ANALYSIS OF BACKGROUND VARIATIONS IN COMPUTED SPECTRAL VEGETATION INDICES AND ITS IMPLICATIONS FOR MAPPING MANGROVE FORESTS USING SATELLITE IMAGERY

Transcription

1 ANALYSIS OF BACKGROUND VARIATIONS IN COMPUTED SPECTRAL VEGETATION INDICES AND ITS IMPLICATIONS FOR MAPPING MANGROVE FORESTS USING SATELLITE IMAGERY Beata D. BATADLAN 1, Enrico C. PARINGIT 2, Jojene R. SANTILLAN 2 Alexander S. CAPARAS 2 and John Louie FABILA 2 1 Remote Sensing & Resource Data Analysis Department, National Mapping & Resource Information Authority, Taguig City, Philippines 2 Department of Geodetic Engineering, College of Engineering, University of the Philippines, 1101 Diliman, Quezon City, Philippines Abstract : This study investigates the effects of various background conditions typically found in mangrove communities on the relationships on various spectral vegetation indices and leaf area index. ASTER image was used to identify/classify mangrove areas within the study area. Mangrove canopy spectral reflectance and leaf area index (LAI) were measured in the field using a spectrometer and Photosynthetically Active Radiation (PAR) sensor. These data were then used to calculate Spectral Vegetation Indices such as RVI, NDVI, SAVI, SAVI2, PVI, OSAVI, TSAVI and DVI. The relationships between several spectral vegetation indices measured from the field and derived from ASTER image and field LAI have been assessed particularly the effects of background variation typically found beneath mangrove canopies. As expected soil influences are prevailing in partially vegetated canopies, they are more significant in LAI below 1.5. Based on the correlation coefficient, the Vegetation Indices which consider soil parameter normalized the soil-background effects such as SAVI 2, OSAVI, TSAVI and SAVI with corresponding regression coefficient of 0.81, 0.74, 0.73 and 0.71 respectively. Keywords: Mangrove, Substrate variation, Spectral Vegetation Indices, Leaf Area Index 1. INTRODUCTION Mangroves are biologically diverse and fragile coastal ecosystems that are severely threatened by human activities in the coastal zone. While there is an urgent need to manage restore and rehabilitate remaining mangrove areas, these interventions are either short-sighted or severely constrained because of insufficient information on the current biophysical conditions of mangrove ecosystems (Primavera et al, 2005). The use of remote sensing in spatial prediction and modelling of vegetation biophysical properties are prominently used in assessing terrestrial forest but limited in mangroves (Rasolofoharinoro et. al, 1998; Saito et al, 2003). Vegetation indices (VIs) derived from satellite data are one of the primary sources of information for operational monitoring of the Earth s vegetative cover. These indices are radiometric measures of the spatial and temporal patterns of vegetation photosynthetic activity that are related to canopy biophysical variables such as leaf area index (LAI). LAI is define as half the total area per unit ground surface area which is an important parameter needed for many physiological and ecosystem studies. Most vegetation indices combined information contained in two spectral bands: red and near-infrared. Some VIs have been designed to be sensitive to vegetation but insensitive to other variables affecting the remotely sensed signal. These VI s do not involve any external factors other than the measured reflectance, such as Normalized Difference Vegetation Index (NDVI), Ratio Vegetation Index (RVI) and Difference Vegetation Index (DVI) (Rouse 1974, Jordan, 1969 and Tucker, 1979). These indices have performed well in many applications but were also found to have limitations because of their sensitivity to different substrates (Huete, 1988). Several soil adjusted vegetation indices are developed to minimize and reduce soil background effects particularly at sparse vegetation and low leaf area index such as Soil Adjusted Vegetation

