Tropical forest mapping and monitoring is a key application domain for EO due to the need for recurrent and frequent data to produce annual information on forest cover in humid and seasonal domains and regular information on forest disturbance processes. It benefits from the long-term consistent archive of Landsat imagery for forest area change and from the start of operation of Sentinel-2 in 2016.

Deforestation is not only associated with a significant part of the global CO2 emissions, but also with a number of adverse consequences including loss of biodiversity and soil erosion. In addition, degradation (i.e. the loss of carbon and ecosystem services in remaining forests, due to harvesting and small-scale disturbances) remains difficult to quantify but could add further emissions of the same magnitude as deforestation for some countries. It is therefore of prime importance to provide better information on forest dynamics to tropical countries and to the international scientific community.

REDD+ provides a framework for policy approaches and positive incentives supporting forest mitigation in developing countries, which requires robust, flexible, consistent, accurate and transparent national forest monitoring systems (NFMS) to measure performance against reference levels and to assess different types of forest in the country, including "natural forest".

Only a few countries in the tropics have developed operational forest cover monitoring systems capable to meet REDD+ reporting requirements (e.g. Brazil for the Brazilian Amazon, Mexico, Indonesia and India). In the past, a number of initiatives (including from EC) have aimed at developing National Forest Monitoring Systems. However, a large majority of these monitoring systems remains not fully compatible with REDD+ requirements yet.

The quality, spatial resolution and frequency make Sentinel-2 data essential for all countries of the three main tropical forest basins in order to help establish national and/or jurisdictional forest reference levels and to develop adequate NFMSs that allows allow for identifying small-scale disturbances and slow conversion changes.

The Sentinel-2 Global Mosaic (S2GM) is a service within the Copernicus Global Land Monitoring service that has been tailored to fulfill the requirements of REDD+ monitoring. It provides convenient and free access to scientifically sound, analysis-ready surface reflectance data from Sentinel-2, which are offered as temporal composites with different compositing periods from one day to one year (Level 3). The S2GM products are based on ESA’s standard L2A product as inputs, i.e. an atmospheric correction has been applied. Before mosaicking and compositing, the input data undergoes a rigorous quality control and all identified clouds are removed. By this means, S2GM substantially facilitates the availability of adequate EO data for REDD+ forest monitoring.

S2GM products are generated on request via a state-of-the art web service, which allows for a tailoring by the user with respect to a number of properties including the area of interest, compositing period, spatial resolution, spectral bands, projection, file format, and metadata content. The S2GM algorithm then selects an original observation per pixel, which optimally represents the time-series of observed spectra in the given compositing period. Depending on the number of valid observations in the compositing period, either a multi-dimensional median, the MEDOID algorithm, or a Short-Term composite algorithm, which has been adapted from Landat’s WELD algorithm, is used to select the optimal observation. It is worth noting here, that no further processing is applied on the data to manipulate the spatial variability of the S2GM products.

We present here how the S2GM service supports REDD+, explain how to use it, and demonstrate its operational readiness as well as its limitation by showing typical examples of land use and forest monitoring at different locations on the planet.

Area A - Board 1

Forest degradation and regrowth make significant contributions to the Earth’s global carbon cycle. The subtle loss of tree cover associated with degradation is responsible for a large proportion of the anthropogenic induced carbon emissions to the atmosphere and has insidious negative effects on global climate. Conversely, tree cover re-growth can sequester carbon from the atmosphere at a faster rate than stable forests and may serve to mitigate the effects of climate change locally and regionally. Methods for detecting tree cover loss from remotely sensed data sources, in the form of complete canopy removal, are well advanced in the literature and in operational practice. However, subtle changes in tree cover are difficult to detect using traditional remote sensing techniques and operational methods for characterizing and monitoring such changes are in their infancy. Time-series analysis of remotely sensed Landsat imagery, with its relatively frequent cadence and long historical record, adds dimensionality to satellite image-based classifications that enables the detection of more subtle forest landscape changes. The cloud computing platform, SEPAL, provides access to analysis ready data and user friendly interfaces to perform time series analysis. Users can explore parameterization and algorithms, notably the Breaks for Additive Season and Trend (BFAST) algorithms, which have the capacity to capture the magnitude of change at the national scale. Verification and identification of the drivers of change are collected through probability sample based assessments using a time series of very high resolution imagery. Results on forest degradation and regrowth are combined with information of the biophysical land characteristics, land use and population to identify areas with potential for forest restoration. Such tools and methods are invaluable for country practitioners to incorporate forest degradation and regrowth in international reporting such as that required by the United Nations Framework Convention on Climate Change (UNFCCC) and to reach emission reduction targets proposed in National Determined Contributions (NDC) under the Paris Agreement.

