基于支持向量机的土壤湿度模拟及预测研究

基于中山大学珠海校区气象观测站日平均风速、日平均气温、日平均空气湿度、日平均水汽压、日平均总辐射量、日平均地表温度、日平均降雨量、日平均蒸发量以及日平均10 cm、20 cm、30 cm土层土壤的含水量,利用支持向量机方法建立气象因子与土壤湿度统计关系,并以此为基础建立土壤湿度模拟与预测模型.结果表明,土壤湿度对气象因子有一定滞后相关性,不同土层土壤湿度对气象因子的滞后相关性不同.研究发现考虑滞后相关性的预测模型在精度上要高于不考虑滞后相关性的预测模型.此外,利用气象因子对地下10 cm的土壤湿度模拟与预测精度较高,而对地下20 cm、30 cm的土壤湿度模拟精度较低.利用地下10 cm与20 cm、20 cm与30 cm的土壤湿度相关性大的特点,可以考虑利用支持向量机方法以10 cm土壤湿度模拟与预测20 cm的土壤湿度,以20 cm的土壤湿度模拟与预测30 cm的土壤湿度,分析结果表明模拟精度较高.

Simulation and prediction of soil moisture based on Support Vector Machine technique

Based on observed meteorological data, such as daily mean wind speed, daily mean air temperature, daily mean air humidity, daily mean water vapor pressure, daily mean total radiation, daily mean land surface temperature, daily mean rainfall, and daily mean evaporation, and daily mean soil moisture at 10 cm, 20 cm and 30 cm in depth, statistical relationships were established between meteorological variables and soil moisture using the Support Vector Machine (SVM) technique, and on such a basis, models for simulation and prediction of soil moisture were built up. It was found that responses of soil moisture to meteorological variables somewhat lagged behind, and were affected by soil depth. The model for prediction of soil moisture taking into account the lag correlation was more accurate than the one that did not count the lag correlation. Besides, using the meteorological variables, the model was more accurate in simulating and predicting the soil moisture at 10 cm in depth than in doing the soil moisture at 20 cm or 30 cm in depth. By taking into account the close relationships between the soils at 10 cm and 20 cm and between the soils at 20 cm and 30 cm in soil moisture, it is advisable to use the support vector machine technique in simulating and predicting soil moisture at 20 cm or 30 cm on the basis of the soil moisture at 10 cm or 20 cm. The findings indicate that the model for simulation of soil moisture is very high in accuracy.