Remote Sensing of Environment 183 (2016) 186–197 Contents lists available at ScienceDirect Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse Multi-index time series monitoring of drought and fire effects on desert grasslands Miguel L. Villarreal a,⁎, Laura M. Norman b, Steven Buckley c, Cynthia S.A. Wallace b, Michelle A. Coe d a U.S. Geological Survey, Western Geographic Science Center, 345 Middlefield Rd MS #531, Menlo Park, CA 94025, United States U.S. Geological Survey, Western Geographic Science Center, 520 N. Park Avenue, Suite #102G, Tucson, AZ 85719, United States National Park Service, Southwest Exotic Plant Management Team, 12661 E. Broadway Blvd, Tucson, AZ 85748, United States d School of Geography and Development, University of Arizona, Box 210076, Tucson, AZ 85721, United States b c a r t i c l e i n f o Article history: Received 28 September 2015 Received in revised form 17 May 2016 Accepted 28 May 2016 Available online 6 June 2016 Keywords: Time series analysis Fractional cover Landsat SATVI NDVI Grasslands Fire management Drought a b s t r a c t The Western United States is expected to undergo both extended periods of drought and longer wildfire seasons under forecasted global climate change and it is important to understand how these disturbances will interact and affect recovery and composition of plant communities in the future. In this research paper we describe the temporal response of grassland communities to drought and fire in southern Arizona, where land managers are using repeated, prescribed fire as a habitat restoration tool. Using a 25-year atlas of fire locations, we paired sites with multiple fires to unburned control areas and compare satellite and field-based estimates of vegetation cover over time. Two hundred and fifty Landsat TM images, dating from 1985–2011, were used to derive estimates of Total Vegetation Fractional Cover (TVFC) of live and senescent grass using the Soil-Adjusted Total Vegetation Index (SATVI) and post-fire vegetation greenness using the Normalized Difference Vegetation Index (NDVI). We also implemented a Greenness to Cover Index that is the difference of time-standardized SATVITVFC and NDVI values at a given time and location to identify post-fire shifts in native, non-native, and annual plant cover. The results highlight anomalous greening and browning during drought periods related to amounts of annual and non-native plant cover present. Results suggest that aggressive application of prescribed fire may encourage spread of non-native perennial grasses and annual plants, particularly during droughts. Published by Elsevier Inc. 1. Introduction Grasslands provide society with a number of important ecosystem services, but overuse and poor management of these landscapes in the past has greatly diminished their spatial extent and quality (Sala & Paruelo, 1997; Samson & Knopf, 1994; White, Murray, & Rohweder, 2000). Fire is the primary natural disturbance in semidesert grasslands, and its removal from the landscape in the past contributed to widespread changes in vegetation structure, composition, and function (McPherson, 1997). Fire has been reintroduced to degraded grasslands as a way to restore key ecosystem services like forage production, pollinator services, soil processes, and wildlife habitat (Neary, Klopatek, DeBano, & Ffolliott, 1999; Pyke, Brooks, & D'Antonio, 2010; Wright, 1974). Prescribed fire is believed to reduce invasibility of grasslands by non-native plants (Ditomaso et al., 2006; Smith & Knapp, 1999), reduce woody plant encroachment and desertification (Bahre, 1991; McPherson, 1997), and increase native species diversity (Valone & ⁎ Corresponding author. E-mail addresses: mvillarreal@usgs.gov (M.L. Villarreal), lnorman@usgs.gov (L.M. Norman), steve_buckley@nps.gov (S. Buckley), cswallace@usgs.gov (C.S.A. Wallace), macoe@email.arizona.edu (M.A. Coe). http://dx.doi.org/10.1016/j.rse.2016.05.026 0034-4257/Published by Elsevier Inc. Kelt, 1999). Large-scale application of fire on private and public grasslands is not easily implemented as it requires considerable expertise and financial investment, has many inherent risks and liabilities, and is difficult, if not impossible, to administer across landscapes with numerous land owners (Yoder, Engle, & Fuhlendorf, 2004). In addition to logistical obstacles, the restorative effects of prescribed fire in grasslands may be negligible in areas already occupied by non-native grasses (Bock & Bock, 1992; Bock, Kennedy, Bock, & Jones, 2007; Geiger & McPherson, 2005; Halpern, Haugo, Antos, Kaas, & Kilanowski, 2011; McGlone & Huenneke, 2004) if not used in combination with additional management actions (Havstad & James, 2010). Much of the uncertainty surrounding prescribed fire in arid grasslands is related to the scarcity of information describing fire effects on ecosystems over the temporal and spatial scales at which they operate; many of the studies that have influenced current management of grasslands in the southwestern US were conducted on small experimental plots, or over short time periods (Bestelmeyer, Goolsby, & Archer, 2011; Sayre, deBuys, Bestelmeyer, & Havstad, 2012). Readily accessible, mid-resolution, multispectral satellite time-series data offers promise for addressing these fundamental questions about the long-term effects of fire and climate on grasslands at a spatial and temporal scale that can provide land managers with actionable information. M.L. Villarreal et al. / Remote Sensing of Environment 183 (2016) 186–197 1.1. Fire and drought in grasslands Many federal, state and private land managers have embraced the idea of prescribed fire as a restoration tool for grasslands, but there remains some ambiguity around how to best implement long-term fire plans given the uncertainty of the combined impacts of climate change and fire on vegetation recovery. In the southwestern United States, early attempts to improve range conditions involved the seeding of non-native grasses like Eragrostis lehmanniana (Lehmann lovegrass) for erosion control purposes, and the plant is now present, if not dominant, in most desert grasslands of the region. E. lehmanniana is a warm season African perennial bunchgrass that is adapted to both drought and fire; E. lehmanniana recovers quickly after fire with greater seed germination, which gives it an advantage over native grasses (Anable, McClaran, & Ruyle, 1992; McGlone & Huenneke, 2004; Ruyle, Roundy, & Cox, 1988). The timing of prescribed fire in semidesert grasslands is critical for reducing the likelihood of unwanted outcomes. The interaction of fire with anomalous weather conditions has been shown to push already degraded grassland systems past thresholds into new states (Drewa & Havstad, 2001; Parmenter, 2008). In southeastern Arizona for example, a drought during 1988–1989 caused considerable loss of both native and non-native grass cover, but above-average late-season precipitation in 1989 favored reestablishment of non-native grasses over drought-damaged native grasses, turning many areas into non-native monocultures (Robinett, 1992). Drought may also hinder post-fire recovery of native perennial grasses while encouraging annual plants and woody plants (Cable, 1967). Long-term grassland monitoring data in Arizona also indicate that the drought of 2006 caused a state shift from native grasses to forbs, which were eventually replaced by non-native perennial grasses (Polyakov et al., 2010). 1.2. Remote sensing of semidesert grassland disturbances Questions about landscape and community-level effects of fire and drought on semidesert grasslands can be addressed using satellite time-series data, but the complexity of post-disturbance recovery requires careful consideration and interpretation of plant community dynamics. Satellite estimates of growing season aboveground net primary production (ANPP) in desert grasslands are generally correlated with annual precipitation, and positive ANPP outliers have been identified 187 as drought-induced shifts from native grass to annual forb cover (Moran et al., 2014). Post-fire vegetation monitoring has been effectively accomplished in non-grassland systems using vegetation index timeseries to measure recovery rates (Goetz, Fiske, & Bunn, 2006; Lentile et al., 2006; Röder, Hill, Duguy, Alloza, & Vallejo, 2008; Van Leeuwen et al., 2010), but post-fire cover changes in grasslands are unique in that the recovery can be quite rapid: herbaceous cover will often fully recover in the first growing season following a burn, and given optimal precipitation the resulting growth can be more vigorous and green than prefire conditions. Vegetation state changes may occur when post-fire conditions are unfavorable for native grass recovery (grazing pressures, extreme