Yield Estimation Of Summer Maize In Hebei Province Based On Solar-induced Chlorophyll Fluorescence Data

Maize is one of the most important food crops,and its demand has increased dramatically in recent decades.China is the second largest producer and consumer of maize in the world,contributing nearly one-fifth of the world’s grain production.Therefore,timely and accurate forecasting of China’s maize yield is essential to ensure global food security.In recent years,the frequency of extreme weather and the lag of traditional vegetation indices(e.g.,normalized difference vegetation index,NDVI,enhanced vegetation index,EVI)in drought conditions have greatly affected the estimation of maize yield.Therefore,solar-induced chlorophyll fluorescence(SIF),which is more sensitive to environmental stress than traditional vegetation indices(VIs),became a new predictor,but the potential of SIF for maize yield prediction under drought conditions still needs to be further investigated.The performance of machine learning(ML)and deep learning(DL)based methods in yield prediction is also not well understood that needs to be studied in more depth.To this end,the present takes a major corn production region(Hebei Province)in China as an example to investigate maize yield estimation under normal and drought years.We utilize three satellite data(SIF,NDVI,and EVI),meteorological data and soil data from2000 to 2019 to predict summer maize yield at the county level,using a ML method(random forest,RF)and a DL method(long and short-term memory,LSTM).The main conclusions of the study are as follows:(1)during yield prediction,the contribution of meteorological data was gradually replaced by that of satellite data(SIF/VIs)as the maize growing season progresses;(2)both methods allowed maize yield to be reasonably estimated approximately 1 month before harvest and could achieve an overall accuracy of90%in predicting yield,with the ML method performing slightly better than the DL method in predicting summer maize yield;(3)under normal weather conditions,high-resolution SIF(R~2=0.76,RMSE=634.06 kg/ha)was similar to NDVI(R~2=0.73,RMSE=656.52 kg/ha)and EVI(R~2=0.75,RMSE=646.86 kg/ha)in predicting summer maize yield;(4)in drought years,high-resolution SIF(R~2=0.74,RMSE=743.87 kg/ha)outperformed both NDVI(R~2=0.70,RMSE=814.48 kg/ha)and EVI(R~2=0.72,RMSE=791.88 kg/ha)vegetation indices in the estimation of summer maize yield,indicating that SIF has a certain superiority in predicting summer maize yield under drought conditions.As such,our study indicates the potential of using SIF data for crop yield prediction and believes that higher quality SIF products may have a significant advantage to improve crop yield prediction in the future.