A GIS-based plant prediction model for wetland ecosystems

An existing non-spatial model for the prediction of response of wetland pla nt species on ecological factors has been transformed into a GIS-based pred iction model which produces spatial output at the landscape scale. The inpu t, spatial patterns of the ecological factors, were constructed with geosta tistical spatial interpolation (kriging). With this GIS-based model the spa tial patterns of presence and absence of 78 wetland plant species are predi cted for an area with wetlands in the Netherlands of approximately 500 squa re kilometers. The GIS-based model has been validated, and the estimated un certainty of the input has been propagated through the model. At the specie s level the output shows spatially coherent and non-random patterns. The va lidation is affected by the propagation of input errors through the model. The number of valid predictions declines approximately 10-20% when 95% conf idence intervals are used in the validation. This study shows that it is fe asible to use a geostatistical interpolation method to construct spatial pa tterns of ecological factors on a landscape scale and to use these patterns as input for a GIS-based prediction model. The added uncertainty on the in put values however, affects the number of valid predictions of the model.