Towards a Generic Theoretical Framework for Spatially Explicit Statistical Lucc Modelling, Allocation Revisited: Formal Foundations and Bias Identification

Abstract

Spatially explicit statistical  land use and land cover change (LUCC) modelling is implemented in various software, such as the CLUE family, LCM or Dinamica EGO. These tools are now relatively mature, but their conceptual foundations are little discussed in the literature. In particular, these modelling environments exhibit substantially different behaviors for the same problem and the same calibration data.These inter-model inconsistencies are revisited ab initio, focusing on allocation. The concept of allocation bias is defined and allows us to pinpoint a number of conceptual errors, biases and algorithmic inaccuracies in existing LUCC modelling environments. We describe error- and bias-free allocation methods, implemented in our own software CLUMPY, which is shown to significantly outperform existing software in terms of formal correctness and accuracy. It is crucial that validation be performed in explanatory variable space, not on maps, contrarily to the most common practice in the field.

Keywords: land use change, Land Cover Change, Model Development, Model validation, Model Accuracy, Allocation