Deriving Water Content of Chaparral Vegetation from AVIRIS Data

www.elsevier.com/locate/sna Deriving Water Content of Chaparral Vegetation from AVIRIS Data Lydia Serrano,* Susan L. Ustin,† Dar A. Roberts,‡ John A. Gamon,§ and Josep Peñuelas* Spectral imaging data acquired with Advanced Visible Infrared Imaging Spectrometer over Point Dume (Los Angeles County, CA, USA) were used to assess the ability of hyperspectral reflectance data to estimate canopy Relative Water Content (RWC) at the landscape level. The study was performed on 23 vegetation stands comprised of three characteristic chaparral plant communities, with contrasting phenological stages and canopy cover. Several estimates of water content based on the near-infrared (NIR; reflectance indices and water thickness derived from reflectance and radiance data) and shortwave infrared (SWIR) water absorption bands were compared to measurements of vegetation structure and water content made on the ground. The Water Index (WI) and Normalized Difference Water Index (NDWI), reflectance indices formulated from the NIR water absorption bands, were the best indicators of canopy RWC estimated from combining leaf relative water content with measures of canopy structure. A stepwise multiple regression revealed that canopy structure explained 36% and 41% of the variation in WI and NDWI, respectively. The explained variance in WI and NDWI increased to 44% and 48% when leaf relative water content was included in the model. By contrast, the inclusion of leaf relative water content did not contribute significantly to the explained variance in indices formulated using SWIR water absorp- * Centre de Recerca Ecològica i Aplicacions Forestals, Universitat Autònoma de Barcelona, Barcelona, Spain † Department of Land, Air and Water Resources, University of California, Davis ‡ Department of Geography, University of California, Santa Barbara § Department of Biology and Microbiology, California State University, Los Angeles Address correspondence to Lydia Serrano, Universitat Autònoma de Barcelona, Facultat de Ciències, Centre de Recerca Ecològica i Aplicacions Forestals, 08193 Bellaterra, Barcelona, Spain. E-mail: [email protected] Received 18 November 1999; revised 10 May 2000. REMOTE SENS. ENVIRON. 74:570–581 (2000) Elsevier Science Inc., 2000 655 Avenue of the Americas, New York, NY 10010 tion bands and in those based on water thickness. The relationship between WI and the canopy RWC significantly improved when only data from plots with green vegetation cover .70% were considered (r250.88, p, 0.001). All the indices studied had an important structural component (as indicated by the strong correlation with NDVI), yet only the indices WI and NDWI additionally responded to water content. These results indicate that the WI and NDWI are sensitive to variations in canopy relative water content at the landscape scale. Elsevier Science Inc., 2000 INTRODUCTION Determination of vegetation water status from spectral reflectance has important applications in the fields of agriculture and forestry (Gao and Goetz, 1995). Water stress is one of the most common limitations to photosynthesis and plant primary productivity, and its measurement has importance for irrigation practices and in the drought assessment of natural communities (Peñuelas et al., 1993). Moreover, the water content of vegetation is a key factor in its susceptibility to fire (Chandler et al., 1983; Pyne et al., 1996). Besides the thermal region of the electromagnetic spectrum (8–14 lm), two spectral regions have been found to be useful for the detection of plant water content: the near-infrared region (NIR, 700–1,300 nm) and the shortwave infrared region (SWIR, 1,300–2,500 nm). Both regions exhibit liquid water absorption features, and these are evident in the reflectance spectra of a fully turgid leaf measured in the laboratory (Carter, 1991; Danson et al., 1992). Several studies have shown relationships between leaf water content and spectral reflectance at the liquid water absorption wavelengths in the SWIR (Hunt and Rock, 1989; Bowman, 1989; Danson et al., 1992) and the NIR region (Peñuelas et al., 1993; Gao, 1996; Peñuelas and Inoue, 1999). 0034-4257/00/$–see front matter PII S0034-4257(00)00147-4 Deriving Water Content of Chaparral Vegetation At the canopy level, several narrow-band indices have been developed to assess water content. Previous studies have shown linear relationships between leaf Relative Water Content (RWC) and the reflectance ratio R900/R970 (Water Index, WI) for complete canopies where Leaf Area Index (LAI) does not vary greatly (Peñuelas et al., 1996). Similar results were found by Shibayama et al. (1993) and Rollin and Milton (1998) using derivative analysis and ratio indices in the NIR region. Indices based on the SWIR region water absorption bands (Normalized Difference Infrared Index, NDII) have also provided good correlations with canopy water content (Hardisky et al., 1983). Other studies have found it difficult to estimate water content at the canopy level due to large reflectance variation among leaves at similar levels of water stress (Cohen, 1991) and associated with changes in canopy geometry resulting from turgor loss (Kimes et al., 1984), as well as due to the small reflectance differences at different levels of water stress (Riggs and Running, 1991). At increasingly larger spatial scales, remote sensing of vegetation water content from liquid water absorption features is more challenging. A major difficulty when estimating canopy water content from remotely sensed data comes from the fact that LAI and background area (soil and nonphotosynthetic components) vary spatially, temporally, and with sensor view angle. Additionally, remote sensing of vegetation water content using airborne and satellite sensors is affected by water vapor in the atmosphere, which prevents the measurement of reflectance at the liquid water absorption bands around 1,450 nm and 1,920 nm (Rollin and Milton, 1998). Some signal remains in the NIR region, where there is a wavelength separation of about 30 nm between peak absorption of water in vapor and liquid forms. Over the last decade, the development of high resolution spectrometers has permitted the separation of the water vapor in the atmospheric column from the liquid water of the surface (Green et al., 1989; Green et al., 1993; Gao and Goetz, 