The conservation and reduction of rainforest loss are critical for climate change mitigation and preserving ecosystem services. International and national initiatives, driven by the Paris Climate Agreement, aim to avoid tropical forest loss, but their success hinges on accurate information about where and why forests are changing. Current forest carbon monitoring often lacks spatial detail and timeliness, hindering progress towards climate mitigation goals. This is particularly true for Africa's humid forests, where changes remain poorly understood and quantified. The age of national forest inventories (NFIs) in the region (4.5 ± 3.2 years old) underscores the inadequacy of existing data for action-oriented mitigation schemes. Rapid deforestation detection has already shown promise in reducing deforestation and associated carbon emissions. The main causes of forest disturbance in the Congo Basin are small-scale agriculture and selective logging, with regional variations. West and East African forests, including Madagascar, have experienced significant forest loss, while the DRC's remaining large forest fragments face immediate threats. Remote sensing offers increasingly feasible options for monitoring forest carbon changes, with radar-based approaches proving effective in overcoming cloud cover issues and assessing small-scale disturbances. This study aims to provide a high-resolution, spatially explicit, rapid monitoring system for local carbon loss in Africa's humid tropical forests by combining radar-based forest change alerts and carbon stock estimates.

Existing literature highlights the importance of accurate and timely forest carbon monitoring for effective climate change mitigation strategies. Studies have shown the effectiveness of near-real-time deforestation alerts in reducing deforestation rates and associated economic costs. Research has identified the primary drivers of forest disturbance in various regions of Africa, including small-scale agriculture and selective logging, with notable regional variations. Several studies have explored the use of remote sensing technologies, particularly radar-based approaches, to monitor forest cover change and biomass, showcasing their potential to overcome challenges posed by cloud cover and provide high-resolution data. However, a significant gap remains in the availability of high-resolution spatiotemporal data on carbon loss in African rainforests, necessitating the development of more advanced monitoring systems.

This study analyzed the spatiotemporal dynamics of carbon loss in Africa's primary tropical humid forests during 2019 and 2020. The methodology involved combining aboveground carbon estimates (derived from remote sensing and field data) with near-real-time radar-based forest disturbance alerts (RADD alerts) at a 10-meter spatial scale and monthly intervals. The RADD alerts, based on Sentinel-1 data, were categorized into high and low confidence levels, and uncertainty estimates were incorporated at both the pixel and country levels, considering the uncertainties of the carbon map and the commission and omission errors of the alerts. Twenty-three African countries with diverse spatiotemporal patterns of carbon loss were analyzed. The aboveground biomass map for 2018 from the ESA Climate Change Initiative (CCI) was used, with adjustments made to address local biases. A bias model was developed using a random forest approach, incorporating covariates such as biome, topography, and forest fractional cover. The bias-adjusted AGB was converted to carbon values. Carbon loss was calculated for both 10-meter and 100-meter spatial scales, differentiating between core and boundary disturbance alerts. Uncertainty in carbon loss estimates was calculated by propagating the AGB standard deviation and the commission and omission errors of the alerts using specific equations that accounted for both disturbed and undisturbed pixels. Country-level uncertainties were calculated by assuming complete dependence between pixel-level uncertainties. The analysis focused on trends at continental, country, and local scales, recognizing that absolute carbon loss numbers should be treated with caution.

The analysis revealed a total carbon loss of 42.2 ± 5.1 MtC yr⁻¹ in 2019 and 53.4 ± 6.5 MtC yr⁻¹ in 2020 for Africa's primary tropical humid forests. Nine countries accounted for 95% of the total gross losses. The Democratic Republic of Congo (DRC) and Cameroon were the largest contributors, with significant annual increases between 2019 and 2020. Madagascar showed the highest annual increase in carbon loss (+153.9%), while Equatorial Guinea showed a decrease (-20.1%). High temporal detail revealed diverse monthly patterns of carbon loss, strongly correlated with local rainfall patterns. Countries like Cameroon, Liberia, Nigeria, Central African Republic (CAR), and Madagascar exhibited clear dry-wet seasonal variations, while the DRC and Republic of Congo showed two dry-wet season variations. The seasonal variations are likely due to increased forest accessibility during dry months. The study identified significant differences between months with the highest and lowest carbon losses, with some countries exhibiting a concentration of carbon loss within a few months (e.g., 75.7% of CAR's annual loss occurred within the first three months of 2020). Monthly carbon loss patterns showed high correlations between 2019 and 2020 for several countries. Spatial analysis revealed several hotspots of carbon loss, with high spatial and temporal details illustrating different drivers such as logging roads, selective logging, mining, oil palm plantations, urban expansion, and smallholder agriculture. The detailed spatiotemporal data enables improved targeting and prioritization of enforcement activities and prediction of future carbon loss patterns.

This study's high-resolution spatiotemporal assessment of carbon loss in African rainforests offers significant improvements over existing global-scale tabular statistics. The monthly carbon loss estimates provide valuable insights for local, national, and international forest initiatives and global carbon policy goals. The use of open-source datasets and cloud computing platforms enables cost-effective national-level monitoring. The near-real-time reporting allows for prompt action to protect threatened forests and allows countries to adjust practices to meet climate change mitigation commitments. The findings underscore the importance of integrating high-resolution spatial and temporal data into carbon accounting and reporting, enhancing transparency and accuracy. The system's relatively low cost and reliance on open-source data make it readily adaptable and scalable for use in other regions.

This study presents a novel framework for estimating tropical forest aboveground carbon loss with high spatiotemporal resolution, providing valuable information for improved implementation and enforcement of forest conservation efforts. The spatially explicit analysis enhances transparency, transferability, and speed of reporting carbon losses. The continentally comprehensive dataset can be easily adapted to incorporate new data, offering a benchmarking approach for tracking progress toward the Paris Agreement goals. Future research could focus on refining biomass estimation techniques, improving the accuracy of forest disturbance alerts, and integrating additional data sources (e.g., land cover succession, drivers of carbon loss) to enhance the precision and robustness of carbon loss estimates.

The study used RADD alerts with a minimum mapping unit (MMU) of 0.2 ha, meaning events smaller than this are not captured. The global nature of the forest baseline product may result in local inconsistencies. Persistent cloud cover in 2018 may have caused an overestimation of carbon loss at the beginning of 2019. The alerts do not distinguish between human-induced and natural disturbances. Uncertainty remains in quantifying degradation levels, although the use of core and boundary alerts helps to address this. The aboveground biomass baseline map was from 2018, potentially introducing uncertainty. Biomass estimation is challenging in mountainous terrain and complex canopies. Despite efforts to address biases in the aboveground biomass map, some uncertainties remain. The study focused on local carbon loss patterns rather than providing precise area estimates for forest disturbances.