Available for download from
http://wrcwebsite.azurewebsites.net/mdocs-posts/establishing-remote-sensing-toolkits-for-monitoring-freshwater-ecosystems-under-global-change/
EXECUTIVE SUMMARY:
BACKGROUND
Freshwater ecosystems are globally considered one of the most threatened ecosystems in the world. More than > 85% of wetlands are estimated to be lost or degraded since the 1700s and the rates of decline in their extent and ecological condition have increased in the past 50 years. Increasing pressures related to water abstraction, water pollution, habitat degradation, the fragmentation that results from these, as well as invasive species and climate change, have a multiplying effect on the degradation of wetlands. Owing to the inadequate connectivity of wetlands in the landscape, the consequences are severe. Efforts initiated within the past 50 years to ensure sustainable use of resources therefore needs to be intensified to avoid further losses.
Similarly to the global assessments, fine-scale studies in South Africa have also showed high levels of wetland degradation with an estimated 50% of wetlands transformed by 1988. Several climate change parameters are likely to exacerbate the current pressures on wetlands. The concerns around climate change include the increase in temperatures observed within the past 50 years, predictions of further temperature increases between now and 2050 and the increasing intensity of storms and droughts. These changes have a multiplying effect on the intensity of current pressures and their interactive effect. In South Africa, the last two National Biodiversity Assessments of 2011 (NBA 2011) and 2018 (NBA 2018) showed that wetlands are inadequately mapped, highly threatened and poorly protected. Inventorying and monitoring of this ecosystem should therefore be a top priority. For nearly 70% of the extent of South Africa, it is estimated that the latest National Wetland Map version 5 (NWM5) represents only 50% of the wetlands, and that these may be mostly vegetated or arid.
Remote sensing has for the past 50 years been an important data source in mapping wetland extent, either through heads-up digitising or image classification. New sensors launched since 2008 have improved the capability of remote sensing to detect wetland vegetation and separate it from upland vegetation, through improved spatial resolution and extra bands representing the red-edge and shortwave infrared regions of the electromagnetic spectrum. These space-borne sensors provided additional capability, beyond that of the freely available Landsat images. Since 2014, the European Space Agency (ESA) has been launching the Sentinel-1, -2 and -3 sensors. The data from the new sensors are freely available to the public, with free software such as the Sentinel Application Platform (SNAP) for preprocessing and analysis. Conditions are therefore ideal for assessing the capabilities of these sensors for aiding in mapping and monitoring of wetlands in South Africa.
AIMS
The aim of this research project was to assess the ability of the new space-borne sensors, such as Sentinel, for the inventory and long-term monitoring of vegetated wetlands, as part of freshwater ecosystems, in the face of global change. The approach was to assess the capabilities of the new multispectral and SAR space-borne sensors in combination with a variety of wetland vegetation indices, for reporting on structure, function and condition of vegetated freshwater ecosystems.
The objectives of the project were to:
1. Ascertain the range of the feature extents of wetlands that could be detected by various types of sensors;
2. Calculate the contribution of field visits to improving the representation of the extent and HGM units of inland wetlands;
3. Determine the capability of the new space-borne sensors to separate wetlands from uplands, using either vegetation or thresholding of soil moisture content;
4. Assess the capability of SAR technology to detect the temporal dynamics of vegetation structure (above ground biomass);
5. Investigate whether optical imagery can detect the seasonal or annual variation of freshwater ecosystems (reporting variation in function); and
6. Evaluate the utility of remote sensing tools for the inventory and monitoring of freshwater ecosystems under global change as part of the national programmes such as the National Wetland Monitoring Programme.
METHODOLOGY
1. The degree to which sensors would be able to detect different wetland sizes was evaluated for three regions of South Africa in the Grassland, Fynbos and arid (Nama-Karoo) biomes, as well as using all wetlands mapped in the National Wetland Map version 5 (NWM5).
2. In-field visits and desktop mapping using the WorldView-3 images acquired for the Tevreden Pan (Mpumalanga Province) and Hogsback (Eastern Cape) study areas were used to improve the extent and hydrogeomorphic unit classification of wetlands. The results were compared to the wetlands from the National Freshwater Ecosystems Priority Areas (NFEPA) project, which were also used in the National Biodiversity Assessment of 2011 (NBA 2011). The output of these datasets were contributed to the NWM5 and used for the NBA 2018 assessment of wetland ecosystem types.
3. The Random Forest classification algorithm was used to assess the classification errors and accuracies of the Sentinel-2 and WorldView-3 images for the Tevreden Pan (Mpumalanga Province) and Hogsback (Eastern Cape) study areas to determine whether wetland and upland vegetation were separable, and whether different wetland vegetation communities could be mapped.
4. Empirical regression modelling was used to compare the ability of the Sentinel sensors to that of WorldView-3 in the estimation and mapping of the above ground biomass of wetland and upland vegetation for the Tevreden Pan (Mpumalanga Province) and Hogsback (Eastern Cape) study areas.