climactic events, etc.), contributing to the spread of non-native grasses and woody shrubs/trees (Fig. 1, hypothesized state changes are adapted from McPherson, 1997; D'Odorico, Okin, & Bestelmeyer, 2012). Persistent drought and continued lack of fire may shift a non-native savannah into an annual plant/bare ground assemblage (Fig. 1). Likewise, if an early-season fire is followed by extreme precipitation, the resulting soil loss, erosion and gully down-cutting can lead to degraded sites dominated by annual plants and/or bare ground. These complex disturbance-related vegetation responses in semidesert grasslands can contribute to misinterpretation of vegetation index data and the process of community recovery, particularly if a positive measure of greenness is interpreted as healthy vegetation. However, in most cases, combining satellite-data with vegetation monitoring plots and field measurements can provide a more informed interpretation of landscape vegetation change. Vegetation estimates from MODIS and AVHRR have been extensively and successfully used to study grassland vegetation dynamics (Piñeiro, Oesterheld, & Paruelo, 2006), but complex spatial patterns that result from the interaction of disturbances with soils, topography, and vegetation are often undetectable with moderate-low resolution sensors (0.25–1 km). The release of data from the Landsat archive is driving new research into detecting and monitoring vegetation changes using a high-frequency time series of mid-resolution (30-m) multispectral data (Melaas, Friedl, & Zhu, 2013; Vogelmann, Xian, Homer, & Tolk, 2012; Zhu & Woodcock, 2014). In many parts of the globe, high-frequency Landsat time-series may be difficult to assemble because of data gaps caused by a combination of return interval (16-day) and cloud cover. These obstacles can be overcome using seasonal averages, best-pixel image composite algorithms, and other statistical filtering Fig. 1. Illustration of hypothesized vegetation state changes related to drought, fire, and grazing. Native grassland sites (1) can transition (A) to non-native dominated (2) when the fire regime is altered and grazing occurs. Reintroduction of fire (B) can help restore native grasses and reduce woody plant cover (3), especially with optimal timing and amount of precipitation (C, 1). Alternatively, non-native (2) sites may transition (D) to annualized/bare states (4) during fire-free periods of prolonged drought. 188 M.L. Villarreal et al. / Remote Sensing of Environment 183 (2016) 186–197 approaches (Brooks, Wynne, Thomas, Blinn, & Coulston, 2014; Hermosilla, Wulder, White, Coops, & Hobart, 2015; Huang et al., 2010). Arid lands however are particularly well suited to Landsat time-series analysis due to relatively low cloud cover though out the year and the rapid response of desert vegetation to available precipitation, which allow for seasonal, annual, or decadal monitoring of land cover trends (Sonnenschein, Kuemmerle, Udelhoven, Stellmes, & Hostert, 2011). Monitoring interannual vegetation dynamics of perennial grasses with remote sensing is difficult due to a short growing season. Satellite spectral vegetation indices (VI) derived from the near-infrared (NIR) bands that are based on chlorophyll absorption like the Normalized Difference Vegetation Index (NDVI), tend to reliably approximate vegetation greenness during photosyntheically-active periods, but do not correspond well to senescent vegetation and litter outside of the growing season (Nagler, Inoue, Glenn, Russ, & Daughtry, 2003; Xu, Guo, Li, Yang, & Yin, 2014). Greenness measured using NIR-based indices may not always correspond to aboveground biomass or primary production, and in grasslands in particular it can be difficult to determine the contribution of annual and perennial herbaceous cover to the index value (Karnieli et al., 2013). The mid- and shortwave-infrared bands are therefore useful for monitoring non-growing season vegetation changes over both interannual and decadal periods, especially for herbaceous lifeforms with short, climate-driven pulses of “green-up” (Asner & Lobell, 2000; Guerschman et al., 2009; Nagler et al., 2003; Qi et al., 2002). The Soil Adjusted Total Vegetation Index (SATVI), a midand shortwave-based index that corresponds with both live and senescent vegetation cover, has been shown to capture temporal changes in grassland vegetation cover more effectively than other vegetation indices (Goirán, Aranibar, & Gomez, 2012; Hagen et al., 2012; Marsett et al., 2006). Combinations of VIS/NIR (e.g. NDVI) and SWIR-based indices (e.g. SATVI) can be applied to time-series data to monitor dynamics of different land-cover types (Andela, Liu, van Dijk, de Jeu, & McVicar, 2013; Hill, 2013; Zhu & Woodcock, 2014), and potentially identify state changes related to drought, fire, and other land disturbances. In this study we used time series remote sensing and field data to monitor the effects of fire and drought on vegetation dynamics of desert grasslands in Southern Arizona. The specific objectives were to: 1) Use field and satellite vegetation cover estimates to assess the effectiveness of fire as a tool to restore invaded desert grasslands 2) Assess differences in vegetation greenness and cover over time across soil types for grasslands that have burned multiple times during a 27 year period 3) Develop and implement a multi-index vegetation monitoring approach to characterize fire and drought induced shifts in vegetation cover We used a high-frequency time series of Landsat TM data to monitor and characterize vegetation-cover changes and phenology at an ungrazed desert grassland in southern Arizona where over 180 fires, both prescribed and wild, occurred in the study area since 1984, producing a complex burn mosaic that offers insight into recent effects of climate and fire on grasslands at a landscape scale. Vegetation cover was assessed using fractional cover estimates calculated from SATVI, which were paired with NDVI time series to examine how each index responds to fire and drought disturbances. We hypothesize that NDVI alone may fail to adequately capture post-disturbance cover changes from native or non-native grass to annualized states that still exhibit considerable “greenness”. We therefore implement a multi-index monitoring approach to quantify coincident changes in fractional cover and in greenness of vegetation over time, in order to capture more subtle changes in dominant plant lifeforms. 2. Methods 2.1. Study area The Buenos Aires National Wildlife Refuge (BANWR) is located in southern Arizona, USA along the international border with Mexico (Fig. 2). The 117,500-acre Buenos Aires Ranch was purchased by United States Fish and Wildlife Service (USFWS) in 1985 under the authority of the Endangered Species Act in order to protect and enhance habitat of the masked bobwhite quail (Colinus virginianus ridgwayi). The refuge is situated at the western edge of the Madrean Archipelago Ecoregion (Omernik, 1987), and the majority of the refuge falls within the bottomlands of the alluvial basin separating the Baboquivari and Atascosa mountains. 85% of the refuge land is semidesert grassland/savannah that is classified and characterized as “Shrub-invaded non-native grassland with 10–35% total shrub cover and mesquite cover b15% and nonnative perennial grasses are common or dominant…a defining characteristic for this type is its potential for shrub reduction using prescribed burns and restoration” (Gori & Enquist, 2003). Objectives of the BANWR Multi-unit Burn Plan are to restore, conserve, and manage the natural abundance and diversity of wildlife and habitat utilizing strategies that focus on environmental and biological integrity, which include: 1) reducing and/or maintaining woody plant canopy cover to b15%, 2) managing fire to reduce introduced Eragrostis lehmanniana (Lehmann lovegrass) frequency to between zero and 10%, and favor native grasses over Eragrostis lehmanniana, and 3) top-killing 90% or more of Eragrostis lehmanniana. Fires prescribed since 1986 have been numerous and generally small (n = 116, mean size =486 ha) and constrained to fire-management units bounded by access roads. Wildfires were generally smaller due to quick suppression (n = 66, mean = 229 ha), but included 13 fires greater than 400 ha that burned across multiple fire management units. The 21,500 ha Santa Rita Experimental Range (SRER), located approximately 40 km east of the BANWR study site (Fig. 2) was established in 1903 and is the oldest continuously operating experimental range in the world (McClaran, Angell, & Wissler, 2002). Longterm monitoring data collected on the range have contributed to a wide-range of ecological research and have informed land management decisions at grasslands across the region (McClaran et al., 2002; Sayre, 2003). Major vegetation communities at BANWR and SRER are the same, facilitating cross-site comparison of vegetation cover and satellite data (both sites fall within Landsat Path 36/Row 38). Some major land use differences exist between the two sites, related primarily to their contrasting objectives of wildlife management and experimental research; while fire is used to manage and manipulate vegetation at BANWR, fire, livestock grazing, and extensive vegetation manipulations are implemented at SRER in order examine the effects of experimental treatments on long-term vegetation changes (McClaran et al., 2002). 