Savannas are important to South Africa as a major reservoir of carbon, biodiversity and economic potential. As such, savannas play a key role in regulating biodiversity (loss), the exchange of carbon between plants, soil and atmosphere and generating revenue from tourism.

Although African savannas such as the Kruger National Park are often seen as pristine natural ecosystems, modern (and potentially eaerlier) humans have made their mark on the landscape using fire for at least 100 ka and probably for much longer. Fire is an important driver of evolutionary and ecological processes in grassy ecosystems and has major implications for carbon dynamics, vegetation structure and thus biodiversity.

In addition to regulating fire, humans drove many of the larger vertebrate taxa to local extinction over the past two centuries. Since large vertebrates can have strong effects on the amount and spatial distribution of vegetation biomass, they can affect the frequency, intensity and spatial spread of grass fires. By manipulating fire, large herbivores may thus regulate carbon dynamics over much wider areas than the areas that are directly grazed or browsed. However, this interaction remains largely unquantified.

Quantifying effects of large herbivores on vegetation structure and fire is timely, as savannas globally are undergoing expansion of woody plants, which is partly attributed to changing climates and elevated atmospheric CO2 concentrations. Furthermore, today’s faunal assemblages in most protected savannas are the consequence of active management, often in the form of translocations and reintroductions. Large vertebrates are thus already a feasible tool in managing African ecosystems and are increasingly used on other continents.

In this study, we ask how large vertebrates affect vegetation structure and fire regimes at the landscape scale. We use mobile terrestrial laser scanning to quantify attributes of vegetation structure at local scales. We then upscale estimates of vegetation structure to the landscape scale, using a combination of very high resolution WorldView3 spectral data and Sentinel 2 imagery. We explore two avenues toward successful upscaling: 1) class-based extrapolation of vegetation structural attributes using land cover classifications derived from WorldView3 imagery, and 2) continuous extrapolation of vegetation structural attributes using empirical relationships with spectral signatures from both WorldView3 and Sentinel 2.

This study was performed in a South American savanna with similar structural attributes to African savannas, i.e. a more-or-less continuous C4 grass layer mixed with savanna trees. The lessons learned from this study are therefore expected to be highly applicable to (South) African savannas, which will be tested in the near future.

Natural resource planning is as much a priority for conservationist as it is for livestock farmers. Whereas the farmer may require this information for optimum livestock production, conservationist needs to maintain biodiversity and ensure that the diversity of ungulates have access to preferred food resources. The interpretation of vegetation data to obtain homogenous mapping units are important for the planning process, but always a challenge since vegetation boundaries are seldom abrupt. Landsat has been a source of data for vegetation mapping for many years. The launch of Sentinel-1 and Sentinel-2 satellites, however, provides an advanced capability to map and monitor vegetation communities and processes.
The Karoo National Park was proclaimed in 1979. A vegetation description of the first sections that were proclaimed was done by Rubin and Palmer (1996). Subsequently, the park has extended considerably westwards. Further work is therefore needed to map the vegetation communities and includes these new areas within the natural resource management plan. The use of long-term remote sensing data for vegetation condition monitoring should also be explored, which could have relevance for surrounding agricultural areas.
The study investigated the use of Sentinel-2, Sentinel-1, and other terrain morphological data to map structural differences in vegetation and broad vegetation communities. These communities were identified using phytosociological data collected at more than 230 sampling localities during the past 25 years. Sixteen vegetation communities have been identified which occupy landscapes within the park that have particular terrain, geological and microclimatic attributes. The sampling localities provided areas from where spatial data regarding terrain geology and spectral and backscatter characteristics could be sampled. The spatial data were thereafter compared with vegetation data employing multivariate analysis ordination and random forest classification. The Karoo-National Park and surrounding areas has since 2016 experienced significant drought periods. A time series of 100m Proba-V NDVI data were used to asses the drought resilience for the mapped landscape types using 2015 as the reference year.

References
Rubin, F. and Palmer, A.R., 1996. The physical environment and major plant communities of the Karoo National Park, South Africa. Koedoe, 39(2):25-52.

