Stéphane Mermoz's research while affiliated with French National Centre for Scientific Research and other places

More than half a decade after the launch of the Sentinel-1A C-band SAR satellite, several near real-time forest disturbances detection systems based on backscattering time series analysis have been developed and made operational. Every system has its own particular approach to change detection. Here, we have compared the performance of the main SAR-based near real-time operational forest disturbance detection systems produced by research agencies (INPE, in Brazil, CESBIO, in France, JAXA, in Japan, and Wageningen University, in the Netherlands), and compared them to the state-of-the-art optical algorithm, University of Maryland’s GLAD-S2. We implemented an innovative validation protocol, specially conceived to encompass all the analysed systems, which measured every system’s accuracy and detection speed in four different areas of the Amazon basin. The results indicated that, when parametrized equally, all the Sentinel-1 SAR methods outperformed the reference optical method in terms of sample-count F1-Score, having comparable results among them. The GLAD-S2 optical method showed superior results in terms of user’s accuracy (UA), issuing no false detections, but had a lower producer accuracy (PA, 84.88%) when compared to the Sentinel-1 SAR-based systems (PA∼90%). Wageningen University’s system, RADD, proved to be relatively faster, especially in heavily clouded regions, where RADD warnings were issued 41 days before optical ones, and the one that better performs on small disturbed patches (<0.25 ha) with a UA of 70.11%. Of all the high-resolution SAR methods, CESBIO’s had the best results regarding UA (99.0%). Finally, we tested the potential of three hypothetical combined optical-SAR systems. The results show that these combined systems would have excellent detection capabilities, exceeding largely the producer’s accuracy of all the tested methods at the cost of a slightly diminished user’s accuracy, and constitute a promising and feasible approach for the forthcoming forest monitoring systems.

There is growing interest in using passive microwave observations and Vegetation Optical Depth (VOD) to study the above ground biomass (AGB) and carbon stocks evolution. L-band observations in particular have been shown to be very sensitive to AGB. Here, and thanks to the multi-angle capabilities of the SMOS mission, a new approach to estimate AGB directly from multi-angular L-band brightness temperatures (TBs) is proposed, thus surpassing the use of intermediate variables like VOD. The European Space Agency (ESA) Climate Change Initiative (CCI) Biomass maps for years 2010, 2017 and 2018 are used as AGB reference. AGB estimates from Artificial Neural Networks (ANN) using a purely data-driven approach explained up to 88% of AGB variability globally; Even so, a decrease in retrieval's performance was observed when models are applied to data from years different than the year used for their training. A new training methodology based on multi-year training sets is presented, leading to results showing more stability for temporal analyses. The best set of predictors and an optimal learning data-set configuration are proposed based on an assessment of the accuracy of the estimates. The ANN methodology using brightness temperatures is a promising alternative with respect to the common method of using a parametric function to estimate AGB from VOD. ANNs AGB estimates showed higher correlation with CCI AGB maps (R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> $\sim$ 0.87 instead of $\sim$ 0.84) and presented a stronger agreement with their spatial structure and less differences in residual maps.

Anthropogenic climate change is now considered to be one of the main factors causing an increase in both the frequency and severity of wildfires. These fires are prone to release substantial quantities of CO2 into the atmosphere and to endanger natural ecosystems and biodiversity. Depending on the ecosystem and climate regime, fires have distinct triggering factors and impacts. To better analyse this phenomenon, we investigated post-fire vegetation anomalies over different biomes, from 2012 to 2020. The study was performed using several remotely sensed quantities ranging from visible–infrared vegetation indices (the enhanced vegetation index (EVI)) to vegetation opacities obtained at several passive-microwave wavelengths (X-band, C-band, and L-band vegetation optical depth (X-VOD, C-VOD, and L-VOD)), ranging from 2 to 20 cm. It was found that C- and X-VOD are mostly sensitive to fire impact on low-vegetation areas (grass and shrublands) or on tree leaves, while L-VOD depicts the fire impact on tree trunks and branches better. As a consequence, L-VOD is probably a better way of assessing fire impact on biomass. The study shows that L-VOD can be used to monitor fire-affected areas as well as post-fire recovery, especially over densely vegetated areas.