2 Index (SAVI), SAVI2, Transformed SAVI (TSAVI), Optimized SAVI (OSAVI) and Perpendicular Vegetation Index (PVI) (Huete 1988, Major et al., 1990, Baret 1991; Richardson 1977). Vegetation indices formula and description were presented in table 1. Mangrove environments are subject to a wider range of variation in background conditions and over a tidal cycle at a specific location. Aside from being inundated during high tide and dry during low tide mangrove also grow in various type of substrate such as white sand and dark muddy soil. Therefore, it is important to systematically investigate the influence of variations in background reflectance properties, created by different background types and changes in moisture content or inundation, on the relationships between spectral indices and canopy biophysical properties in order to identify robust predictive approaches (Phinn, et al. 2003). The influences of substrate conditions on mangroves have been previously tested in simulated conditions (Diaz and Blackburn., 2003) but in real world mangrove forests or stands. The objectives of this study are: to characterize the behaviour of spectral vegetation indices for mangrove canopies in the presence of various background types; to identify the approach or vegetation index which would appear to be the most robust for estimating canopy biophysical properties of mangrove forests in an operational remote sensing scenario for monitoring mangroves; and to understand the effect of variations in background moisture or inundation on vegetation indices. The study area is located in Taklong Island National Marine Reserve (TINMAR) in Nueva Valencia, province of Guimaras. The geographic location of the area lies within 10º24 to 10º26 N latitude and 122º29 to 122º31.15 E longitude. Table 1 Formula for the spectral vegetation indices investigated in this study. Parameters a and b refer to the slope and intercept values derived from field measurement and from ASTER image: in this study a , b for in situ measurement and a , b for ASTER image Index Formula Description References RVI NIR Ratio Vegetation Index. The use of band ratio to Jordan 1969 eliminate various albedo effects NDVI NIR Normalized Difference Vegetation Index. Related Rouse et al. to changes in amount of green biomass, pigment (1973) NIR+ content and concentration and leaf water stress etc. DVI NIR Difference Vegetation Index Tucker 1979 PVI SAVI SAVI2 OSAVI NIR a b 2 1+ a NIR (1 + L) NIR+ + L L0.5 NIR + b / a NIR + NIR+ + L where L was set to 0.16 ( 1 L) TSAVI a( NIR ar b) R+ anir ab+ + a (1 ) Perpendicular Vegetation Index, orthogonal to the soil line. Attempts to eliminate differences in soil background and is most effective under conditions of low LAI Soil Adjusted Vegetation Index. Minimizes soil brightness-induced variations. L0.5 can reduce soil noise problem Soil Adjusted Ratio Vegetation Index Optimization of Soil Adjusted Vegetation Index, to optimize the adjustment factor for general applications resulted in a recommended adjustment factor of 0.16, rather than 0.5.of SAVI Transformed Soil Adjusted Vegetation Index to compensate for soil variability due to changes in solar elevation and canopy structure. Richardson & Wiegand 1977 Huete et. al Major et. al 1990 Rondeaux et al. (1996) Baret and Guyot 1991; Baret et al. 1989

3 2. METHODOLOGY 2.1 Image Processing An image taken by Advance Spaceborne Thermal Emission and Reflection Radiometer (ASTER) on April 13, 2005 were used in this study. ASTER has a total 14 bands 3 in VNIR, 6 in SWIR and 5 in TIR with 15, 30, 90 meters resolution respectively but only VNIR and SWIR were used in this study. Prior to image pre-processing layer stacking were done to combine VNIR and SWIR bands to build a new multiband file with common output resolution of 15 meters. The exact boundary of the study area were masked on the image to sub-set the area of interest. Then the masked image was digitally classified to separate mangrove from nonmangrove (Fig. 1). Spectral vegetation indices (Fig. 2) were extracted from the ASTER image using the formula listed in table 1. Soil-line parameter were also constructed from identified bare soil in the image prior to calculation of the VI s. 2.2 Field Data Collection Mangrove trees in different background conditions were subjected for simultaneous measurement of canopy spectra and light flux density. Predetermined sampling sites were selected based on various soil background, canopy closures and other mangrove conditions. Reflectance spectra were measured within each selected sample above the canopy using Ocean Optics Spectrometer. The system detects and records data with spectral range from 345 nm to 1047 nm. To measure the radiance, a fiber optics was attached to the sensor and was positioned just above the canopy. The spectrometer is connected to a laptop computer, which initiates the scanning procedure, graphically displays the reflectance values and stores the reflectance data. Prior to actual measurement, the spectrometer was calibrated using a white reference panel. This reference panel was measured twice at each sampling site before and just after the measurement. The average value was used in calculation of reflectance. Measurements of target radiance were repeated at least twice at each sampling site and the average value was used for the calculation. Data were collected close to solar noon, between 9:30 AM to 2:30 PM. All the gathered data were converted to Microsoft Excel format and the reflectance were calculated using Lcanopy R (1) L panel where R is the canopy reflectance, L canopy is the measured normal radiance above canopy (average), and L panel is the radiance measured for the calibration panel. The measured spectral reflectance was plotted using Microsoft Excel, reflectance on the X-axis and wavelength on Y- axis. The average values of wavelength that corresponds ASTER band 1, 2, 3 (Green, Red, NIR) was computed. This was used for the computation of spectral vegetation indices. Photons flux density was measured using PAR detector (Onset Applications). This detector was designed to detect photons between nm in wavelength. The detector was mounted on a bracket with a pole and the cable was connected to data logger (HOBO TM by Onset Application). All the measured data were stored in the data logger. The instrument was positioned vertically outside the mangrove canopy to measure photon flux density on the top of the canopy (I o ), then after several seconds the detector was positioned just about a meter above the ground/substrate to measure the photon flux density beneath the canopy (I). The time of measurements, location, canopy closures, and substrate conditions was noted in a field form. LAI was calculated using the following formula (Price 1995): LAI ( I I ) ln π k cos θ (2) where: LAI leaf area index (m 2 leaf area m -2 ground area) I photon flux density beneath the canopy I o photon flux on the top of the canopy k canopy light extinction coefficient θ Sun Zenith Angle (calculated from time, date, location) in degrees. Fig. 1. Set-up of instrument for simultaneous measurement of canopy spectral and light flux density. 2.3 Soil Line Calculation Spectral reflectance values of various bare substrates such as: inundated at various levels, white sand and dark sand were measured from the field. From these data it is possible to define the

