Source apportionment of soil heavy metals using robust absolute principal component scores-robust geographically weighted regression (RAPCS-RGWR) receptor model

The traditional source apportionment models, such as absolute principal component scores-multiple linear regression (APCS-MLR), are usually susceptible to outliers, which may bewidely present in the regional geochemical dataset. Furthermore, the models are merely built on variable space instead of geographical space and thus cannot effectively capture the local spatial characteristics of each source contributions. To overcome the limitations, a new receptor model, robust absolute principal component scores-robust geographicallyweighted regression (RAPCS-RGWR), was proposed based on the traditional APCS-MLR model. Then, the new method was applied to the source apportionment of soilmetal elements in a region ofWuhan City, China as a case study. Evaluations revealed that: (i) RAPCS-RGWR model had better performance than APCS-MLR model in the identification of the major sources of soilmetal elements, and (ii) source contributions estimated by RAPCS-RGWR model were more close to the true soil metal concentrations than that estimated by APCS-MLR model. It is shown that the proposed RAPCS-RGWR model is a more effective source apportionment method than APCS-MLR (i.e., nonrobust and global model) in dealing with the regional geochemical dataset. (c) 2018 Elsevier B.V. All rights reserved.