Forests are one of the key elements in ecological transition policies in Europe. Sustainable forest management is needed in order to optimise wood harvesting, while preserving carbon storage, biodiversity and other ecological functions. Forest managers and public bodies need improved and cost-effective forest monitoring tools. Research studies have been carried out to assess the use of optical and radar images for producing forest height or biomass maps. The main limitations are the quantity, quality and representativeness of the reference data for model training. The Global Ecosystem Dynamics Investigation (GEDI) mission (full waveform LiDAR on board the International Space Station) has provided an unprecedented number of forest canopy height samples from 2019. These samples could be used to improve reference datasets. This paper aims to present and validate a method for estimating forest dominant height from open access optical and radar satellite images (Sentinel-1, Sentinel-2 and ALOS-2 PALSAR-2), and then to assess the use of GEDI samples to replace field height measurements in model calibration. Our approach combines satellite image features and dominant height measurements, or GEDI metrics, in a Support Vector Machine regression algorithm, with a feature selection process. The method is tested on mixed uneven-aged broadleaved and coniferous forests in France. Using dominant height measurements for model training, the cross-validation shows 7.3 to 11.6% relative Root Mean Square Error (RMSE) depending on the forest class. When using GEDI height metrics instead of field measurements for model training, errors increase to 12.8–16.7% relative RMSE. This level of error remains satisfactory; the use of GEDI could allow the production of dominant height maps on large areas with better sample representativeness. Future work will focus on confirming these results on new study sites, improving the filtering and processing of GEDI data, and producing height maps at regional or national scale. The resulting maps will help forest managers and public bodies to optimise forest resource inventories, as well as allow scientists to integrate these cartographic data into climate models.

Most land surface models can, depending on the simulation experiment, calculate the vegetation distribution and dynamics internally by making use of biogeographical principles or use vegetation maps to prescribe spatial and temporal changes in vegetation distribution. Irrespective of whether vegetation dynamics are simulated or prescribed, it is not practical to represent vegetation across the globe at the species level because of its daunting diversity. This issue can be circumvented by making use of 5 to 20 plant functional types (PFTs) by assuming that all species within a single functional type show identical land–atmosphere interactions irrespective of their geographical location. In this study, we hypothesize that remote-sensing-based assessments of aboveground biomass can be used to constrain the process in which real-world vegetation is discretized in PFT maps. Remotely sensed biomass estimates for Africa were used in a Bayesian framework to estimate the probability density distributions of woody, herbaceous and bare soil fractions for the 15 land cover classes, according to the United Nations Land Cover Classification System (UN-LCCS) typology, present in Africa. Subsequently, the 2.5th and 97.5th percentiles of the probability density distributions were used to create 2.5 % and 97.5 % credible interval PFT maps. Finally, the original and constrained PFT maps were used to drive biomass and albedo simulations with the Organising Carbon and Hydrology In Dynamic Ecosystems (ORCHIDEE) model. This study demonstrates that remotely sensed biomass data can be used to better constrain the share of dense forest PFTs but that additional information on bare soil fraction is required to constrain the share of herbaceous PFTs. Even though considerable uncertainties remain, using remotely sensed biomass data enhances the objectivity and reproducibility of the process by reducing the dependency on expert knowledge and allows assessing and reporting the credible interval of the PFT maps which could be used to benchmark future developments.

In this study, we demonstrate the ability of a new operational system to detect forest loss at a large scale accurately and in a timely manner. We produced forest loss maps every week over Vietnam, Cambodia, and Laos (>750,000 km2 in total) using Sentinel-1 data. To do so, we used the forest loss detection method based on shadow detection. The main advantage of this method is the ability to avoid false alarms, which is relevant in Southeast Asia where the areas of forest disturbance may be very small and scattered and detection is used for alert purposes. The estimated user accuracy of the forest loss map was 0.95 for forest disturbances and 0.99 for intact forest, and the estimated producer’s accuracy was 0.90 for forest disturbances and 0.99 for intact forest, with a minimum mapping unit of 0.1 ha. This represents an important step forward compared to the values achieved by previous studies. We also found that approximately half of forest disturbances in Cambodia from 2018 to 2020 occurred in protected areas, which emphasizes the lack of efficiency in the protection and conservation of natural resources in protected areas. On an annual basis, the forest loss areas detected using our method are found to be similar to the estimations from Global Forest Watch. These results highlight the fact that this method provides not only quick alerts but also reliable detections that can be used to calculate weekly, monthly, or annual forest loss statistics at a national scale.

