Evaluating Drought Impact on Postfire Recovery of Chaparral Across Southern California

Abstract

Chaparral shrubs in southern California may be vulnerable to frequent fire and severe drought. Drought may diminish postfire recovery or worsen impact of short-interval fires. Field-based studies have not shown the extent and magnitude of drought effects on recovery, which may vary among chaparral types and climatic zones. We tracked regional patterns of shrub cover based on June-solstice Landsat Normalized Difference Vegetation Index series, compared between the periods 1984–1989 and 2014–2018. High spatial resolution ortho-imagery was used to map shrub cover in distributed sample plots, to empirically constrain the Landsat-based estimates of mature-stage lateral canopy recovery. We evaluated precipitation, climatic water deficit (CWD), and Palmer Drought Severity Index in summer and wet seasons preceding and following fire, as regional predictors of recovery in 982 locations between the Pacific Coast and inland deserts. Wet-season CWD was the strongest drought-metric predictor of recovery, contributing 34–43 % of explanatory power in multivariate regressions (R² =0.16–0.42). Limited recovery linked to drought was most prevalent in transmontane chamise chaparral; impacts were minor in montane areas, and in mixed and montane chaparral types. Elevation was correlated negatively to recovery of transmontane chamise; this may imply acute drought sensitivity in resprouts which predominate seedlings at higher elevations. Landsat Visible Atmospherically Resistant Index (sensitive to live-fuel moisture) was evaluated as a landscape-scale predictor of recovery and explained the greatest amount of variance in a multivariate regression (R² = 0.53). We find that drought severity was more closely related to recovery differences among twice-burned sites than was fire-return interval. Summarily, drought has a major role in long-term shrub cover reduction within xeric chaparral ecotones bounding the Mojave Desert and Colorado Desert, likely in tandem with other global change stressors.

Introduction

The persistence and health of native plant communities are global conservation priorities. Vegetation modulates water, energy, nutrient, and carbon fluxes in the Earth system, while supporting critical ecosystem processes (Costanza and others 1997; Baldocchi and others 2001). Human-caused land cover transformation, disturbance, and climatic change have caused widespread loss, fragmentation, and degradation of native vegetation (Vitousek and others 1997). Landscape and global change ecologies underscore the importance of system feedbacks, thresholds, and compound impacts stemming from multiple stressors (Turner 2010). Abnormal regimes of drought and fire may conjointly alter successional processes which determine composition, structure, and ecological functionality of plant communities (Batllori and others 2019). Meteorological phenomena associated with drought and stemming from climatic change are perhaps the most complex and least understood drivers of ecological change.

Unusually severe and protracted drought episodes have impacted the southwestern USA in recent decades (Williams and others 2013). In some areas, these droughts have increased the frequency and severity of fire due to prolonged warm seasons, elevated vapor pressure deficits, and excessive dry fuel production (Westerling and others 2006; Williams and others 2013; Seager and others 2015). This study focuses on the chaparral shrublands of southern California, where droughts occurred in the recent periods 1987–1990, 1996–1997, 2000–2002, 2007–2008, and 2012–2016 (MacDonald and others 2007; Griffin and Anchukaitis 2014; Mao and others 2015). The severity of the most recent (2012–2016) drought was unprecedented in California during the past 1200 years, and climate models predict increase of ‘mega-drought’ episodes during the twenty-first century (Griffin and Anchukaitis 2014; Yoon and others 2015). The evergreen sclerophyllous woody shrubs which comprise chaparral are adapted to seasonal (June-October) dry periods, and their persistence under highly variable inter-annual precipitation linked to El Niño-Southern Oscillation (ENSO) implies some level of drought tolerance (Kolb and Davis 1994; Cayan and others 1999). Episodes of water deficit that exceed thresholds of biological tolerance and adversely modify natural systems are described as ‘ecological drought’ (Crausbay and others 2017).

Thresholds and long-term consequences of ecological drought in chaparral are challenging to identify, especially where compounded by fire disturbance (Batllori and others 2019). Fire activity in southern California is known to vary according to precipitation receipt and wind activity (Moritz and others 2010) and may therefore shift in response to future climatic change. Fires that recur at brief intervals (< 5–15 years) may diminish chaparral recovery by reducing seedbank production (Lippitt and others 2013; Syphard and others 2019b). In this study, we evaluate the effects of drought as well as fire-return interval on levels of mature-stage (≥ 10 years) postfire recovery in chaparral.

Early research showed that shallow- and deep-rooted chaparral species can fail to develop new growth in drought years (Harvey and Mooney 1964). More severe impacts range from branch dieback to widespread mortality (Davis and others 2002; Venturas and others 2016). Drought impacts are linked to hydraulic failure of xylem tissue under evaporative stress, and to carbon storage deficit under impeded photosynthetic conditions (Oechel and Lawrence 1981; Diffenbaugh and others 2015). Multi-annual drought leads to gradual biochemical and soil-moisture deficits, whereas seasonal drought conditions determine thermal extrema and evaporative potentials (compare McDowell and others 2008; Serra-Diaz and others 2016). The timing of drought-induced die-offs suggests that chaparral is impacted more by drought intensity than duration, though both are potentially important (Pausas and others 2016; Venturas and others 2016).

Drought may impact chaparral differentially per functional types: obligate seeders depend on fire-cued seed banks for regeneration, are generally shallowly rooted, and typically occur in xeric sites; obligate resprouters regenerate vegetatively (typically from root-crowns), are deeply rooted, and predominate mesic sites; facultative seeders exhibit resprouting and seeding capacities, and can span wide climatic ranges (see Jacobsen and Pratt 2018 for species information). Chaparral species with relatively shallow root systems (including Adenostoma fasciculatum, Arctostaphylos, and Ceanothus spp.) exhibit prevalent mortality in drought (Paddock and others 2013; Venturas and others 2016).

When coupled with fire disturbance, drought can also result in mortality of deeply rooted resprouters associated with mesic sites (Pratt and others 2014). Both seedlings and resprouts of chaparral may continue to be sensitive to drought in early years of postfire recovery (Parsons and others 1981; Frazer and Davis 1988; Pratt and others 2014). Drought exposes fragile chaparral seedlings to thermal and evaporative stresses (Mills 1983; Frazer and Davis 1988). Resprouting shrubs are most vulnerable to dehydration during regrowth after fire (Jacobsen and others 2016; Pausas and others 2016). Carbohydrate deficit in resprouters can increase susceptibility to pathogens and diminish regrowth vigor (Radosevich and Conard 1980; Jacobsen and Pratt 2018).