1990; Roberts et al., 1997). In airborne studies, several analytical techniques have been widely used to interpret water absorption features. Canopy water content has been derived from Advanced Visible Infrared Imaging Spectrometer (AVIRIS) surface reflectance data by inversion of radiative transfer models (Jacquemoud et al., 1995; Jacquemoud et al., 1996) and using spectral matching techniques (Gao and Goetz, 1995; Zhang et al., 1996; Roberts et al., 1998). In the latter approach, the absorption spectrum of green vegetation in the liquid water bands is usually expressed as the thickness of a sheet of water (or Equivalent Water Thickness, EWT). Roberts et al. (1997) showed the potential of the retrieved EWT (a product that results from the atmospheric correction of AVIRIS-measured radiance to apparent surface reflectance) to monitor temporal and spatial variations in water in herbaceous, shrub, and coniferous vegetation. Pinzon et al. (1998) used the 571 hierarchical foreground/background analysis, a cascade method regressing the full spectrum in the singular value decomposition space, and showed good predictions on leaves obtained from a wide range of vegetation types. Ustin et al. (1998b) showed that each of these methods could be used to estimate canopy water content of chaparral shrubs using the full AVIRIS reflectance spectrum. Moreover, significant seasonal differences in canopy water content could be detected in AVIRIS images (Ustin et al., 1998a; Ustin et al., 1998b; Ustin et al., 1999). However, some satellite sensors (such as Advanced Spaceborne Reflection Radiometer (ASTER) and Moderate Resolution Imaging Spectroradiometer (MODIS) do not have contiguous spectral coverage in the 0.4–2.5lm region, thus requiring the use of spectral feature mapping or ratio techniques (Gao, 1996). As a consequence, the use of narrow-band-based indices to assess water content at the landscape scale has gained interest. Based on AVIRIS imagery, Gao (1996) proposed monitoring of vegetation water content by combining reflectances at 860 nm (a reference wavelength) and 1,240 nm (a liquid water absorption band). The resulting Normalized Difference Water Index (NDWI) was found to be sensitive to changes in vegetation water content. WI, also called the Water Band Index (WBI), has been reported to be a robust index of water content at the leaf and canopy levels (Peñuelas et al., 1993; Peñuelas et al., 1996; Peñuelas et al., 1997; Peñuelas et al., 1999). Similar results were obtained when deriving WI values from combined leaf and forest reflectance models (Dawson et al., 1997; Dawson et al., 1999). Studies with AVIRIS and field spectrometers indicate that WI can differentiate changing water content associated with season and vegetation type in southern California scrub vegetation (Gamon et al., 1998; Gamon and Qiu, 1999). However, quantitative tests of this index against actual, field-measured water content are needed. We tested the ability of the NIR-based indices WI and NDWI derived from AVIRIS apparent surface reflectance spectra to determine vegetation water content at the landscape level for three distinct plant communities , under varying vegetation cover. We compared the WI and NDWI with surface water estimates (EWT and water thickness) and other reflectance indices sensing water content derived from AVIRIS spectral data in the SWIR region. We also aimed to evaluate the relative contributions of vegetation structure and water content to the reflectance signal. MATERIAL AND METHODS Field Site and Vegetation Description The study area was located in the Santa Monica Mountains (California, USA), a range trending east–west from the Los Angeles metropolitan area toward Ventura County, along the Pacific Coast (34859 N 1188 409 W). 572 Serrano et al. This region is characterized by a Mediterranean climate, having cool, wet winters and hot, dry summers. Three distinct chaparral communities were considered in this study: ceanothus chaparral (dominated by Ceanothus spp.), chamise chaparral (dominated by Adenostoma fasciculatum), and coastal sage scrub (dominated by Salvia and Eriogonum spp. and Artemisia californica; Hanes, 1988). These chaparral communities exhibit different patterns of water relations (Miller and Poole, 1979) and ignition/combustion of fuel (Barro and Conard, 1991). Ceanothus sp. are classified as drought tolerant and exhibit changes in leaf orientation as a response to seasonal loss of turgor (Comstock and Mahall, 1985). Adenostoma fasciculatum (a needle-leafed species) undergoes moderate seasonal senescence, while Salvia sp. is drought deciduous and exhibits pronounced seasonal leaf senescence and changes in canopy water content. Two sites representative of the previously mentioned plant communities were selected for canopy measurements: Zuma ridge and Castro Crest. Zuma ridge is located near the Pacific coast and is composed of coastal sage scrub and mixed chaparral vegetation (the latter not considered in this study), while Castro Crest has characteristic ceanothus and chamise chaparral vegetation. Five plots of ceanothus chaparral, eight plots of chamise chaparral, and ten plots with characteristic coastal sage scrub were studied. Stands of homogeneous vegetation, with uniform aspect and slope, were chosen in the field as the basic unit of study and their boundaries were traced on a map (Harrison, 1996), thus defining a series of polygons. Areas of individual polygons ranged from 4,300 m2 to over 29,000 m2. Transects were located within homogeneous polygons. The point-intercept transect method (California Native Plant Society field sampling protocol; Sawyer and Keeler-Wolf, 1995) was used to assess vegetation composition (and, thus, plant community) and structure. Transects were variable in length, ranging from 22 m to 55 m due to differences in accessibility and size of polygons, with point measurements every 1 m. At each point intercept, plant species was recorded, and height was measured to the nearest 0.1 m. When no vascular plant was hit, it was recorded as soil. Measurements of vegetation composition and structure were taken within weeks of the AVIRIS flight. The absolute cover of a plant species was determined as the percentage area of total polygon area it occupied, including nonvegetated areas, and species relative cover was calculated as the percentage of polygon vegetated area it