kelloggii changes to one of Quercus kelloggii with scattered pine. . . The pure Pinus ponderosa type passes to a Ceanothus integerrimus, Arctostaphylos viscida association. . . In the mixed-conifer type. . .infrequent fires will favour the pines against the firs, arresting the succession to the fir climax. Repeated fires will invariably reduce the high coniferous forests to sprouting shrub associations.'' High-severity fire led to oak and chaparral Leiberg (1902:124) South Fork of the Yuba River Q78: ''The forest has been extensively ravaged by fires and is almost everywhere badly choked with underbrush. Up to 4,500-foot level there are large quantities of oak mixed with it; above that altitude the oak soon thins out and disappears.'' High-severity fire led to oak and chaparral Leiberg (1902:59) North Fork of the Feather River Q79: ''On the slopes of Clermont Hill, south of American Valley, the forest is thin and scattered, and is composed largely of small red fir [Douglas-fir], being set on rocky ground among great masses of undergrowth.'' High-severity fire led to small trees in dense chaparral Leiberg (1902:77) Middle Fork of the Feather River Q80: ''A few miles west of Lumpkin the forest rapidly begins to thin out. . . The stands . . . have been extensively thinned by the ravages of repeated fires. In much of this section the timber is set in thick brush, which has, as elsewhere, increased enormously as a sequel to the fires.'' High-severity fire led to thin forests in dense chaparral Leiberg (1902:77) Middle Fork of the Feather River High-severity fire led to thin forests in dense chaparral Leiberg (1902:78) Middle Fork of the Feather River Q82: ''From Buckeye south . . . a large percentage of the fir has been damaged by recent fires, and nearly all of it stands in thick growths of underbrush.'' High-severity fire led to thin forests in dense chaparral Leiberg (1902:95) South Fork of the Feather River Q83: ''Northeast of Lumpkin . . . the basin of the South Fork shows, by the uniform thinning of the forest and the abundance of chaparral, ample evidence of widespread fires.'' High-severity fire led to thin forests in dense chaparral Leiberg (1902:100) North Fork of the Yuba River Q84: ''The yellow-pine forest, as it follows up the lateral canyons. . .is mostly thin and small. . . The larger proportion consists of red fir [Douglas-fir], here and there white fir, incense cedar, and oak, generally set in thick underbrush, the result of many and severe fires.'' High-severity fire led to thin forests in dense chaparral Leiberg (1902:165) Middle Fork of the American River Q85: ''The forest on the slopes and in the canyon of Rubicon River below the mouth of Grayhorse Canyon is extremely thin. . . On the southern declivities the forest occurs as isolated blocks of thin growth standing in heavy underbrush on bowlder-[sic] strewn slopes. Yellow pine forms 25 to 40 per cent; white fir most of the remainder'' High-severity fire led to thin forests in dense chaparral Leiberg (1902:80) Middle Fork of the Feather River Q86: ''. . .small quantities of yellow pine and red fir [Douglas-fir] in canyons, all of poor quality and generally small size; set in dense chaparral. . .'' High-severity fire led to small trees in dense chaparral Leiberg (1902:81) Middle Fork of the Feather River Q87: ''Region around Mount Jackson and Grizzly Peak: Chiefly white fir, some small and stunted yellow pine, occasional bunches of Shasta fir; all in thick chaparral . . . much damaged by repeated fires.'' High-severity fire led to small trees in dense chaparral Leiberg (1902:77) Middle Fork of the Feather River Q88: ''From Mount Ararat west to Buckeye the forest occurs in more or less isolated stands surrounded by chaparral.'' High-severity fire leaves unburned patches in dense chaparral Leiberg (1902:82) Middle Fork of the Feather River High-severity fire led to thin forests and scattered trees or patches in dense chaparral Leiberg (1902:164) Middle Fork of the American River Q91: ''The forest is mostly yellow-pine, not having been logged; the stands are thin and scattered, everywhere broken by tracts of chaparral or rocky exposures with little soil and hardly any tree