Surprisingly, little research exists on the topic of drought as a control on postfire chaparral recovery (Jacobsen and Pratt 2018). A temporal analysis of drought impact on recovering chaparral may explain variations that are unresolved by fire history and other spatial predictor variables. Repeated burning at intervals of less than 10 years has been shown through field-based studies to diminish chaparral recovery and to promote invasion by exotic plants (Zedler and others 1983; Jacobsen and others 2004; Keeley and Brennan 2012). However, field-based studies may not have captured processes that operate across broader spatial domains (Meng and others 2014). Our review of subregional investigations suggests that fire-return interval explains recovery variation more directly in some areas than others (cp. Lippitt and others 2013; Meng and others 2014; Syphard and others 2019a, b). Chaparral areas that receive high mean annual precipitation tend to exhibit high shrub cover and low exotic plant cover, attributed to high fire severity covariate with biomass (Smith and others 2019). Previous large-scale studies of chaparral recovery have not analyzed the influence of drought in a temporal manner, in conjunction with fire-return interval. Estimation of vegetation change in a wide range of sites representing varied disturbance histories is required for such an analysis (compare Lippitt and others 2013).

Multi-spectral image data from satellite and aerial systems are useful to characterize large-scale patterns of recovery after fire in Mediterranean-type shrublands (Viedma and others 1997; Shoshany 2000; Díaz-Delgado and Pons 2001; Díaz-Delgado and others 2002; Riaño and others 2002; Wittenberg and others 2007; Minchella and others 2009; Gouveia and others 2010; Solans Vila and Barbosa 2010; Vicente-Serrano and others 2011; Lanorte and others 2014; Meng and others 2014; Petropoulos and others 2014; Fernandez-Manso and others 2016). Archived imagery from Landsat (4, 5, 7, and 8) is particularly useful to track inter-annual change in heterogeneous chaparral, owing to 30-m spatial resolution, 16-day revisit time, and a record spanning from 1984 to present for southern California. Signals of vegetation change are enhanced by Landsat surface reflectance products, which reduce atmospheric and solar irradiance artifacts (Masek and others 2006). Multi-annual trends of Landsat-derived Normalized Difference Vegetation Index (NDVI) capture trajectories of change in fractional shrub cover (FSC) associated with postfire recovery (Hope and others 2007; Röder and others 2008). High spatial resolution aerial imagery supports empirical calibration of Landsat NDVI data to units of FSC for the purpose of ecological change analysis (Fraser and others 2011).

Objectives and Scope

The major objective of this study is to evaluate impacts of drought on postfire recovery of chaparral across the southern California coastal eco-region (116° 25′−119° 25′ W; 32° 30′−34° 40′ N). We leveraged the regional coverage and multi-decadal time span of Landsat image series, in order to track recovery in the greatest possible number of sites that burned since the earliest-available image from 1984. Fires within this study period (1984–2018) correspond to unique drought and non-drought episodes, and a variety of fire-return intervals in areas that burned twice. The study areas are in chamise, mixed, and montane chaparral community types and exhibit varied aridity, terrain, and recovery patterns. We conducted spatial-temporal evaluations of drought effects on recovery, based on fire history, terrain, mean climate, and three different metrics of drought intensity in seasons preceding and following the fire events. This study addresses three major research objectives according to following questions:

1. Does postfire recovery at mature stage vary significantly due to drought in summer and wet seasons preceding or following singular fire events?

2. Is drought a significant regional predictor of postfire recovery among sites that burned twice, at intervals ranging from 1 to 23 years?

3. Which geospatial terrain, vegetation, soil, and aridity variables best predict landscape-scale variation in drought-fire impact?

Methods

Overview

Figure 1 illustrates the methodological framework of this study. Historical fire perimeters obtained from California Fire Resource and Protection (FRAP) were overlaid upon maps of plant community type, in order to select potential study areas consisting of chaparral that burned after the first year of available Landsat data (1984). Refined maps of burned area in the study region were then constructed using Landsat imagery in order to the reliability of sample locations. Detailed maps of shrub cover derived from aerial ortho-imagery were used to generate linear regressions between FSC and Landsat NDVI in widely distributed sample plots. Landsat NDVI values thereby related to FSC were used to derive spatially extensive prefire (1984–1989) and postfire (2014–2018) shrub cover estimates, which were compared arithmetically in order to map variations in recovery at mature (10–33 years) postfire stages. Postfire recovery variations were evaluated based on random point samples stratified within three climatic zones, using linear regressions involving multiple spatial and temporal variables. As a temporal analysis, drought conditions were indexed according to wet and summer seasons of the 2 years preceding and 3 years following each fire event. A group of neighboring sites that exhibited low recovery due to drought were used for the spatial component of this study, in which we evaluated landscape-scale variations in terrain and aridity as modulators of drought impact. The results were synthesized to evaluate the relative importance of drought versus fire-return interval in postfire chaparral recovery across southern California.

Figure 1.

Flowchart depicting how key data sets are used in order to address the spatial and temporal components of drought impact upon postfire chaparral recovery. Lighter shaded boxes represent source data, while darker ones indicate data processing and analytical steps.

Landsat Imagery

The Landsat images utilized in this study were processed to directional surface reflectance by the Land Surface Reflectance Climate Data Record (LSRCDR) of the United States Geologic Survey (USGS) (espa.cr.usgs.gov) (Masek and others 2006; Vermote and others 2016). These Landsat images cover the chaparral areas of southern California and coincide with Worldwide Reference System (WRS) path-rows 40–36, 40–37, 41–36, and 42–36. Shrub cover dynamics were evaluated using series of annual Landsat observations between 10 June and 14 July, a period of high solar illumination and senescence in under-storey herbs. These near-anniversary image series covered two 5-year periods which we defined as prefire (1984–1989) and postfire (2014–2018). A separate collection of Landsat images from various years and seasons were used to map burned areas. All Landsat images were masked for clouds and cloud-shadows based on Fmask quality-assurance maps provided by LSRCDR (Zhu and Woodcock 2012). Landsat pixels flagged by Fmask were supplanted with cloud-free pixels from images of proximal acquisition dates.

