Comparison of the hybrid of radiative transfer model and machine learning methods in leaf area index of grassland mapping

The leaf area index (LAI) of grassland is critical for estimating the balance of livestock and livestock production, understanding the dynamics of climate change, and providing feedback for achieving sustainable development. The currently available LAI products have some uncertainties and need to be further improved. Previous studies proposed that integrating the physical model and machine learning (ML) has great potential for the rapid and accurate retrieval of grassland LAI. However, there are few comparative studies on LAI forecast models for different grassland cover to assess the potential of the different hybrid models. Therefore, in this study, five hybrid models based on PROSAIL and ML including deep neural network (DNN), random forest (RF), gradient boosting regression tree (GBRT), support vector machine (SVR), and artificial neural network (ANN) and five mixed models averaging are applied to compare the performance with different forecast models for grassland LAI estimation in Tianzhu County. According to the multiple training, validation, and testing, the results demonstrate that five mixed models averaging and DNN model with a complex network structure are reliable and have higher accuracy and better performance than the estimates from the other four hybrid models, except for its computational efficiency. SVR achieves the best performance in computational efficiency, which it has great potentials to deliver near-real-time operational products for grassland LAI management. Our results show that the hybrid model based on machine learning algorithm coupled with physical process model has great application potential in grassland leaf area index inversion.