2.2. Satellite estimates of vegetation cover We identified 250 cloud-free Landsat Thematic Mapper scenes (Path 36/Row 38) dating from 1985–2011 that cover the study area (Fig. 3). There were on average three scenes during the growing season (July– October) and 7 during the remainder of the year (Fig. 3). Year 2002 had the greatest number of cloud free images (15) and 1990 the fewest (5) (Fig. 3). Images were atmospherically corrected using the Modtran 5 atmospheric radiative transfer model in ATCOR-3 and converted to surface reflectance. The Soil Adjusted Total Vegetation Index (SATVI) employs Landsat Thematic Mapper band 3 (0.63–0.69 μm), band 5 (1.55–1.75 μm) and band 7 (2.08–2.35 μm) and is calculated as: SATVI ¼ ρband5−ρband3 ρband7 ð1 þ LÞ− ρband5 þ ρband3 þ L 2 ð1Þ M.L. Villarreal et al. / Remote Sensing of Environment 183 (2016) 186–197 189 Fig. 2. Location map of the Buenos Aires National Wildlife Refuge and Santa Rita Experimental Range relative to desert grasslands in the Apache Highlands Ecoregion (left top); the number of fires at Buenos Aires National Wildlife Refuge between 1986 and 2011 at the refuge (left bottom); distribution of sampling locations relative to major soil types (right top); and a detail showing surface patterns at different burn areas from a 2001 aerial photograph (bottom right). where ρ is surface reflectance of each band, and L is a soil adjustment factor set to 0.5 (Huete, 1988, Marsett et al., 2006). Minimum and maximum SATVI values are used to calculate the Total Vegetation Fractional Cover (TVFC), which is an estimate of the percent cover within a pixel, both of green and senescent vegetation (0% cover as bare ground and 100% cover as full canopy). SATVI was scaled to TVFC using average minimum and maximum (−0.14 and 0.12 respectively) values measured at sites that were selected to represent a range of annual growing conditions and cover amounts during the study period. TVFC is calculated as: TVFC ¼ SATVI−SATVImin  100 SATVImax−SATVImin ð2Þ In addition to SATVI-TVFC, we calculated the NDVI and compared dynamics of the indices over time. To assess changes in greenness relative to changes in cover we used a z-transformation of both NDVI and SATVI-TVFC that subtracts an individual observation (xi) from the mean (μ) of the time-series divided by the standard deviation (σ): zNDVI ¼ ðxi−μ Þ=σ ð3Þ and zTVFC ¼ ðxi−μ Þ=σ ð4Þ A standardized greenness-to-cover index (GCI) was calculated as the difference between the two z-normalized indices: GCI ¼ zNDVI−zTVFC Fig. 3. Annual and seasonal distribution of Landsat TM scenes used in this study. ð5Þ The GCI provides a per-pixel measure of the magnitude of greenness at a given date relative to estimated cover at that same point in time, and allows us to track these dynamics over time. High GCI values indicate that the location exhibited photosynthetic activity above the expected amount given estimated plant cover from the SATVI-TVFC at 190 M.L. Villarreal et al. / Remote Sensing of Environment 183 (2016) 186–197 that same location and time. In theory, increases and decreases in grass cover and greenness should occur synchronously over time in response to available resources for plant growth (an exception being the growing season immediately post-fire when an area might have high greenness but low cover); extremely high GCI values during the growing season may therefore represent a pulse of short-lived annual plants that typically do not commit resources to developing substantial above and below ground biomass but have high rates of photosynthesis and greening. Annual plant pulses, often consisting of herbaceous forbs, typically occur on sites where native and non-native perennial grasses have not yet recovered after fire, or where drought, fire and post-fire erosion may have affected the site's ability to support perennial bunch grasses. To produce consistent annual measurements, GCI were calculated for Landsat scenes acquired during the peak growing season (August and September, Julian days ranging from 215–271)) for 1988–2009. Due to excessive cloud cover obscuring parts of BANWR during peak growing season, we did not calculate GCI for 1998, 2000, 2001 and 2004. We compared time-series of SATVI-TVFC, NDVI, and GCI for burn sites and adjacent control sites over three soil types that cover most of the study area: a sandy loam, a shallow upland, and a loamy upland (Fig. 2). Soil information was acquired from the United States Department of Agriculture's Soil Survey Geographic (SSURGO) Database (http://websoilsurvey.sc.egov.usda.gov./App/HomePage.htm). We analyzed time-series within areas that had burned 1–5 times using a spatial database of fire perimeters provided by USFWS and assess variation in satellite-measured dynamics using field-collected vegetation cover data. Fire maps were developed using a Landsat protocol similar to Monitoring Trends in Burn Severity (MTBS) approach, but for any fire larger than 4 ha. We averaged index values from 25 randomly distributed sample locations within burn and control areas. Random samples were distributed in upland areas to avoid drainages and depressions that accumulate moisture and therefore may have different herbaceous species assemblages and more xeroriparian woody cover. Drought periods were identified from the annual Palmer Drought Severity Index (PDSI) calculated for Arizona Zone 7 (Fig. 4), which was z-transformed to compare the GCI time series. PDSI is a standardized, time-integrated drought index that takes into account precipitation, temperature, and available moisture capacity of soils (Alley, 1984; Palmer, 1965). Positive PDSI numbers indicate wet periods and negative numbers indicate drought. To evaluate differences in both cover and greenness over time and between different soil groups and number of fires, SATVI-TVFC and NDVI values from each growing season were considered as repeated measurements using two-way repeated measures ANOVA with years as the repeated factor. Repeated measures ANOVA were also calculated within soil groups to test for fire and year effects on SATVI-TVFC and NDVI. Interactions were tested with Pairwise Multiple Comparison Procedures (Holm-Sidak method). 2.3. Validating satellite cover estimates with long-term monitoring data SATVI-TVFC values were calibrated and validated independently with a subset of Landsat scenes that corresponded to repeat measurements at BANWR and SRER. We used twenty-five, 30.4 × 0.31-m monitoring (1994, 1997, 2000, 2003, and 2006) belt transects from SRER and twelve 30-m line-intercept monitoring transects from BANWR (1987, 1993, 1997, and 2002). Species-level data were aggregated into percentages of grass cover, shrub cover, tree cover, and total cover. We categorized large woody plants (Prosopis velutina, Parkinsonia spp.) as “trees”, in order to separate them from low-statured and small canopy desert shrubs (e.g. Mimosa biuncifera) and cacti (e.g. Opuntia spp.), which have different spectral and spatial characteristics that are likely to affect satellite VI measurements. SATVI-TVFC values at pixels that overlapped transect lines were extracted and averaged from a TM scene that temporally matched the field collection dates. We note that the limited areal coverage of belt transects are not ideal for validation of VI measurements taken from square 30-m pixels, especially since vegetation cover of these grasslands can be highly variable over short distances, but when averaged the data effectively characterize broad cover changes over time. To test the relationship between GCI and different plant life-forms, we used cover measurements from 33 vegetation plots collected across BANWR in 2002 (a drought year). These plots were measured during the peak growing season (August–September) using the point-intercept method on line transects within the plot area. Species data were aggregated into percent cover classes for 3 categories defined by the dominant vegetation type: annual forb and grass, perennial native grasses and perennial non-native grasses. 