Grain yield prediction using multi-temporal UAV-based multispectral vegetation indices and endmember abundance in rice

•A multi-stage time-series GY prediction model combining VI and EM abundance.•MESMA method solves the problem of spectral variability in rice fields.•GY response is contributed by both leaves and panicles.•The model used NDRE at booting and ABD from heading to initial filling can give the best results.