ABSTRACT
The raster data structure stores categorical and continuous field data for spatial analysis, environmental modeling, and resource planning. With rapidly advancing sensor networks, the spatial resolution of data is increasing, sometimes outpacing the optimum resolution for applications. Overcoming granularity differences between raw and “analysis ready” data often requires upscaling source data to a desired target map with the goal of maintaining the structure and spatial variance of the higher resolution data. Common strategies for resampling categorical data (nearest neighbor and majority rule) force users to choose between preserving map structure and map variety. A new method is presented here that integrates global and zonal class proportions to guide the optimal allocation of classified cells. This technique provides more representative maps with respect to variety and structure, better retains minority classes, and produces higher (or equal) levels of user’s and producer’s accuracy than the traditional methods. An R-based implementation is provided that has serviceable run times, and the performance of the algorithm is shown to be scalable, proving the tool widely usable.
Acknowledgments
This paper benefited greatly from comments by two anonymous reviewers to whom we are grateful.
Disclosure statement
No potential conflict of interest was reported by the authors.
Data and code availability
The data supporting the findings of this study are made available by the Multi-Resolution Land Characteristics (MRLC) Consortium and were derived from resources available in the public domain: [https://www.mrlc.gov/data]. The code that supports the findings of this study is currently available on GitHub [https://github.com/mikejohnson51/resample]. An archive of the software at time of publication is available here [https://doi.org/10.5281/zenodo.4543984] which was version 0.1.0.