4 soil line concept which is the line representing the relationship between and NIR substrate reflectance as shown in Fig. 4. The slope and intercept of this soil line was used to define the coefficients required for calculating the soiladjusted vegetation indices, which have maximum sensitivity when these coefficients are specific to the backgrounds of the vegetation under investigation. 2.4 Vegetation Isolines The reflectance of Red and NIR bands were plotted to retrieve their relationship. Linear regression lines were fitted through the points corresponding to each level of LAI, this lines are called vegetation isolines. The slopes and intercepts of vegetation isolines in the R-NIR plane are related to the optical properties of the canopy medium. The bare soils the slope and intercept was used to define the coefficient required for calculating various soil adjusted vegetation indices. 3. RESULTS AND FINDINGS 3.1 Image Processing Using a subset of the ASTER image (Fig. 2) within the study area Image classification result in Fig. 3 shows that mangrove areas were identified and delineated successfully from non-mangrove. A total of 60 hectares were present within this marine reserve. Spectral vegetation indices (RVI, NDVI, DVI, PVI, SAVI, SAVI 2, TSAVI and OSAVI) were also calculated from this image Field Data Collection Results Field data collection was conducted last January 28 to 30, 2009 and April 20 to 25, The result of spectral measurement of water at various level of inundation was plotted on a graph in figure 3. It was observed that as water depths increases the reflectance value decreases or the shallower the water the higher the reflectance. The ratio between spectral bands (such as and NIR), is independent from soil moisture or water depth. The spectral values of red and NIR were then used to calculate soil line together with other bare soil reflectance. 3.3 Measurement of Canopy Spectra & Light Flux Density The results of light flux density measurement and computed LAI are shown in Table 2. Spectral vegetation indices were computed based spectral signatures measured from the field using spectrometer. The spectral values of Red and NIR bands were achieved by computing the average reflectance value within a wavelength range of an ASTER image (Red nm, NIR nm) as shown in Fig. 4. Table 3 shows the results of these vegetation indices MAP SHOWING MANGROVE AREAS of TINMAR Nueva Valencia Guimaras W N S E N W E S N W E S Legend: Mangrove Sand bars Non-mangrove Water body Meters Fig. 3. Classified image showing mangrove areas Meters Fig. 2. ASTER Image RGB 321, covering Taklong Island National Marine Reserve

