Spatial structure characterization of the geographic environment for intelligent spatial prediction model construction

Keywords: Spatial structure, Index design, Spatial prediction, Intelligent model construction
Abstract Type: Paper Abstract

Abstract

Geographic conditions are essential factors to be considered in the intelligent construction of geographic analysis models. However, existing intelligent model construction methods only characterize geographic conditions with generalized metrics, such as elevation range, average slope. Such generalized characterization ignores the impact from varied spatial structure of the geographic environment on model performances. To achieve consideration of spatial structure in model construction, we designed an index to quantify the spatial structure characteristics of given tasks. The designed index can also be used to specify the model’s requirements for spatial structure characteristics. In this research, a Multivariate Spatial Structure (MuSS) index is designed to quantify the spatial structure of the geographic environment characterized with multiple geographic variables. The designed index quantifies the spatial structure by comparing if locations with closer spatial relationships are more similar in geographic conditions. A set of spatial prediction case base of a consistent analytical scale is constructed to evaluate the effectiveness of the designed index. Results attest that, when MuSS index suggests autocorrelated pattern of the geographic environment, higher index is related to better prediction accuracies of the ordinary kriging model compared to the random forest model. The designed index can indicate the suitability of spatial prediction models for given spatial tasks and assist the intelligent model construction process.