Presence, abundance and health conditions of aquatic plants assume crucial importance within freshwater and wetland ecosystem, because of their role in biogeochemical processes in water column and sediments, their interactions with other autotrophic and heterotrophic organisms, and the provision of food and shelter to animal communities, such as fish and birds.
The objective of this work is to assess the impacts on the aquatic vegetation, or macrophytes, of Lake Mezzola (North Italy: 46°12’01’’N; 9°26’17’’E) of a rock avalanche event which took place upstream of the water basin in August 2017. On 23 and 25 August 2017 two major rock landslides detached from the Swiss side of Piz Cengalo, in Val Bondasca (CH) with a volume estimated between 2 and 4 million m3. In the following days, the detrital material has firstly reached a man-made reservoir that was already emptied in July as prevention measure. The water, flowing in the downstream catchment of Mera River, then reached Lake Mezzola, causing an increase in suspended solids and water turbidity and that lasted for some months.
Following the events, the Environmental Protection Agency of Lombardy region (ARPA) has put in place a monitoring program of the area with the contribution of EOMORES project (EU H2020: grant 730066), in order to assess the evolution of water quality in Lake Mezzola and its impacts on the ecosystem, focusing on valuable macrophyte communities of Pian di Spagna (a Natura 2000 site). Submerged vegetation here is mainly constituted by Potamogeton spp. and Elodea canadendis Michx., while the surrounding wetland (Pian di Spagna natural reserve) is dominated by common reed beds (Phragmites australis (Cav.) Trin. Ex Steud), with some sparse presence of Schoenoplectus spp., Littorella uniflora (L.) Asch, and Ranunculus spp.
A time series of multi-spectral optical satellite data covering three years (2016-2018), acquired from the satellites of the Sentinel-2 constellation managed by the European Space Agency (ESA) and the EU Copernicus program, with cloud cover above 50% on Lake Mezzola were discarded has been gathered and processed. In addition, hydrological and meteorological parameters (temperature, incoming radiation, precipitation and water level) were retrieved from sampling stations located nearby the study area, and an in situ survey has been carried out (26 September 2018) for documenting the macrophyte conditions in the southern portion of the lake in terms of species presence, density, phenological stage, and traces of impact.
Sentinel-2 dataset was converted to surface reflectance after atmospheric effect correction with SEN2COR (Louis et al., 2016). Starting from Sentinel-2 surface reflectance dataset, time series of water turbidity and two spectral indices sensitive to vegetation features have been derived. Water turbidity has been mapped using ACOLITE code, run with the Dogliotti et al. (2015) algorithm. The two spectral indices are the Water Adjusted Vegetation Index (WAVI), developed specifically to maximize the sensitivity to the density and biomass of aquatic vegetation (Villa et al., 2014), and the Chlorophyll Index green (CIgreen), a proxy of chlorophyll content of the canopy (Hunt et al., 2011). In addition, benthic substrate type maps were derived from peak of season Sentinel-2 scenes for each of the three years (2016, 2017, 2018), using the bio-optical model inversion implemented in the BOMBER code.
The analysis of time series of Sentinel-2 derived maps have shown that both submerged and riparian macrophyte communities of Lake Mezzola and Pian di Spagna wetland were impacted by increased turbidity and sediment transport due to the August 2017 landslide events. Submerged vegetation of the southern part of the lake has exhibited a decrement in 2018, compared to previous years, in terms of total area covered and plant density, together with some signs of shifts in community composition in the littoral zone. Riparian helophytes instead have shown evidence of stress during the growing season following the landslide, translated into a delay in green-up for 2018 season, i.e. reaching WAVI = 0.3, varying between 10 and 40 days compared to previous years. Although last year has shown and unusually cold early Spring in North of Italy, the late development of riparian reed beds in 2018 is not attributable only to this weather anomaly, since terrestrial reeds of Pian di Spagna wetland did not show the same growth delay from time series of WAVI scores.
The impacts of landslide aftermath on riparian reed beds of Lake Mezzola is most visible in areas of reeds completely surrounded by water, thus more sensitive to the effects of turbidity and sediment deposition, a decoupling between density-sensitive WAVI and greenness-sensitive CIgreen was observed in July-August 2018. This was probably caused by the growth of a turf populated by Ranunculus spp. below the reeds, a side effect of the weakening and the loss of biomass of this reed bed, which allowed this understory to develop and expand.
In addition to the results highlighted here above, the Sentinel-2 based approach we implemented will enable the continuous monitoring of ecosystem of Lake Mezzola and surrounding areas during 2019 growing season, and thus to verify the evolution of aquatic vegetation conditions and ecosystem equilibrium towards recovery or further degradation.

Area B - Board 167

Introduction
Wetlands ecosystems provide valuable services and also host high biodiversity, but are nevertheless threatened by hydrological and land cover change. Meanwhile, mapping their exact extent, vegetation types and structure is highly challenging since they are heterogeneous and often form transitions with grasslands, forests and open water. Biodiversity assessments often focus on only species diversity, while beta diversity and especially structural diversity are not quantified. Ecosystem service mapping is typically limited to modelling of services based on land cover, or rarely a detailed multiparameter assessment of a single service depending on stakeholder interest. Decision support would need spatially explicit information on multiple ecosystem services and biodiversity, with high accuracy. Quantification of accuracy in earth observation studies is only possible against in situ field reference data. However, especially in wetlands the collection of field samples and measurements is highly labour intensive and therefore rarely practiced. Additionally, nearly all earth observation evaluation attempts suffer from the scale mismatch between field samples and data pixels: field plots rarely exceed the size of a few meters (especially if they are considered to be homogeneous), while satellite pixels are usually in the order of 20-30 meters. This introduces an uncertainty in the very figures we use to describe the accuracy of a dataset.