growth.'' High-severity fire led to thin forests and scattered trees in dense chaparral Leiberg (1902:156) North Fork of the American River High-severity fire led to thin forests and scattered trees or patches in dense chaparral Leiberg (1902:171-172) Middle Fork of the American River Q94: ''In the valley of Long Canyon 20 to 30 per cent of the timber has been destroyed. . . Everywhere the undergrowth, where the timber has not been wholly destroyed, has more than quadrupled in density, while on some of the southern slopes, like the ridges between Long Canyon and Rubicon River, soil aridity has followed to such an extent that the chaparral is scarcely able to obtain a foothold. All the areas around French Meadows tell the same tale and show the same picture of scattered broken stands of timber set in dense undergrowth, or separated by lanes of chaparral. All the way down the main canyon of the Middle Fork of the American River there is a succession of these fire glades, alternating with heavy stands which serve to indicate the former density of the forest. All the slopes of Duncan Canyon from its head down show the same marks of fire-dead timber, dense undergrowth, stretches of chaparral, thin lines of trees or small groups rising out of the brush, and heavy blocks of forest surrounded by chaparral. North of Duncan Peak and connecting with the burns on the northern slopes of North Fork of American River Canyon the forest has been burned out in narrow lanes and patches. In some places brush has replaced the timber, in other localities the ground has been too rocky, soil aridity has set in, and low shrubs or coarse weeds thinly cover the ground.'' High-severity fire led to thin forests and unburned patches in dense chaparral, along with some areas of herbaceous vegetation Show and Kotok (1924:42-43) Whole Sierra Q95: ''Estimates made after years of study of brush fields indicate that about two-thirds of their area is reproducing sufficiently to establish eventually a commercial forest. The extent to which tree reproduction is taking place depends naturally on the number and distribution of seed trees available, for regeneration can be counted on to a distance of only a few hundred feet from seed trees. Smaller brush fields, generally speaking, are restocking in a satisfactory manner. It is chiefly in the very large brush areas of 5,000 acres [2024 ha] or more . . . that complete restocking will be a matter of several tree generations, or of planting.'' Forest recovery in chaparral after high-severity fire v www.esajournals.org Source Location Quote Interpretation Show and Kotok (1924:43) Whole Sierra Q96: ''The amount of reproduction present in the brush fields to-day is very much greater than would seem on superficial examination, for in many places the young trees are just beginning to break through the brush canopy and to become easily visible. This condition is wholly the result of 15 to 20 or 25 years of fire exclusion.'' Forest recovery in chaparral after high-severity fire Moore (1913:22) Plumas National Forest Q97: ''The brush, particularly on the drier sites, appears to assist rather than to hinder reproduction. In the yellow pine-incense cedar type the brush areas are not extensive and generally contain scattering trees, particularly yellow pine. The brush here does not form a complete covering, but leaves spots of exposed rocky soil. . .In the mixed conifer type the brush areas, though extensive, generally contain scattered old yellow pine and sugar pine individuals which have resisted the fire. The brush here forms a nearly complete canopy hindering, though not entirely preventing the reproduction of yellow pine. Sugar pine, because of its greater tolerance, comes up well through the thinner places in the brush.'' Forest recovery in chaparral after high-severity fire Show (1924:83) Northern Sierra Q98: ''Perhaps the most striking characteristic of the timber region of northern California . . . is the very large area occupied by brushfields. The brushfields, for the most part, are the result of fires which have destroyed the timber and allowed the brush to occupy the ground; in round numbers 1,500,000 acres are now in this condition.