5. The monthly extent of inundation of depressions in the Mpumalanga Lakes District (MLD) was mapped from the Sentinel-2 sensors. Thereafter the maximum extent of inundation and hydroperiod categories were derived by processing statistical information in Excel.
6. A literature review was done to (a) assess the impact of climate change on wetlands in South Africa and (b) determine the capabilities of remote sensing for quantifying and monitoring wetland attributes (Chapter 2). Thereafter a number of themes were listed as priorities for implementation in the National Wetland Monitoring Programme (NWMP) that is about to commence.
An additional chapter (Chapter 7) was included based on a student project funded by the Water Research Commission and associated with this project, which determined whether Soil Moisture Content can be predicted from the Sentinel sensors and used for inventorying and monitoring of wetlands. For this study, empirical modelling was used to determine the correlation between in-field measurements of Volumetric Water Content and the Sentinel images for the Colbyn Wetland in Pretoria, South Africa.
RESULTS AND DISCUSSION
The results and discussion of the various objectives are summarised below as key findings of the literature and subsequently this report:
Key findings from the literature review:
• Remote sensing has been an important data source during the past 50 years for the mapping and monitoring of wetland extent.
• Land cover datasets dominate the product range that was derived from remote sensing sensors in South Africa, with fewer products related to the mapping of wetland vegetation.
• The Landsat sensors are amongst the most used space-borne sensors for the mapping of wetlands and inferring the ecological condition of wetlands.
• The Sentinel optical sensors, launched since 2015, now contribute to finer spatial resolution of mapping and monitoring features in South Africa.
• To date, open water or inundated wetlands were more easily mapped than palustrine wetlands.
• More effort is required to map and monitor palustrine wetlands.
• Unmanned Airborne Vehicle (UAV or ‘drone’) images can contribute to specific interventions required in wetlands, such as thermal imagery of sub-surface fires in peat wetlands (Grundling et al., 2019).
• Research funding contributes to significant discoveries relating to the use of remote sensing for wetland mapping and monitoring.
• Further work will be required for determining how remote sensing indices can be used to represent and monitor ecological condition of wetlands over time.
• Several global and South African products derived from remote sensing that are under development could benefit both the wetland and other realms.
Key contributions and knowledge generated from this study:
• Sentinel sensors can make a valuable contribution to the mapping and monitoring of wetlands in South Africa.
• Wetland vegetation is highly separable from upland vegetation in the Grassland biome for two study areas, including Tevreden Pan and Hogsback.
• Wetland vegetation communities were more separable using WorldView-3 images than with Sentinel-2 images.
• Theoretically, current, freely available multispectral space-borne sensors with a spatial resolution of ≥ 10 m are able to detect > 69% of the aerial extent of South African wetlands as represented in the NWM5.
• Fine-scale studies showed that wetlands on slopes greater than 10% will likely require images of higher spatial resolution compared to those on slopes less than 10%.
• The Sentinel sensors were able to predict above ground biomass (AGB) of wetland vegetation with accuracies comparable to those of WorldView-3.
• The Sentinel optical and Synthetic Aperture Radar (SAR) sensors show potential for estimating soil moisture content (SMC) across a wetland-upland gradient.
• The increased temporal frequency of Sentinel-2 compared to Landsat images, refines the determination of the maximum extent of inundated depressions for reporting to SDG 6.6, as well as characterising hydroperiod classes for the inventory of wetlands.
GENERAL
The Sentinel sensors provide an improved data source for the quantification and monitoring of wetland characteristics vulnerable to global and climate change.
CONCLUSIONS
In conclusion, remote sensing has demonstrated its usefulness for the inventorying and monitoring of wetlands in South Africa. South Africa also has the expertise to facilitate implementation and further research to optimise the use of the now freely available Sentinel sensors for this purpose. Earlier studies were constrained by data processing capabilities, storage and cost of images (e.g. Thompson et al., 2002), but imagery and software are now more freely available for processing, while the capabilities of information systems for processing large images have improved. In our opinion, the only challenge we have today, is to secure funding for implementation and operation of monitoring systems for wetlands.
RECOMMENDATIONS
In summary, we suggest that managers of the National Wetland Monitoring Programme (NWMP) consider the use of remote sensing in five areas of response, related to:
A) the improvement of the representation of wetland extent in future versions of the NWMP and for Sustainable Development Goal (SDG) reporting;
B) Improved characterisation and description of biodiversity of wetland ecosystem types in the NWMP;
C) Intervention strategies that will be increasingly required under predicted climate change scenarios;
D) The modelling of ecological condition, required for assessment and planning; and
E) Monitoring of wetland attributes.
Different themes are summarised, which we consider a top priority, ranked according to the ease of implementation and with recommended sensors. Reporting channels have also been identified through known national departments and international reporting obligations.