Estimation Method And Application For Grassland Aboveground Dry Biomass Based On Assimilation Of Remote Sensing Data And Crop Growth Model

Grassland aboveground biomass is the key index to reflect the health status of the ecological system.Therefore,it is of great practical significance for grassland ecosystem monitoring to study the real-time dynamic estimation method of grassland biomass and obtain the spatial and temporal distribution.At present,it can obtain the biomass distribution of discreted phase based on remote sensing.Unfortunately,they can not reveal the continuous evolution of its internal growth and development,while crop growth model can do.In this paper,a method of assimilating leaf area index(LAI)derived from PROSAIL radiative transfer model and look-up table into WOrld FOod STudies(WOFOST)model was presented for estimating grassland aboveground biomass(AGB)in Wutumeiren prairie located in Qinghai province based on multi-source remote sensing images and field measurements.Finally,the biomass estimation precision before and after assimilation was evaluated based on field measurements.The main research results are listed as follows:(1)In order to overcome the significant uncertainty caused by the input parameters of crop model,the Sobol' sensitivity analysis method was utilized in filtering the crop parameters which were sensitive to the WOFOST model outputs,with the purpose to identify the crop parameters to be calibrated.Then three schemes(using LAI observation only,using biomass observation only and using both LAI and biomass observations)were designed for establishing cost function.Calibration result indicates integrating multi source observation contributes to model calibration and better simulation of local vegetation growth in potential conditions.This laid the foundation for the subsequent simulation of biomass in region.(2)Using Landsat and MOD09A1 reflectance data,multiple phase LAI were retrieved based on PROSAIL model and look-up table algorithm.In order to solve the problem of uneven distribution of vegetation in the study area and the ill posed inversion problem,some categories are adopted,which are the classification of the study area,the sensitivity analysis of the model parameters,and the construction of different look-up tables.Finally,the inversion results are evaluated based on the feild measurements.The Landsat and MOD09A1 retrieved results were acceptable compared with the field measured LAI that the R2(the deterministic coefficient)were 0.89,0.78,and the RMSE(Root Mean Square Error)were 0.56,0.61,respectively.The retrieved LAI would be used into assimilation process.(3)Based on Ensemble Kalman filter algorithm,multi phase retrieved LAI values were assimilated into WOFOST model calibrated in order to update model state variable.Then the temporal and spatial distribution of biomass in the study area were obtained.Finally,the estimation biomass was evaluated based on the field measured data and statistical method.The result showed that at verification plots,RMSE reduced by 511.44 kg/ha in 2011 and 1009.88 kg/ha in 2014 after assimilation of Landsat-LAI,and RMSE reduced by 198.7 kg/ha in 2011 after assimilation of MOD09A1-LAI.In order to evaluate the spatial distribution of biomass after assimilation,the biomass of different coverage region was further statistically analyzed based on fractional vegetation coverage.The results showed that the biomass value and the spatial distribution was more close to the measured value and reality after assimilation in different coverage region.The validity of the mentioned method in this paper for grass biomass estimation is well demonstrated.