Of this million and a half acres probably 75 per cent is restocking naturally, scattered individuals and groups of trees having survived the fires of the past, and can be depended on to take care of themselves. . .'' Forest recovery in chaparral after high-severity fire Boerker (1915:15) Northern Sierra Q99: ''Unlike the chaparral regions of southern California, this brush is only a temporary type and is, in most cases, the result of fire having destroyed the forest cover. . . In most cases, in from 5 to 10 years after the fire has consumed the timber, the brush takes possession of the land; the length of time depending upon the severity of the fire, the presence of brush plants in or near the fire area, and other conditions. After the brush has established itself, if seed trees are nearby, seedlings will get started and fight their way through the brush. It takes from 15 to 30 years for a seedling to get large enough to overtop the brush . . . if the stand of saplings becomes dense enough, the brush underneath will be killed for lack of sunlight and a forest cover will begin to form. This is nature's very slow process of reestablishing a forest cover. If there are no seed trees near the burned area, it is only a matter for conjecture how long it will take a forest to reestablish itself.'' Forest recovery in chaparral after high-severity fire Show and Kotok (1924:36) Whole Sierra Q100: ''Further, the gradual thinning of the forest allows the invasion of brush and other inflammable cover so that succeeding surface fires more readily develop into disastrous crown fires.'' Chaparral fires are also high severity Show and Kotok (1924:43) Whole Sierra Q101: ''Fires in the brush fields . . . are of serious moment, not because they destroy merchantable timber, but because at one stroke they may sweep the new forest from thousands of acres and even destroy the scattered seed trees that are necessary to maintain the forest type.'' Chaparral fires are also high severity Show and Kotok (1924:43) Whole Sierra Q102: ''Not only do fires in brush attain a greater average size than in timber, but for equal areas of timber land and brush land nearly seven times as many acres of brush land are being burned each year as timber.'' Chaparral fires are larger and more frequent than forest fires Show and Kotok (1924:44) Whole Sierra Q103: ''Recent studies have shown . . . that fires in brush are far more difficult to control than those in virgin forest, and attain a much larger average size. Once started, also, they are likely to sweep into adjoining timber stands with an intensity that results in wiping out the immediately adjoining timber belt, thus extending the brush fields themselves. Fires in brush fields are typically crown fires and partake of the nature of crown fires in timber.'' Chaparral fires are also high severity, larger, and expand into forest v www.esajournals.org Table A1. Continued.

Source Location Quote Interpretation
Repeated fires led to chaparral: the attrition theory Show and Kotok (1924:29) Whole Sierra Q104: ''On such as burn as the Ham Station fire . . . not only was a certain percentage stripped of timber, but this area as a whole is reverting to brush, and the brush will in turn increase the intensity of subsequent fires. . . Another example . . . is the Howard fire. . . This area has been subject for many years to repeated forest fires and by this process of attrition has been reduced in density, thereby increasing the amount of undergrowth. The latest fire on this area, which has caused tremendous damage, is the logical and inevitable result of previous fire history.
All such examples lead to the same conclusion: Fire in forest areas invariably breeds still more serious fires.'' Attrition theory of chaparral formation by repeated fires Show and Kotok (1924:78) Whole Sierra Q105: ''Fires in the virgin forests of the California pine region rarely are catastrophes, for they do not wipe out at one stroke the entire stands over a large area. Indeed, they are generally distinguished by the fact that much of the damage is relatively inconspicuous and not immediately evident. But a study of the fires of the past and those of the present shows unmistakably that attrition is the inevitable concomitant of repeated fires. This wearing down of the forest is remarkably exhibited in all its varied stages in the California pine region to-day, from the wellstocked areas of mature timber to the nontimberproducing chaparral. The fire-scarred virgin forest, the broken, patchy timber stand of no present merchantability; the brush fields with scattered, isolated trees, and small groups of trees; the continuous brush fields occupying potential timberland and restocking only slowly; and finally, pure brush or chaparral, the end product, are but the different chapters of the story of attrition.'' Attrition theory of chaparral formation by repeated fires, but consistent with highseverity fire leaving scattered trees and unburned patches of forest in dense chaparral Show and Kotok (1924:41-42) Whole Sierra Q106: ''The most convincing proof that the brush fields are the result of fire is that within a comparatively short distance there may be found all the gradations from a stand of merchantable virgin timber to a stand of brush with no living trees. . . Other evidence that the brush fields were formerly timberland, and have reverted to their present condition chiefly through fires, may be summarized as follows: 1. In the largest brush fields occur scattered patches or islands of virgin forest in naturally protected spots, where our knowledge of present fires shows that timber would be least susceptible to complete destruction. Also living stands of old-growth virgin timber are found immediately adjacent to brush fields and occupying similar sites. 2. Scattering living trees and snags, bearing the evidence of many fires, are not unusual in even the largest brush fields. Even in brush fields with no standing trees or snags a careful search nearly always reveals burnt remnants of tree trunks, stumps, or hollows formed by the complete burning out of stumps. 3. Repeated burnings are shown in charred remains of brush found in brush fields. 4. The woody species occurring as underbrush in the virgin forest are the same as those constituting the cover of adjacent brush fields, and brush is known to sprout after fires in which conifers have been destroyed. 5. Reproduction of coniferous species becomes established in the brush fields wherever seed trees are present and fires are absent.
