Rapid remote monitoring reveals spatial and temporal hotspots of carbon loss in Africa's rainforests

The study addresses the need for timely, spatially explicit monitoring of tropical rainforest aboveground carbon to inform conservation actions, enforcement, and transparent reporting under climate frameworks such as the Paris Agreement. Existing forest carbon monitoring has improved but often lacks sufficient spatial detail and timeliness, particularly for Africa’s humid forests where national forest inventories are several years out of date. Prior work shows that rapid disturbance detection can reduce deforestation, and that main disturbance drivers in the Congo Basin include small-scale agriculture and selective logging, with regional variation. Advances in open-access radar (Sentinel-1) and biomass mapping enable frequent, cloud-penetrating, high-resolution monitoring. The authors aim to combine near-real-time radar-based disturbance alerts with aboveground carbon stock maps to estimate local carbon loss at 10 m and monthly resolution across Africa’s primary humid forests, quantify uncertainties, and reveal spatial and temporal hotspots to support targeted interventions and reporting.

Background highlights include: (1) documented impacts of rapid deforestation alerts on reducing deforestation probability and the significant social cost of carbon benefits; (2) quantified drivers of Congo Basin forest disturbance (e.g., ~84% small-scale agriculture, ~10% selective logging overall, with >60% logging in Gabon and >90% small-scale agriculture in DRC and CAR); (3) extensive historical loss of West and East African tropical forests, with remaining large fragments in DRC under threat; (4) projections of a declining African tropical forest carbon sink; and (5) advances in remote sensing—multi-source time series, biomass mapping, and operational radar-based disturbance detection at 10 m overcoming cloud cover. These studies motivate the need for high-resolution, timely carbon loss monitoring and contextualize expected seasonal and regional patterns.

Study area: Primary tropical humid forest across 23 African countries. A 2018 reference mask was constructed from 2001 primary tropical forest extent minus 2001–2018 losses, excluding mangroves. Forest disturbance alerts: RADD (Radar for Detecting Deforestation) alerts derived from Sentinel-1 time series for 2019–2020. Forest disturbance is complete or partial tree cover removal within a 10×10 m pixel. Alerts are issued via Bayesian updating with two confidence levels: low confidence (>0.85 probability) and high confidence (>0.975), confirmed within up to 90 days. An MMU of 0.2 ha was used (validation scale). Validated user’s and producer’s accuracies for high-confidence alerts ≥0.2 ha were 97.6% and 95.0%, respectively. Pixels were classified as core (8-connected neighbors; assumed complete removal) or boundary (fewer neighbors; assumed partial removal). Aboveground biomass (AGB): ESA CCI Biomass v2 map (2018) at 100 m with per-pixel standard deviation, based on Sentinel-1 C-band and ALOS-2 PALSAR-2 L-band SAR with semi-empirical inversion and allometry. Known biases (e.g., underestimation >250 Mg/ha) were modeled and corrected using research/forestry plot data, textural metrics, biome, terrain, forest cover, and the AGB SD layer via random forest (10-fold CV; RMSE 42.24 Mg/ha; MAE 29.25 Mg/ha). Statistically significant predicted bias (via infinitesimal jackknife SEs) was used to correct AGB. AGB and SD were converted to carbon using a 0.47 factor. Carbon loss estimation: Carbon loss was computed at 10 m (0.01 ha) and 100 m (1 ha) to align with alerts and AGB. At 10 m, loss equals the fraction (1%) of the 1 ha AGB pixel associated with the alerted 10 m pixel, with complete loss (100%) for core alerts and 50% for boundary alerts. At 1 ha, loss equals the fraction of alerted 10 m pixels times the 1 ha carbon stock, with the same core/boundary treatment. Monthly aggregation was performed for 2019–2020, with separate reporting for high and low confidence alerts (low confidence prevalent in the last three months of 2020 due to confirmation lag). Uncertainty: Per-pixel uncertainty propagated AGB SD with alert commission (2.4%) and omission (5%) errors. For disturbed pixels, var(CL) = AGC^2×0.0234 + 0.976^2×stdev_AGC^2 + 0.0234×stdev_AGC^2. For non-disturbed pixels, var(CL) = AGC^2×0.0475 + 0.05^2×stdev_AGC^2 + 0.0475×stdev_AGC^2. Country-level uncertainties assumed complete dependence (conservative), aggregating pixel SDs and using an expanded product-variance formulation: var(CL)_country = (stdev(AGC)^2 + AGC^2) × (0.0234 + 0.976^2) × (0.0475 + 0.05^2) − AGC^2 × 0.976^2 × 0.05^2. Results are expressed as mean ± SD. Drivers were illustrated via local examples (logging roads/selective logging, mining, oil palm, urban expansion, smallholder agriculture). Implementation used open data and Google Earth Engine.