occupied (Gordon and White, 1994). water content was determined by sampling five branches for each species having a relative cover larger than 10%. Branches were collected from the upper part of the canopy. Leaf water content was determined using leaf disks. Five replicates of five disks of 0.45 cm2 area each were punched on different leaves of each branch. Disks were cut using a cork borer. For needle-leafed species and for small leaves, the entire leaf was sampled. Disks (or leaves) were immediately placed on a preweighed vial filled with water and were perfectly sealed and kept in an ice chest until carried to the lab. Vials were stored at 58C in the dark for 72 hours to allow disks (or leaves) to regain full turgor. The weight of the vial with the sample inside was determined using an analytical balance (Mettler AT261 Delta Range, Mettler Instruments Corporation, Hightstown, NJ, USA) and the difference between this final weight and the initial weight (vial without sample) was recorded as Fresh Weight (FW). Afterward, disks were removed from the vial and the excess water was removed using a paper tissue, and full turgor weight (TW) was determined. Leaf disks were dried at 608C in an oven until constant weight to determine dry weight (DW). Leaf water content was expressed as RWC, and was calculated as shown in Eq. (1): Canopy Water Content Samples to determine leaf water content were taken within 4 days of the AVIRIS flight. Within each polygon, Vegetation RWC5[ o (RWCi3RCi)]3(C3H) RWC5(FW2DW)/(TW2DW) (1) Assessment of canopy relative water content (i.e., the water content resulting from the overall species present in a polygon) was conducted by weighing RWC of each species by its absolute cover (C) and height (as a surrogate of LAI) and summing the resulting values for all the species considered in a polygon. Thus [see Eq. (2)], i5n Canopy RWC5 o (RWCi*Ci*Hi) (2) i51 where RWCi is the leaf RWC for the species i, and Ci is the fraction of total area that the species i occupies in a polygon, and Hi the plant height. Due to time and labor constraints, only those species having an absolute cover larger than 10% were considered for each polygon. To evaluate the relative contributions of vegetation water content and vegetation structure to the AVIRIS derived indices, the Canopy RWC expression was reformulated (herein referred as Vegetation RWC) to isolate the structural and water components. Vegetation RWC resulted from combining an expression of RWC (obtained by weighing RWCi of each species by its relative cover, RCi) and a composite expression of vegetation structure [that is, the product of vegetation cover (C) and average canopy height (H) obtained from transect measurements]. Thus, Vegetation RWC was calculated as seen in Eq. (3): i5n (3) i51 The first term in this expression represents weighted Deriving Water Content of Chaparral Vegetation leaf RWC of the species, while the second term represents the canopy structure and is related to the volume of leaves within the canopy. Similar approaches using LAI values instead of height have been used in other studies to derive vegetation water content (Dawson et al., 1999). Table 1. Spectral Reflectance Indices and the Precise AVIRIS Bands Used in This Study AVIRIS Data AVIRIS data were acquired over the Santa Monica Mountains on 17 October 1996. AVIRIS collects spectra in 224 bands from 370 nm to 2500 nm with a sampling interval of 10 nm (Vane et al., 1993; Green et al., 1998). It was flown on an ER-2 aircraft at an elevation of approximately 20 km, collecting a cross-track swath of approximately 11 km, with a ground instantaneous field of view of 20 m. A typical scene consists of 614 samples, and 512 lines covering a 1139-km area. Data presented in this study were obtained from flight 961017B, run 4, scene 5, centered over Point Dume, Los Angeles County. AVIRIS data were radiometrically corrected and processed to retrieve apparent surface reflectance using a modified version of the MODTRAN radiative transfer code (Green et al., 1993). In this code, modeled radiance is fitted to upwelling measured radiance by using a nonlinear least-squares fitting technique. The fit is performed using 20 bands centered around 940 nm to correct for atmospheric water vapor absorption. In addition to water vapor, a simple surface reflectance model is used over the narrow spectral region of the fit. This model approximates surface reflectance as a function of liquid water absorption, reflectance intensity, and slope (dq/dk; see Roberts et al., 1997 for a detailed description). Output images of best-fit estimates of column water vapor and liquid water (herein called EWT and expressed in micrometers of a sheet of water covering a pixel) are generated in this process, and apparent surface reflectance is retrieved. The AVIRIS-retrieved surface reflectance and EWT scenes were registered by selecting ground control points using the georeferenced trail map (where the polygons were traced). Ground control points included road intersections and bends, coast borders, and other clearly defined sites on the image. Image registration was performed with ENVI (Research Systems Inc., Boulder, CO, USA) using linear equations, and resampling was based on the nearest neighbor method (to preserve the spectral values). The root mean square registration error was less than one pixel size (20320 m). Polygons traced in the base map were reconciled to the registered scenes. Averaged EWT values and reflectance spectra were extracted for each polygon. Then, reflectance indices were calculated from each polygon’s averaged reflectance spectrum. See Table 1 for description of the indices studied and the precise bands used for their formulation. 