Area A - Board 2

Context/Purpose
The Paris Agreement calls on developed and developing countries to further REDD+ actions, including providing performance-based finance. Cameroon is advancing from readiness towards results-based payments and has elaborated an emission reduction project idea note, which was submitted and approved by the FCPF Carbon Fund. The ERP area is in the South Cameroon plateau straddling the bimodal and monomodal rainfall agro-ecological zones and covering an area of 93.328 km² corresponding to about 20% of the total area of Cameroon and representative of the range of tropical moist forest.
In light of the challenges of designing and operationalizing such a program, the country is actively working on consolidating a methodology to provide reliable statistics related to historic forest loss in order to establish the program forest reference (emission) level and build a robust monitoring system.
This study presents a 2-stage sampling approach to estimate forest loss in the ER Program area for the period 2000 to 2015.
Methods
A 2-stage sampling approach is used to estimate forest loss uncertainty in the ERP area.
The first stage consists of mapping forest cover and forest loss based on the interpretation of Landsat images for the period 2000 to 2015. A reference map is created using a combination of a semi-automated classification procedure and manual enhancements to ensure the greatest possible accuracy .
The second stage consists of the validation of map statistics by stratified random sampling method based on the interpretation of segments of high resolution satellite data: SENTINEL and Landsat . 100 × 100 m segments or sampling units representing about 0.01% of the study area are visually interpreted by a team of photo-interpreters independently from the map production team to produce a reference dataset.
The sampling unit outputs are used to estimate the proportion forest cover and forest cover loss using regression estimators.
IMPACT software was used to differentiate forest loss from deforestation and forest degradation; also to estimate carbon stocks.
Results
The use of the 2-stage approach is expected to provide robust statistics of forest cover loss with more precision as to the uncertainties. Expected results with Sentinel data using regression estimator should provide uncertainty, also confirm whether the two-stage sampling approach was justified for collecting a reliable forest cover reference dataset for estimating proportion forest cover area and net proportion forest loss. Furthermore, it brings out the proportion of intact forests between 2000-2015, deforested and degraded areas and their degree of confidence.
Interpretation: The precise uncertainties will give more credibility to the forest cover loss statistics and hence to the established reference levels.
Conclusion
The work is current and uses new data and will provide new information regarding forest cover loss and uncertainty.

Area A - Board 4

One of the main challenges in monitoring humid tropical forests so far was the frequent cloud cover limiting the use of optical data. To overcome this problem, the application of SAR data has received a lot of attention lately. While using SAR data for large and medium-sized deforestation mapping has been successfully done during the last years (e.g. Bouvet et al., 2018), its application for monitoring very small forest disturbances such as those caused by selective logging is so far missing. Limitations of applicability due to the geometric properties of sidelooking SAR systems in mountainous areas further limits SAR data use, because persistent cloud cover is often linked to mountainous terrain.
During the H2020 project EOMonDis, we evaluated the combined use of optical and SAR time series data in four different areas and found very diverse results. For areas with frequent cloud cover and for large deforestation plots, SAR data gives a significant added value in terms of detection timeliness and classification accuracy (e.g. in Peru). However, the added value of integrating SAR data in the processing chain for small structured changes (forest degradation) is only marginal; SAR data integration can even cause additional errors, as we could show in Gabon.
For Gabon, we generated a forest/nonforest mask for the year 2013, followed by a time-series based disturbance mapping based on Sentinel-2 and Landsat8 until 2017. The main difficulty in Gabon is its near-constant cloud cover, where misclassifications due to clouds are a frequent and serious problem, as also advanced cloud detection algorithms fail. The application of the EOMonDIS confirmation/rejection approach (Hirschmugl et al., 2017) led to a very good result covering all forest disturbance types (both deforestation and forest degradation, which can be separated later on by applying a MMU). In this case, all disturbances have been joined into one layer. The individual detection date was stored in a separate attribute.
The accuracy assessment for Gabon was done based on several VHR images and manual-visual delineation of degradation patterns visible in existing VHR data. In total, 855 distinct forest disturbance plots were mapped visually, of which six are large deforestation patches. All disturbance plots have a total area of 8400 ha, when excluding the large deforestation, the area is 181.2 ha. Only 2 % of the plots are larger than 0.5 ha. 87 % of the plots are smaller than 0.25 ha. The evaluation was done both plot-based and area-based. The plot based approach calculates the detected area percentage for each reference plot. If more than 25 % of the plot area is detected by the disturbance mask, the plot is considered as “detected”. One reason for the threshold of 25 % is the geometric inaccuracy between the S-2/L8 and VHR data set in combination with the high amount of small and very small reference disturbance plots. Plots larger than 0.5 ha are detected very reliably (95%), while the detection rate decreases with plot size. However, for plot between 0.05 and 0.1 ha, still more than half of the plots were correctly detected. The area based approach compares the area of the disturbance maps with the forest disturbance area from the reference data and led to an accuracy of 91.41%.
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 685761 (EOMonDIS).