- Continental totals: Carbon loss in Africa’s primary humid forests was 42.2 ± 5.1 MtC/yr in 2019 and 53.4 ± 6.5 MtC/yr in 2020. - Concentration among countries: 9 of 23 countries accounted for 95.0% (2019) and 94.3% (2020) of gross losses; these hold ~95.7% of the forest area. DRC accounts for 52.8% of forest area; Gabon 11.8%; Republic of the Congo 11.0%; Cameroon 9.8%. - Country shares and changes: DRC and Cameroon contributed 49.3% and 19.1% of losses in 2019, and 44.7% and 20.6% in 2020, respectively. Annual increase 2019→2020: DRC +15.0%, Cameroon +36.5%. Among countries with ≥1 MtC emissions across the two years, Madagascar had the highest annual increase (+153.9%), while Equatorial Guinea decreased (−20.1%). - Temporal hotspots: CAR had 75.7% of 2020 annual loss concentrated in Jan–Mar; Nigeria 73.9% (Jan–Mar 2020); Liberia 73.1% (Feb–Apr 2020); Madagascar 70.7% (Sep–Nov 2020); Cameroon 62.2% (Jan–Mar 2020). DRC and Republic of the Congo showed lower concentration due to mixed seasonality (DRC 36.7% and Republic of the Congo 32.8% in Jan–Mar 2020). Madagascar showed a 72× difference between the highest and lowest loss months in 2019 (March vs November). - Seasonality: Countries like Cameroon, Liberia, Nigeria, CAR, and Madagascar had clear dry–wet seasonal patterns; DRC and Republic of the Congo exhibited two seasonal cycles due to latitudinal extent. - Year-to-year consistency: Monthly loss patterns between 2019 and 2020 were highly correlated: CAR r=0.94, DRC r=0.92, Madagascar r=0.91, Gabon r=0.90, Cameroon r=0.83; Republic of the Congo r=0.51. - Cumulative profiles: Liberia, Nigeria, CAR, and Cameroon reached 70–90% of annual loss by April; Madagascar reached 60% by October. DRC, Gabon, Republic of the Congo, Equatorial Guinea, and Ghana showed more gradual accumulation. - Local hotspots and drivers: High-resolution maps revealed disturbances linked to logging roads/selective logging (CAR), mining (Republic of the Congo), oil palm development (Cameroon), urban expansion (Equatorial Guinea), and smallholder agriculture (DRC). - Uncertainty: Balanced, small omission and commission errors of alerts and bias-adjusted biomass product yielded uncertainties propagated at pixel and country scales; uncertainties visualized alongside losses.

The integrated near-real-time radar alert and biomass approach delivers spatially explicit, monthly carbon loss estimates that address critical information gaps for Africa’s humid forests. The findings reveal distinct seasonal and interannual patterns and concentrated temporal hotspots, enabling targeted enforcement during peak months and regions. High year-to-year correlation of monthly profiles for many countries suggests predictable patterns that can inform proactive interventions and early estimation of annual losses. The spatial detail links losses to likely drivers (e.g., logging, agriculture, mining), and, combined with auxiliary datasets (e.g., concessions, protected areas), can support assessments of legality and compliance. The approach enhances transparency, completeness, and timeliness for national REDD+ MRV and UNFCCC reporting, and supports rapid decision-making to meet climate mitigation commitments. Open-source data and cloud computing facilitate cost-effective national implementations and large-area comparative analyses.

The study introduces a high-resolution, rapid monitoring framework for estimating aboveground carbon loss across Africa’s primary humid forests by combining Sentinel-1-based disturbance alerts with bias-adjusted biomass maps. The framework produces spatially explicit, monthly estimates with uncertainty quantification, revealing spatial and temporal hotspots and consistent seasonal patterns. These outputs can improve on-the-ground enforcement, accelerate and enhance transparency of reporting for climate policies, and provide a continent-wide benchmark adaptable to new datasets and missions. Future research should integrate improved, up-to-date biomass maps, multi-sensor (optical–radar) alerts, and forthcoming missions (e.g., GEDI, BIOMASS, NISAR) to refine degradation quantification, reduce biases, and further increase accuracy and timeliness.

Key limitations include: (1) Alert MMU of 0.2 ha—events smaller than 0.2 ha are more uncertain and may be underrepresented; (2) Use of a global 2018 humid forest baseline derived from optical data may introduce local inconsistencies and missed late-2018 losses (due to cloud cover) that were detected in early 2019, potentially inflating early-2019 losses; (3) Alerts do not distinguish human-induced vs natural disturbances; (4) Boundary pixels assume 50% carbon loss, which may not capture variable degradation severity; (5) Biomass map limitations include underestimation at high biomass (>250 Mg/ha), local biases in wet and complex canopies, and challenges in mountainous terrain; (6) Uncertainty propagation assumes complete dependence at country scale (conservative), and the analysis does not model land cover successions or explicitly apportion drivers; (7) Reliance on model-based bias correction due to limited national plot data; (8) Low-confidence alerts present in the last three months of 2020 due to confirmation lag. Anticipated improvements include integrating optical and radar alerts, higher-resolution products, and leveraging new missions (GEDI, BIOMASS, NISAR) for better biomass and degradation estimates.