有机化合物K_(OC)片段常数估算模型的误差与稳健性分析

Robustness of A KOC Prediction Model Basedon Fragment Constants*

分析了估算有机化合物吸着系数的片段常数模型的误差．并用４种不同方法检验了估算有机化合物吸着系数（ＫＯＣ）的片段常数模型的稳健性．重复随机抽取单一化合物、逐一抽取估算值与实测值差别较大的个别化合物、重复随机抽取５０个化合物以及逐一抽取特定化合物类别．检验结果表明，模型估算的绝对误差随ｌｏｇＫＯＣ实测结果增大而增大，而误差较大的多为亲水化合物．估算误差中的相当部分实际上来源于实测数据的误差波动．无论是单一抽取或成组抽取化合物的ｊａｃｋｋｎｉｆｅ检验结果证明模型具有较高的稳健性．相对而言，不同类别抽取的ｊａｃｋｋｎｉｆｅ可决系数波动较大。

Modified jackknife tests were applied to examine the robustness of a prediction model for organic carbon normalized sorption coefficient (KOC) based on fragment constants and structural correction factors. Similar trends were observed using various kinds of jackknife tests. Fairly high robustness of the model was demonstrated by the calculated jackknife coefficients of determination (R2). The highest variation in the jackknife coefficient occurred when a group of chemicals within the same category was eliminated from the data set.