573 Index Formulation Reference NDVI WI NDWI Water Index EWT WT MSI NDII (R8952R675)/(R8951R675) R895/R972 (R8572R1241)/(R8571R1241) Peñuelas et al. (1993) Peñuelas et al. (1993) Gao (1996) R867 through R1049 R867 through R1088 R1599/R819 (R8192R1649)/(R8191R1649) Green et al. (1993) Roberts et al. (1998) Hunt and Rock (1989) Hardisky et al. (1983) In addition, we tested an alternative approach to retrieve liquid water (herein referred as Water Thickness, WT) from AVIRIS reflectance data. This approach was proposed by Roberts et al. (1998) and uses a spectral matching technique similar to that used for retrieving EWT from AVIRIS radiance data. Briefly, the natural log of reflectance is modeled as a linear function of the pure water absorption coefficient over the 867–1,088-nm region, and WT is derived from the slope of this function. WT estimates are reported in millimeters of a sheet of water covering a pixel. Statistics Data were analyzed using Pearson correlation and linear regression routines in StatView 4.5 (Abacus Concepts, Berkeley, CA, USA). To better understand the relative contributions of foliage water content and canopy structure to the overall signal detected remotely, we employed stepwise multiple regression. The stepwise multiple regressions were run in the forward mode with the several indices studied as dependent variables and “leaf” RWC and canopy structure as independent variables. Stepwise multiple regression and analysis of variance (ANOVA) were performed with Systat 5.2 (Systat Inc., Evanston, IL, USA). Reflectance data were plotted using Igor (Wavemetrics Inc., OR, USA) and KaleidaGraph 3.08 (Synergy Software, Reading, PA, USA). RESULTS AND DISCUSSION Plant Community Spectral Responses The AVIRIS scene of the Point Dume area is depicted as a three-band pseudocolor image (1,659 nm, 838 nm, and 666 nm: RGB) in Fig. 1. Vegetated areas with high water content appear greener to the eye. Conversely, areas of sparse and senescent vegetation are seen as a brownish-red color, more abundant nearby the coast, but also conspicuous at Castro Crest. These patterns are due to the presence of coastal sage scrub, which grows mainly nearby the coast and experiences senescence following summer drought, and chamise chaparral, which forms sparse canopies and grows further inland. Roads and trails are seen in gray and white. Characteristic spec- 574 Serrano et al. Figure 1. A false-color (1659 nm, 836 nm, and 668 nm: RGB) AVIRIS scene over Point Dume area (Los Angeles County, CA, USA). Sampling sites are labeled (Castro Crest and Zuma Canyon). Image-based characteristic spectral signatures for the three main plant communities are also depicted. tral signatures for the three major plant communities are also depicted in Fig. 1. Reflectance in the visible and NIR regions showed distinctive patterns among plant communities according to the degree of development and canopy greenness. Thus, the spectral signatures of evergreen shrubs (ceanothus and chamise chaparral) showed a sharp transition in the red-edge region (i.e., the transition between the red and the NIR region), which is characteristic of green vegetation. This contrast was more accentuated in ceanothus chaparral than in chamise chaparral due to higher average canopy height (and thus larger biomass) and canopy cover. In contrast, coastal sage scrub, as a result of summer senescence, showed a steady, almost linear increase in reflectance in the red-edge region, indicating that there was little green canopy cover in this fall scene. Differences in NIR reflectance at the liquid water absorption wavelengths were also conspicuous (Fig. 1). The spectral absorption feature at the liquid water absorption bands (relative to the nearby nonabsorbing re- gions) was larger in ceanothus chaparral than chamise chaparral, at both liquid water absorbing wavelengths in the NIR region (i.e., 970 nm and 1,240 nm). On the other hand, coastal sage scrub showed a minimal development of the absorption feature at these liquid water absorption bands, due to its extreme seasonal senescence. EWT values, derived from AVIRIS radiance data, showed a clear contrast among plant communities. These changes mainly respond to varying phenological stages. Coastal sage scrub polygons yielded average EWT around 800 lm (800634.3 lm, mean6SEM), according to the pronounced autumn senescence. In contrast, ceanothus and chamise chaparral showed larger EWT values (1,175653.7 lm and 1,4016135.9 lm, respectively). EWT values differed significantly among plant communities (ANOVA, p,0.0001) and were within the lower range of previously reported EWT values for the plant communities considered (October 1994; Ustin et al., 1998b) and agreed with those reported by Roberts et al. Deriving Water Content of Chaparral Vegetation 575 (1997) for senescent coastal sage scrub and chamise chaparral vegetation for the autumn season (October 1992). Thus, retrieved EWT effectively discriminated among plant communities. Similarly, all the reflectancebased indices sensing water content showed significant differences (ANOVA, p,0.0001) among the plant communities considered (data not shown, but see Fig. 2). In Table 2 a detailed description of the vegetation status, soil cover, and NDVI values for each polygon is provided. As for the indices sensing water status, NDVI values also showed significant differences among plant communities (ANOVA, p,0.0001). Average NDVI values were 0.5560.05 (SEM) in Ceanothus chaparral, 0.4636 0.02 (SEM) in chamise chaparral, and 0.36660.12 (SEM) in coastal sage scrub. AVIRIS-Derived Data and Canopy RWC Relationships When comparing AVIRIS-derived data with the actual canopy water content (field measurements), the strongest correlations with Canopy RWC were found for those indices derived from NIR region water-absorbing bands. The WI and NDWI showed similar merit in estimating Canopy RWC: there was a significant correlation between the reflectance-derived WI (r250.38, p,0.05; Fig. 2A) and NDWI (r250.44, p,0.001; Fig. 2B) and the Canopy RWC. Indices formulated using liquid-absorbing bands in the SWIR region also showed significant correlations with the Canopy RWC. NDII showed a slightly higher correlation with the Canopy RWC (r250.40, p,0.001; Fig. 2C) than the more widely used Moisture Stress Index (MSI; Hunt and Rock, 1989) (r250.30, p,0.05; Fig. 2D). EWT showed significant correlation with the Canopy RWC (r250.30, p,0.10; data not shown). This correlation improved when considering WT. Thus, WT was also significantly correlated with Canopy RWC (r250.36, p,0.05; Fig. 2E). Since both parameters EWT and WT are derived from the same water absorption feature, using similar approaches, discrepancies might be related to the robustness of the data. To address this potential effect, we calculated the coefficient of variation (CV) at each polygon (SD/mean) for retrieved EWT values and for reflectance data over the spectral range used to derive WT (i.e., across the 865-nm to 1,035-nm region). Then, we calculated the average CV for all the polygons studied. Assuming that both water estimates (i.e., EWT and WT) respond to the same surface properties, we Figure 2. Relationships of reflectance indices WI (A), NDWI (B), NDII (C), MSI (D), and WT (E), and the Canopy RWC. Canopy RWC has arbitrary units (see Material and Methods for more details). The determination coefficients (and significance) are indicated in each panel (n523). 