Consistent Pantropical Deforestation Monitoring in Dense Humid Forests from Landsat Time Series (2000–2025)

Land use and land cover change (LULCC) remains one of the largest and most persistent sources of uncertainty in the global carbon budget, limiting confidence in estimates of terrestrial carbon sources and sinks. Within LULCC, deforestation in tropical dense humid forests contributes a substantial share of emissions due to high biomass stocks and continued land conversion across major tropical regions. Despite the availability of several global forest change datasets, estimates of annual deforested area differ widely. Variations in forest definitions, disturbance detection methods, and temporal attribution lead to inconsistent estimates of both the magnitude and timing of forest loss, which propagate directly into uncertainty in LULCC emission estimates used in global carbon budget (GCB) assessments and Earth system models.

Consistent monitoring of tropical deforestation is particularly challenging because of persistent cloud cover, heterogeneous disturbance processes, and strong spatial and temporal variability in forest loss dynamics. These challenges are most pronounced in regions such as the Congo Basin, where observational limitations lead to uneven detection performance and reduced comparability across datasets. Improving the spatial and temporal consistency of deforestation estimates across tropical regions is therefore critical for reducing uncertainty in LULCC emissions and for supporting model evaluation within the GCB.

The objective of this study is to improve the spatial and temporal consistency of pantropical deforestation estimates derived from optical satellite data over the last 25 years. We present a consistent, high-resolution deforestation monitoring approach based on Landsat Analysis Ready Data, with application to Amazonia, Central Africa, and Southeast Asia.

Deforestation is detected within masks of intact tropical forests. To improve robustness in persistently cloudy environments, standard Landsat Quality Flags are complemented by a regionally adaptive, cloud-tolerant masking strategy, enabling the construction of continuous spectral time series suitable for long-term analysis. Deforestation signals are identified using Normalized Burn Ratio time series combined with forest-based local standardization. This yields a statistical change indicator designed to balance sensitivity to disturbance with robustness to noise and data gaps. Candidate deforestation events are further refined using complementary spatial and temporal metrics, including data availability constraints, disturbance amplitude, and spatial proximity to neighbouring events, enhancing coherence while limiting spurious detections. Parameter calibration and optimisation are conducted independently, with reference to existing operational monitoring systems, including Global Forest Change and Tropical Moist Forest products.

Evaluation using a targeted two-level validation framework combining spatial intersections and temporal stratification indicates reduced bias and root-mean-square error in annual deforestation estimates relative to widely used global datasets. Improved detection performance is particularly evident in observationally challenging regions such as the Congo Basin. Analyses in Amazonia and Southeast Asia are ongoing and already show coherent spatial patterns and realistic temporal dynamics.

Overall, this work demonstrates that harmonized, high-resolution optical time series can provide more consistent estimates of tropical deforestation, supporting improved quantification of LULCC emissions. By reducing discrepancies in annual forest loss estimates, the approach provides a more stable observational basis for GCB assessments and Earth system model evaluations like IPCC assessments.