Dynamic Yield Estimation of Winter Wheat Using Time-Series Features and Machine Learning
Ye Haotian
Duan Jianzhao
Li Mengxia
Li Junling
Tian Hongwei
Abstract：Winter wheat yield estimation is crucial for ensuring food security.To enhance the effi-ciency and accuracy of yield prediction,this study integrates multi-source data from 2000 to 2021,inclu-ding county-level winter wheat yield,remote sensing data,meteorological data,and geographical coordi-nates in Henan Province,to analyze the correlation between characteristic variables and yield.By pro-gressively inputting feature variables in a time-series manner,the dynamic yield estimation performance of five models——Linear Regression,Random Forest,Gradient Boosting,LightGBM,and K-Nearest Neighbors——is compared and validated using 2021 data.The results indicate that winter wheat yield ex-hibits a significant positive correlation with NDVI and is strongly associated with geographical location and meteorological factors such as light,temperature,and water.All five models effectively estimated yield,with the LightGBM model(27 feature variables)performing the best(prediction period:May,R2=0.91,RMSE=435 kg·hm-2,NRMSE=7.0%,RE=6.5%).Based on this model,the prediction accuracy for the 2021 yield is significantly high(R2=0.97,RMSE=249 kg·hm-2,NRMSE=5.1%,RE=2.7%),and the spatial heterogeneity of relative errors is related to topographical factors and mixed pixels in remote sensing data.
Keywords：winter wheatyield predictionmachine learningdynamicmulti-source data
Publication Date：2026-05-30
Online Publishing Date：2026-09-12(First online date of this platform, not the publication date of the document)
Pages：9( 46-54 )