2.4. Contemporary field data to characterize current conditions Contemporary vegetation conditions were assessed across BANWR with vegetation plot data collected at thirty-two locations during the 2012 growing season (July–October). These field data were collected approximately10 months after the unanticipated decommissioning of Landsat Thematic Mapper 5 in November 2011 yet before the launch of Landsat 8, which precludes their use as validation of 30-m spectral data at a specific point in time. These data are therefore used to characterize vegetation cover and community composition “endpoints” that correspond closely to the final scene of the Landsat 5 TM time series developed for this project (October, 2011). At each site six 20-m line transects spaced 10-m apart were extended perpendicular from a 50-m tape. Vegetation intercepting the vertical plane of the line was recorded at four height strata (Herrick, Van Zee, Havstad, Burkett, & Whitford, 2005), and all plants were identified to species except annuals, excepting non-native species, which were all identified to species regardless of duration. Percent cover of perennial native grasses, perennial non-native grasses, annual grasses, annual forbs, woody plants and bare ground were estimated for each transect. 3. Results 3.1. Validation of SATVI-TVFC cover estimates Fig. 4. Annual Palmer Drought Severity Index (PDSI) values from 1985–2012 calculated for Arizona Zone 7. SATVI-TVFC was positively correlated with grass cover measured from long-term line-intercept transects at SRER, negatively correlated with shrub cover, and weakly but positively correlated with tree cover and total cover (Fig. 5). R2 values for grass cover ranged from 0.61 in 1994 to 0.07 in 2003, shrub cover ranged from 0.02–0.60, tree cover 0.04–0.23 and total cover 0.02–0.23 (Fig. 5). The negative relationship between SATVI-TVFC and shrub cover and the weak correlation with tree cover are likely related to the small leaf size and diffuse canopy structure of desert shrubs and trees (Senegalia and Vachiella spp., M.L. Villarreal et al. / Remote Sensing of Environment 183 (2016) 186–197 191 Fig. 5. Scatterplots of ground measurements from the Santa Rita Experimental Range and satellite-derived SATVI-TVFC. Transects were measured during the senescent period in late-fall. 2006 was an extreme drought year, and this is reflected in the variability between the two measurements (note grass cover is about 3 times lower than in 1994). Propopis spp) in the study area. Total cover measured at transects was generally the product of tree cover which dominated transects at SRER, and helps to explain the weak correlation between SATVI-TVFC and total cover. Overall at SRER, there was a decline in average vegetation cover of the 25 transects from 24.4% in 1994 to 16.3% in 2009, and decline in average grass cover from 4.1% in 1994 to 0.91% in 2009 (Fig. 5), which explains the decreasing correlation coefficients over time. Grassland monitoring transects at BANWR were fewer, and less consistently collected and managed than those at the SRER, so we pooled the 12 BANWR locations and estimated the mean and standard deviation of perennial grass cover for the years 1987, 1993, 1997 and 2002, and compared these to SATVI-TVFC pixels at those same locations. Both the monitoring and satellite data show an increase in cover from 1987–1993 followed by a decline in 1997 and slight increase in 2002 192 M.L. Villarreal et al. / Remote Sensing of Environment 183 (2016) 186–197 (Fig. 6). Mean cover estimates from satellite and field data are highly correlated (R2 = 0.98), but the field transect data have greater deviation (e.g., 1987 Grass cover μ = 43.4, σ = 22.2, 1987 SATVI-TVFC μ =53.9, σ = 4.3) with the exception of 1997 (Grass cover μ = 18.2, σ = 12.9, SATVI-TVFC μ = 51.2, σ = 6.08) (Fig. 6). The results at BANWR and SRER showed similar vegetation trends between SATVI-TVFC and onthe ground vegetation cover dynamics. 3.2. Temporal dynamics of cover estimates related to drought and fire SATVI-TVFC time-series captured annual and seasonal cover changes and post-fire vegetation recovery trajectories (Fig. 7). In general, cover declined at BANWR from 1985–2011 on all soil types, but the decline was not linear: cover at most sites was stable or increased from 1990– 1996, then declined at most sites beginning in 1996 (Fig. 7). These dynamics generally follow the PDSI pattern characterized by consecutive years of positive PDSI related to above average annual winter precipitation between 1985–1995, followed by longer and more consistent drought periods beginning in 1996 (Fig. 4). Repeated measures ANOVA indicate significant differences in SATVI-TVFC values between soil types (F = 5.773, P = 0.024), but not between number of fires (F = 0.658, P = 0.669). However there was a statistically significant difference between years (F = 15.887, P b 0.001) and significant interaction between number of fires and year (F = 1.488, P = 0.027). Major pairwise differences were observed between non-drought years 1993, 1994, 1995 and drought years 2002, 2003, 2005, and 2006 across all sites. These drought- and fire-related differences are illustrated in Fig. 7, where across all sites and over all soil types, SATVI-TVFC cover estimates were similar in the early non-drought period, but cover began to vary more considerably between sites during the drought periods of 2001– 2003 and 2006–2011, especially at sites with multiple burns during these periods. For example, on the sandy loam soil types (Fig. 7a), a fire in 2006 had little immediate effect on vegetation at the two sites that burned (SL.1F and SL.4F), and most of the cover changes happened as drought progressed through 2007. However, post-fire trajectory of site SL.1F, with only a single fire in 2006, diverged from the other sites, with SATVI-TVFC values increasing from 2007–2011, while the other sites declined (Fig. 7). Transects measured at SL.1F in 2012 indicate average 40% annual and 33% perennial non-native herbaceous cover, which is compositionally much different than the other three sandy loam sites (Table 1). Similarly site SL.2F, which had highest cover of all sites in the early non-drought period (1985–1996), had considerable cover loss and slow recovery after the 1996 and 2001 fires (Fig. 7a). Shallow upland soils (Fig. 7c) had less cover variability during the drought periods when compared to the loamy soil sites (Fig. 7a and b). The 2012 field-measurements indicate the shallow upland sites currently have more perennial grass cover, primarily non-native, and less bare ground and annual cover than the sandy loam sites (Table 1). Outside of the immediate fire-recovery periods following the 1989 and 1997 fires, cover estimates at sites with three and four fires (SU·3F and SU·4F) are generally indistinguishable from the reference site (SU·0F) (Fig. 7c). Cover loss at these sites after the 2001 fire, which occurred during onset of the drought, is more pronounced and recovery times considerably longer (3–4 years) than for the earlier fires (Fig. 7c). SU·5F, with five fires, shows the greatest variability of cover over the entire period, and after the 1995 fire the inter-annual changes become particularly amplified, likely due to a state change from perennial to annual life forms caused by that fire (Fig. 7c). The 1995 fire and all consecutive fires after were during or followed by drought periods (Fig. 4), and SU·5F does not fully recover like SU·3F and SU·4F, sites that burned 1997 rather than 1995. Transect data measured within the SU·5F site burn area in 2012 indicate average cover of 19% bare ground and 18% annual forbs and grasses, the highest out of the four units sampling on this soil type (Table 1). 2012 transect data also indicate the control site SU·0F had the highest average cover of native perennial grass, and SU·3F and SU·4F had higher non-native grass cover in 2012 than SU·0F and SU·5F (Table 1, Fig. 7c). 