Pixel time series-based quantification and validation to improve spatiotemporal analysis of land cover data

• Map errors in land cover time series manifest as Switching and Multi-transitions. • Post-Classification Comparison (PCC) cannot detect Switching and Multi-transitions. • Pixel Time Series (PTS) can isolate potential errors from these trajectories. • Error propagation amplified through PCC can be prevented through PTS. • Wider adoption of PTS needed to better analyse and validate land cover time series. As land cover data continue to be produced at higher resolutions and over longer time series, methods of quantifying land change that can account for potential errors from misclassification are increasingly needed. Using a case study of 30-m annual forest change across 3,937,655 km 2 of Southeast Asia, we compared analytical outcomes from two techniques: Post-Classification Comparison (PCC), the pairwise differencing of land cover maps, and Pixel Time Series (PTS), wherein the entire collection of transitions across a time series is simultaneously assessed. We found that PTS isolated two time series phenomena—switching (oscillations between two land cover classes) and multi-transitions (≥ 2 transitions over ≥ 3 classes)—of which 92.3% were erroneous. By contrast, those errors remained undetected through PCC and were instead directly propagated into analysis, leading to considerable overestimation of gross and net forest change. Thus, incorporation of PTS will readily benefit a variety of higher-order spatial applications, including within the Measurement, Reporting, and Verification (MRV) domain. A paradigm shift away from PCC towards a PTS-based mode of geographical analysis is needed to more accurately leverage the increasingly dense spatiotemporal information provided from new land cover time series databases.