Area A - Board 57

Reliable information about the extent of tree canopy cover (TCC) is essential for knowledge-based policy-makers and different research fields such as environmental modelling. Latter are interested in this topic, due to the crucial importance of forests as co-determinators for the regulation of the global carbon cycle (Avitabile 2012). Hence, optimal forest cover detection requires input data that provides both high spatial and temporal resolution, allowing a more realistic spatial representation of forests cover, particularly in sparsely forested regions.
These criteria are met by the Sentinel-1 and -2 satellites, which are operated by the European Space Agency (ESA). In order to derive accurate maps of forest cover in different latitudes and thus ecosystems, preprocessing and classification was firstly carried out using optical Sentinel-2 data only to derive TCC within this study. Following, a comparison with results from radar data (Sentinel-1) alone and from a synergistically classification approach (Sentinel-1 and -2) was accomplished, to evaluate the beneficial effect of data fusion for forest cover derivation in varying ecosystems. As study sites, Thuringia (Germany) and Kruger National Park (South Africa) were chosen, to assess the robustness of the machine-learning classification procedure. Therefore, the combination of spectral bands, vegetation parameters such as Fraction of Absorbed Photosynthetically Active Radiation (fAPAR) and Leaf Area Index (LAI) from Sentinel-2 as well as multi-temporal metrics derived from Sentinel-1 were used in an non-parametric Random Forest (RF) approach, leading to considerably high classification accuracies between 80 and 95 %. In order to evaluate performance, cross-validation (CV) was carried out using an innovative approach applying the concept of spatial autocorrelation to account for model overfitting which is not part of traditional CV’s and thus, causing the introduction of a BIAS into the classification.
Results of spatial and non-spatial CV were then compared to each other to evaluate the impact of spatial dependence within the data. Analysis of the variable importance revealed that Sentinel-2 channels 11 & 12 (SWIR) as well as the minimum backscatter of the multi-temporal Sentinel-1 metrics where the most important parameters for the discrimination between forest and non-forest. The results indicate that Sentinel-1 and –2 fusion provides valuable information for forest cover derivation in varying latitudes and ecosystems.

Area A - Board 55

The savanna ecosystems in South Africa, which are predominantly characterized by woody vegetation (e.g. shrubs and trees) and grasslands with seasonal changes, are vulnerable to impacts from droughts, fires, herbivory, etc. Therefore, monitoring of the vegetation structure and dynamics is one of the essential components for the management of complex savanna ecosystems such as the Kruger National Park (KNP).
To map the vegetation structure in the KNP, data from ESA’s Copernicus Sentinel-1 (dual-polarized - VV/VH) for the period between 2015 and 2017, with 10 m spatial resolution and repeated acquisitions each 12 days, were utilized. Multi-temporal statistics (e.g. median, percentiles, etc.) were derived from the time series to identify changes in the backscatter signal over time. A high resolution Light Detection and Ranging (LiDAR) data set from Smit et al. (2016) was reclassified to produce vegetation height classes and was consequently used as reference information. The classification of the vegetation height classes was carried out using the random forest package ranger in R. The estimation of the model accuracy was accomplished via spatial-cross validation and resulted in a value of around 70 %. In addition, the final product has been compared to other reference maps (e.g. Bucini et al. (2009)) as well as high-resolution google earth imagery.
The methodology of this study was designed for mapping static savanna vegetation cover classes based on height estimates. Future studies will expand this approach to derive woody cover percentage information as well as their dynamics using Sentinel-1 time series information.

References:

Bucini, G.; Saatchi, S.; Hanan, N.; Boone, R. B.; Smit, I. Woody cover and heterogeneity in the savannas of the Kruger National Park, South Africa. In International Geoscience and Remote Sensing Symposium (IGARSS); IEEE, 2009; Vol. 4, p. IV-334-IV-337.

Smit, I. P. J.; Asner, G. P.; Govender, N.; Vaughn, N. R.; van Wilgen, B. W. An examination of the potential efficacy of high-intensity fires for reversing woody encroachment in savannas. J. Appl. Ecol. 2016, 53, 1623–1633, doi:10.1111/1365-2664.12738.

Area A - Board 56

The project ‘South Africa Land Degradation Monitor (SALDi)’ will be addressing the behaviour and functioning of multi-use landscapes with respect to land use, land cover change, water fluxes, and the implications for habitats and ecosystem services. Its overarching goal is to implement new and sustainable, adaptive land degradation assessment tools in close collaboration between the German and South African partners.
In cooperation with our South African partners (ARC, CSIR, ESKOM, SANSA, Universities of Pretoria, Stellenbosch, Free State), the overarching goal of SALDi is to implement novel, adaptive, and sustainable tools for assessing land degradation in multi-use landscapes in South Africa. Particularly, SALDi aims:
1) to develop an automated system for high temporal frequency (bi-weekly) and spatial resolution (10 to 30 m) change detection monitoring of ecosystem service dynamics,
2) to develop, adapt and apply a Regional Earth System Model (RESM) to South Africa and investigate the feedbacks between land surface properties and the regional climate,
3) advance current soil degradation process assessment tools for soil erosion, as this process represents an intrinsic limiting factor for biomass production and other regulating, supporting and provisioning ecosystem services.
The overarching goal of the parallel EMSAfrica-project is to establish jointly with the South African partners (SAEON, EFTEON) an interdisciplinary research infrastructure to define effective management strategies for South African ecosystems considering aspects of climate change, biodiversity and land management by implementing:
1) a long-term ecological observation infrastructure combined with associated short term experimental manipulation,
2) a combination of advanced modelling approaches (ecosystems, plant communities, multi-agent) and remote sensing techniques for multi-scale and transdisciplinary scenario generation,
3) data exploration and visualization by a web-based Decision Support System that supports a multi-modal dialogue between decision makers and domain experts, and
4) multi-disciplinary/inter-disciplinary/broad-ranging capacity building.

Focus of both projects SALDI and EMSAfrica, funded through the German ministerial programme BMBF-SPACES, is on refining existing approaches by explicitly taking into account the annual and seasonal variations in a spatially explicit and high-resolution approach as well as taking into account the socio-economic dimension.

The South African National Parks project CO-LD-EMS serves the two new SPACES projects SALDI and EMSAfrica by combining their remote sensing activities. Its remote sensing concept aims 1) to develop an automated system for high temporal-frequency (bi-weekly) and spatial resolution (10 to 30 m) change detection of ecosystem service dynamics through 2) implementation of a long-term ecological observation infrastructure, 3) a combination of advanced modelling approaches and multi-scale remote sensing techniques, 4) data exploration and visualization with a web-based Decision Support System and 5) respective multi- and inter-disciplinary capacity building activities.
Out key questions are: 1) What are the relevant ecological observables in space and time from the new generation of Sentinel-1/-2 and Landsat 8 optical and Earth observation data with respect to structure, composition, soil and fuel moisture to serve KNP’s and neighbouring territories’ monitoring and management needs? 2) Can new spatio-temporal indicators be defined for e.g. grassland and drought monitoring inside KNP and land degradation outside through the availability of the new generation of radar-optical Sentinel-data?
Our KNP research site is located southwest of Skukuza and is one of the six 100x100 km2 study areas. Activities include: selected vegetation surveying with terrestrial laser scanning (TLS) during dry and wet seasons to quantify grass, shrub and tree biomass dynamics; radar and optical remote sensing data exploitation to determine structural parameters and to map seasonal surface moisture changes; installation of an autonomous soil moisture probing network.

This presentation gives an overview of the involved methodological procedures and data streams.

Area A - Board 58