Fire History, Site Selection, and Stratification

Study areas were selected throughout southern California based on shrub community type and fire history criteria. Community types included montane, mixed, and chamise chaparral. Chamise is dominated by Adenostoma fasciculatum (in excess of 60% relative cover); mixed chaparral is co-dominated by A. fasciculatum, Ceanothus spp., Arctostaphylos spp., and Quercus berberidifolia; montane chaparral is composed largely of Arctostaphylos patula, C. cordulatus, C. integerrimus, C. velutinus, Cercocarpus ledifolius, and Chrysolepis sempervirens. These community types are most prevalent in southern California according to CALYEG maps developed in 1980, prior to the period of our study (Matyas and Parker 1980; fs.usda.gov). More recent FVEG maps show the same chaparral community types as more extensive in some areas (fs.usda.gov). Therefore, we utilized a spatial union of the FVEG and CALVEG chaparral classes, affording priority to CALVEG maps in cases of conflict.

All areas that burned more recently than 1974—and that coincide with the chaparral community types of interest—were identified using fire perimeter data of FRAP (calfire.gov). Areas that burned in the periods 1975–1984 and 2009–2018 were then excluded from our study (with exceptions noted below) because such areas may contain shrubs that were immature during prefire (1984–1989) or postfire (2014–2018) assessment periods. This criterion for excluding immature sites is based on findings that postfire recovery of chaparral is a 10-year process (McMichael and others 2004; Hope and others 2007; Storey and others 2016). We included matured chaparral areas that burned in the period 1985–2008 as prospective study areas. We selected the maximum possible extent of study area that burned in this 24-year period.

Fire history since 1984 in the prospective study areas was reconstructed based on Landsat Normalized Bum Ratio (NBR) data (Key and Benson 1999), for the reason that FRAP fire perimeters are shown to contain inaccuracies (Syphard and Keeley 2017). We derived prefire-to-posifire NBR difference (dNBR) and relative-difference (RdNBR) bum severity indices, using formulas given by Miller and Thode (2007). These burn severity indices were used conjointly to identify burned areas, which were then classified using value-exceedance criteria that we identified through iterative pixel queries and inspection of binary map results. In this procedure, we classified a total of 155 Landsat images, many of which recorded several fire scars.

We excluded human-built features from the study areas by manual delineation based on 0.6-m spatial resolution, color-infrared ortho-imagery captured in 2016 by the National Agriculture Imagery Program (earthexplorer.usgs.gov). After masking out built features and immature chaparral sites, we obtained a set of 240 contiguous areas (hereafter: stands) which are internally uniform with respect to shrub community type and fire history during the study period. These stands range from 1 to 557 km² (19 ± 44 km², $\overline{x}$ ± SD). Approximately 92% of the collective study area (173 stands) burned once, whereas eight percent of it (67 stands) burned twice at intervals of 1–23 years (Figure 2).

Figure 2.

Map of burned chaparral stands included in this study, displayed per number of burns in the period 1985–2008. Montane zone boundaries coincide with 500 mm mean annual precipitation contours, which are based on a 30-year gridded climate normal (prism.oregonstate.edu) displayed by a min-max contrast stretch (values exceed 800 mm per year in some locations).

Stands were organized into three custom-defined climatic zones based on geographic position and mean annual precipitation (prism.oregonstate.edu). We defined the montane zone as high-elevation areas that receive at least 500 mm y⁻¹ of precipitation (Figure 2). The cismontane zone receives less than 500 mm yr⁻¹ of precipitation and spans from the Pacific Coast to the montane zone. The transmontane zone also receives less than 500 mm y⁻¹ in precipitation, but is located on the leeward (rain shadow) sides of the Transverse and Peninsular Ranges, and includes the xeric chaparral ecotone which intergrades into desert vegetation.

Estimation of Vegetation Change

We evaluated recovery on the basis of change in shrub cover between the prefire period and the end of the postfire period, which allowed 10 or more years of recovery in the study areas. Estimates of recovery were based on temporal trajectories of Landsat NDVI, as illustrated in Figure 3. Trajectories derived in five-year periods are less sensitive to the effects of variable precipitation on annual leaf production than are single-year values (Riaño and others 2002). Postfire trajectories (2014–2018) were characterized by linear best-fit functions, to capture potential increase of shrub cover at sites that burned in 2004–2008. Postfire NDVI values were extracted from termini of the best-fit trajectories, which coincide with 21 June 2018. Prefire values were based on mean NDVI from the prefire period (1984–1989) in which vegetation abundance was assumed to be stable in the study areas. Several areas of interest had burned after 2016, or in the early years of the prefire period. Accordingly, we excluded data points of years following these fire events from the NDVI trajectories. For areas that burned in 1985, 1986, 1987, or 1988, we included only the NDVI values that precede these years into the prefire mean calculations.

Figure 3.

Examples of Normalized Difference Vegetation Index (NDVI) series based on pixels from two exemplary burned sites (A and B) in eastern San Diego County. Metrics applied to the prefire 1984–1988 (5-year means) and postfire (2014–2018) periods (linear best-fit functions) are shown along with fire event timings at each site. The NDVI series represented in this figure were normalized for phenological artifacts using unburned control sites.

Detailed maps of shrub cover were generated from the 2016 ortho-imagery in order to derive estimates of FSC from Landsat NDVI. Chaparral within the study region was delineated into 49 distinct polygonal segments, defined by visual interpretation of the CALVEG maps and ortho-imagery as having internally consistent shrub community types and canopy density patterns. At least one calibration plot of 3-km × 3-km dimensions was placed in each segment. Shrub cover maps of the plots were generated using decision-tree, threshold-based classifiers which we developed using ERDAS IMAGINE® image processing software. The classifications were based on spectral image transforms including NDVI, intensity (mean of visible bands), and red-to-intensity ratio (diagnostic of exposed soils). We selected and iteratively refined thresholds based on pixel query and inspection of trial output results. This mapping approach yields 87–95% accuracy when applied to chaparral (Storey and others 2016). We present an example of the ortho-image data (Figure 4A, B) used to derive a three-class land cover map (Figure 4C).

Figure 4.

Illustration of fine-scale aerial ortho-image data used to derive empirical relations between fractional shrub cover and Landsat NDVI. A color-infrared ortho-image of 0.6-m spatial resolution displayed in true color, B false-color composite of spectral transformations used in classification, and C decision-tree classification used to distinguish green shrub cover from exposed earth and non-shrub vegetation.

