融合多源时空数据的冬小麦产量预测模型研究

为提高大尺度冬小麦产量预测精度,以2005—2019年河南省遥感数据、气象数据、土壤含水率等多源时空数据为特征变量,分析其与小麦单产的相关性,并基于随机森林算法对特征变量进行了重要性分析,构建了融合多源时空数据的冬小麦产量预测模型。结果表明:增强型植被指数（Enhanced vegetation index, EVI）、日光诱导叶绿素荧光（Solar-induced chlorophyll fluorescence, SIF）与高程为小麦产量预测的重要因子,与小麦产量呈高度正相关,对小麦产量预测的重要性指标均超过0.45,远大于土壤含水率、降水量、最高温度、最低温度等因子;基于随机森林算法构建的小麦不同生长阶段产量预测模型中,以10月—次年5月和10月—次年4月为特征变量的产量预测模型精度较高,R²分别为0.85和0.84,RMSE分别为821.55、832.01 kg/hm²,在空间尺度上,豫西和豫南丘陵山地模型预测相对误差高于平原地区。该研究结果可为大尺度作物产量预测提供参考。

Prediction of Winter Wheat Yield Based on Fusing Multi-source Spatio-temporal Data

In order to improve the prediction accuracy of winter wheat yield in large scale region, taking remote sensing data, meteorological data, soil moisture data of Henan Province from 2005 to 2019 as characteristic variables, the correlation between them and wheat yield was analyzed. The importance of characteristic variables was analyzed based on random forest algorithm. And a wheat yield prediction model was established by means of fusing multi-source spatio-temporal data. The results showed that enhanced vegetation index (EVI), solar-induced chlorophyll fluorescence (SIF) and elevation was an important factor for remote sensing estimation of wheat yield, which was highly positively correlated with wheat yield. The importance of EVI, SIF and elevation to wheat yield exceeded 0.45, far greater than soil moisture, rainfall, maximum temperature, minimum temperature and other factors. The yield prediction model based on random forest algorithm and constructed with the wheat growth stage from October to next May and October to next April as the characteristic variables had higher accuracy, coefficient of determination (R2) were 0.85 and 0.84, and respectively, the root mean square error (RMSE) were 821.55kg/hm2 and 832.01kg/hm2. The prediction relative errors in hills and mountains of western and southern Henan was higher than that in plain areas. The research results provided a reference for large-scale crop yield.