Globally approximately 50.000 salt pans exist. These landscape features are highly dynamic environments that are difficult to study by in-situ and conventional methods because of their harsh climatic conditions and large spatial areas. Remote sensing can help to study their environmental dynamics and provide important constraints on their sedimentological, mineralogical, and hydrological evolution. This study utilizes Landsat 8 data combined with a spectral endmembers characterization to derive the spatial distribution of surface crust types over time on the Omongwa Pan located in the Namibian Kalahari. For this purpose, 49 surface soil samples have been collected for spectral and mineralogical characterization during multiple field campaigns reflecting different seasons and surface conditions of the playa environment. An approach was developed to allow the spatio-temporal analysis of the salt pan crust dynamic in a dense time-series consisting of 77 Landsat 8 cloud-free scenes between 2014 and 2017, covering at least three major wet-dry cycles. The Sequential Maximum Angle Convex Cone (SMACC) extraction method was used to derive image endmembers from the Landsat time-series stack. Evaluation of the extracted endmember set revealed that the multispectral data allowed the differentiation of four endmembers associated with mineralogical mixtures of the soil crust’s composition in dry conditions and three endmembers associated with flooded or muddy pan conditions. The dry soil crust endmember spectra have been identified in relation to Visible, Near Infrared and Short Wave Infrared (VNIR SWIR) spectroscopy and X ray diffraction (XRD) analyses of the collected surface samples. According these results the spectral endmembers are interpreted as efflorescent halite crust, mixed halite-gypsum crust, mixed calcite quartz sepiolite crust, and gypsum crust. For each Landsat scene the spatial distribution of these crust types was mapped with the Spectral Angle Mapper (SAM) method implemented in the Google Earth Engine to allow the fast extension of the time-series to newly acquired data to support a regular monitoring of pan processes. Further, the surface crust dynamic was analysed in comparison with the pan’s moisture regime and other climatic parameters. The results show that the crust dynamic is mainly driven by flooding events in the wet season, but it is also influenced by temperature and aeolian activity in the dry season. The approach utilized in this study combines the advantages of multitemporal satellite data for temporal event characterization with advantages from VNIR-SWIR spectroscopy of image and ground data analyses that allow improved mineralogical differentiation and characterization, as well as the capabilities of current cloud-processing services for access and processing of remote sensing data.

Savanna ecosystems are an important, yet poorly understood component of the global carbon cycle. Their spatio-temporal monitoring is a particularly challenging task because of the fine-scale heterogeneity and dynamics of these environments. As a result, spatially explicit and continuously updated information on savanna vegetation parameters like herbaceous biomass (also called fuel biomass or fuel load) are urgently needed. Among others, these thematic maps are particularly useful for sustainably managing conservation areas as they can help decision makers identifying areas of high fire risk and/or forage availability.

With the advent of the European Copernicus satellite program (i.e., the Sentinel missions), new opportunities for detailed, area-wide, frequent, and up-to-date observations of savannas have become available that still remain largely unexplored. This study investigates the potential synergies between multi-temporal Sentinel-1 and -2 data for wall-to-wall herbaceous biomass mapping in Kruger National Park (KNP), South Africa. To this end, annual time series of radar (Sentinel-1) and optical (Sentinel-2, Landsat-8) data as well as a number of image products derived thereof (e.g., interferometric coherence, vegetation indices) are employed. For calibration and validation purposes, use is made of in situ herbaceous biomass values measured as part of the veld condition assessment (VCA) and experimental burn plot (EBP) studies conducted within KNP. From a methodological perspective, the potential of Sentinel-1 and -2 data is assessed on the basis of a random forest (RF) machine learning approach that is applied within the framework of three dedicated experiments. Each experiment quantifies herbaceous biomass modeling performance in dependence of the input data considered: (1) using Sentinel-1 data only (e.g., backscatter intensity, interferometric coherence); (2) using Sentinel-2 data only (e.g., spectral reflectance; soil, water and vegetation indices); and (3) using Sentinel-1 and -2 data in combination. The results of this study provide detailed insights into the individual and combined potential of Sentinel-1 and -2 time series data for wall-to-wall herbaceous biomass mapping in KNP.

South Africa is a water scarce country and predictions indicate that this scarcity is likely to intensify due to a growing population and climate change. The recent drought in the Western Cape Province of South Africa has highlighted the urgent need for effective water resource management in the face of a changing climate. While large reservoirs and dams provide centrally managed sources of water for drinking and irrigation, small, privately owned dams located in predominantly rural areas contribute significantly to total water storage. These small dams are largely used to harvest runoff for use in irrigation, water for livestock and other agricultural activities. While they serve as important sources of water for agriculture, they have also been found to reduce stream runoff and function as sediment traps which, when breached, threaten downstream water systems. Currently, it is unclear how many of these reservoirs exist in the Western Cape.