3.3. Greenness and cover dynamics NDVI and SATVI-TVFC values calculated from the same Landsat image and measured at the same location were not strongly correlated; R2 values ranged from 0.057 on sandy loam soils to 0.318 on shallow upland soils. Unlike SATVI-TVFC, time-series NDVI does not respond predictably to fire-related cover loss and recovery periods at the shallow upland soils sites (Fig. 8a). These four shallow upland sites generally exhibit similar temporal NDVI profile of annual green-up and senescence, with very little separation between sites at any given point in time despite differences in number of fires. This contrasts considerably with the more dynamic changes in the SATVI-TVFC time series of burned sites in Fig. 8b, which relative to the control site (SU·0F) display cover declines after fires and subsequent periods of recovery (Fig. 8b). Results from the repeated measures ANOVA further illustrate the weak response of NDVI to number of fires (F = 0.242, P = 0.929), and showed no strong interactions between NDVI, number of fires, and year (F = 1.162, P = 0.233). There were however notable differences in NDVI between soil types (F = 3.510, P = 0.075), significant differences between years (F = 23.958, P b 0.001), and significant interactions between soil types and years (F = 9.630, P b 0.001). The Greenness to Cover Index (GCI) was implemented to amplify the differences between NDVI and SATVI-TVFC time series dynamics and expose potential information on vegetation changes following disturbances. GCI data were validated with 2002 field measurements of percent cover for 33 plots categorized by the dominant herbaceous cover (annual, native perennial and non-native perennial) (Fig. 9). Data indicate that extreme positive GCI values (N0.5) are associated with plots dominated by annual plants but with low-moderate vegetation cover (20–40%), and plots dominated by E. lehmanniana, also with low- Fig. 6. A) Summary of perennial grass cover from monitoring transects (n = 12) measured at Buenos Aires National Wildlife Refuge (BANWR) in 1987, 1993, 1997 and 2002; and B) SATVITVFC values measured at the same transect locations and dates. M.L. Villarreal et al. / Remote Sensing of Environment 183 (2016) 186–197 193 from negative GCI values in 1989 and 1991, to predominantly positive values beginning after 1996 – indicating a likely shift from more dense Lehmann-dominated cover to an annual or mixed-annual E. lehmanniana type after burning in the 1990s and 2000s (Fig. 10a). During the drought of 2008, all sites had GCI N +1, with SL.1F reaching a GCI of + 1.9. SL.1F burned in 2006 and the intervening years were all drought years, and this peak likely represents a post-fire a shift from negative GCI (dense cover of perennial non-native grass) to annual dominated cover, which is confirmed with 2012 field data (40% annual cover) (Fig. 10a, Table 1). Loamy upland sites show an amplified positive GCI response to the 2002–2003 drought, which is consistent with both the SATVI-TVFC dynamics (cover declines from 1996–2011), and with field data that indicate these sites had considerable bare ground and annual plant cover in 2012 (Fig. 01b, Table 1). The loamy upland site with four fires (LU·4F) had considerably different GCI dynamics than the other loamy upland sites, with generally low GCI values until the fires in 1997 and 2001, where GCI exceeds + 1 for 3 consecutive years, then declines during the 2006–09 drought while other sites show positive GCI. 2012 field data indicate this site had almost no annual cover (4%), considerable non-native (64%) and bare cover (19%), while the other three sites have 10–20% annual cover (Table 1). Shallow upland sites were characterized in 2012 by high-moderate non-native perennial grass cover and relatively low annual cover (Table 1). Sites on shallow upland soils had the least variability of GCI between sites and relative to the no fire site (SU·0F), despite multiple burns during the period (Fig. 10c). Shallow upland sites shift from generally negative GCI during the 1990s to positive GCI after the 2002 drought, but the relationship between GCI and PDSI is generally weak (Fig. 10c). SU·5F responded with GCI N + 1 after a burn in 2008, a year with low PDSI, and had considerably higher annual plant cover in 2012 compared to the other shallow upland sites (Table 1). 4. Discussion Fig. 7. SATVI-TVFC measurements at sites on: A) sandy loam, B) loamy upland and C) shallow upland soils. Cover variability across sites and soil types increases over time as the droughts become more common starting in 1996 and as more fire occurs. It should be noted that the Landsat time-series is incomplete during some years, and intraannual data gaps may contribute to some sharp drops in cover from one date to the next that are not related to a disturbance. moderate cover (20–60%) (Fig. 9). Extreme negative GCI values (b−0.5) are associated with sites with moderate-high E. lehmanniana cover (N 50%) (Fig. 9). Plots dominated by native grass, but with lowmoderate cover (10–30%) had negative GCI values and all other native grass plots with moderate to high cover had GCI values between −0.5-0.0. Native dominated sites did not have any positive GCI values (Fig. 9). When plotted as a time series alongside a z-normalized PDSI, GCI dynamics can further illustrate vegetation dynamics related to disturbance history and drought (Fig. 10). For example, sandy loam sites trended In this paper we describe a multi-index approach for monitoring semidesert grassland vegetation change in response to drought and fire. The key component of this analysis is the coupling of cover and greenness estimates made from multispectral Landsat images, that when used together can illuminate shifting ecosystem conditions and inform long-term fire management plans, timing of annual prescribed burns, and invasive species management. Remote sensing time-series analyses have many applications for fire management and conservation planning in grasslands, and are potentially applicable at much larger scales than this case study. With Landsat 8 these approaches can be operationalized in near-real time by producing bi-monthly estimates of live or senescent cover with SATVI-TVFC, and annual peak growing-season GCI values to identify and monitor possible state changes. The GCI, while applied here during peak growing season, could also be applied at different seasons to monitor peak growth dynamics of winter annuals, or non-native species that may have slightly different phenology than native species (Bradley, 2014; Casady, van Leeuwen, & Reed, 2013). Satellite-based VIs complement ground data collected for long-term monitoring programs, providing additional information to help characterize recovery trajectories through time and across the landscape. Validation results suggest that while SATVI-TVFC responds negatively to increased shrub and tree cover, the index provides an effective measure of amounts of herbaceous vegetation cover within a pixel. When implemented as a time series SATVI-TVFC showed clear differences in grass cover dynamics between soil types, especially during the later drought periods. By comparing burned areas to control sites, SATVI-TVFC provided information on post-fire recovery times across soil types and through time, which when compared to ground monitoring data can be used to estimate points in time when considerable cover changes are occurring. The BANWR desert grassland study site has low woody plant cover 194 M.L. Villarreal et al. / Remote Sensing of Environment 183 (2016) 186–197 Table 1 Average cover measured at plots situated within sample units on sandy loam, shallow upland, and loamy upland sites. Standard deviations are in parenthesis. Woody cover was estimated from intercepts at all height classes (0 to N3.0 m). Site ID Plots (n) Soils Bare Annual grass/forb Native grass Non-native grass Woody Fires SL.0F SL.1F SL.2F SL.4F LU·0F LU·4F LU·5Fa LU·5Fb SU·0F SU·3F SU·4F SU·5F 3 2 4 4 2 2 1 1 4 4 2 3 Sandy loam Sandy loam Sandy loam Sandy loam Loamy upland Loamy upland Loamy upland Loamy upland Shallow upland Shallow upland Shallow upland Shallow upland 37.2 (13.8) 14.8 (5.0) 13.1 (3.9) 27.5 (7.5) 60.4 (5.9) 19.4 (7.9) 19.6 10.0 12.3 (4.2) 14.7 (4.8) 19.6 (15.3) 19.2 (8.8) 22.4 (12.2) 40 (11.2) 16.5 (4.2) 34.8 (19.8) 20.4 (17.7) 3.8 (4.7) 10.4 21.7 5.6 (4.8) 4.2 (3.6) 9.2 (12.4) 17.9 (13.5) 30.5 (14.3) 12.9 (4.1) 13.4 (4.1) 28.6 (14.9) 5.4 (7.7) 13.3 (8.25) 15.4 62.1 19.8 (22.2) 15.4 (15.1) 5 (7.1) 14.2 (13.5) 23.2 (16.2) 33.3 (8.8) 65.7 (15.4) 22.7 (23.5) 9.0 (9.1) 64.2 (20.0) 52.9 8.8 55.9 (23.8) 61.8 (15.5) 62.5 (33.0) 50.6 (16.8) 24.2 (8.8) 19.3 (2.6) 14.4 (9.6) 14.7 (5.9) 15.6 (22.1) 10.0 (2.9) 0.0 5.0 22.7 (3.0) 13.0 (7.7) 7.9 (2.9) 10.3 (3.9) Reference 2006 1996, 2001 1991, 2000, 2002, 2006 Reference 1986, 1997, 2001, 2007 1986, 1995, 2000, 2004, 2008 1986, 1993, 1997, 2000, 2010 Reference 1989, 1997, 2001 1986, 1997, 2001, 2007 1986, 1995, 2000, 2004, 2008 which minimized potential SATVI-TVFC errors resulting from mixedpixels with high shrub and tree cover. At mixed cover sites, time series data should be interpreted with caution: SATVI-TVFC declines could result from either increasing cover of bare ground or increasing cover of shrubs and trees. GCI dynamics were useful for identifying sites dominated by annual plants, particularly on sandy loam soils during drought years, when short-term greening occurs disproportionately to the amount of total vegetation cover. Positive index values related to annual plant cover can be explained by characteristics of their life-history: they are shortlived plants that respond to seasonal rainfall and increased soil