Estimating Leaf Area Index at 250M Spatial Resolution from Modis Data Using General Regression Neural Networks

Multiple leaf area index (LAI) products have been generated from satellite remote sensing data. These products are used to monitor global environmental changes, climate change, and carbon cycles. However, these products have too low spatial resolution to meet various application requirements. For example, crop growth monitoring and yield estimation require the higher spatial resolution LAI product. In this study, a method was proposed to generate 250m spatial resolution LAI product from Moderate-Resolution Imaging Spectroradiometer(MODIS) surface reflectance data using general regression neural networks (GRNNs). A dataset from MODIS surface reflectance product (MOD09Q1), MODIS observation geometry data, and Global Land Surface Satellite (GLASS) LAI product during the period from 2001–2017 was used to train the GRNNs in this study. Because the spatial resolution of these data are inconsistent, we proposed a downscaling method based on pure pixels to unify the spatial resolution of these data to 250m. we can get the GRNNs by training these datasets. Then, the MOD09Q1 surface reflectance and the corresponding observation geometry are entered into the GRNNs to retrieve LAI values. The retrieved LAI values were compared with the Geoland3 (GEOV3) LAI product and validated by ground measured LAI values. The results demonstrated that the method can get better temporal profiles than the GEOV3 LAI product. And the retrieved LAI values are in good agreement with the ground measured LAI values (R2 = 078861 and RMSE = 0.5645).