576 Serrano et al. Table 2. Characteristic Attributes (Site, Plant Community Type, Percentage Soil Cover, and Vegetation Status) and Mean NDVI6SD for the Polygons Studied Site Castro Crest Castro Crest Castro Crest Castro Crest Castro Crest Castro Crest Castro Crest Castro Crest Castro Crest Castro Crest Castro Crest Castro Crest Zuma ridge Zuma ridge Zuma ridge Zuma ridge Zuma ridge Zuma ridge Zuma ridge Zuma ridge Zuma ridge Zuma ridge Zuma ridge Polygon Plant Community Soil Cover (%) 1 2 3 4 5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10 11 Ceanothus chaparral Chamise chaparral Chamise chaparral Ceanothus chaparral Chamise chaparral Ceanothus chaparral Chamise chaparral Ceanothus chaparral Chamise chaparral Ceanothus chaparral Chamise chaparral Chamise chaparral Coastal sage scrub Coastal sage scrub Coastal sage scrub Coastal sage scrub Coastal sage scrub Coastal sage scrub Chamise chaparral Coastal sage scrub Coastal sage scrub Coastal sage scrub Coastal sage scrub 95 94 94 93 67 100 67 96 67 91 69 72 97 91 89 94 86 97 94 97 100 84 83 Vegetation Status green green green green green green green green green green green green senescing senescing senescing senescing senescing senescing green senescing senescing senescing senescing (dry) (dry) (dry) (dry) (dry) (dry) (dry) (dry) (dry) (dry) NDVI (mean6SD) 0.4260.177 0.4960.068 0.4360.111 0.4960.095 0.4060.067 0.6560.045 0.4260.075 0.6960.069 0.5760.120 0.4960.084 0.4460.049 0.4460.069 0.3460.026 0.3360.040 0.3360.076 0.3760.059 0.3860.065 0.3460.049 0.5260.039 0.4560.058 0.4160.019 0.3660.034 0.3560.034 The number of pixels at each polygon ranged from 12 to 72. would expect similar average CV values. Within the spectral range used for deriving WT, the average CV was 12.34%, while for retrieved EWT average CV was 26.68%. Previous studies have shown that dark objects on the surface (Gao and Goetz, 1990; Schlapfer et al., 1998) and errors in wavelength calibration (Green et al., 1990) could promote large errors in water vapor estimates. Since EWT values are retrieved while simultaneously solving for water vapor (among other parameters), errors in the water vapor estimates could affect the EWT estimates. Moreover, errors in water vapor retrieval may translate into a factor of two error in EWT estimates (Roberts, unpublished data). Thus, in our study, reflectance data provided more robust estimates of surface properties when compared to radiance-retrieved EWT. However, within the context of this study, it is difficult to accurately assess all of the previously mentioned error sources. The correlations between the reflectance indices and the Canopy RWC improved slightly when polygons with vegetation cover ,70% were removed from analysis (data not shown). The increase in the correlation coefficients was of similar magnitude for all the reflectance indices considered in this study (r increased <0.10 over the reported values). Even though soil reflectance does not show liquid water absorption features centered at 970 nm and 1,240 nm (Bowker et al., 1985; Zhang et al., 1996), exposed soil may bias or diminish the relationship to canopy water content (Gao, 1996; Ustin et al., 1998b). Other sources of error might be due to the contribution of nonphotosynthetic components to canopy reflectance (van Leeuwen and Huete, 1996). Confounding effects due to background contribution (soil and nongreen canopy components) to the reflectance signal might not be the sole factor that introduced noise in these relationships. Errors associated with field sampling and the scaling approach probably contributed to the scatter. In our study, canopy height and vegetation cover were combined to obtain an estimator of plant biomass and, hence, as a rough approximation of LAI. Thus, canopy height was used as a surrogate of LAI. This assumption is fairly appropriate for homogeneous canopies with random leaf distribution, but might have promoted bias in canopies not meeting this assumption, particularly in those polygons with senescing vegetation, corresponding to coastal sage scrub community. In fact, the composite expression height3cover (H3C) was a fairly good estimator of green biomass, and hence of LAI, as indicated by its correlation with NDVI (r50.56, p,0.05, n513) in ceanothus and chamise chaparral polygons. In contrast, for those polygons with senescent vegetation there was no correlation between H3C and NDVI (r50.00). Thus, when polygons with green vegetation cover ,70% were disregarded, all correlations of the reflectance indices with the Canopy RWC significantly improved. The increase in the correlation coefficients was of greater magnitude for those reflectance indices derived from the NIR region (Figs. 3A and 3B) than for those derived from the SWIR region (Fig. 3D and 3E; r2 increased <0.45 for the NIR-based indices and <0.23 Deriving Water Content of Chaparral Vegetation 577 Figure 3. Relationships of the reflectance indices WI (A), NDWI (B), NDVI (C), NDII (D), MSI (E), WT (F), and the Canopy RWC. Canopy RWC has arbitrary units (see Material and Methods for more details). The coefficients of determination (and significance) are indicated in each panel (n59). Data correspond to polygons with ceanothus and chamise chaparral vegetation cover .70%. for the SWIR-based indices). Similarly, the correlation between the reflectance-derived WT and Canopy RWC slightly increased, although not significantly (Fig. 3F). Thus, a more robust estimator of LAI in senescent stands could have improved the relationships of the indices studied with the Canopy RWC across the different plant communities studied. Despite these potential problems associated with the scaling method, which affected the accuracy of the Canopy RWC estimates, the results reported were successful in comparing the relative ability of several remote measures of water content. An additional source of error could have resulted from the 4-day difference in time between the field measurements and the AVIRIS overflight. The