Attrition theory of chaparral formation by repeated fires, but consistent with highseverity fire leaving scattered trees and unburned patches of forest in dense chaparral Sterling (1904a:3) Southern Sierra Q107: ''Here, as in the northern Sierras, the prevalence of chaparral over large areas suitable for timber growth, and undoubtedly once under forest cover, is due to repeated fires which gradually killed the seed trees and destroyed all young growth.'' Fires kept young reproduction rare Show and Kotok (1924:26) Whole Sierra Q171: ''It is shown that instead of uniformly thinning the stands of seedlings and saplings, a surface fire wipes out a certain portion of the stand wherever it runs. Outside the area actually burned, and yet within the boundary of the burn as a whole, there will be a certain amount of reproduction, even though small, that is able to survive.'' Fires kept reproduction rare, except in unburned areas Show and Kotok (1924:60- Sugar pine reproduction and others were favored along roads and trails Greeley (1907:106) Southern Sierra Q190: ''Where sugar pine seedlings and saplings occur at all they are usually in groups under broken cover or in narrow openings in the stand. On the same sites fir and cedar crowd the sugar pine closely and bunches or large patches of these species occur in among the groups of sugar pine. Yellow pine, in this mixed forest, seeks the drier and warmer sites on the tops of ridges and southern exposures. Here it also is commonly found in large even-aged groups, from open bunches of mature trees to thickets of saplings and seedlings.'' White fir, Douglasfir and incense cedar reproduction was in dense thickets, ponderosa pine rarer, except on driest sites Sudworth (1900:524) Southern Sierra Q194: ''The reproduction of white fir is very general over the range of the species, and in some sections the young growth is exceedingly abundant. Thickets of seedlings and saplings are often found covering many acres, and to the exclusion of all other species. In locations where other young growth is present the white fir may comprise 40 to 60 per cent of the whole growth. The wonder is that mature trees of this species are not more abundant. But when fires occur, the richly resinous foliage and branches of the young growth suffer more severely than the pines or cedars. Owing to thinness of foliage and less resin, a few of the latter may escape fatal burning; but it is rare than any of a low thicket of firs ever survives even a surface fire.'' White fir reproduction was common and very dense over some large areas, unless fires  Analysis of human-effect zones at the time of the surveys I estimated buffer widths that represent the effect zone around each land use, so that I could use this to find the area less affected by Euro-American land uses. I used historical descriptions of land uses (e.g., Sudworth 1900, Leiberg 1902, Laudenslayer and Darr 1990, McKelvey and Johnston 1992, Gruell 2001, Beesley 2004, combined with more general analyses of effect zones for specific land uses (e.g., Forman et al. 2003), and a buffer analysis of an indicator of disturbance. The indicator is the area of patches of montane chaparral, which were observed to be created directly by logging as well as indirectly by high-severity fires spreading from ignitions by people or lightning fires in logging slash (Leiberg 1902, Show andGreeley 1926). I reasoned that, if more chaparral area occurred in the vicinity of a land use than in the larger study area, that is an added indicator of an effect-zone for the land use.
The effect zone is not known for many landuses and only roughly known in general, and I am buffering line data which do not intersect all land-uses. Thus, I erred toward including too much area to increase the probability of approximating the complementary less-affected area.
I began with hypotheses about the width of effect zones for each land use. I expected a 100-m buffer around water-system features (ditches and reservoirs), as trees were likely not extensively removed near these features. I expected 200-m buffers around roads and trails, in part based on Forman et al. (2003), but also early photographs that show varying, but limited impact on forests near roads (Gruell 2001). I expected a 3220-m buffer (2 miles) around railroads, because Leiberg (1902:39) said: ''...a strip about 4 miles wide, from Truckee to Colfax, paralleling the Southern Pacific Railroad, is said to have yielded up its forest chiefly to supply the locomotives...,'' also mentioned in McKelvey and Johnston (1992). However, this may be an overestimate of the effect zone, as Gruell (2001) showed that in six of seven early photographs of the Central Pacific Railroad line, logging occurred in the railroad right-of-way, but not on nearby slopes.