Areal fractions of shrub cover were tabulated using 90-m × 90-m sampling grids which we imposed on the calibration plots. We aggregated by averaging the 3 × 3 Landsat NDVI pixel arrays within these grid elements (subplots), to reduce the spurious effects of geometric misregistration among images from the time series (compare Hope and others 2007). Thirty grid elements were sampled randomly from each plot, at 150-m minimum spacing to prevent selection of adjacent grid elements. Grid data samples were used to relate Landsat NDVI and FSC in each of the 49 segments, using slope and intercept values derived from ordinary least-squares (OLS) regressions. Coefficient of determination (R²) values indicating goodness-of-fit in the plot-based Landsat NDVI-FSC regressions ranged from 0.56 to 0.95, with a mean value of 0.83 and standard deviation of 0.13. Arithmetic differencing of prefire versus postfire FSC produced a regional map of absolute percent change in fractional shrub cover (hereafter: dFSC).

Geospatial Data

Data sets described in Table 1 were used as potential explanatory variables. Previous work suggests that terrain can modulate drought impact according to sun exposure, soil properties, and moisture accumulation (Meentemeyer and Moody 2002). We evaluated seven edaphic variables from the State Soil Geographic (STATSGO) including textural, hydrologic, and thickness properties (nrcs.usda.gov). A 30-m digital elevation model (ned.usgs.gov) was used to derive topographic slope, aspect, solar irradiance from spring to autumn equinoxes, and curvature variables which are potential controls on plant establishment (Davis and Goetz 1990). As riparian chaparral may be affected differently by drought (Coates and others 2015), we inferred riparian zones based on 30-m proximity to hydrologic channels (data.cnra.ca.gov).

Table 1.

Geospatial Data Sets Used to Explain Postfire Chaparral Recovery

Data set	Variable(s)	Data origin site	Spatial resolution	Temporal properties
CALVEG	Plant community type	fs.usda.gov	1.6–3.2 km	Produced in 1977–1979
FVEG	Plant community type	fs.usda.gov	250 m	Revised in 2015
State Soil Geographic	Hydrologically relevant soil characteristics (7)	nrcs.usda.gov	1 km	Produced in 1997
30-year mean annual climate	Precipitation, temperature, vapor pressure deficit	prism.oregonstate.edu	800 m	Based upon monthly means (1981–2011)
Terrain	Elevation, slope, aspect, curvature, surface irradiance	ned.usgs.gov	30 m	–
Riparian zones	30-m proximity to streams	data.cnra.ca.gov	30 m	–
Visible Atmospherically Resistant Index	Sensitive to live-fuel moisture, derived from Landsat 5 images	espa.cr.usgs.gov	30 m	Single observations from 13 June 1986
Monthly precipitation	Precipitation used in the Livneh and others (2013) model	data.nodc.noaa.gov	6 km	Monthly estimates from 1983 to 2011 included
Palmer Drought Severity Index	Water deficit based on soils, temperature, precipitation	wrcc.dri.edu	4 km	Monthly estimates from 1983 to 2011
Climatic water deficit	Total evaporative demand relative to soil moisture	Climate.calcommons.org	270 m	Monthly estimates from 1983 to 2010

Mean annual precipitation, temperature, and vapor pressure deficit were evaluated based on Parameter-elevation Regressions on Independent Slopes Model (PRISM) data gridded at 800-m resolution (Daly and others 1994; prism.oregonstate.edu). Landsat images acquired on 13 June 1986 were used to derive Visible Atmospherically Resistant Index (VARI) (Gitelson and others 2002), which is sensitive to live-fuel moisture (LFM) in chaparral (Stow and others 2005; Roberts and others 2006; Peterson and others 2008). The VARI is based on visible (red, green, and blue) spectral bands and therefore is not a direct correlate of NDVI. We based VARI on Landsat images from early June because transpiration is expected to be high at this time, while the summer ‘dry-down’ phase is just beginning (compare Coates and others 2015). The year 1986 was a near-average precipitation year and preceded fire in the relevant study areas. Because VARI and plant cover are potentially correlated (compare Gitelson and others 2002; Stow and others 2005), we normalized for shrub cover by calculating the ratio of VARI to prefire FSC. We used VARI as a proxy for spatial (but not temporal) variation in LFM.

Drought Metrics

Spatial-temporal drought metrics were based upon seasonal aggregates of monthly Palmer Drought Severity Index (PDSI) (Palmer 1965), monthly Climatic Water Deficit (CWD) (Stephenson 1998; Flint and others 2013), and monthly precipitation (PRCP) (Livneh and others 2013). The PRCP data gridded at 6-km resolution were provided by National Centers for Environmental Information (NCEI) (data.nodc.noaa.gov/livnehmodel). The PDSI index integrates soil properties with precipitation and temperature in order to simulate moisture anomalies associated with drought, and is closely related to potential evapotranspiration (Szép and others 2005). We obtained monthly PDSI data (1983–2011) gridded at 4-km resolution from WestWide Drought Tracker (Abatzoglou and others 2017; wrcc.dri.edu). The CWD represents the difference of potential and actual evapotranspiration, is closely related to plant-available moisture (Flint and others 2013), and may provide the most direct indication of drought stress in chaparral. The 270-m resolution monthly CWD data obtained through USGS are available from 1983 but span only to 2010 (Flint and others 2013; climate.calcommons.org). The monthly PDSI, PRCP, and CWD indices were aggregated per wet seasons (November-May), in which hydrologic pulses enter the chaparral system and regulate annual foliar production (compare Christensen and Muller 1975). Separately, we also aggregated the drought indices per summer ‘dry-down’ seasons (June-August) when shrub foliage and air temperatures are high, but precipitation is minimal (compare Coates and others 2015). Absolute seasonal PRCP totals (mm) were evaluated separately from relative-percentage values, calculated with respect to 28-year means (1983–2011).

Analytical Procedures

We evaluated temporal as well as spatial components of recovery variation due to drought. The temporal component is based on the seasonal PDSI, CWD, and PRCP drought metrics. We defined S₀ as the wet or summer season during which a fire occurs, or as the first season after fire (mainly for wet-season drought metrics, in the many cases that fires occurred in autumn). Drought-metric values corresponding to S₀ are defined herein as PDSI-S₀, CWD-S₀, and PRCP-S₀. We defined seasonal time steps in relation to S₀, including one annual cycle after (S₊₁), two cycles after (S₊₂), one cycle prior (S₋₁), and two cycles prior to S₀ (S₋₂). Seasonal drought metrics were derived for prefire years to evaluate the hypothesis that antecedent drought can diminish recovery. Drought-metric seasons were associated with twice-burned stands according to the fire date of greater drought intensity, based on seasonal PDSI values. Each seasonal time step based on each drought index was evaluated as an independent variable.