Area A - Board 5

Urban population is growing at a tremendous speed especially in Africa and Asia. In combination with climate change and related natural disaster and –risks, the need for cities to become safer and more resilient is larger than ever. This is reflected in several international initiatives and on a high level agenda such as in the United Nations Sustainable Development Goal No. 11 or in the Sendai Framework for Disaster Risk Reduction.
Earth Observation can play a significant role in monitoring the progress towards these goals and also in achieving them. In order to do so, specific attention should be given to the wide variety of different sensors and methods, which need to be integrated to produce the required type of information. We demonstrate this integration by showing an example for the Indonesian city of Semarang, which is affected by land subsidence and flooding. The SDG Target 11.5 “By 2030, significantly reduce the number of deaths and the number of people affected and substantially decrease the direct economic losses relative to global gross domestic product caused by disasters, including water-related disasters, with a focus on protecting the poor and people in vulnerable situations” is addressed. The problem in Semarang has grown over the past decade, as the areas at the coastline, once inhabited by the wealthy, have now turned into quarters for the poor people due to increase flooding. These people are even more prone to suffering from floods due to inadequate housing, limited transport facilities and an overall higher vulnerability due to limited economic resources. This specific case was done for the World Bank City Planning Lab programme as part of the EO4SD Urban project.
In terms of EO data, we used (i) SAR and optical high resolution data as well as crowd sourcing information (urban voice) for the flood history (ii) VHR data and open street map for the up-to-date land use map (iii) interferometric SAR data for assessment of the subsidence rates and (iv) existing SRTM data as the only available elevation data and finally (v) World Population data for the estimation of the amount of people affected. Although out-of-fashion, we used (vi) human intelligence as opposed to artificial intelligence to combine the individual products. Current example results are e.g. spatial statistics, such as for the LU class “Residential and Public Urban Fabric Areas”, where 83 % are not affected by flood risk; about 8 % of the areas are in medium risk areas, 7 % in high and almost 2 % in very high flood risk areas.
A comparison of these results with results from applying artificial intelligence algorithms to generate a similar outcome is foreseen in the next months.
The preparation and work done for this article is part of the European Space Agency supported Programme “Earth Observation for Sustainable Development (EO4SD) - Urban”. Joanneum Research would like to thank especially the World Bank City Planning Labs programme and Team for their support; this includes Gayatri Singh the Task Team Leader, Champaka Rajagopal and Aurora Dias Lokita who facilitated the user interactions and data/information collection from the City.