Anthropogenic climate change is now considered to be one of the main factors causing an increase in both frequency and severity of wildfires. These fires are prone to release substantial quantities of CO2 in the atmosphere and to destroy natural ecosystems while reducing biodiversity. Depending on the ecosystem and climate regime, fires have distinct triggering factors and impacts. To better analyse and describe fire impact on different biomes, we investigated pre and post fire vegetation anomalies at global scale. The study was performed using several remotely sensed quantities ranging from optical vegetation indices (the enhanced vegetation index (EVI)) to vegetation opacities obtained at several microwave wavelengths (X-band, C-band, and L-band vegetation optical depth (X-VOD, C-VOD, and L-VOD)), ranging from 2 to 20 cm. It was found that C- and X-VOD are mostly sensitive to fire over low vegetation areas (grass and small bushes) or over tree leaves; while L-VOD depicts better the fire impact on tree trunks and branches. As a consequence, L-VOD is probably a better way of assessing fire impact on biomass. The study shows that L-VOD can be used to monitor fire affected areas as well as post-fire recovery, especially over densely vegetated areas.

Most land surface models can either calculate the vegetation distribution and dynamics internally by making use of biogeographical principles or use vegetation maps to prescribe spatial and temporal changes in vegetation distribution. Irrespective of whether vegetation dynamics are simulated or prescribed, it is not practical to represent vegetation across the globe at the species level because of its daunting diversity. This issue can be circumvented by making use of 5 to 20 plant functional types (PFT) by assuming that all species within a single functional type show identical land–atmosphere interactions irrespective of their geographical location. In this study, we hypothesize that remote-sensing based assessments of above-ground biomass can be used to refine discretizing real-world vegetation in PFT maps. Remotely sensed biomass estimates for Africa were used in a Bayesian framework to estimate the probability density distributions of woody, herbaceous, and bare soil fractions for the 15 land cover classes, according to the UN-LCCS typology, present in Africa. Subsequently, the 2.5 and 97.5 percentile of the probability density distributions were used to create 2.5 % and 97.5 % confidence interval PFT maps. Finally the original and refined PFT maps were used to drive biomass and albedo simulations with the ORCHIDEE model. This study demonstrates that remotely sensed biomass data can be used to better constrain PFT maps. Among the advantages of using remotely sensed biomass data were the reduced dependency on expert knowledge and the ability to report the confident interval of the PFT maps. Applying this approach at the global scale, would increase confidence in the PFT maps underlying assessments of present day biomass stocks.

French Guiana forests cover 8 million hectares. With 98% of emerged land covered by forests, French Guiana is the area with the highest proportion of forest cover in the world. These forests are home to an exceptionally rich and diverse wealth of biodiversity that is both vulnerable and under threat due to high levels of pressure from human activity. As part of the French territory, French Guiana benefits from determined and continuous national efforts in the preservation of biodiversity and the environmental functionalities of ecosystems. The loss and fragmentation of forest cover caused by gold mining (legal and illegal), smallholder agriculture and forest exploitation, are considered as small-scale disturbances, although representing strong effects to vulnerable natural habitats, landscapes, and local populations. To monitor forest management programs and combat illegal deforestation and forest opening near-real time alerts system based on remote sensing data are required. For this large territory under frequent cloud cover, Synthetic-Aperture Radar (SAR) data appear to be the best adapted. In this paper, a method for forest alerts in a near-real time context based on Sentinel-1 data over the whole of French Guiana (83,534 km2) was developed and evaluated. The assessment was conducted for 2 years between 2016 and 2018 and includes comparisons with reference data provided by French Guiana forest organizations and comparisons with the existing University of Maryland Global Land Analysis and Discovery Forest Alerts datasets based on Landsat data. The reference datasets include 1,867 plots covering 2,124.5 ha of gold mining, smallholder agriculture and forest exploitation. The validation results showed high user accuracies (96.2%) and producer accuracies (81.5%) for forest loss detection, with the latter much higher than for optical forest alerts (36.4%). The forest alerts maps were also compared in terms of detection timing, showing systematic temporal delays of up to one year in the optical method compared to the SAR method. These results highlight the benefits of SAR over optical imagery for forest alerts detection in French Guiana. Finally, the potential of the SAR method applied to tropical forests is discussed. The SAR-based map of this study is available on http://cesbiomass.net/.