中文摘要：

在长期田间试验基础上,分别利用数值模拟方法(Numerical Simulation,NS)和人工神经网络(Artificial Neural Network,ANN)模型构建江南平原土地整治区典型林地的土壤水分运动模型,并对土壤贮水量进行预测。NS模型校正结果表明,该模型虽能较好地预测林地土壤含水量动态变化,但是NS模型对训练期和验证期0~60cm土层贮水量预测的均方根误差(Root Mean Square Error,RMSE)分别为11.09和8.29mm,而ANN预测的RMSE分别为4.17和4.08mm,说明ANN的预测效果好于NS模型。最后,敏感性分析结果表明ANN预测精度对输入参数的敏感程度由高到低依次为:前期土壤贮水量】降水量】最高气温】最低气温。

英文摘要：

Forest hydrological research plays a critical role in alternative land-use practices (e.g.,restoring farmlands to forests) and management.Mathematical simulation models have often been applied to hydrological processes in forest soils.Based on long-term field experimental data,in this study,numerical simulation (NS) method was adopted to construct soil water movement model in typical woodland in land consolidation district of Jiangnan plain,and to predict soil water storage.A total of 100 days of observed soil moisture data at four depths (400 observations) were adopted to calibrate the Richards equation by using the inverse option in HYDRUS-1D.The calibration was for the assessment of the Richards equation to simulate soil moisture content in the experimental site.The result of NS model showed that the coefficient of determination (R2) between simulated and observed value of soil moisture in 10,20,40 and 60 cm depth was 0.863,0.870,0.865 and 0.665,respectively,indicating that soil moisture dynamics in woodland can be effectively predicted by this model.It is noteworthy that the 60-cm depth was found to have the lower accuracy than the other three depths.This may be attributed to the "free drainage" bottom boundary condition at this depth,which often leads to the overestimation of the simulated drainage during the relatively wet periods.Therefore,the simulated soil moisture contents at the lower depth (i.e.,60 cm) tended to be systematically underestimated.In addition to NS,the artificial neural network (ANN) was also applied in this study.Four independent variables including precipitation,daily maximum temperature,daily minimum temperature,and antecedent soil water storage were used to construct ANN model to predict the soil water storage as well.When predicting the soil water storage of 0-60 cm layer,Root Mean Square Error (RMSE) for training period and validation period in NS model was 11.09 and 8.29 mm,respectively,while that in ANN model was 4.17 and 4.08 mm,respectively,indicating that the prediction result