lapse of time between actual leaf water content measurements and reflectance data acquisition could have introduced some variability in the relationships between the spectral indices and the canopy water status. However, this effect was presumably negligible because there was no intervening precipitation at this late stage in the summer drought. The reflectance indices for estimating water content considered in this study were highly intercorrelated. The correlation coefficients were always higher than 0.93 (p,0.001; data not shown). However, we have seen that the WI and NDWI showed a closer correlation with the Canopy RWC than other indices formulated using water absorption bands either in the NIR or SWIR regions. The sensitivity of the absorption features at 970 nm and 1,240 nm seems to be due to the weaker water absorption and, therefore, greater penetration of NIR radiation into the canopy than at longer water absorption wavelengths. Thus, the signal “saturates” less quickly, yielding reflectance readings that appear to be more linearly related to the total moisture content (Peñuelas et al., 1993). Distinguishing Structural Signals from Water Signals To gain further insight into the relationships between the AVIRIS-derived indices and the field-measured vari- 578 Serrano et al. Table 3. Results of the Stepwise Regression of Several Reflectance Indices Derived from AVIRIS Reflectance and the Composite Canopy Structural Variable (Canopy Height3Percentage Cover, HxC) and the Foliage Relative Water Content (RWC) NDVI WI NDWI MSI NDII WT EWT Regression Equation p (HxC) p (RWC) r SEE y50.31910.1093(HxC) y50.898510.03093(HxC)10.07043RWC y520.246810.05113(HxC)10.10193RWC y51.467120.25363(HxC) y520.243510.12903(HxC) y50.55610.3183(HxC) y5651.81314.33(HxC) ,0.001 ,0.001 ,0.001 ,0.01 ,0.01 ,0.01 ,0.01 n.s. 0.10 0.11 n.s. n.s. n.s. n.s. 0.58 0.66 0.69 0.59 0.63 0.61 0.54 0.081 0.019 0.028 0.181 0.082 0.213 260.3 SEE indicates standard error of the estimate (n523). See Table 1 for a description of the reflectance indices. EWT was derived from the atmospheric correction of AVIRIS radiance data. n.s. indicates not significant. ables, stepwise multiple regression was performed considering two independent variables: canopy structure and RWC at each polygon. In this analysis, Canopy RWC was reformulated as Vegetation RWC (see Material and Methods section). For EWT and WT estimates, the model uniquely included the structure as an independent variable and the explained variance was 29% and 38%, respectively. Similar results were obtained when considering the indices MSI and NDII formulated using SWIR bands, and the explained variance was 35% and 40%, respectively. When considering those indices derived from the NIR region, that is, WI and NDWI, the model also included RWC as a second independent variable (Table 3). The explained variance in WI increased from 36% to 44%, and from 41% to 48% in NDWI when RWC was added. Thus, besides indicating canopy structural characteristics, the indices formulated using water absorption bands in the NIR region provided additional information related to the foliage water status. These results are consistent with those obtained by Gamon et al. (1998) and Gamon and Qiu (1999), who found significant changes in the slope of the WI vs. NDVI relationships between spring and fall seasons in chaparral vegetation. However, they disagree with similar studies by Ustin et al. (1998b) and Roberts et al. (1999), which showed that the rela- tionship between measures of green vegetation (in this case the green vegetation fraction) and EWT also changed seasonally, showing a pattern suggestive of decreasing leaf water in fall relative to spring. The studies by Ustin et al. (1998b) and Roberts et al. (1999) can be reconciled with results presented here by considering the possibility that the sensitivity of EWT to RWC is too low based on a single date to be significant across plant communities (relative to structural differences). Seasonal changes in RWC are more striking, and thus produce a measurable change in EWT independent of structure. In contrast, the WI and NDWI appear to be sufficiently sensitive to detect differences in water status across contrasting structural types within a single date. When considering the relationships between the reflectance indices sensing canopy water status and NDVI, an indicator of canopy greenness and LAI (Myneni et al., 1995; Gamon et al., 1995; Verstraete et al., 1996), we observed contrasting responses according to the spectral regions considered in their formulation. All these relationships were linear, but differed in the degree of correlation. The WI and NDWI, formulated using NIR liquid water-absorbing wavelengths, showed correlation coefficients with NDVI<0.77 (p,0.01). The indices formulated using bands in the SWIR were also highly corre- Figure 4. Images of retrieved EWT (left), NDVI (center), and WI( right). EWT, NDVI, and WI were calculated from AVIRIS over Point Dume (Los Angeles County, CA, USA). Deriving Water Content of Chaparral Vegetation lated with NDVI. The correlation coefficients were r520.86 (p,0.001) and r50.88 (p,0.001) for MSI and NDII, respectively. However, all these previously mentioned reflectance-based indices were not as tightly correlated with NDVI as WT (r50.92, p,0.001), in agreement with previous studies that showed the strong structural regulation of WT expression derived from AVIRIS data (Roberts et al., 1998). The relative contribution of the water signal to the indices studied is also illustrated in Fig. 3, where reflectance indices sensing water content for polygons with green vegetation cover .70% are depicted vs. Canopy RWC. For this set of data, this relationship was stronger for the reflectance indices studied (see r2 values in Fig. 3) than that of NDVI (Fig. 3C; r250.35, p,0.10), indicating that these reflectance indices are directly sensing water content (relative to the structural contribution). Further support to the contribution of the water signal to the indices studied is given by its correlation with the residuals from the regression between reflectance indices sensing canopy water status and NDVI (i.e., the correlation between Canopy RWC and the previously mentioned residuals), in agreement with the previous stepwise regression results. The correlation coefficients were significant for the WI (r50.74, p50.023) and