Random vector points and rectangular polygon plots were used to sample the seasonal drought-metric grids, and to evaluate these as explanatory variables of recovery (dFSC) using multivariate linear regressions in IBM SPSS®. We collected five separate sample sets from the dFSC map that serve unique purposes: (1) stand-aggregate values (n = 173), used to evaluate frequency-probability of recovery across the study region; (2) aggregates of 30-m pixels (7 × 7 pixel-kernel means) distributed randomly throughout the single-burn areas at mean density of one per 10 km² (n =491), used in regional evaluation of drought effects; (3) paired sample plots of 0.5-km × 0.5-km dimensions (n =416), used to evaluate recovery in the twice-burned stands relative to nearby single-burn (control) stands, as a function fire-return interval and drought condition; (4) single-pixel samples distributed randomly across a selected set of 10 drought-impacted stands (n =697, 1 per km² mean density), to explain landscape-scale spatial variation in drought impact; and (5) a set of 75 plots (0.5-km × 0.5-km) sampled from a set of 12 stands with highly varied recoveries. These 12 stands were on gradual terrain and close in proximity, which provided control for climatic setting.

The 416 paired-plot samples were placed in the closest possible proximity (2.2 ± 1.3 km, $\overline{x}$ ± SD) and stratified according to terrain aspect and plant community type. This sample arrangement provided control for many spatial variables that can influence recovery (compare Meng and others 2014), although perfect control for terrain slope and variations internal to the 0.25 km² plots was not possible due to limited extent of potential sampling area. The 491 sample points were stratified per climatic zones (Figure 2), and we added climatic variables (mean annual precipitation and temperature) to the OLS regressions to account for climatic gradients within the zones.

Results

Regional Patterns of Recovery at the Stand-Aggregate Level

Mature-stage recovery levels varied substantially between and within the stands. Approximately 65% of the stands recovered to within 5% of mean prefire shrub cover. Declines in mean shrub cover of 10–15% were observed in about 32% of the stands, whereas only about 3% of the stands declined in shrub cover by more than 15%. The stands collectively exhibited strong recovery in most areas and moderate degradation in some areas. Total reduction of shrub cover that would be consistent with total ‘type conversion’ was not observed in any of the stands. Regional and within-stand spatial variations in recovery are discussed further below. Figure 5 illustrates how inter-annual trends in PRCP, CWD, and PDSI are related to area burned and mean stand recovery.

Figure 5.

A Areal coverage of single-bum study areas within the montane, cismontane, and transmontane zones of southern California chaparral; B mean percent shrub cover change in stands aggregated by burn year; average wet-season (Nov. of prior year to May of labeled year) drought-metric values within the stands, including C Palmer Drought Severity Index (PSDI), D percent of the mean precipitation of the period 1983–2011, and E climatic water deficit. Green bars denote non-drought, orange bars denote moderate drought, and red bars suggest extreme drought based on each metric. Drought severity based on PDSI values in years of the time series not represented here includes 1983 (5.3), 1984 (− 0.1), 2009 (− 3.6), 2010 (0.3), and 2011 (2.85).

Based on the criterion of PDSI less than − 1.8 (Figure 5C) coupled with below-average wet-season precipitation (Figure 5D), the major drought periods lasting more than 1 year included 1989–1991, 1996–1997, 2000–2002, and 2007–2008. The three climatic zones exhibit unique fire chronologies and potentially different impacts of drought upon recovery. The montane zone contains the largest extent of burned study area. The most active fire years in each climatic zone were 2002, 2003, 2006, and 2007 (Figure 5A). Limited postfire recoveries are noted in cismontane sites that burned in the drought-proximal years 1990, 2004, and 2008 (Figure 5B). The montane stands exhibit moderate degradation (that is, decline in FSC of more than 5%), most notably in the non-drought years 1994 and 1995, but otherwise recovered well. The transmontane stands show the most prevalent and severe declines in FSC, in association with both drought and non-drought periods. A one-way analysis of variance (ANOVA) indicates a significant difference in recovery (dFSC) between regional groupings of stands that burned in drought versus non-drought years (p = 0.01; F = 6.5; F_crit = 3.86). This ANOVA does not address the potential effect of drought in years before or after fire, which is the focus of subsequent analysis based on seasonal drought metrics.

Objective 1: Evaluating Region Drought Impact in Single-Burned Sites

We evaluated seasonal drought metrics as recovery predictors based on the 491 random sample locations. Table 2 shows results from separate regression tests (refer to R² values) that we applied for each of the drought-metric types, which include five annual cycles (S₋₂, S₋₁, S₀, S₊₂, and S₊₂) of summer and wet seasons. Table 2 also shows fractions of explained variance attributable to the seasonal drought metrics; these were derived by summing the standardized β coefficients of the significant variables and computing their respective fractions (Table 2). Regressions based on data from each climate zone showed that wet-season PDSI was not a significant predictor of recovery, although summer PDSI was significant in several tests (p <0.05) (Table 2). The absolute seasonal PRCP values (mm) were not significant, whereas relative-percentage wet-season PRCP metrics were significant for each climatic zone. The CWD metrics based on summer and wet seasons were significant for the cismontane and transmontane zones (Table 2). Mean annual precipitation and temperature were also significant predictors of recovery but exhibit variable effect sizes among the tests, and were omitted from Table 2 for brevity.

Table 2.

Results of Least-Squares Linear Regressions

Climate zone	Variable	p	t	β Coef.	Fraction of explained variance	Slope (α)	SE	Collinearity tolerance	VIF	Model Adj. R2
Montane	Wet-season PRCP-S−2	0.03*	− 2.25	− 0.16	0.14	− 1.20	0.53	0.60	1.66	0.26
	Summer PDSI-S+2	0.01*	2.51	0.20	0.26	0.01	0.00	0.49	2.02	0.27
Cismontane	Wet-season PRCP-S0	0.04*	2.04	0.18	0.20	3.81	1.87	0.73	1.36	0.10
	Summer CWD-S−0	0.04*	2.04	0.23	0.26	0.30	0.15	0.43	2.32	0.15
	Wet-season CWD-S−1	0.00**	3.00	0.32	0.23	0.17	0.06	0.48	2.09	0.16
	Wet-season CWD-S−2	0.02*	2.42	0.28	0.20	0.13	0.05	0.41	2.44
Transmontane	Wet-season PRCP-S−1	0.00**	− 3.60	− 0.42	0.30	− 7.02	1.95	0.48	2.09	0.35
	Wet-season PRCP-S+1	0.03*	2.20	0.25	0.18	3.53	1.60	0.51	1.96
	Summer PDSI-S−2	0.05*	2.00	0.27	0.14	0.01	0.00	0.33	3.06	0.39
	Summer PDSI-S−1	0.00***	− 3.89	− 0.35	0.18	− 0.01	0.00	0.75	1.34
	Summer PDSI-S+1	0.01**	2.86	0.33	0.18	0.01	0.00	0.45	2.25
	Summer PDSI-S+2	0.04*	2.11	0.26	0.14	0.01	0.00	0.41	2.47
	Summer CWD-S−2	0.05*	1.98	0.24	0.27	0.14	0.07	0.45	2.25	0.35
	Wet-season CWD-S−1	0.00**	4.55	0.56	0.34	0.24	0.05	0.39	2.58	0.41
	Wet-season CWD-S+1	0.01**	− 2.84	− 0.34	0.21	− 0.17	0.06	0.40	2.50

