Search

Previous ecological risk assessments were mainly concentration-oriented rather than source-oriented. Moreover, land use is usually related to source emissions but was rarely used to improve the source apportionment accuracy. In this study, the land-use effects of heavy metals (HMs) in surface (0–20 cm) and subsurface (20–40 cm) soils were first explored using ANOVA in a suburb of Changzhou City, China; next, based on robust absolute principal component scores-robust geographically weighted regression (RAPCS/RGWR), this study proposed RAPCS/RGWR with land-use type (RAPCS/RGWR-LUT) and compared its source apportionment accuracy with those of basic RAPCS/RGWR and commonly-used absolute principal component scores/multiple linear regression (APCS/MLR); then, the source-oriented ecological risks were apportioned based on RAPCS/RGWR-LUT and Hakanson potential ecological risk index method; finally, this study proposed robust residual kriging with land-use type (RRK) for spatially predicting the source-oriented ecological risks, and compared its spatial prediction accuracy with those of robust ordinary kriging (ROK) and traditionally-used ordinary kriging (OK). Results showed that: (i) by incorporating land-use effects, RAPCS/RGWR-LUT obtained higher source apportionment accuracy than RAPCS/RGWR and APCS/MLR; (ii) the two most important external input sources of the ecological risks were 'atmospheric deposition' (PERI surface = 47.11 and PERI subsurface = 35.27) and 'agronomic measure' (PERI surface = 28.93 and PERI subsurface = 20.37); (iii) the biggest ecological risk factor was soil Cd (ER surface = 57.14 and ER subsurface = 47.62), which was mainly contributed by 'atmospheric deposition' (ER surface = 33.14 and ER subsurface = 25.71); (iv) RRK obtained higher spatial prediction accuracy than ROK and OK; (v) the high-risk areas derived from 'atmospheric deposition' were mainly located in the southwest of the study area, and the high-risk areas derived from 'agronomic measure' were scattered in the agricultural land in the north and south of the study area. The above information provided effective spatial decision support for reducing the source-oriented input of the ecological risks of soil HMs in a large-scale area. [Display omitted] • RAPCS/RGWR with land-use type (RAPCS/RGWR-LUT) was proposed to apportion HM source. • RAPCS/RGWR-LUT obtained higher apportionment accuracy than RAPCS/RGWR and APCS/MLR. • Source-oriented ecological risks were obtained by RAPCS/RGWR-LUT and Hakanson PERI. • By incorporating land-use type, RRK got higher prediction accuracy than ROK and OK. • Spatial distribution patterns of the source-oriented risks were revealed by RRK. Capsule : Land-use data were first incorporated into RAPCS/RGWR to improve source apportionment accuracy, and then the spatial distribution pattern of the source-oriented ecological risks of HMs in both surface and subsurface soils were revealed using robust residual kriging. [ABSTRACT FROM AUTHOR]

[Display omitted] • RGWR with land-use types (RGWR-LUT) was proposed to correct in-situ PXRF Pb. • The spatially varying correction coefficient surface was generated by RGWR-LUT. • RGWR-LUT obtained the highest correction quality than traditional correction models. • SGS-PXRF* was proposed to incorporate the corrected PXRF Pb for spatial simulation. • SGS-PXRF* obtained the highest spatial simulation quality. Cost is a priority consideration in soil heavy metal investigation. For the correction of in-situ portable X-ray fluorescence (PXRF), traditionally-used linear regression (LR) is sensitive to outliers and cannot deal with the spatially varying influence of the relevant environmental factors. Moreover, each land-use type usually has a specific impact on the analysis quality of in-situ PXRF but was rarely incorporated into the correction model. This study first proposed robust geographically weighted regression with land-use types (RGWR-LUT) to correct in-situ PXRF Pb in an urban–rural transition area of Changzhou City, China. Next, its correction quality was compared with that of LR, LR with land-use types (LR-LUT), GWR, GWR with land-use types (GWR-LUT), and RGWR. Then, sequential Gaussian simulation with the corrected (by RGWR-LUT) in-situ PXRF as hard data (SGS-PXRF*) was proposed for spatial simulation of soil Pb. Finally, its spatial simulation quality was compared with that of SGS, SGS with the raw in-situ PXRF as hard data (SGS-PXRF), sequential Gaussian co-simulation with the raw in-situ PXRF as co-variable (CoSGS-PXRF), and CoSGS with the corrected (by RGWR-LUT) in-situ PXRF as co-variable (CoSGS-PXRF*). Results showed that: (i) spatially varying correction coefficient surface for in-situ PXRF Pb were obtained by RGWR-LUT; (ii) correction quality for in-situ PXRF Pb was in the order of RGWR-LUT (RI = 81.44%), GWR-LUT (RI = 72.1%), RGWR (RI = 67.83%), GWR (RI = 51.07%), LR-LUT (RI = 48.71%), and LR (RI = 22.72%); (iii) the corrected in-situ PXRF provided more information input than the raw in-situ PXRF for spatial simulation of soil Pb; (iv) SGS-PXRF* obtained the highest spatial simulation quality (RI = 62.5%). It is concluded that RGWR-LUT could effectively correct in-situ PXRF, and SGS-PXRF* could effectively incorporate the corrected in-situ PXRF for spatial simulation of soil heavy metals. [ABSTRACT FROM AUTHOR]