Methods
All these difficulties can be overcome in a purposefully designed multi-level field and remote sensing setup. Upscaling is possible from direct quantification of parameters of interest in the field via quick field indicators through LIDAR-derived proxies to Sentinel-2 data products. Error propagation can be quantified through each of these processing levels, and sufficient amounts of data can be produced for calibrating and evaluating the next level. First of all, biodiversity and ecosystem functions of interest have to be represented by well defined physical quantities. In case of ecosystem services, this is possible at the level of ecosystem service potential, without evaluating societal demand or market value. We defined dominant vegetation species diversity, vegetation class diversity and the vertical distribution of biomass as metrics of biodiversity, and measured all of these by fully harvesting 0.5×0.5m plots in the field and weighing the biomass belonging to individual species within each sample. Additionally, biomass and leaf weight were measured individually in 0.5 height layers within these samples. Ecosystem service proxies quantified were biomass, underwater surface and leaf weight from the same harvested samples. Decomposition was measured by placing standard biomass samples in the sediment that were weighed again after a year. Microclimate related services were measured by placing climate loggers along transects across the vegetation and evaluating the difference in temperature and humidity within the wetland compared to synchronous external controls. Approximately 50 such full samples were collected across a range of Hungarian lake wetlands. At the next level, quick indicators were designed for the same parameters. Biodiversity metrics were estimated in point samples, while biomass and vertical structure were quantified by counting vegetation contacts to a pole in height intervals. Canopy coverage and density was quickly measured from zenith-facing photographs that were also correlated with microclimate metrics and decomposition. Several hundred of these quick indicators were collected and calibrated based on the direct measurements.
From the remote sensing side, airborne LIDAR was used as the main data source. First of all, random forest point cloud classification in OPALS software was applied to outline main wetland vegetation types within the study areas. LIDAR-based 3-dimensional vegetation structure and heterogeneity metrics were combined in GLM-s to estimate scalar quantities of ecosystem service and biodiversity indicators, with separate functions for different vegetation types wherever possible. The collected field reference datasets are sufficient for calibrating LIDAR-based proxies for most of the indicators, and thus maps can be produced for the full study regions (>500 km2) with sub-meter to meter-scale pixel sizes. Accuracy of these remote sensing-based datasets is verified against the field indicator datasets where scale mismatch is not a problem since the pixel size is close or equal to the field plot size.
Finally, the LIDAR-based maps can be used to design and calibrate satellite-based indicators for the same set of ecological parameters. For this purpose, the LIDAR map is an ideal ground truth as homogeneous areas can be selected that match the satellite pixel size. Using standard vegetation metrics and heterogeneity indicators or potentially even seasonality information, satellite-based regressions can be trained that allow estimating these variables and thus extrapolation beyond the spatial and temporal limits of the LIDAR dataset.

Results
A standard-ready methodology was developed using a self-contained and cost-efficient set of tools and measurements, that can be carried out by a single person in a wetland field setting. Reference datasets were collected in the shore wetlands of Lake Balaton and Lake Tisza reservoir (Hungary) and Lake Fertő/Neusiedler (Hungary and Austria). Cross-correlation between direct and quick indicators show that estimated mean biomass from pole contacts correlates relatively well to harvesting-based data, including the height distribution. However, zenith-facing photos provide a less quantitative estimate of biomass as they saturate at vegetation density levels that occur rather frequently. However, they are accurate representations of the ecosystem service "regulation of light pollution". Microclimate measurements pointed out remarkable differences in water temperature within wetland vegetation that depend mainly on distance to the open water. Decomposition depends mainly on local water depth, which is difficult to quantify from LIDAR data. LIDAR allows accurate delineation of old reed stands with a lot of dry biomass at ground level and young vigorously growing reed with most of the leaves in the top of the canopy. Vegetation classification successfully separates Carex, Typha, wet meadows and various growth forms of reed. Radiometric calibration and full-waveform information provide significant contributions to classification accuracy. Satellite upscaling was only attempted for wetland-nonwetland classification so far but shows potential for recognizing subtle changes in wetland extend.

Discussion and outlook
These datasets show that it is possible to quantify biodiversity and ecosytem service indicators directly in the field even under adverse conditions. The multi-level approach we proposed allows accurate upscaling to remote sensing. LIDAR proved crucial for wide-area vegetation calculation due to its independence from external illumination conditions and its ability to quantify both radiometric and structural properties and the scale of interest. The data resolutions used here include cases similar to the nationwide point clouds of several European countries and thus point toward the nationwide application of LIDAR for wetland monitoring and assessment. This would allow satellites to be used only for temporal extrapolation and short-term monitoring. The application of multi-parameter ground truthing and LIDAR mapping together provides a completely new view on wetland biodiversity and ecosystem services, allowing trade-offs and synergies to be investigated on very large sample sizes. For decision support, the regionwide availability of high-resolution information on wetland types, biodiversity and services means that the impact of landscape planning can be accurately quantified and scenarios compared. Therefore, this method is expected to be transformative both for ecosystem science and conservation practice.