Optical satellite remote sensing is routinely used for water body mapping, and the application and methods are fairly mature. More recently, SAR sensors have also been used for this purpose, usually motivated by the issue of cloud cover in optical imagery. However, waterbodies with high turbidity or aquatic plant cover are also challenging to discern in optical imagery. Sentinel-1 provides freely accessible C-band SAR imagery with high temporal resolution, but with medium spatial resolution (~20m in IW mode) and the accompanying issue of speckle. In this study, we evaluate the ability of Sentinel-1 to map small dams in the Western Cape Province, with particular attention to the effect of multitemporal filtering.

A study area of around 780km² was defined around the town of Paarl. A stack of ten Sentinel-1A IW images with a 12-day temporal resolution was used to perform multitemporal filtering of the first image in the stack, dated 9 December 2017. Filtering was performed in the SNAP 5.0 software using a Refined Lee filter. For comparison, an unfiltered version of this image as well as a single-image Refined Lee filtered version was also prepared. For each of these three versions of the image, VV and VH sigma-nought, as well as ten GLCM texture measures were extracted. The images were segmented in eCognition and averaged values for each of the image features extracted per image object. A Random Forest (RF) and Support Vector Machine (SVM) classifier was trained on an accompanying Sentinel-2 image using a binary classification scheme (Water and “Other”). All classifications were repeated 1,000 times to account for training/reference splits and random selection in RF.

The results of the classifications showed that, for both RF and SVM, multitemporal filtering achieved higher overall accuracies and kappa values, and lower errors of omission and commission than for single image filtering and unfiltered imagery. For the RF classification, multitemporal filtering also substantially reduced the variance of the overall accuracies across iterations, though both classifiers achieved similar averaged overall accuracies. For RF, overall accuracies of 84%, 85% and 91% were achieved for unfiltered, single image filtered and multitemporal filtered imagery, respectively. The corresponding Kappa values were 0.69, 0.68 and 0.76. Using multitemporal filtering, errors of omission and commission were consistently below 10%, irrespective of classifier.

While questions remain (such as the smallest dam size that could be detected, the impact of segmentation parameters, the sensitivity of the classification to dam level variations and the transferability of the classifiers) the study yielded useful results. We have shown that multitemporal filtering improves the classification accuracy of small dams using dual-polarised Sentinel-1 imagery, and yields acceptable accuracy and error levels. Given the large extent of Sentinel-1 IW scenes, it would be feasible to develop a regional or catchment-wide small dam mapping system as a first step towards developing a census of these small reservoirs in the Western Cape. The cost-benefit of a combined SAR/optical system should then be investigated.

The monitoring of soil moisture content and spatial-temporal dynamics is of great importance for understanding the earth system, as it is an essential surface parameter for observing the linkage between water, energy and carbon dynamics. Soil moisture is influenced by the amount of precipitation but also affects precipitation due to evapotranspiration.
The El Niño/Southern Oscillation (ENSO), which is driven by periodic fluctuations of ocean temperatures in the equatorial Pacific, is negatively correlated with the amount of rainfall occurring during the South African summer season, which can, in an extreme case, lead to floods or droughts. Droughts are the major catastrophic scenario, making it a great threat to humans, as it can result in severe environmental and economic changes (e.g. agricultural and biomass productivity).
The project SALDi (South African Land Degradation Monitor) aims to implement novel, adaptive and sustainable tools for assessing land degradation. The goal is to take advantages of high spatial-temporal Earth Observation and in situ information as well as modeling approaches for developing a data cube, which will improve land degradation management in politics as well as communal and agricultural planning.
An automated in situ soil moisture probe network will be installed in early 2019 covering various landscapes in South Africa. Each of the six study sites covers an area of 100 km by 100 km. To investigate spatial-temporal dynamics of surface moisture, data from ESA’s (European Space Agency) Copernicus Sentinel-1 satellites will be utilized. The C-band data will be used to derive surface moisture based on a hyper-temporal co-polarized (vertical-vertical - VV) radar backscatter change detection approach. Recent analysis showed the benefit of including vegetation information from Sentinel-2 and Landsat-8 to analyze vegetation structure types and dynamics with respect to surface moisture. We will present early results of the analysis of surface moisture dynamics derived from Earth observation time series for the development of a land degradation monitor.