Area A - Board 6

1.0 Introduction
With the launch of Sentinel-1A/B and Sentinel-2A/B by the European Space Agency (ESA), a new era of frequent coverage of the earth surface by high resolution (HR) satellite imagery was initiated. Together with the Landsat and other satellite missions, it is now possible to build up dense time series with sufficient spectral and geometrical resolution which allow new analysis methods for improved forest and land cover mapping. The application of dense time series of Sentinel and the HR data provides the possibility to overcome mapping inaccuracies in tropical dry forest caused by seasonal changes of forest cover, to compensate data gaps caused by cloud coverage, to improve the analysis of human induced changes and to make an early detection of deforestation and forest degradation events possible. However, the data volumes of dense time series data stacks from Sentinel and other satellite systems are, compared with traditional processing methods (mono- and bi-temporal analysis), tremendously increasing and therefore require a sophisticated IT infrastructure to compute wall-to-wall land cover maps. It has been proven more efficient for the European Service Providers to make use of cloud processing options instead of purchasing, maintaining and constantly upgrading existing IT infrastructure. On the other side, the handling of huge data volumes and the application of complex processing algorithms pose an enormous infrastructure and capacity challenge for developing countries. Thus, technology transfer and capacity building are major pillars of development cooperation programmes but however, the status of having up-to-date hard- and software is almost always lacking behind the requirements of a fast evolving technology. Therefore, working on cloud-based processing chains will be an opportunity for improved technology transfer and fostering REDD+ implementation in developing countries. In order to achieve a continuity and sustainability of all the previous developments for REDD+ Users, it was proposed to prototype, demonstrate and validate cloud-based processing of dense time series of Sentinel-2 data for tropical dry forest monitoring in a realistic IT environment.
The presented activities are supported by the ESA financed project “Innovator Supporting Developing Countries in Cloud-Based Forest Monitoring for REDD+” and are implemented in close cooperation with the Malawian Forestry Department. The applied methods and algorithms built on the achievements of precursor Projects like GSE FM REDD, REDDAF and EOMonDis.

2.0 Implementation
The European Commission (EC) has launched the Copernicus Data and Information Access Services (C-DIAS) initiative. DIAS facilitates access to open Copernicus data by providing data and information access alongside cloud-processing resources, tools and ancillary data. Mundi Web Services (https://mundiwebservices.com) is one out of several existing DIAS and used for hosting and provisioning of the developed services to the Users. It is important to be mentioned that the Copernicus program of the EC with its Sentinel satellites produces huge data volumes of approximately 10 TB of Earth Observation (EO) data per day (Kempeneers & Soille, 2017). Thus, the described developments are addressing the paradigm of bringing the 'User to the data' at a similar pace as the data becomes available. This in turn will facilitate access for scientific, commercial or institutional Users, and will increase the User uptake. The societal impact from the activity is further highlighted by addressing in particular the reporting requirements within the REDD+ policy, which is a key for developing countries to move towards the result based payments phase of the policy. A strong emphasis is put on the User Engagement component by following a bottom-up approach, where User’s needs and capacity were first defined as a major driver for the technical development and implementation of a web-based front office. Malawi was particularly selected as a pilot country for the development, implementation and provision of the cloud-based forest monitoring tools with a special focus on the monitoring of tropical dry forest areas.

3.0 Functionalities
The current project is designing and implementing a web based front office, which will enable Users from developing countries to perform autonomously cloud-based EO data analysis with the scope of tropical dry forest monitoring. The technical implementation of the platform use consolidated, verified methods and algorithms for the pre-processing and for the mapping of land cover changes, while taking advantage of the Mundi functionalities and its back storage.
A main technical advantage of the provided services is the provision of a time features calculation for the envisaged land cover classification. The time features selection and calculation was adapted from Valero et al. (2016) and a set of suitable time features is available on the front office. Time features can reveal the intensity of change information and represent the statistical properties of a dense time series EO data stack. Therefore, they are powerful input features for various thematic classification or regression tasks in the field of land cover monitoring. Time features do not require manual scene selection or prior knowledge of the change event dates and can be flexibly computed from reflectance or index data. Due to the iterative calculation over a user given time span, they are as well suitable features for near real-time change detection systems, which can be dynamically updated with each new available satellite scene.

4. Conclusions
The developed platform application will close the gap between the Copernicus Back office infrastructure and the accessibility of data/products by integrating the C-DIAS Interface services into a Third Party Front Office for Forest Monitoring. The Project will enable Stakeholders and Users from developing countries to create sophisticated applications for forest monitoring within an innovative cloud-based Front Office which unifies the Big Data functionalities of a C-DIAS back storage with verified processing algorithms for tropical dry forest mapping.

References:
Kempeneers, P, Soille, P. (2017) Optimizing Sentinel-2 image selection in a Big Data context, Big Earth Data, 1:1-2, 145-158, DOI: 10.1080/20964471.2017.1407489
Valero, S.; Morin, D.; Inglada, J.; Sepulcre, G.; Arias, M.; Hagolle, O.; Dedieu, G.; Bontemps, S.; Defourny, P.; Koetz, B. Production of a Dynamic Cropland Mask by Processing Remote Sensing Image Series at High Temporal and Spatial Resolutions. Remote Sens. 2016, 8, 55.