Area B - Board 168

While large extensive wetlands, such as those on lake shores are in focus of monitoring and conservation, small patches of wetland vegetation in the mosaic of the landscape receive far less attention: in many cases their extent and location are not known precisely as they are smaller than the typical scale of regional vegetation mapping. Meanwhile, these small backcountry wetlands are especially important as they are often the last undisturbed habitats in landscapes extensively modified by agriculture and urbanization, and as such provide sanctuaries and “stepping stones” for wildlife sensitive to human presence. They also play an important role in regulating water quality and quantity through pollution absorption, groundwater recharge and flood retention. Therefore, backcountry wetlands are essential elements of green infrastructures.
Moreover, Hungary in the Carpathian basin has a special river and wetland structure, where we have remaining wetland habitat patches reduced from large continuous areas of the last centuries. The remaining great habitats function as natural reserve areas (Natura 2000 habitats), but the tracing and tracking of the small habitats should be the next important step in the protection of biodiversity in Hungary. Small wetland patches are prone to rapid change depending on local water and management conditions, therefore up to date information is necessary for their assessment and protection.
During our ESA OWETIS (Observation of local Wetland areas from Satellite Imaging) project, we have built a Hungarian wetland database, including altogether 333 test sites, out of which 210 areas are categorized as wetlands. Beside wetlands, grasslands and croplands are also added to the database, which categories are sometimes confused with wetlands during the segmentation process. The database is categorized on 4 different levels; therefore, subcategories inside wetlands are also included, creating the opportunity to test the developed segmentation and classification methods. Along with the wetland database, satellite imagery data – including Sentinel 2 and SPOT 6/7 multispectral images from different years – were collected for the test sites to make time series analysis and change detection also feasible.
Basically, two approaches were introduced for wetland segmentation. First, we developed a Fusion Markov Random Field (FMRF) model, which has the ability to handle multiple layers (even with different acquisition time), from which the properties of each pixel is determined and segmentation is performed based on the properties of the neighbouring pixels around the given pixel. Second, the collected wetland and imagery database created the opportunity to extract training data for deep neural networks. Thus, different variants of U-Net were trained for wetland prediction. Finally, prediction maps of the most promising trained networks were fused with the FMRF model.

Results were evaluated quantitatively and qualitatively by ecological expert project partners, giving also a biological interpretation of the ongoing processes in the observed vegetation areas.

Figure 1: Wetland mapping on Lake Balaton, Hungary - The test patches of the wetland database are shown in yellow, followed by the MRF-based multitemporal segmentation and the test site in Google Maps.

Area B - Board 169

Wetlands provide many essential ecosystem services, but at the same time they are threatened by processes like e.g. urbanization, expansion of farmland and extraction and pollution of freshwater. Due to their heterogeneous characteristics, spatiotemporal dynamics and topographic features, wetlands are often hard to identify, map and monitor, which hinders their efficient conservation and recognition in policy frameworks and management practices. To improve this situation, it is necessary to make efficient use of remote sensing capabilities and the available imagery especially of freely available satellite programs like the USGS Landsat imagery and the Copernicus program.

GEO-Wetlands is a collaborative framework formed as an initiative within the Group on Earth Observations (GEO). One of its goals is to ease the effort of finding, accessing, and using these datasets. Further, it provides improved possibilities for analysis and implementation of the data in support of decision making and reporting processes, tailored to the needs of the wetland community.

On the global level, GEO-Wetlands mainly strives to support the Ramsar Convention on Wetlands as well as UN Environment, who are both custodian agencies for SDG 6.6 “By 2020, protect and restore water-related ecosystems, including mountains, forests, wetlands, rivers, aquifers and lakes” and responsible for defining the methodology for indicator 6.6.1 “Change in the extent of water-related ecosystems over time”. As the reporting processes of both, the Ramsar Convention and the SDG framework, require countries to report regularly on the status and trend of wetlands (or water related ecosystems) it is very important to improve the capabilities of national agencies and even local wetland managers with regards to the use of EO based methods, tools, information products and datasets. Capacity building and community engagement are key for achieving this.

GEO-Wetlands is based upon the activities and results of several contributing projects over the last years. Examples from these projects highlight the value and importance of Earth Observation for improving the understanding and sustainable management of wetland ecosystems. The collaborative framework GEO-Wetlands provides is meant to allow continuation of these efforts. An important objective for the initiative is to ensure that achievements, results and products of these projects are maintained and stay available for future use and further development.

Therefore, GEO-Wetlands provides an online platform functioning as knowledge-hub for EO of Wetlands and for community interaction and collaboration. Further, the GEO-Wetlands Community Portal, a geoportal focused on providing easy access to wetland maps and statistics as well as case studies and application examples, provides an entry point to relevant spatial datasets and satellite imagery. These platforms fully support capacity building and usability by providing guidelines, training materials and use cases.

The future development of GEO-Wetlands depends strongly on engagement and contributions from the broad community of EO practitioners as well as wetland experts and its open collaborative character is meant to increase the global capabilities for wetland mapping, inventory, monitoring and assessment through co-creation and innovation.

