False discovery rate estimation and control in remote sensing: reliable statistical significance in spatially dependent gridded data

In remote sensing, analysing statistical significance (expressed in terms of <i>p</i>-values) in gridded datasets with thousands of pixels requires addressing the multiple testing problem, which increases the risk of false positives. The false discovery rate (FDR) provides a flexible alternative to traditional correction procedures, yet its application in remote sensing remains underexplored. This research combines FDR estimation via the location-based estimator (LBE) with FDR control using the Benjamini-Hochberg (BH) procedure to enhance the reliability of statistical inference in spatially gridded data. These methods were applied to gridded <i>p</i>-values (<i>p</i>-value map) derived from spatiotemporal Contextual Mann-Kendall (CMK) trend tests using the global MODIS NDVI (Moderate Resolution Imaging Spectroradiometer – Normalized Difference Vegetation Index) MOD13C2 product, highlighting their applicability to scenarios requiring <i>p</i>-value-based corrections. Our findings highlight the complementary strengths of FDR estimation and control, offering a robust framework for addressing large-scale multiple testing challenges in remote sensing under spatial dependence.