Uncertainty prediction of built-up areas from global human settlement data in the United States based on landscape metrics

The level of landscape heterogeneity may affect the performance of remote sensing based land use / land cover classification. However, the relationship between mapping accuracy of built-up surfaces and morphological characteristics of built-up areas has not been analyzed explicitly, and previous studies typically rely on aggregated landscape metrics to quantify the morphology of built-up areas, neglecting the fine-grained spatial variation and scale dependency of such metrics. Herein, we aim to fill this gap by assessing the associations between focal landscape metrics, derived from binary built-up surfaces, and focal data accuracy estimates. We test our approach for built-up surfaces from the Global Human Settlement Layer (GHSL) for Massachusetts (USA), by examining the explanatory power of landscape metrics for predictive modeling of commission and omission errors in the GHS-BUILT R2018A data product. We find that the Landscape Shape Index (LSI) exhibits the highest levels of correlation to focal accuracy measures. These relationships are scale-dependent, and increase with the level of spatial support. Our results are consistent across different regions within the U.S., and we find that the Recall measure has the strongest relationship to measures of built-up surface morphology across different temporal epochs and spatial resolutions. Regression analysis results (R2>0.9) indicate that it is possible to estimate commission errors in the GHSL in the absence of reference data, and that omission errors in the GHSL can be modeled without accessing the data themselves. Lastly, we test the generalizability of our predictive accuracy models to a different version of the GHSL (i.e., the GHS-BUILT-S2) covering a study area in North Carolina. We find varying levels of model transferability that increases with the spatial support at which landscape metrics and accuracy estimates are calculated.