An integrated object-based sampling approach for validating non-contiguous forest cover maps in fragmented tropical landscapes

• The proposed sampling method combines proportional allocation, stratified random sampling, and distance optimization. • The accuracy of the 25 m JAXA PALSAR-2/PALSAR FNF map over Nigeria was assessed. • The estimated overall accuracy (81.1%) was closest to the true accuracy (82.4%) for a subset region. • The method captured landscape variability better than conventional approaches. Validating forest cover maps is essential for evidence-based conservation and sustaining ecosystem services. However, complex spatial patterns in fragmented tropical forest landscapes—often comprising non-contiguous forest patches, interspersed with agricultural lands and other land cover types—pose considerable difficulties for accuracy assessment using conventional techniques. To address this, we developed an integrated object-based sampling (IOBS) method that combines stratified random sampling, proportional allocation, and sample distance optimization. The IOBS method was applied to assess the accuracy of the Japan Aerospace Exploration Agency (JAXA) global 25 m PALSAR-2/PALSAR forest/non-forest (FNF) 2020 map across 14 ecoregions in Nigeria. Its performance was compared to simple random, systematic, and stratified random sampling using the coefficient of variation (CV), heterogeneity index (HI), and true accuracy metrics. IOBS demonstrated substantially higher spatial variability (CV = 109.37) and heterogeneity (HI = 0.21) compared to other methods (CV = 28.84–53.93, HI = 0.05–0.11). The IOBS estimated an accuracy of 81.1 %, closely aligning with the true accuracy of 82.4 % and outperforming other methods (75.3 %–79.7 %). The higher performance of IOBS stems from its ability to capture a broad range of forest conditions—from extensive contiguous cover to small, fragmented patches—while minimizing spatial autocorrelation through distance optimization. By better representing local heterogeneity, IOBS offers a robust and precise framework for validating categorical forest cover maps in complex tropical landscapes, advancing accuracy assessment practices for remote sensing applications.