Reducing the loss of information and gaining accuracy with clustering methods in a global land-use model

Global land-use models have to deal with processes on several spatial scales, ranging from the global scale down to the farm level. The increasing complexity of modern land-use models combined with the problem of limited computational resources represents a challenge to modelers. One solution of this problem is to perform spatial aggregation based on a regular grid or administrative units such as countries. Unfortunately this type of aggregation flattens many regional differences and produces a homogenized map of the world. In this paper we present an alternative aggregation approach using clustering methods. Clustering reduces the loss of information due to aggregation by choosing an appropriate aggregation pattern. We investigate different clustering methods, examining their quality in terms of information conservation. Our results indicate that clustering is always a good choice and preferable compared to grid-based aggregation. Although all the clustering methods we tested delivered a higher degree of information conservation than grid-based aggregation, the choice of clustering method is not arbitrary. Comparing outputs of a model fed with original data and a model fed with aggregated data, bottom-up clustering delivered the best results for the whole range of numbers of clusters tested. (C) 2013 Elsevier B.V. All rights reserved.