Spatial-variability-based algorithms for scaling-up spatial data and uncertainties

Abstract

When using remote sensing and geographic information systems, accurately scaling- up spatial data of a variable and their uncertainties from a finer to a coarser spatial resolution is widely required in mapping and managing natural resources and ecological and environmental systems. In this study, four up-scaling methods were derived based on simple and ordinary cokriging estimators and a sequential Gaussian cosimulation algorithm for points and blocks. Taking spatial variability of variables into account in the up-scaling process made it possible to simultaneously and accurately obtain estimates and estimation variances of larger blocks from sample and image data of smaller supports. With the aid of Thematic Mapper imagery, these methods were compared in a case study where overall vegetation and tree covers were scaled up from a spatial resolution of 30/spl times/30 m/sup 2/ to 90/spl times/90 m/sup 2/ with a stratification method at 90/spl times/90 m/sup 2/. The results showed that the methods Point simple coKriging/spl I.bar/Point co-Simulation scaling UP (PsK/spl I.bar/PSUP) and PsK/spl I.bar/Block co-Simulation (PsK/spl I.bar/BS) led to smaller errors and better reproduced spatial distribution and variability of the variables than the other methods. Choosing PsK/spl I.bar/PSUP or PsK/spl I.bar/BS depends on the users' emphasis on accuracy of estimates and variances, computational time, etc. The methods can be applied to multiple continuous variables that have any distribution. It is also expected that the general idea behind the methods can be expanded to scaling-up spatial data for categorical variables.