Area B - Board 170

Wetlands are considered as biodiversity hotspots and they are commonly valued for water purification, flood retention and carbon sequestration. The need for their protection is undisputed. The urgency of the topic has become more tangible with recent studies demonstrating the rapid decline of global wetland area, where inland wetlands are being most affected. Agricultural wetland use is often described as the main driver of wetland degradation and loss through altering hydrologic characteristics, pollution by chemicals and natural habitat destruction. Particularly in food insecure sub-Saharan African countries, local populations often depend directly and the whole countries’ populations indirectly on agricultural production in wetlands.
A lack of spatially explicit information on the very complex and variable wetland landscapes is seen as particularly problematic on the African continent. Knowledge about location, hydrological and land use and land cover characteristics as well as their changes is needed to inform sustainable wetland management. Earth Observation has proven to be an invaluable technology to close this gap and provide quantified and up-to-date spatial information to conventions like the Ramsar convention, the Convention on Biological Diversity or the Sustainable Development Goals, and national government bodies and non-governmental organizations.
Therefore, this study investigated how Earth Observation data can contribute to sustainable wetland management in the context of both wetland conservation and food security. We developed a comprehensive, flexible and applicable characterization and monitoring framework by matching end users’ information needs with remote sensing- based products. Capabilities of such a framework are showcased at the example of Rwanda, drawing from data products generated within the German BMWi-funded research project ‘Copernicus-based Detection and Monitoring of tropical Wetlands’ (DeMo-Wetlands). DeMo-Wetlands aimed at developing methods for the delineation, assessment and monitoring of wetlands in Rwanda in support of international conventions and in close consultation with national stakeholders by making use of the freely available Copernicus missions data. The present study broadens the intended application for conservation purposes by integrating the pressing issue of food security. The combination of several Sentinel-derived data products allowed for the contribution of information on a national scale in terms of extent, land use and its intensity as well as hydrology and demonstrates applications at the interface of wetland conservation and food production in Rwanda. Among others, patterns of high and low intensity agricultural activity could be identified as well as ecologically important phenomena like floating vegetation. Compared to the identified information needs grouped according to the Driver-Pressure-State-Impact-Response (DPSIR) scheme, applications at the interface of wetland conservation and food production in Rwanda could be demonstrated.

Area B - Board 171

In the context of the Collaborative Data Innovations for Sustainable Development Pilot Funding provided via the Global Partnership for Sustainable Development Data (GPSDD), a project was carried out in Uganda to showcase the full process of implementing the use of Earth Observation (EO) for monitoring and reporting of the Sustainable Development Goals (SDGs). SDG sub-indicator 6.6.1a, on changes in wetland extents over time, was selected for pilot implementation at the national scale.

The objective of project was to explore the potential of EO satellite data for taking stock of and monitoring wetlands, a vital component of the global water resources ecosystem. The activity piloted design and development of a user-friendly digital system for use by the Ugandan Ministry of Water and Environment, to enable national authorities to generate spatial time series statistical data for taking inventory and monitoring national wetland resources, train the government on its use, and produce a roadmap for scaling to other countries in East Africa. This is a unique attempt to demonstrate the potential of satellite-derived EO data to provide a full national wetland inventory in Uganda, which has been a pilot country for monitoring of SDG Target 6.6.

The four main components of the project were:
• Toolbox for wetland inventory mapping, based on the GlobWetland Africa Toolbox, with extended functionality for reporting on SDG 6.6.1a.
• Baseline wetland inventory map for Uganda based on Sentinel-1 and Sentinel-2 data for 2016 and 2017. Validation has been performed incorporating in situ data collected by the Ministry of Water and Environment and RAMCEA.
• Online portal for SDG reporting on indicator 6.6.1a on Wetland extents.
• Training of national stakeholders in Uganda as well as from the other RAMCEA countries on the toolbox and online platform.

In order to make the relevant tools available for national agencies in charge of reporting, the GlobWetland Africa Toolbox was further developed to support national scale inventories of wetland extents. Furthermore, training on the use of Earth Observation in general and specifically on wetlands were carried out with a focus on stakeholders in the involved government agencies within Uganda and the other RAMCEA countries. A total of three training courses were held during the course of the project.

The wetland inventory was produced from time series of high-resolution imagery in the optical and radar spectrum provided by the modern Sentinel satellites (Sentinel 1a, 1b, 2a, and 2b). The images were processed through a highly specialized chain of algorithms to derive information on the probability and type of wetland. The pilot product for Uganda, covering a total area of 241.038 km², was produced from a total of 3500 optical scenes (multi-spectral images) and 1700 radar scenes from 2016 and 2017 with a total raw data volume of 5 TB. The inventory map will be used as baseline for SDG reporting for indictor 6.6.1 for wetland extent monitoring in Uganda.

The online portal has been developed to enable data producers and decision makers to explore the baseline wetland map and extract relevant information on trends. This serves the requirements in terms of reporting on SDG indicator 6.6.1a on monitoring wetland extents, but also allows more detailed analysis on a finer scale. Production of statistics for display in the portal can be performed within the Toolbox by using a custom workflow that will generate all relevant content (statistics and reports). This makes updating the portal with updated statistics, new reporting units or adding new wetland inventories a simple task that can be performed by, e.g., national statistical agencies or line ministries. The code for the portal is based on open source components and has been released on GitHub and can hence be reused or adapted by other countries.

