A STUDY ON FAST ESTIMATION OF VEGETATION FRACTION IN THREE GORGES EMIGRATION AREA BY USING SPOT5 IMAGERY

Transcription

1 A STUDY ON FAST ESTIMATION OF VEGETATION FRACTION IN THREE GORGES EMIGRATION AREA BY USING SPOT5 IMAGERY Zhu Lian, WU Bin-fan, ZhouYue-min, Men Ji-hua, Zhan Nin Institute of Remote Sensin Applications, Chinese Academy of Sciences, Beijin 111, P R China Datun Road No 3, P O Box 9718, 111, Beijin, P R China - zhulian4@sohucom KEY WORDS: Remote Sensin; Veetation; Estimation; SPOT Imaery; Mosaic Pixel ABSTRACT: Veetation Fraction is a comprehensive quantitative index in forest manaement and veetation community cover conditions, and it s also an important parameter in many remote sensin ecoloical models, such as RUSLE model in soil erosion study Traditional round measures method to et the reional-scale veetation fraction is very difficult due to the costs, labor and time involved Therefore, it s meaninful to study on fast estimation of veetation fraction from remotely sensed data In this study, we make use of modified Gutman model to et veetation Fraction in Three Gores emiration area by usin SPOT5 Imaery It assumes each pixel is a mosaic pixel The spectral information is composed of veetation and soil spectral information Throuh computin the normalized difference veetation index (NDVI) which is sensitive to veetation, we define the soil and veetation threshold accordin to round field data Then, usin Gutman model we et veetation Fraction The result shows the accuracy of this method is fine, and it is feasible in fast estimation of reional veetation fraction 1 INTRODUCTION Veetation Fraction is defined as the percentae of veetation occupyin the round area in vertical projection It s a comprehensive quantitative index in forest manaement and veetation community cover conditions, and it s also an important parameter in many remote sensin ecoloical models, such as RUSLE model in soil erosion study Chanes in veetation cover directly impact surface water and enery budets throuh plant transpiration, surface albedo, emissivity, and rouhness (Aman et al, 1992) Field measurement, as a traditional method of veetation fraction, can be divided into three kinds of methods accordin to principles: field sample method, instrument method and visual estimation method (Sutherland, 1999) Traditional field measurement method to et the reional-scale veetation fraction is very difficult due to the costs, labor and time involved Furthermore, the reliability of some field measurement methods for the veetation fractional coverae is questionable (Curran et al, 1986) Thus, traditional method is not feasible in reional-scale estimation of veetation fraction In order to et reional-scale estimation of veetation fraction, we can utilize remote sensin data Satellite data provides a spatially and periodic, comprehensive view of land veetation cover (Chen Yunhao et al, 25) The common remote sensin method is to use empirical reression formulation or veetation index to estimate veetation fraction Graetz et al (1988) utilized the fifth band of Landsat MSS data and field data to establish empirical relationship, and estimated the veetation fraction in semi-arid soil reion Peter (22) used four bands of ATSR-2 remote sensin data (555nm 87nm and 163nm) to reress separately with veetation fraction and LAI, the result shows estimation precision usin the linear mixture model with four bands is hiher than sinle band or veetation index Empirical reression model can be well applied in specified district, especially with small areas However, it needs lots of precise field data and can not be applied in other district Gutman and Inatov put forward a method to et the estimation of veetation fraction It assumes each pixel is a mosaic pixel The spectral information is composed of veetation and soil spectral information The method is fit for ettin the estimation of veetation fraction inlare-scale reion 2 STUDY AREA The reservoir area of the Three Gores Project (TGP) is situated at a point 16º ~111º5 east lonitude, 29º16 ~31º25 north latitude in the lower part of the upper reaches of the Yantze It starts from Yichan of Hubei Province and ends at Jianjin City of Chonqin Municipality The inundated areas affected by return currents and the places involvin resettlement cover 2 counties or cities The reservoir area covers 57,9km 2, includin 9,777km 2 of cultivated land, accountin for 169%, averain 76 mu (one mu 1/15 of a hectare) The emiration area is part of the reservoir area It covers 16,km2 It is located within 1 kilometers distance from Yantze River The population of the emiration area is 725,5 It covers middle and low mountain valleys in Sichuan and Hubei and the low hilly areas in eastern Sichuan There is the Daba Mountain to the north and the Yunnan-Guizhou- Guizhou Plateau to the south, with the toporaphy taperin from west to the east The location of the emiration area in the reservoir area of the TGP is illustrated in Fiure 1 987