I expected a 1000-m buffer around farms, ranches, and buildings, assuming that individuals may have most sought and transported fuel wood and building materials within this distance, but would be unlikely to burn near their infrastructure. Around sawmills, I expected a 2414-m buffer (1.5 miles), based on Beesley (2004), who suggested that early Sierran sawmills typically would cut timber within a 2.5-to 3-mile circle before moving to a new location, also suggested by Berry (1917). However, Sudworth (1900:513) said: ''A common practice of mill operators is to consume all saw timber in a radius from the plant of 2 1/2 to 3 miles, and then move to another site.'' Around mining operations, I hypothesized only a 1000-m buffer. Mine timbers and wood for other uses were often supplied by sawmills, not by the mining opera-  37 m). à Meandist is a measure of local tree density, based on the mean distance among the four trees at the section corner.
§ Equations are shown for Q. kelloggii, but the crown-radius fit was poor, thus the ''all species'' equation was used instead.
v www.esajournals.org tions themselves (Beesley 2004). Early photos show that local removal of timber near some larger mining operations could extend beyond 1000 m, but in many other cases the forest appears to have remained unaltered even within 100 m of mining operations (Gruell 2001), thus 1000 m may be a high estimate of the average width of the effect zone.
To refine these hypotheses, I buffered each occurrence of every land use (Table E1) using buffers of 100 m, 200 m, 500 m, and 1000 m and larger buffers mentioned above for specific land uses, then tested whether the percentage of chaparral area inside the buffered area exceeded the percentage of chaparral area in the study area, suggesting a concentration in the buffer area. To ensure a sufficient sample, I only completed the analysis if there were at least 10 occurrences of the land use and the area inside the buffered land use exceeded 1000 ha. If an effect was observed at a particular buffer width, I expanded the buffer width to further investigate the extent of an effect zone. After reviewing the results of this buffer analysis of chaparral patches, I finalized buffer widths, buffered each land use and measured its area, then merged all the buffers into a single map of buffered human effects.
Some potentially indirect effects on forest structure cannot be spatially modeled. Fires were reportedly started by sheep herders moving across these landscapes in the late-1800s (Leiberg 1902, McKelvey andJohnston 1992), although this effect may have been overestimated by early observers (Vankat and Major 1978). Excessive grazing by livestock, widely evident by A.D. 1900 (Leiberg 1902, Vankat andMajor 1978) may have reduced fine fuels enough by the time of the surveys to have also reduced fire spread. I could not model effect zones from these mobile or undefined sources, as they extend for unknown distances or from unknown locations.
Combined human effects, in terms of their length along section lines, are more than ten times greater in the northern Sierra than the southern Sierra (Table E1). The buffer analysis of chaparral percentage shows sawmills had a substantial effect near the mill and a detectable effect extended beyond the hypothesized 2414 m, so I extended the buffer to 4000 m, where the level of chaparral is similar to the study area (Fig.  E1a). This 4000 m distance (about 2.5 miles) is Areas inside buffers may overlap among land uses, so cannot be summed at the bottom of the column. à An ellipsis (. . .) indicates that data were not available or an entry would not be appropriate.
v www.esajournals.org remarkably similar to the 2.5 to 3 mile radius that Sudworth (1900:513) suggested, which validates his observation and also this buffer analysis approach. Slightly more chaparral occurs in the railroad buffer of 1000 m but not in the hypothesized 3220 m buffer (Fig. E1a), but because of Leiberg's quote, I decided to leave the buffer at 3220 m. Although ranches and farms showed slightly elevated chaparral within 1000 m in the southern Sierra, they showed much less chaparral than expected in the northern Sierra, so I left the buffer at 1000 m. People may generally, but not always have been successful in avoiding fires and other disturbances near their ranches and farms. Similarly, there was no elevated chaparral within 1000-3000 m buffers in the northern Sierra, but there is slightly elevated chaparral in 2000-5000 m buffers in the southern Sierra. Looking at the map, it appears possible that one fire could have been ignited and escaped from near buildings in the southern Sierra, but more likely the buildings happened to be near a wildfire. Since there were four times as many ranches and farms in the analysis of the northern Sierra, I left the buffer at 1000 m for buildings. Road-and-trail buffers of 100 m and 200 m have very slightly elevated chaparral in the northern Sierra, but not in the southern Sierra, thus I left the buffer at 200 m. As hypothesized, there is no detectable effect from water systems or mining, but I left buffers at 100 m and 1000 m, respectively, to err on the side of The map is a model and of course has limitations, but it does have some validation in early scientific reports and maps of humanaffected areas. The human-effect area from the survey data matches well with the area that Leiberg (1902) mapped as having been logged/ culled (Fig. E2), likely with many other associated land uses, including roads and trails, a water system etc. There are areas inside Leiberg's mapped area that had likely not been logged or otherwise altered by the time of the surveys, but v www.esajournals.org were altered later, by the time of Leiberg's mapping. I do not have a digital version of Fitch's (1900a, b) maps of 30 0 quadrangles, which overlap a few townships of my southern Sierra study area in and just west of Yosemite. However, these maps show that there was no area in the overlap mapped as logged/culled, and the survey data also document no logging in this overlap area.

