基于混合模型的草原土壤湿度动态预测

Dynamic Prediction of Grassland Soil Moisture Based on a Hybrid Model

摘要: 本文以内蒙古锡林郭勒草原为典型研究区，针对草原生态系统管理中土壤湿度精准预测的需求，提出了一种融合时间序列分析与BP神经网络方法的回归混合模型，旨在实现多深度土壤湿度的动态预测。基于2012~2021年长期观测的土壤湿度、蒸发量及降水量数据集，研究采用三阶段建模方法：(1) 构建单隐含层BP神经网络模型，以蒸发量和降水量作为输入变量，直接输出不同深度土壤湿度预测值；(2) 应用X11模型对各深度土壤湿度时间序列进行分解，提取季节成分、趋势成分和随机成分，采用Holt-Winters三参数指数平滑法建立时间序列预测模型；(3) 通过最优加权融合策略整合两种单一模型的预测结果，构建最终的回归混合模型。模型验证结果表明，该混合模型在预测精度上显著优于单一模型(MSE = 70.07, R² = 0.93)，能够准确捕捉土壤湿度的垂直分布特征和季节性动态变化规律，为草原生态系统的精准化管理及水资源优化配置提供了坚实的科学依据与技术支撑。

Abstract: This paper focuses on the Xilingol Grassland in Inner Mongolia as a representative research area. In response to the need for precise predictions of soil moisture within grassland ecosystem management, we propose a regression hybrid model that integrates time series analysis with BP neural network methods. The objective is to achieve dynamic predictions of soil moisture at various depths. Utilizing long-term observed datasets of soil moisture, evaporation, and precipitation from 2012 to 2021, this study employs a three-stage modeling approach: (1) A single-hidden layer BP neural network model is constructed, using evaporation and precipitation as input variables to directly output predicted values of soil moisture at different depths; (2) The X11 model is applied to decompose the time series data of soil moisture at each depth into seasonal components, trend components, and random components. Subsequently, the Holt-Winters three-parameter exponential smoothing method is employed to establish the time series prediction model; (3) The prediction results from both individual models are integrated through an optimal weighted fusion strategy to construct the final regression hybrid model. The verification results indicate that this hybrid model significantly outperforms individual models in terms of predictive accuracy (MSE = 70.07; R² = 0.93). It effectively captures both vertical distribution characteristics and seasonal dynamic change patterns of soil moisture. This provides a robust scientific foundation and technical support for precise management practices in grassland ecosystems as well as optimal allocation strategies for water resources.