Dynamic modelling of rice leaf area index with quad-source optical imagery and machine learning regression models

Optical satellite imagery has been widely used to monitor leaf area index (LAI). However, most studies have focussed on single- or dual-source data, thus making little use of a growing repository of freely available optical imagery. Hence this study has evaluated the feasibility of quad-source optical satellite imagery involving Landsat-8, Sentinel-2A, China's environment satellite constellation (HJ-1 A and B) and Gaofen-1 (GF-1) in modelling rice green LAI over a test site located in southeast China at two growing seasons. With the application of machine learning regression models including Random Forest (RF), Support Vector Machine (SVM), k-Nearest Neighbour (k-NN) and Gradient Boosting Decision Tree (GBDT), results indicated that regression models based on an ensemble of decision trees (RF and GBDT) were more suitable for modelling rice green LAI. The current study has demonstrated the feasibility of quad-source optical imagery in modelling rice green LAI and this is relevant for cloudy areas.

Automated summary

This study evaluated the feasibility of using quad-source optical satellite imagery in modeling rice green LAI. Results showed that regression models based on an ensemble of decision trees were more suitable for this task. The findings are significant for areas with high cloud coverage.