The project demonstrated the full cycle of SDG monitoring and reporting on a SDG sub-indicator, from supplying the free access to the data (from the Copernicus Programme) and the tools (the GlobWetland Africa Toolbox), through capacity building and establishment of baseline conditions, to dissemination of the results in a compliant reporting format. This will enable better monitoring of wetland extents within Uganda in the future, and hopefully lead the path for uptake by other RAMCEA countries or further afield.

Area B - Board 172

Wetlands, which represent only 6% of the Earth's surface, nevertheless play a crucial role in the global water cycle and climate dynamics. Nowadays these sensitive parts of the world are threatened; both by climate change and by changes in the nature of the soil, including drainage and deforestation. Despite the need for precise information on these areas in order to be protected to the best, today there is still few detailed documentation concerning these regions, which are sometimes politically unstable and difficult to access. This is the case of the « Cuvette Centrale » of the Congo Basin, also called the « Flooded Cuvette », where peat bogs have recently been discovered. This discovery has made of this region a place that requires special attention. Indeed, peat bogs are very sensitive ecosystems that contain a very large amount of carbon and are mainly found in swamps. Two studies recently focused on the identification of the different areas within the Cuvette according to their flooding. These two studies, which combined Palsar-1 and 2 SAR images respectively with optical data, have demonstrated significant limitations in the use of the radar data. Our study is therefore interested in a detailed characterization of the different zones within the Cuvette according to their flood dynamics using radar data. Two sensors were used: the PALSAR-2 image sensor from the JAXA ALOS satellite, and the Jason-2 satellite Poseidon altimeter radar sensor jointly developed by NASA and CNES. The use of high-resolution time series (25 m) from the ALOS PALSAR-2 sensor allowed a first characterization of the different zones according to their flooding over time. A second characterization was performed by analyzing time-series of radar altimetry-based water levels automatically generated from the Ku-band backscattering from Jason-2. This second analysis allowed not only to support the characterization coming from the image SAR data but also to provide information concerning the shape of the flood wave. The results given in this study were then compared to the results from the earlier studies performed over the Congo « Cuvette Centrale » using PALSAR-1 and 2 images to discuss it.

Area B - Board 173

Information about wetland dynamics remains a major missing gap in understanding changes in atmospheric methane and terrestrial water storage, which inflicts large uncertainties in climate projections. Moreover, rapid urbanization has exacerbated wetland change and exert an urgent demand on wetland monitoring. Observations of wetland change requires improved remote sensing technologies not only to consistently and accurately provide global inventories but also to examine changes spanning multiple decades. Among the limitations of the current observing system, wetland monitoring is strongly affected by the opacity of dense vegetation, which hinders mapping water extent and measuring water level underneath these canopies.

The use of signals transmitted by the Global Navigation Satellite Systems (GNSS) reflected off inundated areas was proposed recently in the literature. Signals from spaceborne TDS-1 and CyGNSS GNSS-Reflectometry (GNSS-R) missions demonstrated that inundated wetlands can be identified under different vegetation conditions including dense rice canopy and thick forest with tall trees, where optical sensors and monostatic radars provide limited capabilities. In these studies the power levels of GNSS-R over water bodies were higher than in dry areas, because of the higher reflectivity of the water and its smoothed surface when hidden below vegetation.

A new study is here presented, to assess the skills of spaceborne GNSS-R to sense wetland water under one of the densest wetland canopies of the mangroves in the Cần Giờ Biosphere Reserve in south Vietnam (see Figure 1), which is recognized as a UNESCO biosphere reserve site. In contrast with observations over wetlands under less dense vegetation covers, GNSS signals reflected off these densely vegetated wetlands present low power levels; therefore, this observable may not be applicable to map inundated areas in this very dense vegetation case. New observables are analyzed in this study, taking advantage of the raw acquisition mode of the CyGNSS mission, which took dedicated and unprocessed data streams over the Cần Giờ Reserve following an in-situ experimental campaign across this wetland area in 2018. The coherence of the signals is then searched through the phase of the carrier signal to trach the phase signatures over calm waters in the dense mangrove wetland. Signs of coherence in the carrier phase have been obtained along the GNSS-R tracks along the thick mangroves of Cần Giờ, showing the potential of the technique to monitor extremely dense wetlands from spaceborne platforms. The study will present the technique, experimental campaign and preliminary results.

Figure 1: Dense mangrove conditions observed during an experimental campaign in the UNESCO Cần Giờ Biosphere Reserve, Vietnam.

Area B - Board 174

Background:
Deposits of peat underneath peat swamp forests are among the world's largest reservoirs of carbon. Although tropical peatlands occupy only about 0.3% of the global land surface, they could contain as much as 20% of the global soil carbon stock. It should be noted that tropical peat swamps are unevenly distributed across the globe. The tropical peat swamp forests of Southeast Asia account for approximately 26.5 Mha of the total tropical area of 38 Mha, with Indonesia alone contributing an estimated 17-27 Mha.