These were used to evaluate seasonal precipitation (PRCP), climatic water deficit (CWD), and Palmer Drought Severity Index (PDSI) drought metrics as explanatory variables of postfire recovery (dFSC). Only significant metrics and seasons are shown. Mean annual precipitation and temperature are also significant in each test but were excluded from this table. In bold font are Adjusted R² values that are highest for each climate zone, including montane (n = 223), cismontane (n = 161), and transmontane (n = 107).

^*p < 0.050;

^**p < 0.010;

^***p < 0.001.

Statistical tests relating to the montane zone were less reliable than those relating to the other climatic zones, which contain substantially greater variations in recovery (Figure 5B). Only wet-season PRCP-S₋₂ and summer PDSI-S₊₂ were statistically significant for the montane zone (Table 2). Both PRCP-S₀ and CWD-S₀ were significant for the cismontane zone, which may imply drought impact was important in the first growth season after fire. However, cumulative water deficits in wet seasons preceding fire (CWD-S₋₁ and CWD-S₋₂) exhibited the strongest R² with respect to recovery for the cismontane zone. For the transmontane zone, the regression model that includes CWD-S₋₁ and CWD-S₊₁ exhibited the highest R² value (Table 2). Summer CWD-S₋₂ and PDSI in most seasons (S₋₂, S₋₁, S₊₁, and S₊₂) were also significant for the transmontane zone, but yielded low R² values. Wet-season CWD exhibited the strongest overall relationship to recovery, and contributed 34–43% of the explanatory power in these multivariate tests, when β coefficient fractions were tabulated cumulatively across wet seasons.

Objective 2: Evaluating Regional Impacts of Fire-Return Interval Versus Drought

We first evaluated differences in recovery with respect to fire-return interval and number of burns, irrespective of drought. A comparison of the single- and multiple-bum plot sets implies no significant difference between their sample mean values (paired, one-tailed t test; p = 0.11; F_sig = 0.15). A subset of these data including only those plots that declined by more than 5% in FSC, exhibits a significant difference between the single- and multiple-burn samples (p =0.01; F_sig = 0.001). The most useful metric of recovery in this application is the difference in dFSC between counterpart sample plots. This metric showed no significant association with fire-return interval (p = 0.79).

Drought effects were evaluated based on metric values from individual plots (i.e., absolute values), and separately based on relative differences in drought-metric values between the counterpart plots. Results of statistical tests which evaluate the absolute and relative-difference drought metrics (with respect to relative-difference dFSC) are given in Table 3. Relative-difference seasonal PRCP metrics (S₋₂ and S₊₁) were significant in explaining recovery differences among the paired plots. Absolute and relative-difference PDSI based on summer and wet seasons were also significant predictors of relative-difference dFSC, and yielded R² values higher than the PRCP metrics. The highest R² value (0.09) resulted from prefire summer CWD (S₋₂). Fire-return interval was not a significant variable when included in the drought impact evaluations.

Table 3.

Results of Landscape-Scale Regression Tests

Variable	p	t	β Coef.	Slope (α)	SE	Collinearity tolerance	VIF	Model Adj. R2
Difference in wet-season PRCP-S−2	0.01**	2.62	0.22	2.33	0.89	0.64	1.58	0.05
Difference in wet-season PRCP-S+1	0.04*	2.05	0.20	1.78	0.87	0.49	2.06
Summer PDSI-S−1	0.04*	2.02	0.12	0.00	0.00	0.69	1.44	0.06
Summer PDSI-S0	0.00**	3.02	0.15	0.00	0.00	0.89	1.12
Summer PDSI-S+2	0.00***	3.74	0.18	0.00	0.00	0.96	1.04
Difference in summer PDSI-S−2	0.00***	4.05	0.32	0.00	0.00	0.73	1.38	0.07
Summer CWD-S−2	0.00**	3.29	0.23	0.13	0.04	0.44	2.29	0.09
Wet-season CWD-S+2	0.00***	− 3.54	− 0.19	− 0.06	0.02	0.78	1.28	0.05

These were used to evaluate differences in recovery (n = 208) between 416 paired single and multiple-bum sites, as a function of fire-retum interval and drought metrics based on wet and summer season precipitation (PRCP), Palmer Drought Severity Index (PDSI), and climatic water deficit (CWD).

^*p < 0.050;

^**p < 0.010;

^***p < 0.001.

Objective 3: Evaluating Landscape-Scale Variations in Drought Impact

Ten stands were used to evaluate landscape-scale spatial predictors of recovery variation due to drought impact (Figure 6). These stands were selected by the following criteria: (1) the most recent fires in these stands were preceded or followed by a drought year; and (2) declines of shrub cover greater than 15% were observed in heterogeneous patterns across substantial portions of these stands (Figure 6). Most of these selected stands are in the transmontane zone (cp. Figures 2, 6F).

Figure 6.

Spatial patterns of postfire recovery (based on Landsat NDVI trajectories) within selected, drought-impacted stands. Frame f indicates locations of all study areas relative to selected drought-impacted stands: transmontane xeric sites include A, B, and D; site A is cismontane; site e is high-elevation montane. Half of the stands used in temporal analysis (F) were well recovered.

Exploratory OLS regression analysis led us to eliminate nonsignificant variables including solar irradiance, terrain (aspect, slope, and curvature), presence of riparian features, and most of the STATSGO edaphic variables (available water capacity, thickness, organic matter, clay content, conductivity, and hydrologic group). Mean annual precipitation and vapor pressure deficit were eliminated due to collinearity with elevation.

