Components of uncertainty in primary production model: the study of DEM, classification and location error

The use of geographic information system (GIS)-based ecological models is increasing and input datasets of these models are improving daily. Still, there is a notable gap in quantifying the uncertainty related to these models. Quantifying uncertainty in spatial ecology is indeed crucial because it may improve the support that GIS provides for decision support systems. This article aims to quantify uncertainty and error propagation in a dynamic GIS model that predicts ecosystem productivity in dry environments. This was done through the following operative objectives: (1) comparing the contribution to model uncertainty of topographic error with classification error; (2) testing whether the uncertainty contributed by the secondary topographic index (radiation layer) is greater than the uncertainty contributed by the primary topographic indices (aspect or slope); and (3) quantifying the contribution of the location error to model uncertainty. The research was applied in four steps: (1) spatial database design and collection of validation data; (2) standard error determination, based on statistical indices for simulation; (3) development of simulation codes to assess the uncertainty and error propagation of the environmental variables; and (4) determination of the hierarchy of uncertainty factors. The results show that the contribution of the DEM layer to the model uncertainty is substantial, as opposed to the negligible uncertainty contributed by the rock map. The error simulation results were found to be different among subregions and were dependent on slope gradient and error magnitude. Error propagation from the secondary topographic index (radiation layer) was occasionally found to contribute less to the model uncertainty than the primary topographic index (aspect). It was also found that location error correction has only a small positive effect on the model's predictability. The reason is related to the limited ability to determine location error, because here the correction method was spatially uniform and based on visual interpretation. Future research can focus on the assessment of model behavior with different DEM spatial resolutions to find the best resolution for prediction. There is a need to analyze the effect of the model's climatologic variables to better understand their uncertainty effect on the model's temporal dimension. In addition, there is a need to develop a unique algorithm that will make an optimal assessment of spatial nonuniform correction of the location error.