NDWI (r50.73, p50.027), slightly significant for the SWIRbased indices NDII (r50.64, p50.06) and MSI (r50.60, p50.09), and not significant for the WT (r50.4, p50.29). Thus, besides responding to the canopy structure, WI and NDWI were more sensitive to changes in Canopy RWC than those indices formulated using SWIR bands and WT. These results are also in agreement with those reported by Gao (1996), who found that the NDWI was able to detect variations in the canopy water content independently of NDVI. Also in accordance with these results, NDVI and EWT presented similar spatial patterns of variation in vegetated areas (Fig. 4, except for the city of Malibu at the Point Dume area). In contrast, WI did not parallel so closely the spatial patterns in NDVI, further confirming that WI partially sensed an independent vegetation condition than NDVI (Fig. 4). CONCLUSIONS Although a number of leaf and canopy scale studies exist, few studies have compared ratio techniques and physical methods to direct water content at the landscape level. All the indices sensing water status, as well as NDVI, effectively discriminated among plant communities. Through statistical techniques (stepwise regressions and analyses of residuals) we identified the canopy structure as the main source of signal variation in the reflectance indices studied. However, WI and NDWI appeared to respond not only to canopy structure but also to the canopy water content. Furthermore, WI and NDWI were 579 more independent of conventional vegetation indices (NDVI) than those formulated using SWIR liquid waterabsorbing bands. We conclude that the WI and NDWI are reliable indicators of relative water content not only at the leaf and canopy levels but also at the landscape scale. These indices could be useful tools to assess vegetation water status, particularly when using spectrometers without contiguous spectral coverage. The authors are grateful to J. Valeriano from CREAF for assistance in image processing. This research was supported by a National Aeronautics and Space Administration grant. The manuscript was completed under support by Spanish CICYT grants CLI97-0344 and AMB97-0332. A fellowship to L.S. from Ministerio de Educación y Cultura (Spain) is greatly appreciated. REFERENCES Barro, S. C., and Conard, S. G. (1991), Fire effects on California chaparral systems—An overview. Environment Int. 17: 135–149. Bowker, D. E., Davis, R. E., Myrick, D. L., Stacy, K., and Jones, W. T. (1985), Spectral reflectances of natural targets for use in remote sensing studies. NASA Ref. Pub. 1139, Washington DC. Bowman, W. D. (1989), The relationship between leaf water status, gas exchange, and spectral reflectance in cotton leaves. Remote Sens. Environ. 30:249–255. Carter, G. A. (1991), Primary and secondary effects of water content on the spectral reflectance of leaves. Am. J. Botany 78:916–924. Chandler, C., Cheney, P., Thomas, P., Trabaud, L., and Williams, D. (1983), Fire in Forestry, Vol. I. Forest Fire Behaviour Effects. Wiley, New York. Cohen, W. B. (1991), Temporal versus spatial variation in leaf reflectance under changing water stress conditions. Int. J. Remote Sens. 9:1865–1876. Comstock, J. P., and Mahall, B. E. (1985), Drought and changes in leaf orientation for two California chaparral shrubs: Ceanothus megacarpus and Ceanothus crassifolius. Oecologia 65:531–535. Danson, F. M., Steven, M. D., Malthus, T. J., and Clark, J. A. (1992), High-spectral resolution data for determining leaf water content. Int. J. Remote Sens. 13:461–470. Dawson, T. P., Curran, P. J., North, P. R. J., and Plummer, S. E. (1997), The potential of understanding the biochemical signal in the spectra of forest canopies using a coupled leaf and canopy model. In Physical Measurements and Signatures in Remote Sensing (G. Guyot and T. Phulpin, Eds.), Balkema, Rotterdam, pp. 463–470. Dawson, T. P., Curran, P. J., North, P. R. J., and Plummer, S. E. (1999), The propagation of foliar biochemical absorption features in forest canopy reflectance: a theoretical analysis. Remote Sens. Environ. 67:147–159. Gamon, J. A., Field, C. B., Goulden, M. L., Griffin, K. L., Hartley, A. E., Joel, G., Peñuelas, J., and Valentini, R. (1995), Relationships between NDVI, canopy structure, and photosynthesis in three Californian vegetation types. Ecological Applications 5:28–41. 580 Serrano et al. Gamon, J. A., Lee, L.-F., Qiu, H.-L., Davis, S., Roberts, D. A., and Ustin, S. L. (1998), A multi-scale sampling strategy for detecting physiologically significant signals in AVIRIS imagery. In Summaries of the 7th Annual JPL Earth Science Workshop, 12–16 January. Pasadena, CA, JPL publication 97–21, Vol. 1, pp. 111–120. Gamon, J. A., and Qiu, H.-L. (1999), Ecological applications of remote sensing at multiple scales. In Handbook of Functional Plant Ecology (F. I. Pugnaire and F. Valladares, Eds.), Marcel Dekker, Inc., New York, pp. 805–846. Gao, B.-C. (1996), NDWI—A normalized difference water index for remote sensing of vegetation liquid water from space. Remote Sens. Environ. 58:257–266. Gao, B.-C., and Goetz, A. F. H. (1990), Column atmospheric water vapor and vegetation liquid water retrievals from airborne imaging spectrometer data. J. Geophys. Res. 95: 3549–3564. Gao, B.-C., and Goetz, A. F. H. (1995), Retrieval of equivalent water thickness and information related to biochemical components of vegetation canopies from AVIRIS data. Remote Sens. Environ. 52:155–162. Gordon, H. J., and White, T. C. (1994), Ecological Guide to Southern California Chaparral Plant Series. Technical publication R5-ECOL-TP-005 USDA, Forest Service, Pacific Southwest Region, San Francisco, CA. Green, R. O., Carrere, V., and Conel, J. E. (1989), Measurement of atmospheric water vapor using the Airborne Visible/ Infrared Imaging Spectrometer. Proc. ASPRS conference on Image Processing, Reno, Nevada, 23–26 April. Green, R. O., Conel, J. E., Margolis, J. S., Carrere, V., Bruegge, C. J., Rast, M., and Hoover, G. (1990), In-Flight Validation and Calibration of the Spectral and Radiometric Characteristics of the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS), SPIE Vol. 1298, Imaging Spectroscopy of the Terrestrial Environment, pp. 18–36. Green, R. O., Conel, J. E., and Roberts, D. A. (1993), Estimation of aerosol optical depth and additional atmospheric parameters for the calculation of apparent reflectance from radiance measured by the airborne visible/infrared imaging spectrometer. In: Summaries of the 4th Annual JPL Airborne Geoscience Workshop, 25–29 October, Vol. 1, AVIRIS workshop, Washington, DC, pp. 73–76. Green, R. O., Eastwood, M. L., Sarture, C. M., Chrien, T. G., Aronsson, M., Chippendale, B. J., Faust, J. A., Pavri, B. E., Chovit, C. J., Solis, M., Olah, M. R., and Williams, O. (1998), Imaging Spectroscopy and the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). Remote Sens. Environ. 