5 Reflectance (a) GREEN NIR dry white sand wet white sand 10 cm depth 23 cm depth 56 cm depth 86 cm depth 100 cm depth (b) Fig. 4. Spectral reflectance curve of water (a) at various level of inundation including wet and dry sand with (b) its equivalent R-G-NIR band value. 3.4 Soil Line Calculation Red and NIR reflectance of all types of bare substrate were plotted as shown in Fig. 5. A regression line was fitted and result show that correlation coefficient R The slope and intercept which was used for calculating some VIs were a and b NIR y x R² Fig. 5. Soil Line with slope and intercept a and b used for computation of some vegetation indices. NIR Reflectance R² R² R² R² R² Reflectance y x y x y x y x y x LAI Fig. 6. NIR versus Red reflectance for canopies within different ranges of LAI. Table 2. The computed value of leaf area index (LAI) measured from the field using PAR sensor. Site Over Canopy Below Canopy LAI Vegetation Isolines Regression lines were fitted on NIR and Red reflectance for mangrove canopy and bare soil in figure 6. It can be noticed that with increasing amounts of vegetation or as LAI increases the lines shift upward towards a higher NIR and a lower red reflectance. The lines also decrease in length with the bare soil baseline being the longest.

6 Table 3. Results of spectral vegetation indices derived from feld spectral measurements. # SAV RVI NDV SAV OS DVI TSA PVI correlation coefficient results were summarized in Table 4 and illustrated in Fig. 7. NDVI exhibit the highest correlation value followed by PVI and SAVI 2 with R , and respectively. NDVI exhibit best result in this case compare to analysis based on field measured spectra. 3.8 Mangrove Canopy Reflectance at Various Type of Substrate Spectral signatures of mangrove canopy of the same type in terms of canopy closure sparsity over different types of substrate. Its reflectances were evaluated to illustrate how substrate variations affect the overall reflectance of mangrove canopy. Fig. 8 shows mangrove canopy spectral reflectance over different types of substrate. It can be noticed that the behavior of spectral reflectance of mangrove canopy were affected by the properties of substrate. This might be the case that mangroves at low LAI have the tendency to be misclassified due to background or substrate optical properties (Van Leeuwen and Huete 1996). 3.6 Correlation Analysis LAI and Field VIs With the Red and NIR reflectance obtained from the field and the soil coefficient values, all the vegetation indices were calculated and Table 3 shows the results. The VI s values were then correlated with LAI measured from the field and its corresponding plots were shown in Fig. 6. It can be observed that values of vegetation indices increases with LAI. A polynomial regression was fitted in each plot to show correlation between VIs and LAI. The result shows that SAVI 2, OSAVI, TSAVI and SAVI shows a higher correlation results with corresponding regression coefficient of 0.81, 0,74, 0.73 and 0.71 respectively. These VI s were developed to minimize soil background effect. As expected soil influences are prevailing in partially vegetated canopies, they are more significant in LAI below 1.5. In the case of SAVI 2 which got the best VI for lower LAI (below 1.5) it introduced larger noise for higher LAI (above 2). Fig. 7. A polynomial regression fit to show relation between field leaf area index (LAI) and spectral vegetation indices from the field. 3.7 Correlation Analysis LAI and ASTER VIs Vegetaion indices map were constructed based on ASTER bands using the same formulation in Table 1. Soil parameter for some VI formula were obtained from soil line constructed based on bare soil/substrate in the image. The geographic locations of all the 28 sites that were measured in the field were plotted on these vegetation indices maps. The relationship between VI s values and LAI measured from the field were analyzed and

