Use of Empirical Mode Decomposition based Denoised NDVI in Extended Three-Temperature Model to estimate Evapotranspiration in Northeast Indian Ecosystems

Abstract:

Evapotranspiration (ET) is an essential component involved in the energy balance and water budgeting methods, and its precise assessment are crucial for estimation of various hydrological parameters. Traditional point estimation methods for ET computation offer quantitative analysis, but lag in spatial distribution. The use of Remote Sensing (RS) data with good spatial, spectral and temporal resolution having broad spatial coverage, could lead the estimations with some advantages. However, approaches which requires data rich environment, demands time and resources. The estimation of spatially distributed soil evaporation (E_s) and transpiration from canopy (E_c) by RS data, followed by their combination to provide the total ET, could be a simpler approach for accurate estimates of ET flux at macro-scale level. The ‘Extended Three Temperature Model’ (Extended 3T Model) is an established model based on same approach and is capable to compute ET and its partition of E_s and E_c within the same algorithm. A case study was conducted using Extended 3T Model and MODIS products for the Brahmaputra river basin within the Northeast India for years 2000-2010. The extended 3T model was used by including its pre-requisite the land surface temperature (T_s), which was separated into the surface temperature of dry soil (T_sm) and the surface temperature of vegetation (T_cm), decided by a derivative of vegetation index (NDVI) called fractional vegetation cover (f). However, NDVI time series which is nonlinear and nonstationary can be decomposed by the Empirical Mode Decomposition (EMD) into components called intrinsic mode functions (IMFs), based on inherent temporal scales. The highest frequency component which was found to represent noise was subtracted from the original NDVI series to get the denoised product from which f was derived. The separated land surface temperatures (T_sm and T_cm) were used to calculate the E_s and E_c followed by estimation of total ET. The spatiotemporal variation of E_s, E_c and ET from denoised NDVI and original NDVI were analyzed in a comparative manner for different land covers in the study area. This scheme was used to justify the ET flux variation in magnitude and interannual/seasonal patterns during the years 2000-2010 in diversified land covers across ecosystems within the study area.