基于多元回归方法的土壤水分预测模型研究

Study on the Simulation Model of Soil Moisture with Multiple Regression Method

通过对气象因子与土壤水分相互关系分析,并采用多元回归法中“后退法”和“逐步前进法”进行自变量筛选.两种回归方法所选择标准虽然不同,但却得到了基本相同的结果,同时,研究结果也显示两种方法各有优点,“后退法”善于发现联合作用较强的变量,而“前进法”则善于发现单独作用较强的变量.通过多元回归法构建了延安地区0-50 cm共计5层土壤湿度的预测模型,分别是：Y1=68.154 9＋0.619x1-0.2110x2＋1.294 8x3＋0.1875x6;Y2=108.9065＋0.537 4x1-0.226 9x2＋0.716 9x3＋0.339 0x6＋0.132 5x7;Y3=131.937 2＋0.441 2x1-0.241 5x2＋0.2653x3＋0.4231x6＋0.2257x7＋0.1444x8;Y4=149.4898＋0.3859x1-0.3143x2-0.2404x4＋0.3033x5＋0.5014x6＋0.3079x7＋0.2393x8＋0.1774x9;Y5=144.8923＋0.3033x1-0.3020x2-0.2769x4＋0.3592x5＋0.4436x6＋0.3124x7＋0.2686x8＋0.2078x9.通过验证发现模型预测值和实测值可以发现二者之间具有较好的一致性.

By analyzing the relationship of climate factors and soil moisture, it selects the independent variables with the backward elimination and forward regression of multiple regression method. Although the selection standards of the two methods have some differences, the selection results are basically the same. At same time, we also found the two methods have different advantages, and it is easy to get the unite variable with the back regression, and solitude variable with forward regression. With the methods,the five equations constructed are as follows： Y1=68.154 9＋0.619x1-0.2110x2＋1.294 8x3＋0.1875x6;Y2=108.9065＋0.537 4x1-0.226 9x2＋0.716 9x3＋0.339 0x6＋0.132 5x7;Y3=131.937 2＋0.441 2x1-0.241 5x2＋0.2653x3＋0.4231x6＋0.2257x7＋0.1444x8;Y4=149.4898＋0.3859x1-0.3143x2-0.2404x4＋0.3033x5＋0.5014x6＋0.3079x7＋0.2393x8＋0.1774x9;Y5=144.8923＋0.3033x1-0.3020x2-0.2769x4＋0.3592x5＋0.4436x6＋0.3124x7＋0.2686x8＋0.2078x9. And the validated result shows that the forecasting value and the model and measured value are consistent.