Regression outputs based on the significant landscape variables are given in Table 4. Chaparral community type and soil permeability were positively related to recovery, with relatively small effect sizes according to β coefficient values (0.12 and 0.14, respectively). The positive β coefficient of the community type variable indicated greater recovery in montane (Stratum 3) and mixed (Stratum 2) as compared to chamise (Stratum 1) chaparral. Elevation exhibited a strong and negative relationship to recovery (β = − 0.35). The normalized VARI exhibited the strongest relationship to recovery (β = − 0.38). Model fit (R² = 0.53) suggests that the spatial variables in Table 4 explain about 53% of the variance in recovery within this selection of stands.

Table 4.

Results from Landscape-Scale Spatial Analysis

Variable	p	t	β Coef.	Slope (α)	Standard Error	Collinearity tolerance	VIF
Elevation	< 0.001	− 6.00	−0.35	−0.01	0.00	0.60	1.66
Plant community	< 0.001	10.27	0.12	2.01	0.70	0.78	1.29
Soil Permeability	< 0.001	−7.85	0.14	−0.45	0.26	0.48	2.09
Visible Atmospherically Resistant Index	< 0.001	18.45	−0.38	31.87	1.73	0.88	1.14

An ordinary least-squares regression test used to significant explanatory variables of postfire recovery samples (n = 697) at ten drought-impacted chaparral stands in southern California. Adj. R² = 0.53; SE = 2.58; df₁ = 4; F_sig < 0.001.

A separate group of 12 stands (also depicted in Figure 6F) was used to evaluate temporal effects of drought at a relatively localized scale. These stands burned in years (1990, 1995, 1999, 2002, 2003, or 2004) that represent periods with a variety of drought as well as non-drought conditions (Figure 5). These stands exhibited varied recovery levels but relatively uniform terrain and internal recovery patterns. The close proximity and similar elevations of these stands provided some control for mean climate and other environmental variables. Wet-season CWD was used in this analytical component owing to its high significance in other tests (Table 2). Unlike the regional analysis, the most recent fire among these stands occurred in 2004, which permitted evaluation of recovery metrics over five seasons following fire (2005–2010).

As shown in Table 5, CWD in several prefire wet seasons (S₋₃, S₋₂) and the first postfire wet season (S₀) were significant predictors of recovery, and represent substantial fractions of the explained variance (0.19–0.31). The strongest predictor of recovery was CWD-S₀ based on its low variable inflation and individual R² value of 0.13. Long-term (28-year) mean wet-season CWD also exhibited a significant, positive relation to recovery (Table 5). CWD-S₋₁ yielded a high individual R² value, but was insignificant in the multivariate test due to high variable inflation. Drought metrics based on CWD in seasons after S₀ were insignificant (Table 5). The significant drought variables in this analysis together explained 42% of the variation in recovery, based on this small mosaic of stands that were evidently more impacted by drought than others in the montane and cismontane zones.

Table 5.

Regional Analysis of Drought Impact on Recovery

Variable	p	t	β Coef.	Slope (α)	SB	Collinearity tolerance	VIF	Individual R2	Fraction of variance explained
Mean CWD	0.000	4.75	0.89	1.10	0.23	0.22	4.55	0.06	0.24
CWD-S−3	0.002	−3.24	−0.96	−1.01	0.31	0.09	11.31	0.00	0.26
CWD-S−2	0.001	−3.52	−1.17	−0.44	0.12	0.07	14.15	0.04	0.31
CWD-S−1	0.111	−1.62	−0.51	−0.21	0.13	0.08	12.56	0.09	0.00
CWD-S0	0.002	−3.14	−0.73	−0.35	0.11	0.14	6.92	0.13	0.19
CWD-S+1	0.192	1.32	0.58	0.38	0.29	0.04	24.90	0.11	0.00
CWD-S+2	0.619	0.50	0.14	0.06	0.11	0.10	9.94	0.00	0.00
CWD-S+3	0.083	1.76	0.95	0.37	0.21	0.03	37.06	0.11	0.00
CWD-S+4	0.25	1.17	1.55	0.44	0.16	0.15	17.50	0.00	0.00
CWD-S+5	0.78	−0.27	−0.41	−0.10	0.33	0.27	14.63	0.00	0.00

Results are from univariate (individual-variable) and multivariate linear regressions used to evaluate postfire recovery (change in fractional shrub cover) relative to climatic water deficit (CWD) in wet-season cycles relative to time-of-fire at twelve stands (sample plot n =75), and long-term (27-year) wet-season mean CWD values. Adj. R² = 0.42; Standard Error (SE) = 5.4; df₁ = 8; F_sig < 0.001.

Discussion

We utilized a 35-year series of multi-spectral Landsat surface reflectance images in order to characterize and explain variations in postfire recovery of southern Californian chaparral in relation to drought intensity over the period 1983–2011. The major goals of this study were: (1) to evaluate the effect of seasonal drought intensity on chaparral recovery among single-burned stands in montane, cismontane, and transmontane zones; (2) to comparatively evaluate the impacts of fire-return interval and drought; and (3) to evaluate a suite of geographic variables as potential predictors of landscape-scale recovery variation within areas impacted by drought.

The evidence of coupled drought-fire impacts produced in this study corroborates previous findings (Parsons and others 1981; Frazer and Davis 1988; Pratt and others 2014). Key insights gained from this study concern the magnitude, spatial variation, and relevant timing(s) of drought impact on chaparral recovery. Our results suggest that impacts of drought on recovery in recent decades were most severe in transmontane chaparral, associated with an ecotone which separates the regional arid (desert) and semi-arid (Mediterranean-type) eco-climatic zones. Paddock and others (2013) also report chaparral mortality in this ecotone due to drought, in absence of fire. We find that drought-fire impacts are most acute in higher-elevation transmontane sites composed of chamise chaparral. This result may reflect sensitivity in young A. fasciculatum resprouts, which predominate over seedlings at higher elevations (Keeley and Soderstrom 1986). An alternative explanation is that high-elevation congener plants are less adapted to thermal anomalies during drought (McCullough and others 2016). Drought also enhances freeze-cavitation susceptibility at high elevation due to moisture deficit (Davis and others 2007). Paradoxically, we find that recoveries are robust and drought effects were lesser in the montane climatic zone, which consists mainly of mixed and montane chaparral community types. Growth season precipitation antecedent to fire was a significant predictor of recovery within the montane zone. This finding may imply vulnerability to protracted drought, due to cumulative deficits of plant carbohydrates or deep soil moisture (compare Radosevich and Conard 1980; Oechel and Lawrence 1981). Based on a local mosaic of single-bum stands that exhibited highly variable recoveries, we find the clearest evidence of protracted drought impact (R² = 0.42) in wet seasons preceding and immediately following fire.

