Efficient treatment of cross-scale interactions in a land-use model

Abstract

Due to significant advances in information technology in the last decades complex computer models have become a common tool in various disciplines. One application is the modeling of global land use. A major challenge in modeling is the linking of processes on different scales such as in land use the local production of agricultural commodities and their global trading. Neglecting these cross-scale interactions leads to significant biases in model projections while a 1:1 representation is computational infeasible. Therefore, a good balance between accuracy and abstraction is essential. In this thesis I investigate efficient implementations of cross-scale interactions in agricultural land-use models. Based on the global land-use model MAgPIE (“Model of Agricultural Production and its Impact on the Environment“) I focus on two dominant aspects: First, the inclusion of spatially explicit data in a global optimization model; second, the proper representation of technological change as a major crossscale interaction and dominant driver for agricultural land use change. As a consequence of limitations in complexity of global optimization models the problem arises that spatially explicit, high-resolution data cannot be used directly as model input. Typically, the spatially explicit data is upscaled by using a static upscaling rule which leads to a significant loss of information. As an alternative I discuss the use of clustering methods for upscaling. I provide a general framework including the creation of clusters, the upscaling of inputs, and the downscaling of outputs. My investigations show that the information loss due to upscaling in the upscaled data itself, but also in the model outputs derived with upscaled data decreases significantly compared to the information loss in upscaling with static grids. Technological change is another important cross-scale interaction. In agriculture technological change means local yield growth induced by supra-regional investments in Research and Development (R&D). Whereas in the past increases in agricultural production were often mainly achieved by expansion of agricultural land, nowadays most increases in total production are outcome of R&D. I present an implementation of this process in MAgPIE including a feedback of land-use intensity on the effectiveness of R&D. To model this feedback I introduce an output-oriented measure for agricultural land-use intensity which takes all sources of intensification into consideration. Based on this measure I show that the effectiveness of investments in R&D decreases with the agricultural land-use intensity. I use this finding to provide an implementation of technological change in MAgPIE. My findings imply that apart from detailedness especially the implementation has a significant impact on general model quality. Therefore, in model development the framework used for implementation should be emphasized to a greater extent.