Peat swamp forests are among the world’s most threatened and least known ecosystems. In Southeast Asia large areas of peat swamp forest have been deforested (for timber), converted for agricultural projects (even though the soil is too acid), or are converted into plantations (such as oil palm, Borneo rubber), even though peat systems are fragile and sensitive to hydrological disturbance. Drainage through canalisation has frequently severely disrupted water table level dynamics, causing the peat layers to dry out and trees to collapse over large areas. Besides resulting in CO2 emissions due to oxidisation this process makes them particularly vulnerable to fire, especially during ‘El-Niño’ years. Emissions from the fires in Indonesia during 1997-1998 for example, have been estimated at 0.8-2.5 gigatonnes (Gt) of carbon.

Water management is essential in addressing these disturbances. However, the relationship between spatial and temporal dynamics of peat swamp forest hydrology, carbon content and forest health would need further study. Such understanding would not only support the conservation of peat swamp forest, but also the rehabilitation of degraded tropical peatlands, which may significantly reduce carbon emission and fire risk.

Need for near real-time monitoring:
The Indonesian government is committed to improve peatland management and has established the peat restoration agency (BRG) to coordinate its efforts. This commitment includes the rehabilitation of 3 million ha of degraded peatlands, blocking 10,000 km of canals and construction of 10,000 dams in the next few years. To implement its mandate, BRG is critically dependent upon up-to-date and accurate data on the location of existing and new canals, occurrence of flooding, drainage and occurrence of fires in Indonesian peat. Remote sensing data is needed to obtain such information in a timely manner over large areas.

Project objectives:
BRG requested the Indonesian Space Agency (LAPAN) to provide remote sensing data to meet this aim. Wageningen University and SarVision support LAPAN and BRG to build up technical capacity for near real-time monitoring through the ESA EO Science for Society project “Tropical Peat Watch” (ESA AO/1-9101/17/I-NB). This project will demonstrate to LAPAN, BRG and other potential users and stakeholders a peat monitoring system that provides information on drainage canals, hydrology and fires through innovative use and integration of multiple earth observation data sources from the European Space Agency (Sentinel-1, Sentinel-2, Sentinel-3) and other third party missions (PALSAR, Landsat, MODIS). The project focuses on demonstration sites in Central Kalimantan and Riau. After the project, it is expected that the demonstrated system will be scaled up and implemented by LAPAN, supporting peat restoration policy analysis and monitoring in BRG for all peatlands in Indonesia.

Solutions for near real-time information delivery (presentation of first results):
Hydrology – L-band radar is uniquely suitable to monitor flooding under vegetation canopies, such as peat swamp forests, and is sensitive to variations in peat soil moisture. It is suitable to detect excess drainage along canals. Relationships (mainly qualitatively) between soil moisture, canopy cover and L-band radar backscatter have been studied using historical data (JERS-1 and PALSAR-1). The currrent study will make efforts to better quantify this relationship using PALSAR-2 and a new network of ground water level gauges. Hydrological models for ombrogenous peat (domes) are utilised to describe the temporal dynamics of the hydrological patterns observed and to identify areas at risk.
Drainage canals – Sentinel-1 time series are used to detect changes in the network of drainage canals in near real-time. Detection of smaller canals (less than 5-10 m wide) may pose a challenge.
Fire damage – Fire damage will be mapped using a combination of sensors, utilizing unique sensor capabilities. Sentinel-1 will be used for near real-time mapping of fire damage, while Sentinel-3 (and other sensors) will be used for fire detection. The radar systems (both L-band and C-band) are capable of detecting underground peat fires, because trees topple after their roots are burned, which is not seen by optical sensors (because of severe smoke and absence of flames).
Subsidence – Radar interferometry may give information on peat soil subsidence (mainly thorough oxidisation) in strongly degraded areas (which often have only sparse remaining vegetation).
The system – Data download, pre-processing as well as thematic processing is done in a fully automated way (as much as possible) to warrant near real-time information delivery. Access is done through a dedicated web viewer. Specifically, the system would result in alerts on new canal construction, monitoring land cover changes (deforestation, forest degradation, plantation development), fire alerts, alerts for excessive peat drainage, floods alerts and floods frequency mapping for open areas and areas under forest canopy (the latter is important because the flood patterns determine the appropriate reforestation strategies).

Area B - Board 175

Almost 2% of Mongolia is covered by peatlands, which constitutes about 27,000 square kilometres, out of 2,5 million square kilometres of the total territory of the country. The peatlands are mainly spread along the river basins, in close proximity to the lakes and other water bodies and on the top ridges of the middle high mountains (heights up to 3800 meters). The peatlands in Mongolia are important for Mongolian nomads and used for livestock pasturing during spring and autumn and hay making during autumn.