moisture and commit considerable resources to seed production, rather than Fig. 8. Example illustrating an NDVI (A) time series at four sites on shallow upland soils with different burn intervals compared to SATVI-TVFC (B) measurements at those same sites. plant biomass. Conversely, areas with high amounts of cover and low greenness (i.e. low GCI) were indicative of sites dominated by the perennial E. lehmanniana, which typically green up earlier in the season. E. lehmanniana is more dependent on winter precipitation than native grasses, and will often green-up with sufficient winter precipitation (Cable, 1971; Huang & Geiger, 2008) but maintain considerable aboveground biomass throughout the year. This method of using plant cover, greenness, and phenology to differentiate between native and non-native species could be implemented to map other non-native grasses such as cheatgrass (Bromus tectorum) in the Great Basin, buffelgrass (Pennisetum ciliare) in the Sonoran desert, or Sahara mustard (Brassica tournefortii) in the Mojave desert. Prescribed fire has been used to successfully achieve BANWR's management goals of reducing woody plant cover, and in general, total vegetation cover has also declined over time at all sites measured with satellite imagery. The use of fire as a means to increase native perennial grass cover has not been successful, and both field and remote sensing data suggest that most sites have experienced increases in non-native perennial, annual grasses, and bare ground. Two of the remote sensing sites with multiple burns (SL.4F and LU·5Fb) had high native grass Fig. 9. Relationship between herbaceous cover and the greenness to cover index for 33 vegetation plots measured in 2002 (drought year). Annual (AN), Lehmann lovegrass (E. lehmanniana) (LL), and Native (N). In general, AN dominated sites with low-moderate cover have positive index values. Sites dominated by native grasses have higher index values as the cover increases, but most of the sites have low cover and index values between − 0.5 and 0. LL dominated sites had a linear relationship with the index, however, areas with dense monoculture of LL had negative index values. M.L. Villarreal et al. / Remote Sensing of Environment 183 (2016) 186–197 195 both of which appear to be occurring at BANWR. E. lehmanniana was introduced into the southwestern US and northern Mexico in the 1930s and had doubled its spatial extent in Arizona by the 1980s (Cox & Ruyle, 1986). Species niche modelling suggests the spread of E. lehmanniana will continue in the future (Schussman, Geiger, MauCrimmins, & Ward, 2006), and remote sensing techniques that can be used to assess its distribution will be useful for management and restoration. Landsat NDVI has been successfully used to monitor temporal changes in semidesert grasslands (Elmore, Mustard, Manning, & Lobell, 2000; Nouvellon et al., 2001), and differencing of multi-date Landsat NDVI has led to successful mapping of invasive grasses that display early greening compared to native species (Peterson, 2005). Based on results presented in this study, NDVI does not appear to be an optimal index to monitor semidesert grassland dynamics in areas with: 1) frequent fire and 2) a complex mix of native, non-native, and annual plants. Complicating the matter is the fact that many woody and succulent desert plants generally have small leaf area and photosynthetic biomass resulting in lower absorption in the red wavelengths when compared to grasses, especially during the growing season. NIR/redbased indices do not adequately capture vegetation dynamics during senescent periods, which can be more than ¾ of the annual time-series in semidesert grasslands. Results from the repeated measures ANOVA on growing season NDVI indicate that NDVI varies over time in response wet/dry periods and across soil types, but the interactive effects of drought and fire did not produce significant differences in the NDVI signal as it did with SATVI-TVFC. Sites that burned 4–5 times showed little difference in temporal dynamics of NDVI when compared to control sites during recovery periods. TVFC calculated from the SWIR-based SATVI, however, showed clear and immediate cover losses after fire. Vegetation transect data collected at the study sites in 2012 show considerable variation in composition of annual and perennial herbaceous cover and percentage of bare ground. Annual plant cover may contribute to higher greenness at a site during their short lifespans, but actual plant cover on these sites may be much lower, a discrepancy that was captured at the end of the SATVI-TVFC time series but not the NDVI. 5. Conclusions Fig. 10. Graph of growing season Greenness to Cover Index and Palmer Drought Severity Index (PDSI) changes over time at sites on A) sandy loam, B) loamy upland, and C) shallow upland soils. cover relative to their paired sites; however it should be noted that two of the fires that burned these sites were wildfires that were not timed and managed to accomplish restoration objectives. Additionally, and unexpectedly, unburned control sites SL.0F and SU·0F had the highest average native grass cover of their soil groups. Recent meta-analysis of prescribed fire effects on native and non-native grasses suggest that non-native cover is rarely reduced after fire (Alba, Skálová, McGregor, D'Antonio, & Pyšek, 2015) given the perpetuation of the grass/fire cycle (D'Antonio & Vitousek, 1992), and because fire often favors annual growth over perennial growth forms (Gómez-González et al., 2011), Desert grasslands are economically and ecologically important components of southwestern US landscapes that are facing multiple ongoing stressors. In this study we used field data and 27 years of Landsat imagery to monitor grassland vegetation changes in response to fire and drought. In particular we were interested in understanding how the number of fires, and the timing of those fires relative to drought, might affect the relative cover of native grasses, non-native grasses, annual plants, and woody plants across different soil types. We found that general cover dynamics were similar across soil types during the nondrought period from 1985–1995, but cover trajectories varied considerably in the early 21st century as drought progressed. Repeated burning of these sites during the drought period further complicated vegetation recovery: some sites that burned during the drought had variable response and recovery times, indicative of vegetation state changes (i.e. change from perennial grassland to barren/annualized state). Differences in post-fire dynamics between SATVI-TVFC and NDVI showed that the SWIR-based index better captures post-fire cover changes, and better characterized long-term recovery trajectories of desert grasslands. When grasslands are burned in early summer, as they typically are, they will often green-up a short time later with summer rainfall, but this greening is often the result of opportunistic annual grasses and forbs and does not reflect recovery of perennial grasses. The SATVI-TVFC, which exploits SWIR absorption of senescent vegetation, offers a relatively consistent annual time series of cover changes irrespective of plant phenology cycles and therefore provides a better measure of post-fire recovery. Differencing the time-series standardized NDVI and SATVI-TVFC with the GCI provided a metric to highlight the 196 M.L. Villarreal et al. / Remote Sensing of Environment 183 (2016) 186–197 deviation of these two indices, which we found corresponded to different amounts of native, non-native, and annual herbaceous cover. Research and management projects interested in real-time monitoring of drought and fire effects on grasslands, rangelands, and other semiarid vegetation communities can benefit from implementing a multi-index approach like the one presented in this paper. Acknowledgements This research was funded through a Mendenhall Fellowship provided by the Land Remote Sensing and Land Change Science Programs of the USGS. Michelle Coe was supported by a University of Arizona/ NASA Space Grant Undergraduate Research Internship. The authors would like to thank Steven Sesnie (US Fish and Wildlife Service) for an early review of the manuscript, and three anonymous reviewers for their insightful comments. We also appreciate the contributions provided by Dan Cohan, Juliette Gutierrez, Lacrecia Johnson, Alycia Parnell, and Emily Yurcich (US Fish and Wildlife Service), Philip Heilman and Chandra Holifield Collins (USDA Agricultural Research Service), Stephen Hagen (Applied Geosolutions), Kristen Bonebrake and Sarah Studd (National Park Service Sonoran Desert Network), and Jonathan Smith, Susan Benjamin, Mara Tongue, and Matthew Jamieson (US Geological Survey). Any use of trade, product, or firm names in this publication is for descriptive purposes only and does not imply endorsement by the US government. References Alba, C., Skálová, H., McGregor, K. F., D'Antonio, C., & Pyšek, P. (2015). Native and exotic plant species respond differently to wildfire and prescribed fire as revealed by meta-analysis. Journal of Vegetation Science, 26(1), 102–113. http://dx.doi.org/10. 