DSSAT作物模型的统计和图形校验方法

Statistical and graphical methods for evaluation of DSSAT crop model

本文以玉米,大豆,马铃薯和冬小麦试验为例,详细讨论了DSSAT模型的统计校验和图形校验方法及其支持软件EasyGrapher的应用。结果指出:R2不是理想的模型校验统计,因为它是测量线性回归y=α+βx+ε的拟合度,其随机误差ε建立在正态性、相互独立和方差同质的假设之上。模型校验主要是检验残差di=yi-xi(y为测量值,x为模拟值),不是估计回归系数α、β。RMSE、E、EF和d都是比较好的校验模型残差的统计量,它们不需要遵循回归模型的基本假设,统计量的物理意义明确,大样本容量下统计校验更为可靠。图形校验是模型校验必不可少的辅助工具,当有测量数据时,时间序列和残差校验图形是两种最基本的图形校验方法;没有测量数据,模拟图形仍然能够检验模型输出之间或与时间的动态关系,辨析模拟误差或者错误。应用EasyGrapher软件可以快速进行DSSAT模型的图形校验和统计校验。

This paper discussed statistical and graphical evaluation methods for DSSAT model using field experiments of maize, soybean, potato and winter wheat. The results indicated that R2 is not a good statistic for model evaluation because it tests the goodness of fit of a linear regression y = α ＋ βx ＋ ε where random error, ε, was assumed to follow normality, independence and equal variance. Model evaluation aims at testing residual error d = y -x（y measured data, x simulated data）,but not estimating regression coefficients, α, β, RMSE, E, EF and d are all good “difference measures”, they do not need follow three assumptions, and they have clear physical meaning. Large sample size increase reliability of statistics. Graphical evaluation is a necessary method for model evaluation. Time series and residual error graphs are two basic graphical methods for model evaluation if there are measured data. Simulation graphs can also be used to display relationships among outputs or against time, to analyze residual errors or mistakes even if no measured data. EasyGrapher program is a useful tool for statistical and graphical evaluations of DSSAT model’s output.