Error assessment in a soil acidification modelling study: efficiency issues and change of support

The soil acidification model SMART2 requires 25 input variables for each point support location it is run. For the regional modelling of soil acidification, the model is run at a dense grid of point locations, and point support model output (aluminium concentration below the root zone) is spatially aggregated to values for 10 km × 10 km blocks. Monte Carlo analysis was used to assess the uncertainty in model output due to uncertainty in model input variables. Spatial correlation of input variables had to be incorporated to get realistic values for the uncertainty in block aggregated model output. To increase the efficiency of the Monte Carlo analysis, Latin hypercube sampling of Gaussian random fields was used. For several sample sizes the efficiency of Latin hypercube sampling is compared to simple random (iid) sampling.