Using Remote sensing from both optical Landsat-8 and radar ALOS PALSAR sensors is new application not only in Mongolia, but in wetlands and peatlands mapping and monitoring. Several studies are cited successful application of the ALOS PALSAR data in mapping the distribution of the peatlands (Morishita and Hanssen 2015), including peatland delineation under forest canopy (Antropov and others 2012). In our research, we used images from Landsat-8 satellite acquired in 2015 in resolution of 30 meters, to produce the outline (shape files) of the peatlands. The dates were chosen to match the dates of the field trip, concluded during October-November 2015. The 350 GPS points in areas of Tuvshruuleh, Solongot and Orkhon in Central and West of Mongolia, to capture the delineation and to analyse the characteristics of peatlands were collected. The ALOS PALSAR images in resolution of 25 meters were acquired during winter 2014-12-09 to summer 2015-09-29 and color composites produced from horizontal and vertical polarizations. In addition, thematic maps on Soil, Vegetation (pasture type), Forest, Ecosystems and Permafrost maps in scale of 1:1 000 000 from Environmental Information Center, Ministry of Environment and Green Development of Mongolia were used as a reference. During field trip the observations are shown that Mongolian peatlands are mainly located on slopes lower than 15 degrees (the most favorable location to accumulate the water) and these locations are “peatland specific” and it is related with the topography. Also, peatlands in Mongolia are typically extended along river beds and foot of the mountains.

The Landsat-8 data used to calculate the wetness (amount of water, moisture), greenness (the vegetated state), brightness (density of vegetation/land cover) and Normalized Difference Vegetation Index NDVI (total biomass). The obtained parameters were used as input into General additive distribution GAM model, to derive the outline of the peatlands. The ALOS PALSAR data processing algorithm followed approach mentioned in (Hoekman and others 2010), the Bayesian approach based on (unsupervised) mixture modelling and Markov Random Field (MRF) classification. The approach of radiometric slope correction of SAR images followed the approach mentioned in (Hoekman and Reiche 2015) for slope correction in complex terrain.

The obtained results are shown the classification accuracy of the water (river streams) almost 80%, with correctly classified/modelled water, followed by forest in a 97% accuracy. The accuracy of agricultural land was 58%, due to problems of mixed pixels and unclear differentiation between dry land and abandoned agricultural land. As for peatlands it was about 70% respectively. The ALOS PALSAR images presented very clear difference between forest and peat areas on images acquired in winter and summer month. The final results from radar ALOS PALSAR images coincide with predicted locations derived from Landsat-8 images and matched with the locations of the peatlands digitized from 1:100 000 topomaps, with some differences. However, it is difficult to explain such differences between the outline of peatlands derived from topographic maps produced in 1946 and recent remote sensing data and it is not it is not clear, how the Mongolian peatlands were affected in the past years by climate change. There are no studies mentioning the effect of climate change to Mongolian peatlands and it is important to know, whether the areas of peatlands in Mongolia were intensively drying or expanding. The study results are in progress to publish.

Area B - Board 176

As a combination of terrestrial and aquatic ecosystem, wetland is an important human living environment and has prized as the kidney of the earth. As an inland wetland, Honghe National Nature Reserve is the miniature of Sanjiang Plain wetlands, which is highly typical and representative. However, over a long period of time, unreasonable development and utilization have caused serious ecological environment problem, the biodiversity gradually reduces and wetland function has been seriously degraded. Therefore, it is of great significance for the scientific management of wetland resources to carry out the analysis of wetland landscape pattern and driving force. This paper takes Honghe Nature Reserve as the research object, with the support of RS and GIS technology, combination with object-oriented classification method and the knowledge of landscape ecology, through the six phases of the remote sensing images from 1984 to 2014, and other related statistical data. Then the change process and driving mechanism of the wetland landscape pattern in the study area were analyzed comprehensively in the past three decades. The main conclusions are as follows:
(1) Using object-oriented classification method based on multiple remote sensing indexes to classify the information from remote sensing images. Then we improve the results with GIS rules to get better classification results. Object-oriented classification method is effective to reduce the classification of fragmentation phenomenon and get better result than normal classification methods. Based on the principal component analysis method, the band weight parameters in the process of object-oriented classification image segmentation were determined. DEM layer was added in the process of segmentation, which can effectively make use of the difference of information between different bands to obtain the better segmentation effect of image.
(2) From 1984 to 2014, wetland landscape has greatly changed. The advantage landscape during 1984 to 2000 is marsh landscape. The area of meadow increased year by year and become dominant landscape beyond the marsh at 2005. The area of typical wetland which includes marsh and water-body decreased year by year and the proportion dropped from 51.1% to 31.7%. Because of the construction of water conservancy, the area of water-body shrunk by 68.4%, and at the end of the study period, the proportion of water-body is only 5.8%.Bush area increased gradually and then decreased after getting the peak 40.81 square kilometer at 2005. The area of island forest is relatively stable, and the area decreased in the later period due to the rise of dam construction water level and the death of a large number of silver birches. The area of paddy field and dry land increased gradually, and the area of dry land increased rapidly. By the end of the study, it reached 10.51km2, still accounting for a small proportion in the whole.
(3) From 1984 to 2005, landscape types mainly transformed from water-body， marsh and meadow to each other, and from meadow to bush and to island forest, and meadow served as the link between water-body and marsh to bush and island forest. After 2005, the conversion tends to be clear between each types, the conversion direction is from water-body to marsh to meadow to bush and then to island forest. Due to the small impact of natural factors, paddy and dry land didn’t change much. During 1989 to 2000, the rate of wetland landscape change is the lowest with the integrated dynamic is only 4.7%. The change speed of the other four periods was relatively faster, comprehensive wetland landscape dynamics reaches around 11%. From the single landscape dynamic we can see that: paddy, dry land, meadow and bush have a rising trend, and rate decrease in turn. Water-body, forest and marsh have a decreasing trend, dropping gradually decrease. Meadow landscape area has the characteristics of wetland degradation increases, shows that wetlands are being degraded year by year, but the degradation rate getting smaller and smaller. The artificial landscape with paddy fields and dry land as the main types is disturbed by human activities, and the dynamic change is much greater than the natural landscape.
(4) Typical wetland landscape tends to be more complex and fragmentized. The dominance of landscape gradually lost. The area of meadow landscape increased year by year, the number of plaques has increased 1.94 times, and became the matrix landscape of the study area. The integrity of dry land is strong, island forest and bush landscapes tend to be fragmentized. From the perspective of landscape, the degree of landscape fragmentation is higher and higher, the complexity of landscape heterogeneity is higher and higher, the diversity of landscape is more and more abundant, and there are certain dominant landscape types in landscape.
(5) The changes of wetland landscape pattern are the result of natural factors and human activities factors. The amount of rainfall plays a decisive role in the change of water area. With the decrease of rainfall, the wetland landscape changes from water-body to marsh to meadow. The surrounding climate of the reserve tends to be warm and dry, which provides convenience for the surrounding agricultural development. The rapid expansion of rice planting area leads to the increase of agricultural water consumption, which leads to the decrease of underground water level in the reserve. The construction of the dam and the drainage trunk of the river cut off the upstream water sources of the reserve. Estimated by the water balance equation, the increase of agricultural water consumption caused by the rapid increase of rice planting area is the main factor for the change of Honghe wetland landscape. The construction of water conservancy projects and the surrounding drainage have increased this trend, with little impact from natural factors such as rainfall and temperature.

