A method and application of multi-scale validation in spatial land use models

The majority of the large number of existing land use models lack a proper validation, often because of data problems. Moreover, despite recognition o f the necessity to incorporate a multi-scale analysis, scale dependencies a re normally not considered during validation. In this paper, a multi-scale land use change modelling framework, conversion of land use and its effects (CLUE), is calibrated for Costa Rica and validated at five spatial resolut ions for Honduras and Costa Rica. Both countries experienced locally very s trong actual land use changes. Calibration runs show that the model is very sensitive to changes in the autonomous development parameter, which define s the influence of the finest resolution. Validation results are very satis factory for both countries. Especially, changes in major land use types are reproduced with the model. Changes in localised land use types are more di fficult to project. The magnitude of gains and magnitude of losses are slig htly underestimated in all cases. The multi-scale validation demonstrates t hat results improve strongly, and exponentially, with decreasing spatial re solution. Strong reduction of the number of observations results in a corre lation between actual and modelled changes that approximates the perfect va lue of 1. The study demonstrates that the CLUE modelling framework can repr oduce changes as they took place in Central America in the 1970s and 1980s, and shows how conclusions can differ depending on the scale at which valid ation is performed. (C) 2001 Elsevier Science B.V. All rights reserved.