Maize Leaf Area Index Retrieval Using FY-3B Satellite Data by Long Short-Term Memory Model

The Medium Resolution Imaging Spectrometer (MERSI) onboard FY-3 satellite possesses the Characteristics of wide scanning range, short revisit period and high spectral resolution, which can provide wide-area and long-term sequence data for leaf area index (LAI) retrieval research. Long Short-Term Memory (LSTM) has great advantages of nonlinear fitting and can utilize the relationships between samples, can also solve the problem of large dimension of sample features. It is of great significance to apply it for LAI retrieval. Based on the FY-3B/ MERSI data simulated by hyperspectral data for five stages of maize canopy, this study explored multi-layer LSTM for LAI retrieval. Then, the results were compared with those of stepwise regression, partial least squares regression (PLSR) and single-layer LSTM method. The retrieval accuracies of Multi-layer LSTM model were better than those of other three models. Multi-layer LSTM provides a methodological reference for LAI retrieval studies.