Climatic water deficit (CWD) is the strongest drought-metric predictor of recovery in the cismontane and transmontane zones. Remarkably, prefire CWD is negatively associated with recovery in the transmontane zone (Table 2). This anomaly could be explained by a surge in plant growth producing abundant fuel, which may increase subsequent fire severity and cause mortality of resprouters (Moreno and Oechel 1993). More indicative of drought impact, CWD, PRCP, and PDSI in seasons after fire show positive associations with transmontane recovery. It would be speculative to attribute this postfire drought sensitivity to plant functional traits, although Paddock and others (2013) report the greatest drought mortalities in seeding shrubs at a low-elevation, transmontane site within our study region. Potentially differential fire-drought responses among shrub species within sample areas represent a major uncertainty in this study.

Few studies have evaluated impacts of repeated fire in conjunction with drought (Jacobsen and Pratt 2018). Smith and others (2019) explained 8–21 % of the variation in shrub cover (a variable that is distinct from recovery) within the Simi Fire and Old Fire sites, and reported that most recent fire intervals were significantly related to shrub cover. However, Smith and others (2019) (compare Lippitt and others 2013) did not directly quantify shrub cover prior to the initial fires punctuating the return intervals, and the authors imply that precipitation represents a confounding influence due to the lower fire probability at relatively mesic sites. Similarly, Syphard and others (2019a, b) found that fire-return interval contributes to only 11–23% of the explained variation, and that micro-climatic and macro-dimatic site variables cumulatively explain a greater portion of recovery variation. Our statistical results based on 416 paired sample plots suggest that, whereas postfire recovery was significantly impacted by drought, short-interval fire effects were nonsignificant. We attribute the discrepancy between our results and the latter studies (regarding short-interval fire effects) to differences in methodology including sampling chronology, measurement of prefire vegetation, and whether temporal drought influences were statistically accounted. We find that recovery variations resulted from intense drought soon after or before fire, culminating in a weak but significant effect (R² = 0.05–0.09; p = 0.001–0.04) that was more pronounced after multiple fires. In discussing postfire recovery variations generally, we acknowledge that sensitive chaparral species or areas were perhaps already impacted by drought or short-interval fire prior to the study period (compare Meng and others 2014). Also, this study excluded sites that burned more than twice, because this would complicate attribution of drought intensity according to fire timing. Our finding that drought can impact recovery in single-burn areas may warrant greater attention to drought than fire-return interval for chaparral ecology and management.

We find that landscape-scale variations in recovery among the drought-impacted sites are somewhat predictable (R² = 0.53) based on terrain, vegetation, soils, and visible reflectance index data (Table 4). The LFM-sensitive VARI derived from Landsat imagery is a strong and negative predictor of recovery within the study areas, particularly when normalized for shrub cover variation. This outcome may support the interpretation that resprouting chaparral in mesic locations is most subject to drought-fire impact. Notably, Peterson and others (2008) showed that differences between chaparral landscapes reduce the spatial correlation of VARI to LFM (R² = 0.59), as compared to within-site serial correlation through a season (R² = 0.78). Our use of single prefire Landsat images may yield different VARI patterns than would images from other years or seasons, but we deemed the early-June observations from a non-drought year (1986) to be representative of LFM under high-growth conditions. With clearer understanding of the LFM-drought effect relationship (and additional predictor variables), the regression model herein could be adapted to prioritize locations for ecological study or conservation along xeric chaparral ecotones.

The vast majority (58–69%) of variation in postfire recovery is unexplained by seasonal drought metrics and other geospatial variables evaluated in this study. Part of this uncertainty is attributable to the Landsat-based dFSC data, part may stem from spatial misalignments and scale differences among the data sets, and part is likely due to a lack of data on potentially significant yet unquantified variables. Monthly drought indices aggregated into seasonal periods do not capture important meteorological variations that occur on diurnal or weekly time-scales. The CWD index incorporates data on soil properties and meteorological estimations that are not fully reliable (Flint and others 2013). Application of localized calibration plots was effective in normalizing spatial biases in the NDVI-FSC relation due to aridity and plant community type differences across the region. However, Landsat NDVI-based metrics of FSC exhibited R² values of 0.56–0.95 in this study, and thus entail uncertainty when sampled in a spatially explicit manner. Cross-comparison of dFSC to other data sets of differing spatial resolutions and positional accuracies can also cause error in spatial analysis (Foody 2007). Additionally, NDVI trajectories that are more temporally resolved (e.g., intra-seasonally) are required to better elucidate the chronological aspect of drought impact on recovering chaparral. Our analysis of landscape-scale patterns of recovery (Figure 6) suggests that detailed maps of plant species may enable a more substantive analysis than was feasible using maps of plant community types. Finally, it was beyond the scope of this study to evaluate the 2012–2016 drought—which may have impacted chaparral stands that were completely or mostly recovered from prior fires by 2012—due to the limited time-frames of the CWD and PRCP data. Our finding that drought in seasons close to fire events is most impactful on recovery suggests that—although the 2012–2016 drought may have led to shrub mortalities—such impacts are most likely not due to interruption of postfire recovery in our study areas. Our Landsat NDVI trajectory data suggest that annual shrub growth declined in 2014–2016 but rebounded to normal levels in 2017–2018, and was not outside the historical range of variability in 2012–2018.

Despite the uncertainties noted above, this study is a contribution to the ecology of disturbance in southern Californian chaparral. It provides empirical evidence consistent with the following hypothesis: the locations of fires relative to climatic gradients—and stochastic timing of fire relative to drought intensity fluctuation—conjointly influence patterns of recovery, which determine long-term structural and compositional stability of chaparral. This study should raise further questions tractable by remote sensing, field work, modeling, or laboratory experiment. Analyses that are temporally resolved at intra-seasonal scales and supported by spatially explicit plant species data are needed to better explain drought-fire sensitivity in chaparral. Seasonal Landsat NDVI trajectories during early postfire recovery and between sequential fire events may show critical thresholds with respect to drought intensity, and clarify the relationship of drought to short-interval fire impact. Evaluation of large-scale patterns in vegetation change based on time-sequential remote sensing will continue to be important in future decades, as disturbance regimes and plant communities continue to evolve.

Highlights.

• Postfire recovery of chaparral was evaluated across southern California (1984–2018).

• Drought in wet seasons surrounding fire events significantly affected recovery.

• Geographic range of chaparral may be contracting due to drought-fire in xeric areas.