1111/jvs.12212. Alley, W. M. (1984). The palmer drought severity index: Limitations and assumptions. Journal of Climate and Applied Meteorology, 23, 1100–1109. Anable, M. E., McClaran, M. P., & Ruyle, G. B. (1992). Spread of introduced Lehmann lovegrass Eragrostis lehmanniana nees. In southern Arizona, USA. Biological Conservation, 61(3), 181–188. http://dx.doi.org/10.1016/0006-3207(92)91114-8. Andela, N., Liu, Y. Y., van Dijk, A. I. J. M., de Jeu, R. A. M., & McVicar, T. R. (2013). Global changes in dryland vegetation dynamics (1988–2008) assessed by satellite remote sensing: Comparing a new passive microwave vegetation density record with reflective greenness data. Biogeosciences, 10(10), 6657–6676. http://dx.doi.org/10.5194/ bg-10-6657-2013. Asner, G. P., & Lobell, D. B. (2000). A biogeophysical approach for automated SWIR Unmixing of soils and vegetation. Remote Sensing of Environment, 74(1), 99–112. http://dx.doi.org/10.1016/S0034-4257(00)00126-7. Bahre, C. J. (1991). A legacy of change: Historic human impact on vegetation in the Arizona borderlands. Tucson: University of Arizona Press. Bestelmeyer, B. T., Goolsby, D. P., & Archer, S. R. (2011). Spatial perspectives in state-andtransition models: A missing link to land management? Journal of Applied Ecology, 48(3), 746–757. http://dx.doi.org/10.1111/j.1365-2664.2011.01982.x. Bock, J. H., & Bock, C. E. (1992). Vegetation responses to wildfire in native versus exotic Arizona grassland. Journal of Vegetation Science, 3(4), 439–446. http://dx.doi.org/10. 2307/3235800. Bock, C. E., Kennedy, L., Bock, J. H., & Jones, Z. F. (2007). Effects of fire frequency and intensity on velvet mesquite in an Arizona grassland. Rangeland Ecology & Management, 60(5), 508–514. http://dx.doi.org/10.2111/1551-5028(2007)60[508:EOFFAI]2.0.CO; 2. Bradley, B. A. (2014). Remote detection of invasive plants: A review of spectral, textural and phenological approaches. Biological Invasions, 16, 1411–1425. http://dx.doi.org/ 10.1007/s10530-013-0578-9. Brooks, E. B., Wynne, R. H., Thomas, V. A., Blinn, C. E., & Coulston, J. W. (2014). On-the-fly massively multitemporal change detection using statistical quality control charts and Landsat data. IEEE Transactions on Geoscience and Remote Sensing, 52(6), 3316–3332. http://dx.doi.org/10.1109/TGRS.2013.2272545. Cable, D. R. (1967). Fire effects on semidesert grasses and shrubs. Journal of Range Management Archives, 20(3), 170–176. Cable, D. R. (1971). Lehmann lovegrass on the Santa Rita experimental range, 1937-1968 (el Zacate Lehmann lovegrass en la Estacion experimental de Santa Rita Durante los Años de 1937-68). Journal of Range Management, 24(1), 17–21. http://dx.doi.org/10. 2307/3896058. Casady, G. M., van Leeuwen, W. J. D., & Reed, B. C. (2013). Estimating winter annual biomass in the Sonoran and Mojave deserts with satellite- and ground-based observations. Remote Sensing, 5(2), 909–926. http://dx.doi.org/10.3390/rs5020909. Cox, J. R., & Ruyle, G. B. (1986). Influence of climatic and edaphic factors on the distribution of Eragrostis lehmanniana nees in Arizona, USA. Journal of the Grassland Society of Southern Africa, 3(1), 25–29. http://dx.doi.org/10.1080/02566702.1986.9648027. D'Antonio, C. M., & Vitousek, P. M. (1992). Biological invasions by exotic grasses, the grass/fire cycle, and global change. Annual Review of Ecology and Systematics, 23(1), 63–87. http://dx.doi.org/10.1146/annurev.es.23.110192.000431. D'Odorico, P., Okin, G. S., & Bestelmeyer, B. T. (2012). A synthetic review of feedbacks and drivers of shrub encroachment in arid grasslands. Ecohydrology, 5(5), 520–530. http://dx.doi.org/10.1002/eco.259. Ditomaso, J. M., Brooks, M. L., Allen, E. B., Minnich, R., Rice, P. M., & Kyser, G. B. (2006). Control of invasive weeds with prescribed burning. Weed Technology, 20(2), 535–548. Drewa, P. B., & Havstad, K. M. (2001). Effects of fire, grazing, and the presence of shrubs on Chihuahuan desert grasslands. Journal of Arid Environments, 48(4), 429–443. http:// dx.doi.org/10.1006/jare.2000.0769. Elmore, A. J., Mustard, J. F., Manning, S. J., & Lobell, D. B. (2000). Quantifying vegetation change in semiarid environments: Precision and accuracy of spectral mixture analysis and the normalized difference vegetation index. Remote Sensing of Environment, 73(1), 87–102. http://dx.doi.org/10.1016/S0034-4257(00)00100-0. Geiger, E. L., & McPherson, G. R. (2005). Response of semi-desert grasslands invaded by non-native grasses to altered disturbance regimes. Journal of Biogeography, 32(5), 895–902. http://dx.doi.org/10.1111/j.1365-2699.2004.01235.x. Goetz, S. J., Fiske, G. J., & Bunn, A. G. (2006). Using satellite time-series data sets to analyze fire disturbance and forest recovery across Canada. Remote Sensing of Environment, 101(3), 352–365. http://dx.doi.org/10.1016/j.rse.2006.01.011. Goirán, S. B., Aranibar, J. N., & Gomez, M. L. (2012). Heterogeneous spatial distribution of traditional livestock settlements and their effects on vegetation cover in arid groundwater coupled ecosystems in the Monte Desert (Argentina). Journal of Arid Environments, 87, 188–197. http://dx.doi.org/10.1016/j.jaridenv.2012.07.011 (0). Gómez-González, S., Torres-Díaz, C., Valencia, G., Torres-Morales, P., Cavieres, L. A., & Pausas, J. G. (2011). Anthropogenic fires increase alien and native annual species in the Chilean coastal matorral. Diversity and Distributions, 17(1), 58–67. http://dx.doi. org/10.1111/j.1472-4642.2010.00728.x. Gori, D. F., & Enquist, C. A. F. (2003). An assessment of the spatial extent and condition of grasslands in central and southern Arizona, southwestern New Mexico and northern Mexico. (Prepared by The Nature Conservancy, Arizona Chapter. 28 pp. http:// azconservation.org/dl/TNCAZ_Grasslands_Assessment_Report.pdf). Guerschman, J. P., Hill, M. J., Renzullo, L. J., Barrett, D. J., Marks, A. S., & Botha, E. J. (2009). Estimating fractional cover of photosynthetic vegetation, non-photosynthetic vegetation and bare soil in the Australian tropical savanna region upscaling the EO-1 Hyperion and MODIS sensors. Remote Sensing of Environment, 113(5), 928–945. http://dx. doi.org/10.1016/j.rse.2009.01.006. Hagen, S. C., Heilman, P., Marsett, R., Torbick, N., Salas, W., van Ravensway, J., & Qi, J. (2012). Mapping total vegetation cover across western rangelands with moderateresolution imaging spectroradiometer data. Rangeland Ecology & Management, 65(5), 456–467. http://dx.doi.org/10.2111/REM-D-11-00188.1. Halpern, C. B., Haugo, R. D., Antos, J. A., Kaas, S. S., & Kilanowski, A. L. (2011). Grassland restoration with and without fire: Evidence from a tree-removal experiment. Ecological Applications, 22(2), 425–441. http://dx.doi.org/10.1890/11-1061.1. Havstad, K. M., & James, D. (2010). Prescribed burning to affect a state transition in a shrub-encroached desert grassland. Journal of Arid Environments, 74(10), 1324–1328. http://dx.doi.org/10.1016/j.jaridenv.2010.05.035. Hermosilla, T., Wulder, M. A., White, J. C., Coops, N. C., & Hobart, G. W. (2015). An integrated Landsat time series protocol for change detection and generation of annual gapfree surface reflectance composites. Remote Sensing of Environment, 158, 220–234. http://dx.doi.org/10.1016/j.rse.2014.11.005. Herrick, J., Van Zee, J., Havstad, K., Burkett, L. N., & Whitford, W. N. (2005). Monitoring manual for grassland, shrubland and savanna ecosystems. Volume I: Quick start. Volume II: Design, supplementary methods and interpretation. USDA-ARS Jornada experimental range. Hill, M. J. (2013). Vegetation index suites as indicators of vegetation state in grassland and savanna: An analysis with simulated SENTINEL 2 data for a north American transect. Remote Sensing of Environment, 137, 94–111. http://dx.doi.org/10.1016/j.rse.2013.06. 