65:227–248. Hanes, T. L. (1988), California chaparral. In Terrestrial Vegetation of California (M. G. Barbour and J. Major, Eds.), California Native Plant Society, Special Publication 9, pp. 417–470. Hardisky, M. A., Lemas, V., and Smart, R. M. (1983), The influence of soil salinity, growth form, and leaf moisture on the spectral reflectance of Spartina alternifolia canopies. Photogramm. Eng. Remote Sens. 49:77–83. Harrison, T. (1996), Trail Map of the Santa Monica Mountains Central. Tom Harrison Cartography, San Rafael, CA, USA. Hunt, E. R., Jr., and Rock, B. N. (1989), Detection of changes in leaf water content using near- and middle-infrared reflectances. Remote Sens. Environ. 30:43–54. Jacquemoud, S., Baret, F., Andrieu, B., Danson, F. M., and Jaggard, K. (1995), Extraction of vegetation biophysical parameters by inversion of the PROSPECT1SAIL models on sugar beet canopy reflectance data. Application to TM and AVIRIS sensors. Remote Sens. Environ. 52:163–172. Jacquemoud, S., Ustin, S. L., Verdebout, J., Schmuck, G., Andreoli, G., and Hosgood, B. (1996), Estimating leaf biochemistry using the PROSPECT leaf optical properties model. Remote Sens. Environ. 56:194–202. Kimes, D. S., Newcomb, W. W., Schut, J. B., Pinter, P. J., and Jackson, R. D. (1984). Directional reflectance factor distribution of a cotton row crop. Int. J. Remote Sens. 5:263–277. Miller, P. C., and Poole, D. K. (1979), Patterns of water use by shrubs in southern California. Forest Science 25:84–97. Myneni, R. B., Hall, F. G., Sellers, P. J., and Marshak, A. L. (1995), The interpretation of spectral vegetation indices, Institute of Electrical Electronics Engineering. Trans. Geosci. Remote Sens. 33:481–486. Peñuelas, J., Filella, I., Biel, C., Serrano, L., and Savé, R. (1993), The reflectance at the 950–970 nm region as an indicator of plant water status. Int. J. Remote Sens. 14: 1887–1905. Peñuelas, J., Filella, I., Serrano, L., and Savé, R. (1996), Cell wall elasticity and Water Index (R970 nm /R900 nm) in wheat under different nitrogen availabilities. Int. J. Remote Sens. 17:373–382. Peñuelas, J., and Inoue, Y. (1999), Reflectance indices indicative of changes in water and pigment contents of peanut and wheat leaves. Photosynthetica 36(3):355–360. Peñuelas, J., Piñol, J., Ogaya, R., and Filella, I. (1997), Estimation of plant water concentration by the reflectance Water Index WI (R900/R970. Int. J. Remote Sens. 18:2869–2875. Pinzon, J. E., Ustin, S. L., Castaneda, C. M., and Smith, M. O. (1998), Investigation of leaf biochemistry by hierarchical foreground/background analysis. IEEE Trans. Geosci. Remote Sens. 36:1–15. Pyne, S. J., Andrews, P. L., and Laven, R. D. (1996), Introduction to Wildland Fire, 2d ed. John Wiley and Sons, Inc., New York. Riggs, G. A., and Running S. W. (1991), Detection of canopy water stress in conifers using the airborne imaging spectrometer. Remote Sens. Environ. 35:51–68. Roberts, D. A., Brown, K., Green, R. O., Ustin, S. L., and Hinckley, T. (1998), Investigating the relationships between liquid water and leaf area in clonal Populus. In: Summaries of the 7th Annual JPL Earth Science Workshop, 12–16 January. Pasadena, CA, JPL publication 97–21, Vol. 1, pp. 335–344. Roberts, D. A., Dennison, P. E., Morais, M., Gardner, M. E., Regelbrugge, J., and Ustin, S. L. (1999), Mapping wildfire fuels using imaging spectrometry along the wildland urban interface. Proc. 1999 Joint Fire Science Conference and Workshop, 17–19 June. Boise, ID, USA. Roberts, D. A., Green, R. O., and Adams, J. B. (1997), Temporal and spatial patterns in vegetation and atmospheric properties from AVIRIS. Remote Sens. Environ. 62:223–240. Rollin, E. M., and Milton, E. J. (1998), Processing of high spectral resolution reflectance data for the retrieval of canopy water content information. Remote Sens. Environ. 65:86–92. Sawyer, J. O., and Keeler-Wolf, T. (1995), A Manual of Califor- Deriving Water Content of Chaparral Vegetation nia Vegetation. California Native Plant Society, Sacramento, CA. Schlapfer, D., Borel, C., Keller, J., and Itten, K. I. (1998), Atmospheric precorrected diffrential absorption technique to retrieve columnar water vapor. Remote Sens. Environ. 65: 353–366. Shibayama, M., Takahasi, W., Morinaga, S., and Akiyama, T. (1993), Canopy water deficit detection in paddy rice using a high resolution field spectroradiometer. Remote Sens. Environ. 45:117–126. Ustin, S. L., Roberts, D. A., Green, R. O., Zomer, R. J., and Garcia, M. (1999), Remote sensing methods monitor natural resources. Photonics Spectra 33(N10):108–113. Ustin, S. L., Roberts, D. A., and Hart, Q. J. (1998a), Seasonal vegetation patterns in a California coastal savanna derived from Advanced Visible/Infrared Imaging Spectrometer (AVIRIS) data. In Remote Sensing Change Detection: Environmental Monitoring Applications and Methods (C. D. Elvidge and R. Lunetta, Eds.), Ann Arbor Press, Ann Arbor, MI, pp. 163–180. 581 Ustin, S. L., Roberts, D. A., Pinzón, J., Jacquemoud, S., Gardner, M., Scheer, G., Castañeda, C. M., and Palacios-Orueta, A. (1998b), Estimating canopy water content of chaparral shrubs using optical methods. Remote Sens. Environ. 65: 280–291. Vane, G., Green, R. O., Chrien, T. G., Enmark, H. T., Hansen, E. G., and Poter, W. M. (1993), The Airborne Visible Infrared Imaging Spectrometer. Remote Sens. Environ. 44: 127–143. van Leeuwen, W. J. D., and Huete, A. R. (1996), Effects of standing litter on the biophysical interpretation of plant canopies with spectral indices. Remote Sens. Environ. 55: 123–138. Verstraete, M. M., Pinty, B., and Myneni, R. B. (1996), Potential and limitations of information extraction on the terrestrial biosphere from satellite remote sensing. Remote Sens. Environ. 58:201–214. Zhang, M., Ustin, S. L., Rejmankova, E., and Sanderson, E. W. (1996), Remote sensing of salt marshes: Potential for monitoring. Ecological Applications 7(3):1039–1053.