7 (a) (b) Fig. 8. Spectral curve of mangrove canopy of the same type over different substrate (a) Low density, (b). Medium density) Table 4. Comparison of Correlation Coefficient R 2 results of LAI versus VI's derived from "in situ" measurement spectra and ASTER image spectra. Vegetation Correlation Coefficient (R 2 ) Indices "in situ" ASTER SAVI OSAVI TSAVI SAVI RVI PVI NDVI DVI CONCLUSION AND RECOMMENDATION The relationships between several spectral vegetation indices and LAI has been assessed particularly the effects of background variation typically found beneath mangrove canopies. It was found that based on coefficient of determination using data from the field for all substrate type and LAI, the effects of background were slightly more pronounced for DVI, PVI, and RVI. Those that were considered to be moderately affected were SAVI and NDVI and the least affected were OSAVI and TSAVI which consider soil/substrate as parameter. While SAVI 2 which got the highest correlation coefficient of (R ), seems to be potential tool to assess mangrove LAI. While examining the vegetation indices derived from Aster image which were also correlated with field LAI, NDVI, PVI and SAVI 2 appears to be highly correlated with LAI with almost equal correlation coefficient of (R , 0.707, respectively). Among those indices that were obtained from both ASTER and in situ spectra, only SAVI 2 appears to be highly correlated with field LAI in both cases. Therefore SAVI 2 turn out to be the ideal or the most robust VI which can be used to calibrate LAI from ASTER image. The independency of a VI to substrate optical properties can be improved if one can adjust to a specific soil line of the target area. Further work are needed to test these findings on other field situation where other factors such as the contribution of woody tissues to the composite canopy response and the effects of sun/sensor geometry and the atmosphere on a remotely sensed signal need to be investigated. ACKNOWLEDGEMENTS This research was partly supported by the first author s Development of Geospatial Techniques to Analyze Impacts of Oil Spills on Coastal Resource and Environment Grant-in-Aid Project from Philippine Council for Advanced Science and Technology of the Department of Science and Technology (PCASTRD-DOST) and the first author s thesis grant from the PCASTRD s Local Graduate Scholarship. The authors also wish to acknowledge Dr. Resurreccion Sadaba, Dr. Nestor Yunque and Mr. Abner Barnuevo of UP in the Visayas (UPV) for their accommodation, assistance and support while conducting field data collection. Engr. Meriam M. Makinano assisted us during the field surveys. REFERENCES Baret, F., and G., Guyot, (1991). Potentials and limits of vegetation indices for LAI and APAR assessment, Remote Sensing of Environment, Vol. 35: Baret, F., G., Guyot, and D.J., Major, (1989). TSAVI: a vegetation index which minimizes soil brightness effects on LAI and APAR estimation. in Proceedings of the 12th Canadian Symposium on Remote Sensing and IGARSS'89, Vancouver (Canada). 3: Diaz, B. M. & Blackburn, G. A., (2003), Remote Sensing of mangrove biophysical properties: evidence from a laboratory simulation of the possible effects of background variation on spectral vegetation indices, International Journal of Remote Sensing, Vol. 24, No. 1,

8 Jordan, C. F., (1969). Derivation of leaf area index from quality of light on the forest floor. Ecology, Vol. 50, pp Huete, A. R., (1988), A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment, Vol. 25, Major, D. J., Baret, F., and Guyot, G., (1990), A ratio vegetation index adjusted for soil brightness. International Journal of Remote Sensing, Vol. 11, Phinn, S. R., Menges, C., Hill, G. J. E., and Stanford, M., (2000), Optimizing remotely sensed solutions for monitoring, modeling and managing coastal environments. Remote Sensing of Environment, Vol. 73, Price, J. C., and Bausch, W. C., (1995), Leaf area index estimation from visible and near infrared reflectance data. Remote Sensing of Environment, 52, Primavera, J. H. (2005) Global Voices of Science: Mangroves, Fishponds and the Quest for Sustainability, Science, Vol. 310, Richardson, A.J., and C.L., Wiegand, (1977). Distinguishing vegetation from soil background information, Photogrammetric Engineering and Remote Sensing, Vol. 43, No. 2 : Rasolofoharinoro Blasco, F., Bellan, M. F., Aizpuru, M., Gauquelin, T., and Denis J., (1998), A remote sensing based methodology for mangrove studies in Madagascar. International Journal of Remote Sensing, Vol. 19, Rondeaux, G., Steven, M., and Baret, F., (1996), Optimisation of soil-adjusted vegetation indices. Remote Sensing of Environment, Vol. 55, Rouse, J. W., Haas, R. H., Schell, J. A., Deering, D. W., (1974). Monitoring vegetation systems in the Great Plains with ERTS. Proceedings of Third Earth Resources Technology Satellite-1 Symposium, Greenbelt: NASA SP-351, Saito H., Bellan, M. F., Al-Habshi A., Aizpuru, M., Blasco, F., (2003), Mangrove research and coastal ecosystem studies with SPOT-4 HRVIR and TERRA ASTER in the Arabian Gulf. International Journal of Remote Sensing, Vol. 21, Tucker, C. J., (1979), Red and photographic infrared linear combinations for monitoring vegetation. Remote Sensing of Environment, Vol. 8, 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 Sensing of Environment, Vol. 55,