基于Stacking模型的土壤综合肥力评价——以富川瑶族自治县植烟区为例

Soil integrated fertility evaluation based on stacking model:A case study of tobacco planting areas in Fuchuan Yao Autonomous county

【目的】实现耕地土壤综合肥力评价的标准化、快捷化和自动化,指导作物种植布局和土地资源利用。【方法】以富川瑶族自治县植烟区为研究对象,选定广西植烟区1038个样点作为训练数据。在考虑土壤条件、气候环境和地形地貌等因素的基础上,选取15种影响因子计算土壤综合肥力指数,采用Stacking算法构建回归预测模型,实现对研究区土壤综合肥力的评价。【结果】基于6种关键因子的Stacking02模型在测试集上性能优秀,平均绝对误差、均方误差和R2决定系数分别为0.0066、0.0084和0.9529。应用Stacking02模型评估研究区综合肥力,结果显示研究区北部综合肥力优于南部,其优质产区占比更高。低综合肥力指数地区土壤pH偏高而全氮速效钾含量偏低。【结论】相较初级学习器,Stacking模型性能提升。采用6种关键因子建立的Stacking02模型具有优异的泛化性能,简化了计算流程并节省算力,该方法可扩展至其它地区及作物的土壤综合肥力评价,从而制定针对性的作物利用方案。

【Objective】By standardizing,expediting,and automating the evaluation of integrated farmland soil fertility,the present paper aimed to optimize crop planting patterns and make use of land resources rationally.【Method】Taking the tobacco planting areas of Fuchuan Yao Autonomous county as examples,1038 sample points from Guangxi tobacco planting areas were selected as training data.Based on the consideration of soil conditions,climate environment,and topography,15 indicators were chosen to calculate the integrated soil fertility index.The Stacking algorithm was applied to establish a regression prediction model for evaluating the integrated fertility of the study area.【Result】The Stacking02 model based on six key factors exhibited excellent performance on the test set,with mean absolute error,mean squared error,and R~2 coefficient of determination of 0.0066,0.0084,and 0.9529,respectively.The Stacking02 model was used to assess the integrated fertility of the study area,showing that the northern part of study area was better than the southern part,with a higher proportion of high-quality production areas.Low comprehensive fertility index regions had relatively high soil pH and low total nitrogen and available potassium contents.【Conclusion】Compared to primary learners,the performance of the Stacking model improved.The Stacking02 model established using six key factors demonstrated excellent generalization performance,simplifying the calculation process and saving computational power.The same method can also be used to evaluate the soil integrated fertility of the study area.It suggests that the method should be extended to the comprehensive fertility evaluation of other regions and crops'soils to develop targeted crop utilization schemes.