基于多元混合线性模型的稻田地力评估

Assessment of paddy soil productivity by using the multiple linear mixed-effect regression model

土壤地力对水稻产量具有较大贡献,关键地力因子筛选和地力模型评估对指导农业管理实践有重要意义。分别选取湖南道县和澧县典型高中低产稻田50 cm深土壤剖面,间隔5 cm分层取样对其理化性质进行分析。依据皮尔森相关系数挑选出影响稻田地力的关键因素,利用多元线性回归和混合线性回归分别对稻田地力建模评估并比较。研究表明,决定稻田地力的土壤基本理化性质重要性排序为:阳离子交换量=钙离子>钾离子>速效钾>pH值>粘粒含量,都呈显著正相关性。模型比较中,按0-20 cm,20-30 cm,30-50 cm深度采样建立的多元混合线性模型对稻田地力预测效果最好,选用的固定效应包括:速效钾,速效磷,总磷,钙离子、钾离子,阳离子交换量,pH值,粘粒和砂粒含量,随机效应为饱和导水率,有机碳,钠离子和容重,所得模型能解释稻田地力总变异的85%,而不分层的多元线性回归模型则只可解释73%。

Soil productivity makes an important contribution to rice yield. Selection of the key factors as well as model a ssessment for soil productivity can provide better guidelines for agricultural activity. Six 50-cm depth undisturbed typical paddy soil profiles from the high, moderate and low-yield fields, respectively, were collected in two locations （Dao county and Li county） in Hunan province and then stratified to measure soil physical and chemical properties. The Pearson＇s correlation analysis was carried out to determine the significant factors, and soil productivity predictive models were developed using the multiple linear regression and the mixed-effect linear regression. Our results showed that the importance of soil properties was in the following order： cation exchange capacity （CEC） = Ca^2＋ 〉 K^＋〉 available K （AVK） 〉 pH 〉 clay particle content （CLAY）; The mixed-effect regression model derived using the sample data by layers of 0-20 cm, 20-30 cm, 30-50 cm performed the best; it contained nine fixed effects （AVK, AVP, total P, Ca^2＋, K^＋, CEC, pH, CLAY and sand particle content） and four random effects （K10, SOC, Na＋ and BD）; and it can explain 85% of variability of soil productivity, in comparison with the determination of 73% produced by the ordinary multiple linear regression model.