Accounting for change of support in spatial accuracy assessment of modelled soil mineral phosphorous concentration

Agricultural activities in the Netherlands cause high nitrogen and phosphorous fluxes from soil to ground- and surface water. A model chain (STONE) has been developed to study and predict the magnitude of the resulting ground- and surface water pollution under different environmental conditions. STONE has three main components, namely: 1) the fertiliser distribution model CLEAN; 2) the atmospheric transport and deposition model OPS; and 3) the soil and soil-water quality model ANIMO. The goal of this study was to assess the accuracy of the STONE model output by comparing model predictions with independent observations, while accounting for spatial variation and differences in spatial support. The specific STONE output considered was mineral phosphorous concentration in the top soil. Predictions from STONE are made at the block support and, specifically, for 500 × 500 m grid cells, whereas the observations are on a point support. Therefore, aggregation of the point observations or disaggregation of the STONE model predictions is needed in order to test the accuracy of the STONE model. This study used the aggregation route. Specifically, ordinary and regression block kriging were used to aggregate the point observations to the block support. Overall, there was a good correspondence between the kriged observations and the STONE predictions, with no evidence of bias in the model predictions. However, there were meaningful differences locally, which was partly attributed to the smoothing of the kriging interpolator. Closer inspection revealed that not all of the differences could be attributed to measurement errors in the phosphorous observations and interpolation errors in the kriged phosphorous maps. Thus, it was demonstrated that the STONE output suffers from input and/or model errors. However, quantification of these errors was restricted by the fairly large interpolation uncertainty of the kriged phosphorous observations. Nevertheless, the spatial pattern of error in the STONE predictions could be partly attributed to combinations of landuse and soil type