Area B - Board 178

Tidal channel networks are one of important factor influencing the morphodynamics of the tidal flat and the distribution of the sedimentary facies. Moreover, tidal channel network features can be useful to surface sedimentary facies classification in tidal flats. To record changes in morphology and quantify network expansion, it is of critical importance to delineate tidal channels precisely and objectively (Liu et al., 2015). Characteristics of tidal channels such as low relief, variable widths, strong anisotrophy of waterway directions are used to delineate tidal channel networks from the remotely sensed data such as Lidar, optical and SAR imagery. Here we investigate the relationship between tidal surface sedimentary facies and channel distribution in Gomso bay tidal flat, Korea. In this study, we determined the tidal channel networks based on the Lidar and TanDEM-X (TerraSAR add on for Digital Elevation Measurement) image with large baseline (>~1km) acquired during science phase, which is optimized for topographic mapping in gentle or flat areas such as coastal land. Tidal channel networks were effectively extracted by the automated extraction of tidal creeks proposed by Liu et al. (2015). Before the automated extraction, Sun’s denoising algorithm was applied to the TanDEM-X image to reduce the noise by speckle effects in intertidal DEM, and smooths the surfaces while keeping main topographic features that is favorable to channel network analysis. The pattern of tidal channel distribution was compared in terms of 3D fractal analysis. Extracted tidal channels from the TanDEM-X image has lower fractal dimension value compared with those from Lidar. Some ripples next to channels in sand flats were also recognized as channels during delineation from the TanDEM-X image However, both tidal channels extracted Lidar and TanDEM-X image in each sedimentary facies had relatively constant meandering patterns, but the density and complexity were distinguishable. Moreover, the topographic difference between similar sedimentary facies were also distinguishable.

Area B - Board 179

Soil Moisture can be described as the level of saturation in the upper soil layer relative to the soil field capacity, regulated by the precipitation and potential evaporation, while being highly variable in space and time. It constitutes a key factor for agricultural management such as optimizing the fertilizer rates and irrigation, applying pesticides or herbicides and crop management. Moreover, soil moisture is considered as valuable information in many sectors such as Hydrology, Biogeography, Geomorphology, Agronomy and Climatology. To this end, the main objective of this research was to evaluate the concurrent use of satellite multispectral and SAR radar data for estimating soil moisture in large spatial scales. In particular, Landsat 8 Surface Reflectance data as well as Sentinel 1 GRD SAR data were employed in the region of Arta across the Amvrakikos Gulf and the Amvrakikos Wetlands Natural Park. Recent studies have indicated that the amplitude derived by SAR data in VV polarization along with information about the Normalized Difference Vegetation Index (NDVI), the Normalized Difference Moisture Index (NDMI) and the Moisture Index (MI) from multispectral data can be proxies of soil moisture. In-situ measurements from the Enhydris project were also acquired spanning three different years. Google Earth Engine was exploited for mining the satellite data through the Javascript API services. Then several experiments were performed in order to establish correlations between the Ιn-Situ and satellite data, based on statistical and machine learning tools like Linear Regression, Polynomial Regression, Generalized Additive Models (based on R Statistical package), as well as deep learning models, using the TensorFlow Framework in association with the Keras library in R. Generally speaking, based on the considered relatively large, multitemporal dataset, the statistical approaches did not manage to establish concrete correlations in any of the performed experiments and combinations. The MI index along with the VV backscatter though were closer to the expressed variation in the In-Situ dataset. Based on the deep machine learning framework, stronger correlations were established between the In-Situ data from Enhydris and a combination of VV amplitude and NDVI satellite observations.

Keywords: Remote Sensing, Earth Observation, Radar, High Resolution, Deep Learning, Correlation, In-situ, Soil, R

Area B - Board 166