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

Abstract

Leaf area index (LAI) products is widely applied for vegetation monitoring, yield estimation, ecological monitoring and global change studies. The demand for high-quality LAI products with different spatial and temporal resolutions is gradually increasing. Multiple LAI products have been generated from satellite remote sensing data. However, most of these LAI products have low spatial resolution. In recent years, deep learning algorithms have played a prominent role in scientific computing. Its powerful feature extraction and nonlinear fitting capabilities are very suitable for parameter estimation. In this paper, a new method based on convolutional neural network is proposed to estimate LAI values at 250m spatial resolution from Moderate-Resolution Imaging Spectroradiometer (MODIS) surface reflectance data by combining the existing Global Land Surface Satellite (GLASS) LAI product. The convolutional neural network consists of three convolutional layers, and its input is two bands (red and NIR) of MODIS surface reflectance at 250m spatial resolution and GLASS LAI at 500m spatial resolution. The results show that the method proposed in this paper can effectively estimate LAI values at 250m from MODIS surface reflectance data by combining the GLASS LAI product. The retrieved LAI values have better temporal continuity and agree well with ground measured LAI values when compared with the MODIS LAI product.