2 The International Archives of the Photorammetry, Remote Sensin and Spatial Information Sciences Vol XXXVII Part B8 Beijin 28 When it is variable density: NDVI = f NDVI + ( 1 f ) NDVI 4 i i L is the leaf area Index and k is the extinction coefficient NDVI and NDVI, are the sinals from bare soil (L->) and dense reen veetation (L-> ) respectively F is the veetation fraction Uniform-pixel model is a model in ideal condition In this model, veetation fraction is 1% NDVI is determined by leaf area index Thus, we can only choose mosaic-pixel model i Fiure 1 Study area 3 MODEL The main objective of this study is to make use of Gutman s mosaic-pixel model to et fast Veetation Fraction estimation in reional scale It assumes each pixel is a mosaic pixel The spectral information is composed of veetation and soil spectral information The same NDVI sinal may result from different sub-pixel structures of a satellite pixel (Price 1992) Accordin to different combinations of horizontal and vertical densities, Gutman summarize two different models: uniformpixel model, mosaic-pixel models Table 1 shows the models and their applied conditions On one hand, the study area is mainly mountainous and hilly area and the veetation cover is well except urban area On the other hand, extinction coefficient k and leaf area index L is difficult to obtain in lare-scale reion, and the precision of k and L will limit final veetation fraction precision Considerin the two aspects, we make use of mosaic-pixel model in dense veetation condition as the approximate calculation model of veetation fraction in veetation cover area In non-veetation cover area, the veetation fraction is zero Land use classification map is used to distinuish veetation and nonveetation area Formulation 2 is transformed to formulation 5 as follows to et f f NDVI NDVI = NDVI NDVI 4 METHODOLOGY 5 model Uniformpixel model Mosaicpixel model Veetation density Uniform full veetation Dense veetation Nondense veetation Variable density veetation Schematic representation The flowchart for the veetation fraction is illustrated in Fiure 2 Table 1 Schematic representation of models When it is uniform full veetation, veetation fraction is 1%, and it conforms modified Beer s law: NDVI = NDVI - ( NDVI NDVI )exp( kl ) 1 When it is dense veetation : NDVI = f NDVI + ( 1 f ) NDVI 2 Fiure 2 Flowchart for veetation fraction estimation When it is non-dense veetation: NDVI = f NDVI + ( 1 f ) NDVI 3 988

3 The International Archives of the Photorammetry, Remote Sensin and Spatial Information Sciences Vol XXXVII Part B8 Beijin imae preprocess Imae preprocess includes radiation calibration, atmospheric correction, imae fusion and eometric rectification Raw imaes record the Diital number (DN) In order to et accurate normalized difference veetation index (NDVI), we should transform DN value to reflectivity After radiation calibration, we can et the imae reflectivity Because of atmospheric condition s difference in horizon, it forms satellite sinal s difference caused by atmosphere scatterin and atmosphere absorption Thus, we should eliminate atmospheric effect In support of Erdas ATCOR2 module, we input imae acquisition date, solar zenith, round elevation, visibility and other related parameters, and carry out atmospheric correction Then, we utilize 1m multi-spectral bands imae and 25m panchromatic to carry out eometric rectification each other The accuracy is less than 1 pixel The next is to determine imae fusion method We compare principal component analysis (PCA) method, multiplicative method, brovey transform method and wavelet transformation method, etc Finally, we choose principal component analysis (PCA) method It is the optimal method and can meet the need of correct NDVI calculation It can not only sharp the imae to 25m, and keep the oriinal multi-spectral reflectivity Other methods will result in that there are noises in fusion imaes or it chanes spectral values Finally, imae-to-map eometric rectification was carried out usin 1:1 toporaphic maps Geometric accuracy is less than 1 pixel The project is Gauss Kruer (Spheriod: Krasovsky; Central meridian: 18 ; False eastin: 5 meters) we made validation and accuracy evaluation The result shows accuracy of land cover classification is as hih as 95% 44 veetation fraction estimation mosaic-pixel model in dense veetation condition is selected to estimate veetation fraction Accordin to formulation 5, f NDVI NDVI = NDVI NDVI We need to calculate NDVI NDVI is defined as follows: B NDVI = B nir nir B + B r r 6 NDVI is scaled to [,255] Scaled NDVI will enhance imae display effect and will save disk space Then, NDVI and NDVI should be determined Both NDVI and NDVI are constant But the NDVI and NDVI are different between different SPOT5 scenes We should find bare soil endmember and dense veetation endmember in each SPOT5 scenes NDVI and NDVI in each SPOT 5 5 RESULTS AND DISCUSSION 42 Cloud Removal In Three Gores Reservoir, clear sky is hard to see Thus, it is almost impossible to et cloudless SPOT5 imaes in the reional-scale It is necessary to mark cloud areas and replace these areas usin other SPOT5 imaes or TM imaes The precondition is that the acquisition date of those SPOT5 imaes or TM imaes is close with oriinal SPOT5 imaes 43 land cover classification and mask NDVI is usually sensitive to veetation However, NDVI of Some special features will be abnormal These features includes roof made of different material such as aluminium, road with special materials, etc The NDVI of those features is similar with veetation Thus, it is difficult to identify veetation and non-veetation areas only usin NDVI, and land cover classification map is needed Land cover classification map is used to identify veetation areas and non-veetation areas Correspondinly, the mask file is produced The method of land cover classification is objectoriented The land use classification system adopts modified FAO land use classification system In support of Econition Software, we carry out imae sementation Throuh different feature indices, such as spectral mean, NDVI, SAVI, etc, we set up feature space Those indices is sensitive to different features Based on those, we produced basic land cover classification map Then, usin some field samples, we modified the land cover classification results manually Usin other field samples, Fiure 3 Estimated veetation fraction map The veetation fraction estimation result (Fiure 3) shows that the veetation fraction in the west of study area and the areas alon Yantze River are relatively lower than other areas The reason is that Chonqin city locates in the west of study area, it is the most important city in the west of China The areas alon Yantze River is areeable for human bein survival and development, human bein activity results in the veetation fraction of these areas is lower and lower The other areas is mainly mountainous areas, residential area is sparsely distributed in these areas, and less human bein activity makes forest in these areas in ood condition Next, we analyze the accuracy of veetation fraction estimation quantitatively Field survey was carried out in September,

