Towards localized accuracy assessment of remote-sensing derived built-up land layers across the rural-urban continuum

TL;DR

This paper evaluates methods for assessing the localized accuracy of remote-sensing built-up land maps across rural to urban areas, addressing challenges like class imbalance and small sample sizes.

Contribution

It investigates the effectiveness of spatial agreement measures for local accuracy assessment of built-up land datasets across the rural-urban spectrum.

Findings

Certain spatial agreement measures are more robust to class imbalance.

Local accuracy varies significantly across rural-urban areas.

The study provides guidelines for better accuracy assessment in remote sensing data.

Abstract

The accuracy assessment of remote-sensing derived built-up land data represents a specific case of binary map comparison, where class imbalance varies considerably across rural-urban trajectories. Thus, local accuracy characterization of such datasets requires specific strategies that are robust to low sample sizes and different levels of class imbalance. Herein, we examine the suitability of commonly used spatial agreement measures for their localized accuracy characterization of built-up land layers across the rural-urban continuum, using the Global Human Settlement Layer and a reference database of built-up land derived from cadastral and building footprint data.