Modelling of Photosynthesis Vegetation Cover Fraction on Upscaling Approaches by using Landsat-8, AWiFs and MODIS Data

Abstract

Green Vegetation Fraction (GVF) is one of the most important land surface products to partition the fraction of the surface into evapo-transpiration and evaporation controlled by vegetation and bare soil respectively. Besides, GVF is a sensitive bioindicator for identifying vegetation anomaly, land degradation and enhanced areas of moisture loading due to evapo-transpiration and input parameter for soil loss equation. Satellite Remote Sensing provides a seemingly obvious data source for quantifying GVF over large areas by virtue of multispectral capability and temporal repitivity. Based on the concept of Gutman and Ignatov (1998) mosaic pixel model Bingfang et al. (2004) developed an improved Dimidiate Pixel Model to estimate vegetation fraction using NDVI values of soil and vegetation after careful selection of thresholds. The main objective of the present study in mixed forest areas of Paschim Midnapur is to generate GVF from Landsat-8 data using Dimidiate pixel model and comparing with ground observation as well as on the concept of upscalling approach compared with estimated GVF values from AWiFs data and coarse resolution MODIS data. Altogether 26grids of 1000 x 1000 m size having different degree of ground vegetation cover are selected and the representative value of GVF was determined using Landsat-8 OLI sensor data. Time composite MODIS NDVI product of 1km was also used for comparative evaluation of prediction accuracy at different spatial scale with respect to measured GVF data. There is good agreement between GVF predicted from Landsat-8 and AWiFS data (R2 = 0.93, RMSE = 3.11 and R2 = 0.88, RMSE = 4.35) and MODIS NDVI products of 250m (R2 = 0.79, RMSE = 5.93), whereas the correlation between MODIS NDVI products of 1km and measured values were less significant. The results show that Landsat-8 and time composite MODIS NDVI can predict GVF with reasonably good accuracy for large area at a time especially in deciduous forest areas. Poor correlation between MODIS NDVI products of 1km and ground observation could be due to coarse ground footprint of MODIS NDVI products of 1km data and data handling or data processing error.