Assimilation of NDVI data in a land surface – Vegetation model for leaf area index predictions in a tree-grass ecosystem

Abstract

ABSTRACT Periodic observations of vegetation index, such as the normalized difference vegetation index (NDVI), can be used for data assimilation in heterogenous ecosystems. Indeed, the new Sentinel 2 Multispectral instrument and Landsat 8 Operational Land Imager sensor data are available at such high temporal and spatial resolutions that can be used to detect the patches of the main vegetation components (grass and trees) of heterogenous ecosystems, and capture their dynamics. We demonstrate the possibility to merge grass and tree NDVI observations and models, to optimally provide robust predictions of grass and tree leaf area index. The proposed assimilation approach assimilates NDVI data through the Ensemble Kalman filter (EnKF) and dynamically calibrates a key vegetation dynamic model parameter, the maintenance respiration coefficient (ma ). In the presence of large bias of the grass and tree ma base values, only the use of the proposed assimilation approach removes the large bias in the biomass balance, dynamically calibrating maintenance respiration coefficients, and corrects the model. The performance of a land surface – vegetation model was improved by assimilating observations of NDVI. The effective impact of the proposed assimilation approach on the evapotranspiration and CO2 uptake predictions in the heterogenous ecosystem is also demonstrated.