Area A - Board 7

We want to present results of the Sentinel4REDD project. The overall aim of the Sentinel4REDD-Project is the development of new remote sensing based methods using Sentinel-1 and Sentinel-2 data to support UNFCC (United Nations Framework Convention on Climate Change) REDD+ MRV (Measurement, Reporting and Verification) Systems.
We show results on the example of two testsites located in Mexico, one is over decidious and evergreen forests in Hidalgo and the other is over tropical dry forests in Yucatan.
For a signature analysis, we first selected forested, deforested and degraded areas based on visual interpretation of multi-temporal very high resolution (1 m) optical imagery. We developed a new speckle filter which retains the spatial resolution by using the temporal domain. We plotted filtered Sentinel-1 time series for the three reference classes and were able to determine the time frame of deforestation and forest degradation. The initial analyses showed promising results regarding the separation of forest and forest-change classes with filtered Sentinel-1 data in contrast to original SAR backscatter images.
To generate REDD+ products (forest/non-forest, deforestation, forest degradation, reforestation maps), we apply two approaches using dense time series from Synthetic Aperture Radar (SAR) and optical sensors. The first method is based on multi-temporal metrics, e.g., mean, standard deviation and different percentiles of the SAR backscatter and surface reflectances for different time periods. The second is based on different seasonalities of land cover classes. For this, the time series of every pixel is decomposed into subsignals with differing temporal frequencies. From these subsignals different statistics are calculated. These statistics can then be used to derive forest/non-forest maps for different time periods to then obtain resulting deforestation and reforestation maps or to directly gain deforestation and degradation maps via the application of machine learning techniques.
Furthermore, we present preliminary deforestation maps for study sites in Mexico and South Africa based on Bayesian probability approach and filtered Sentinel-1 backscatter time series.

Area A - Board 8

With the launch of the two Sentinel-1 satellites, very dense synthetic aperture radar (SAR) time series have become available.

These new time series allow to develop novel speckle supression algorithms.
We propose a new speckle filter for hypertemporal SAR image stacks which only operates in the temporal domain. This is espacially relevant in heterogeneous savannah biomes, because established speckle filters assume spatially homogeneous areas.
The filtering in the temporal domain fully preserves the spatial resolution.
The backscatter time series of a resolution cell is the superposition of subsignals with differing temporal frequency.
Under the assumption, that the speckle is changing on every acquisition,
the high frequency parts of the signal correspond to speckle.
We are using the empirical mode decomposition (EMD), which is a non-parametric alternative to Fourier analysis, to decompose the signal into these subsignals with differing frequencies.
In contrast to the Fourier analysis, the decomposition does not depend on a priori set base functions.
Therefore, the hypertemporal SAR time series is decomposed into components of different frequency which are derived in a fully data-adaptive manner.
The high frequency part of the signal is removed to obtain a smoothed SAR time series.
The lower frequency sub signals are related to actual biophysical processes and are similar on neighbouring pixels of equal land cover.
Therefore, the smoothing over time results also in a spatial smoothing over homogeneous areas.

We compare the results of our new approach with the results by the Quegan filter with 3 x 3 and 5 x 5 spatial windows [1].
For the comparison of the spatial smoothing we compute the equivalent number of looks (ENL) over homogeneous forest areas.
The ENL is the number of different images which would be needed to obtain the same smoothing by multilooking.
With the EMD filter, we obtain a similar ENL compared to the 5x5 Quegan filter and an improved ENL compared to Quegan 3x3.
To compare the edge preservation we chose two homogeneous regions with different average backscatter and computed the width of the edge between those regions after applying different filtering approaches.
We showed that the EMD filter improves edges in original Sentinel-1 and PALSAR-2 data, while the Quegan filter introduces a certain amount of blurring.

Area A - Board 9

Selective logging is an important human activity in tropical forests. In the Brazilian Amazon it can cover an area bigger as that reported as deforested. Imagery from medium resolution satellites, such as Landsat, have been used to map selective logging in tropical forests. However, the Landsat spatial resolution (30 m) has often been considered as too coarse for reliably mapping small-scale selective logging. Imagery from the recently launched Sentinel-2 satellites, with a 10 m spatial resolution, may improve the detection of this type of forest disturbance.