Succession of chaparral patches to forest
Are chaparral patches successional to forests or more permanent because of their environmental setting? I tested this in the two study areas by examining recent aerial photographs of locations that were dominated by chaparral at the time of the surveys.
I selected two townships in the north and two in the south that had the most chaparral at the time of the surveys. These four townships cover about 37,000 ha and include some steep canyon slopes, some ridgetops, as well as more gentle topography, and include both low and high elevations. I downloaded, for each township, the 8-10 digital orthophoto quads (DOQs) from the U.S. Geological Survey (http://earthexplorer. usgs.gov) needed to cover the area of chaparral. These black-and-white photos have about 1-m pixel resolution, sufficient to see moderate-sized individual trees, and are dated from 1993. I overlaid these DOQs on the line-segments that surveyors identified as dominated by chaparral shrubs (e.g., Ceanothus integerrimus, C. cordulatus, Arctostaphylos patula, A. viscida etc.). Surveyors were required to map the location where they left forest and entered chaparral and vice versa.
Human effects were visually evident in the aerial photographs, in some areas, in the form of roads, contouring, logging in nearby forests, and other disturbances. I could not ground-truth to ensure that apparent chaparral was truly still chaparral. I also could not determine whether chaparral patches that persisted were the result of a fire since the time of the surveys. It is also possible that some areas that are now forested had trees that were planted.
The comparisons, over periods of 109-118 years, show that 21.8% of the chaparral present at the time of the surveys was still chaparral by 1993, and 78.2% of the chaparral became forest (Table F1). Greater persistence of chaparral, mostly in T004SR020E, was strongly associated with southerly-facing slopes, as virtually all patches on northerly-facing slopes were forested by 1993. Previous studies have shown that montane chaparral may remain dominant for up to about 60 years after fire in the northern Sierra (Conard and Radosevich 1982), and more than 100 years if fires recur (Wilken 1967). I did not determine whether fires recurred in the chaparral areas I sampled in this study. In the Lake Tahoe area, chaparral area declined by 62.4% on average on mostly xeric southerly-facing slopes during a .120-year period after fire that included no subsequent fires (Nagel and Taylor 2005). Early authors also noted that 2/3 to 3/4 of chaparral areas was recovering to forest (Appendix A: Q95, Q98). This is a little less than the 78.2% decline observed here in 109-118 years, although the more southerly, xeric slopes here showed less than the 78.2% decline. Vankat and Major (1978) also documented, using repeat photography, that chaparral stands in Sequoia National Park that were evident in photographs taken before 1920 Table F1. Percentage of chaparral present at the time of the surveys that was still chaparral or was forested by A.D. 1993 in four townships in the northern and southern Sierra. showed invasion and/or increase in trees by the 1970s. Vankat and Major (1978:382) said ''These findings support the hypothesis of Show and Kotok (1924) that stands of such shrubland vegetation are the result of single intense fires, or are the cumulative effect of repeated fires in areas of potential forest vegetation.'' Based on these findings, I suggest that about 80% of the chaparral in my western Sierra Nevada study areas could succeed to forest on a 60-120 year time-scale, rather than persist because of its environmental setting. However, all montane chaparral likely could be maintained as chaparral over long periods if fires recur (Wilken 1967). This is tempered by the finding that fire may actually be less likely in chaparral than in surrounding forests (Nagel and Taylor 2005), although this was not the suggestion of Show and Kotok (1924), who envisioned that fires in chaparral were more frequent, larger, and more severe than in forests (Appendix A: Q100-Q103). However, all fires that could maintain chaparral by killing trees would be high-severity fires, as commonly noted by early observers (Appendix A: Q100, Q102, Q103). Table G1. Forest-density estimates for historical forests of the western Sierra Nevada Mountains and nearby areas in California cross-validated with survey-based estimates. Estimates are divided into ponderosa pine-Douglasfir, Sierran mixed conifer, and white fir phases. General estimates are those too vague in location to allow specific comparison with survey estimates.