4 The International Archives of the Photorammetry, Remote Sensin and Spatial Information Sciences Vol XXXVII Part B8 Beijin 28 and September, 27 Some abnormal samples were eliminated and 19 valid Samples patches were selected to evaluate the accuracy of veetation fraction estimation, which includes different land-cover type, such as conifer forests, broadleaf forests, cropland, shrub, rassland, water body, urban construction These samples are randomly, proportionally distributed in the study area The result of accuracy analysis is listed in Table 2 ID land cover type O bs Est Abs Rel 1 cropland % 2 broadleaf forest % 3 broadleaf forest % 4 shrub % 5 Water body 6 shrub % 7 conifer forest % 8 cropland % 9 conifer forest % 1 Broadleaf % 11 conifer forest % 12 Grass land % 13 Shrub % 14 broadleaf forest % 15 conifer forest % 16 Grass land % 17 conifer forest % 18 Grass land % 19 conifer forest % 2 urban Av error Av accu racy construction 8% 92% Table2 Accuracy estimation of estimated veetation fraction (Obs Observation value, Est estimation value, Abs absolute error, Relrelative error, Averror, averae error, Av accuracy, averae accuracy ) The analysis of accuracy estimation (Table 2) shows veetation fraction estimation result is fine, the averae accuracy is up to 92% The method of veetation fraction estimation is feasible in lare-scale reion Scatterram of observation and estimation (Fiure 4) was produced and linear reression was carried out Estimation value is hihly related with observation (R 2 =96, n=2) estimation value(%) y = 9223x R 2 = observation value(%) Fiure 4 Estimation value Vs observation value 6 CONCLUSION In this study, we use Gutman model combin with land cover map to et veetation cover estimation in Three Gores emiration area It assumes each pixel is a mosaic pixel The spectral information is composed of veetation and soil spectral information The accuracy of estimation is fine, eneral accuracy is up to 92% The method is feasible to et fast estimation of veetation fraction in lare-scale reion The accuracy of land cover map and imae fusion will effect the accuracy of veetation fraction estimation in some deree REFERENCE Adams, J B, Smith, M O, & Johnson, P E (1986) Spectral mixture modellin: A new analysis of rock and soil types at the Vikin Lander 1 site Journal of Geophysical Research, 91, Sutherland W J Zhan J-T trans1997 Manual of Survey Method in Ecoloy Beijin: Science and Technoloy Literature Press Curran P J, Williamson H D 1986 Sample size for round and remotely sensed data Remote Sense of Environment, 2:31-41 Gutman, G, Inatov, A, 1998 The derivation of the reen veetation fraction from NOAA/AVHRR data for use in numerical weather prediction models International Journal of Remoete Sensin 19, Calera, A, Martinez, C, Melia, J, 21 A procedure for obtainin reen plant cover: relation to NDVI in a case study for barley International Journal of Remote Sensin 22, Chen, J, Chen, Y, He, C, Shi, P, 21 Sub-pixel model for veetation fraction estimation based on land cover classification Journal of Remote Sensin 5, Foody, GM, Cox, DP, 1994 Sub-pixel land cover composition estimatin usin a linear mixture model and fuzzy membership functions International Journal of Remote Sensin 15, Price, JC, 1992 Estimatin veetation amount from visible and near infrared reflectance Remote Sensin of Environment 41,

5 The International Archives of the Photorammetry, Remote Sensin and Spatial Information Sciences Vol XXXVII Part B8 Beijin 28 Price, JC, 1993 Estimatin leaf area index from satellite data IEEE Transactions on Geoscience and Remote Sensin 31, Quarmby, NA, Townshend, JRG, Settle, JJ, 1992 Linear mixture modellin applied to AHVRR data for crop area estimation International Journal of Remote Sensin 13, ACKNOWLEDGMENTS This work was supported by Three Gores Project Construction Committee of the State Council Project The Dynamic and Real-time Monitorin of Ecoloy and Envirionment in Three Gores Project (SX 22-4) 991

6 The International Archives of the Photorammetry, Remote Sensin and Spatial Information Sciences Vol XXXVII Part B8 Beijin