This study had the objective of comparing the performance of Landsat 8 (L8) and Sentinel-2 (S2) data for the detection of selective logging in an area located in the Brazilian Amazon.

The study area is located in a new logging frontier, in the Brazilian State of Amazonas, near the Village of Antonio do Matupi (-7.927069, -61.567564). For better understanding the performance of L8 and S2 imagery for mapping selective logging, the study was carried out in seven areas under sustainable forest management (SFM), with areas ranging from 100 to 1,000 hectares with logging activities carried out in 2017. This allowed us to ensure that the detected forest disturbances were caused by selective logging activities. In order to limit our analysis in the forest areas and to avoid confusion with areas of deforestation, we built a forest mask through an object-based image analysis. Forest disturbances caused by logging activities were mapped by calculating the difference of a self-referenced normalized burn ratio (ΔrNBR) index for the time period of 2016-2017 for both sensors. A bootstrapping procedure was applied to define the optimum ΔrNBR thresholds for the forest disturbance maps. Using the optimum thresholds values we derived binary maps of disturbed/undisturbed forest. These binary maps were validated using a reference data set (300 grid cells of 150 x 150 m size), built through visual interpretation of the corresponding satellite imagery. These reference data sets were used to calculate the accuracy of the forest disturbance maps, which were later corrected to account for the area proportion. Additionally, field data of different logging infrastructure (log landings, felling gaps, logging roads and skid trails) were collected in the seven study areas. These field data were used to evaluate the “detectability” for each type of logging infrastructure for the two sensors.

The imagery of Sentinel-2 performed slightly better than L8, with area-adjusted overall accuracies of 86.5% and 83% respectively. The percentages of logged area, of the seven SFM areas vary from 2.4% to 13.9% for L8 and from 1.4% to 10% for S2, leading to a percentage of disturbance of 5.6% and 4.0% for Landsat 8 and Sentinel-2, respectively, if we consider all study areas. In consequence, Landsat 8 imagery produces an estimated area affected by selective logging that is 38.5% higher than the estimate based on Sentinel-2 data. Logging infrastructure was better detected from S2 data, especially for log landings, confirming the sensor’s potential for mapping small-scale selective logging.

Area A - Board 10

Emissions from deforestation and degradation from tropical forests are important contributor do global climate change due to the release of carbon into the atmosphere when the forest is lost. An initiative to prevent and decrease the carbon emissions from forests is Reducing Emissions from Deforestation and Forest Degradation (REDD+), a global financial mechanism that provides economic incentives to developing countries for carbon stock improvement, enhancement of sustainable forest conservation, and undertaking voluntary actions to reduce their emissions. Since 2004, states in Brazilian Amazon have made efforts to decrease deforested areas, and to promote REDD+ projects into public policies in order to help the forest management and the deforestation rates reduction. Therefore, to achieve significant impacts and greater acceptance from policy makers in Brazil, better approaches to assess and monitor the maintenance of forest areas must be developed, since it has been often a challenging task. In the state of Pará, there are at least seven REDD+ projects registered as voluntary carbon offset projects, but very little information is available for practitioners and policy makers interested in improving the effectiveness of REDD+ Data regarding the current greenhouse gas (GHG) emissions from forests derived from free and easily accessible remotely sensed data products could be very attractive and important to estimate the reference levels and drivers of forest deforestation and degradation in and out REDD+ projects. In this sense, this work presents an application of random forest (RF) algorithm based on Sentinel 2 (S-2) images with Google Earth Engine (GEE) to obtain a set of vegetation index (NDVI, EVI, ARVI, SAVI) and determine the classes and distribution of land cover of REDD+ registered projects in the state of Pará, Eastern Brazilian Amazon. The land cover types were classified using S-2 L1C images for the years 2016 and 2017, and further overlaid to identify the spatial distribution and land cover dynamics to produce land cover maps. For validation purposes, it was used data available from the maps produced by the TerraClass Project (official data for the monitoring of LUCC in Brazilian Amazon). As an additional objective, this work estimates the potential biomass carbon stocks in such REDD+ projects from the land cover maps and ranked, according to this accuracy, the vegetation index most appropriate to point change detection. Finally, the presented framework focused on offering low cost-intensive approaches to track and monitor land cover change over time by using GGE and medium resolution (10m) Sentinel 2 images to promote policy interventions to avoid carbon forest loss.

Area A - Board 11