Comparison of machine learning algotithms for leaf area index retrieval from time series MODIS data

Temporally continuous and high quality leaf area index (LAI) products are urgently needed for crop growth monitoring, yield estimation and other research fields. However, most of the methods used to retrieve LAI just use a single phase satellite observational data to estimate LAI. Because of the impact of clouds and aerosols, the LAI products generated by these methods are temporally discontinuous. In this study, performance of three machine learning algorithms for parameter estimation using time series data is evaluated. The three machine learning algorithms are back-propagation neutral network (BPNN), general regression neutral networks (GRNNs) and multivariate adaptive regression splines (MARS). The results show that these machine learning algorithms have a good performance in time series LAI retrieval and GRNNs outperform the other algorithms.