004 (0). Huang, C., & Geiger, E. L. (2008). Climate anomalies provide opportunities for large-scale mapping of non-native plant abundance in desert grasslands. Diversity and Distributions, 14(5), 875–884. http://dx.doi.org/10.1111/j.1472-4642.2008.00500.x. Huang, C., Goward, S. N., Masek, J. G., Thomas, N., Zhu, Z., & Vogelmann, J. E. (2010). An automated approach for reconstructing recent forest disturbance history using dense Landsat time series stacks. Remote Sensing of Environment, 114(1), 183–198. http://dx.doi.org/10.1016/j.rse.2009.08.017. Huete, A. R. (1988). A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment, 25, 295–309. Karnieli, A., Bayarjargal, Y., Bayasgalan, M., Mandakh, B., Dugarjav, C., Burgheimer, J., ... Gunin, P. D. (2013). Do vegetation indices provide a reliable indication of vegetation degradation? A case study in the Mongolian pastures. International Journal of Remote Sensing, 34(17), 6243–6262. http://dx.doi.org/10.1080/01431161.2013.793865. Lentile, L. B., Holden, Z. A., Smith, A. M. S., Falkowski, M. J., Hudak, A. T., Morgan, P., ... Benson, N. C. (2006). Remote sensing techniques to assess active fire characteristics and post-fire effects. International Journal of Wildland Fire, 15(3), 319–345. Marsett, R. C., Qi, J., Heilman, P., Biedenbender, S. H., Watson, M. C., Amer, S., ... Marsett, R. (2006). Remote sensing for grassland management in the Arid Southwest. Rangeland Ecology & Management, 59(5), 530–540. http://dx.doi.org/10.2111/05-201R.1. McClaran, M. P., Angell, D. L., & Wissler, C. (2002). Santa Rita experimental range digital database: user's guide. Gen. Tech. Rep. RMRS-GTR-100. Ogden, UT: U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station (13 pp.). McGlone, C. M., & Huenneke, L. F. (2004). The impact of a prescribed burn on introduced Lehmann lovegrass versus native vegetation in the northern Chihuahuan Desert. Journal of Arid Environments, 57(3), 297–310. http://dx.doi.org/10.1016/S01401963(03)00109-5. M.L. Villarreal et al. / Remote Sensing of Environment 183 (2016) 186–197 McPherson, G. R. (1997). Ecology and management of North American Savannas. Tucson: The University of Arizona Press. Melaas, E. K., Friedl, M. A., & Zhu, Z. (2013). Detecting interannual variation in deciduous broadleaf forest phenology using Landsat TM/ETM + data. Remote Sensing of Environment, 132, 176–185. http://dx.doi.org/10.1016/j.rse.2013.01.011. Moran, M. S., Ponce-Campos, G. E., Huete, A., McClaran, M. P., Zhang, Y., Hamerlynck, E. P., ... Hernandez, M. (2014). Functional response of U.S. grasslands to the early 21st-century drought. Ecology, 95(8), 2121–2133. http://dx.doi.org/10.1890/13-1687.1. Nagler, P. L., Inoue, Y., Glenn, E. P., Russ, A. L., & Daughtry, C. S. T. (2003). Cellulose absorption index (CAI) to quantify mixed soil–plant litter scenes. Remote Sensing of Environment, 87(2–3), 310–325. http://dx.doi.org/10.1016/j.rse.2003.06.001. Neary, D. G., Klopatek, C. C., DeBano, L. F., & Ffolliott, P. F. (1999). Fire effects on belowground sustainability: A review and synthesis. Forest Ecology and Management, 122(1–2), 51–71. http://dx.doi.org/10.1016/S0378-1127(99)00032-8. Nouvellon, Y., Moran, M. S., Seen, D. L., Bryant, R., Rambal, S., Ni, W., ... Qi, J. (2001). Coupling a grassland ecosystem model with Landsat imagery for a 10-year simulation of carbon and water budgets. Remote Sensing of Environment, 78(1–2), 131–149. http:// dx.doi.org/10.1016/S0034-4257(01)00255-3. Omernik, J. M. (1987). Ecoregions of the conterminous United States. Annals of the Association of American Geographers, 77(1), 118–125. http://dx.doi.org/10.1111/j. 1467-8306.1987.tb00149.x. Palmer, W. C. (1965). Meteorological drought. Research paper no 45, United States weather bureau, Washington, DC. Parmenter, R. R. (2008). Long-term effects of a summer fire on desert grassland plant demographics in New Mexico. Rangeland Ecology & Management, 61(2), 156–168. http://dx.doi.org/10.2111/07-010.1. Peterson, E. B. (2005). Estimating cover of an invasive grass (Bromus tectorum) using tobit regression and phenology derived from two dates of Landsat ETM + data. International Journal of Remote Sensing, 26(12), 2491–2507. http://dx.doi.org/10. 1080/01431160500127815. Piñeiro, G., Oesterheld, M., & Paruelo, J. M. (2006). Seasonal variation in aboveground production and radiation-use efficiency of temperate rangelands estimated through remote sensing. Ecosystems, 9, 357–373. Polyakov, V. O., Nearing, M. A., Stone, J. J., Hamerlynck, E. P., Nichols, M. H., Holifield Collins, C. D., & Scott, R. L. (2010). Runoff and erosional responses to a drought-induced shift in a desert grassland community composition. Journal of Geophysical Research, 115(G4), G04027. http://dx.doi.org/10.1029/2010JG001386. Pyke, D. A., Brooks, M. L., & D'Antonio, C. (2010). Fire as a restoration tool: A decision framework for predicting the control or enhancement of plants using fire. Restoration Ecology, 18(3), 274–284. http://dx.doi.org/10.1111/j.1526-100X.2010. 00658.x. Qi, J., Marsett, R., Heilman, P., Bieden-bender, S., Moran, S., Goodrich, D., & Weltz, M. (2002). RANGES improves satellite-based information and land cover assessments in Southwest United States. Eos, Transactions American Geophysical Union, 83(51), 601–606. http://dx.doi.org/10.1029/2002EO000411. Robinett, D. (1992). Lehmann lovegrass and drought in southern Arizona. Rangelands, 14(2), 100–103 (Retrieved from http://agris.fao.org/agris-search/search/display.do? f=1994/US/US94089.xml;US9418466). Röder, A., Hill, J., Duguy, B., Alloza, J. A., & Vallejo, R. (2008). Using long time series of Landsat data to monitor fire events and post-fire dynamics and identify driving factors. A case study in the Ayora region (eastern Spain). Remote Sensing of Environment, 112(1), 259–273. http://dx.doi.org/10.1016/j.rse.2007.05.001. 197 Ruyle, G. B., Roundy, B. A., & Cox, J. R. (1988). Effects of burning on germinability of Lehmann lovegrass. Journal of Range Management, 41(5), 404–406. http://dx.doi.org/10. 2307/3899577. Sala, O., & Paruelo, J. (1997). Ecosystem services in grasslands. Nature's Services: Societal Dependence on Natural Ecosystems, 237–251. Samson, F., & Knopf, F. (1994). Prairie conservation in North America. Bioscience, 44(6), 418–421. http://dx.doi.org/10.2307/1312365. Sayre, N. F. (2003). Recognizing history in range ecology: 100 years of science and management on the Santa Rita experimental range. In M. P. McClaran, P. F. Ffolliott, & C. B. Edminster (Eds.), Santa Rita experimental range: 100 years (1903–2003) of accomplishments and contributions. Proceedings RMRS-P-30 (pp. 1–15). Fort Collins, Colorado, USA: U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station. Sayre, N., deBuys, W., Bestelmeyer, B., & Havstad, K. (2012). “The range problem” after a century of rangeland science: New research themes for altered landscapes. Rangeland Ecology & Management. http://dx.doi.org/10.2111/REM-D-11-00113.1. Schussman, H., Geiger, E., Mau-Crimmins, T., & Ward, J. (2006). Spread and current potential distribution of an alien grass, Eragrostis lehmanniana nees, in the southwestern USA: Comparing historical data and ecological niche models. Diversity and Distributions, 12(5), 582–592. Smith, M. D., & Knapp, A. K. (1999). Exotic plant species in a C4-dominated grassland: Invasibility, disturbance, and community structure. Oecologia, 120(4), 605–612. http://dx.doi.org/10.1007/s004420050896. Sonnenschein, R., Kuemmerle, T., Udelhoven, T., Stellmes, M., & Hostert, P. (2011). Differences in Landsat-based trend analyses in drylands due to the choice of vegetation estimate. Remote Sensing of Environment, 115(6), 1408–1420. http://dx.doi.org/10. 1016/j.rse.2011.01.021. Valone, T. J., & Kelt, D. A. (1999). Fire and grazing in a shrub-invaded arid grassland community: Independent or interactive ecological effects? Journal of Arid Environments, 42(1), 15–28. http://dx.doi.org/10.1006/jare.1999.0500. Van Leeuwen, W. J. D., Casady, G. M., Neary, D. G., Bautista, S., Alloza, J. A., Carmel, Y., ... Orr, B. J. (2010). Monitoring post-wildfire vegetation response with remotely sensed time-series data in Spain, USA and Israel. International Journal of Wildland Fire, 19(1), 75–93. Vogelmann, J. E., Xian, G., Homer, C., & Tolk, B. (2012). Monitoring gradual ecosystem change using Landsat time series analyses: Case studies in selected forest and rangeland ecosystems. Remote Sensing of Environment, 122, 92–105. http://dx.doi.org/10. 1016/j.rse.2011.06.027. White, R. P., Murray, S., & Rohweder, M. (2000). Pilot analysis of global ecosystems: Grassland ecosystems. (xi + 89 pp.). Wright, H. A. (1974). Range burning. Journal of Range Management, 27(1), 5–11. http://dx. doi.org/10.2307/3896428. Xu, D., Guo, X., Li, Z., Yang, X., & Yin, H. (2014). Measuring the dead component of mixed grassland with Landsat imagery. Remote Sensing of Environment, 142, 33–43. http:// dx.doi.org/10.1016/j.rse.2013.11.017. Yoder, J., Engle, D., & Fuhlendorf, S. (2004). Liability, incentives, and prescribed fire for ecosystem management. Frontiers in Ecology and the Environment, 2(7), 361–366. http://dx.doi.org/10.1890/1540-9295(2004)002[0361:LIAPFF]2.0.CO;2. Zhu, Z., & Woodcock, C. E. (2014). Continuous change detection and classification of land cover using all available Landsat data. Remote Sensing of Environment, 144, 152–171. http://dx.doi.org/10.1016/j.rse.2014.01.011\.