APPENDIX G
Phase, general estimate/ specific comparison, and summary statistic Author Source !60% of the trees were oaks, thus concentrations of oaks, to see how they were distributed and whether oak concentrations were favored by, or damaged by mixed-and high-severity fires. I reasoned that if oaks were damaged by mixedand high-severity fires, then concentrations of oaks would be found most often in areas that had exclusively low-severity fire over the reconstruction period, the 110-years prior to the surveys. However, Cocking et al. (2014) showed that highseverity fire actually promotes persistence and restoration of oaks in competition with conifers, at least on small scales. There is no statistical test because it is ambiguous what the sample units v www.esajournals.org are, some spatial autocorrelation exists, and what I really have is not a sample, but an estimate of the full population of all concentrations for the study area, which itself is representative, but not a statistical sample of the full SMC forest range.
In the northern Sierra, the oak concentrations covered 11,286 ha (8.5%) of the 133,482 ha unaffected area, but in the southern Sierra they covered only 1,368 ha (0.7%) of the 196,461 ha unaffected area. Only the northern Sierra is shown (Fig. H1) and analyzed further because it had sufficient sample size. As shown in Table 4, oaks were most abundant in the ponderosa pine-Douglas-fir phase, thus at lower elevations of the overall Sierran mixed-conifer forest.
Generally, concentrations were each the size of one polygon, thus about 750-800 ha, but occasionally twice that size and occasionally smaller due to clipping by the boundary of the unaffected area (Fig. H1).
Oak concentrations were not most abundant in areas with exclusively low-severity fire ( Fig. H2) over the reconstruction period. That should have occurred if oaks had been widely damaged by higher-severity fire. Instead, oak concentrations were found across all fire severities (Fig. H2), but were a little favored by mixed-severity fire, disfavored by low-severity fire and also disfavored overall by high-severity fire (Fig. H2). Lower overall occurrence in high-severity fire was because of much lower occurrence in recently burned areas, represented by chaparral (High-chaparral). Concentrations were slightly favored in the later successional stages after highseverity fire, represented by scattered trees (High-scattered trees) and by early-successional forest (High-forest area).
The relatively low level of oak concentrations in low-severity fires is consistent with lower vigor and lack of release of oaks in low-severity fires found by Cocking et al. (2014). The slightly positive association with mixed-severity fires is consistent with the observation that oak populations are multi-aged (Garrison et al. 2002, Cocking et al. 2012 but are favored by higherseverity fires (Cocking et al. 2014). Mixedseverity fires are more intense than low-severity fires, but leave more survivors than high-severity fires. The same may be true for high-severity fire areas with scattered trees and in early-successional forests. It is unclear whether much fewer oak concentrations in chaparral represents actual mortality of oaks or if they just had not yet resprouted sufficiently to be visually apparent to surveyors above the chaparral shrubs. The fact that they are favored in later successional stages suggests they were present, but not seen.
Concentrations of oaks occurred commonly after both mixed-and high-severity fires and were slightly favored after mixed-severity fires. Low-severity fires, in contrast, led to fewer concentrations of oaks, but did lead to some, likely because the mechanism of small highseverity fires identified by Cocking et al. (2014) was historically part of the low-severity fire regime. However, overall, low-severity fires were less effective per unit area at creating concentrations than were the historically dominant higherseverity fires, particularly the mixed-severity fires. The post-fire successional sequence after mixed-and high-severity fires is likely to naturally favor oaks early on, followed later by conifers that outgrow and overtop the oaks. Given the 281-year historical fire rotation in the northern Sierra Nevada (Table 7), there would typically be ample time for conifers to naturally recover after fire and overtop the oaks. This is simply the natural recovery of coniferous forest after mixed-and high-severity fire, not ''encroachment'' as labeled by Cocking et al. (2012). Fire exclusion could perhaps increase the hectares over which this process proceeds. Higherseverity fire is essential to maintain the SMC forest and its oaks.