In times of rapid global change, ecosystem monitoring is of utmost importance. Combined field and remote sensing data enable large-scale ecosystem assessments, while maintaining local relevance and accuracy. In heterogeneous landscapes, however, the integration of field-collected data with remote sensing image pixels is not a trivial matter. Indeed, much of the uncertainty in models that use remote sensing to map larger areas lies on the field data integration. In this study, we propose to use fine spatial resolution (5 x 5 m2) remote sensing data as auxiliary data for upscaling field-sampled aboveground carbon data to target (meso-scale, i.e., 30 x 30 m2) image pixels. We also assess the effects of field data disaggregation and extrapolation, with and without using the auxiliary data. We test this on three study sites in heterogeneous landscapes of the Brazilian savannah. We thus compare two methods that use auxiliary data—surface method, which uses a weighting layer, and regression method, which applies a regression model—with one method without auxiliary data—cartographic method. To evaluate our results, we compared observed vs. estimated aboveground carbon values (for known samples) at the pixel level. Additionally, we fitted a random forest regression model with the assigned carbon estimates and the target satellite imagery and assessed the influence of the fraction of extrapolated vs. sampled carbon values on model performance. We observed that, in heterogeneous landscapes, the use of fine spatial resolution remote sensing data improves the upscaling of field-based aboveground carbon data to coarser image pixels. We also show that a surface method is more suitable for spatial disaggregation, while a regression approach is preferable for extrapolating non-sampled pixel fractions. In our study, larger datasets, which included a higher proportion of estimated values, generally delivered better models of aboveground carbon than smaller datasets that are assumed to more reliably reflect reality. Our approach enables to link field and remote sensing data, which in turn enables the detailed mapping of aboveground carbon in heterogeneous landscapes over large areas through the optimized integration of field data and multi-scale remote sensing data.

Global drylands form a major land cover which takes up about 40% of the Earth's surface, and accounts for approximately 40% of the global net primary productivity. These drylands are under large pressure due to global change and human activities, but the impact of this pressure remains unclear as these ecosystems are largely understudied. This is why our research aims to unravel the driving mechanisms of vegetation shifts in the Sahel. In this study, we focus specifically on the impact of intra-seasonal rainfall patterns on the functioning of savanna ecosystems in the Sahel. Indeed, among the climatological drivers, changes in intra-seasonal meteorological patterns may have a significant influence on the vegetation composition and functioning of savanna ecosystems. In particular, understanding how these ecosystems respond to intra-seasonal rainfall variability is crucial to predict their fate under climate change.

In order to unravel the impact of water stress on Sahelian drylands, we combine in-situ measurements with dynamic global vegetation models and remote sensing observations. We first parameterized a global dynamic vegetation model, LPJ-GUESS, to reproduce the vegetation composition and the surface fluxes measured at six Sahel sites, spread out from Senegal to Sudan. By comparing the simulated outcomes of various rainfall scenarios in our parameterized model, we then quantified the relative importance of the different rain season characteristics for five focus areas in the Sahel. We studied the influence of varying the rain season length, start, intensity and the total precipitation on the annual carbon cycling, phenology and vegetation composition of the simulated ecosystem. These different rainfall patterns are constructed using a newly developed approach that sustains meteorological consistency while at the same time providing enough control to change these characteristics independently. Finally, we investigated the link between intra-seasonal rainfall variability and vegetation shifts in the Sahel. For this, we compared the model results with remote sensing data of our focus areas, such as the LAI, FPAR, NDVI and VOD, which provide an additional validation of our model output. This work allows us to gain a process-based insight in the intricate dynamics between meteorological variability and the functioning of savanna ecosystems, and it provides a first step towards unraveling the driving mechanisms of vegetation shifts in the Sahel.

The Brazilian savanna, commonly known as the Cerrado, is characterized by a gradient of natural vegetation structure and density, along with a very diverse species composition. A weak conservation status in combination with a growing demand on land resources for agricultural production have already led to large land conversions of the Cerrado’s natural vegetation. This makes it critical to provide land managers and decision makers with spatially explicit information where conservation actions should be directed. While the usefulness of currently available land use/cover maps are often spatially, temporally or regionally restricted, time series of freely available data from the Landsat archive enable to overcome such limitations. Landsat time series allow for synoptic coverage of large areas at 30 m spatial resolution and open up opportunities to go beyond classic land cover classification e.g. by deriving phenological information of the observed vegetation. Land surface phenology metrics, derived from remote sensing data, capture the vegetation’s seasonal behavior on a pixel level and enable to frequently gather phenological information over large extents, which for most parts of the Cerrado still remain fragmented. While being beneficial for the mapping approach per se, these metrics allow insights into ecosystems that can help to deepen our understanding of ecosystem dynamics and their responses to environmental change. We used an 8-day Landsat collection 1 EVI time series, in which data gaps were closed using a weighted ensemble of radial basis convolution filters. We subsequently derived land surface phenology (LSP) metrics for the period 2013/2014 for the entire Cerrado biome (approx. 2 Mkm²). We identified eight LSP archetypes for the Cerrado based on similarities between LSP metrics using a k-means cluster algorithm. LSP archetypes enabled a detailed description of phenological patterns in existing ecoregions for the entire Cerrado, and demonstrated the complementary benefits of these two concepts e.g. for a spatially explicit design of conservation strategies. We further used boosted regression trees with selected biophysical variables related to climate and terrain, to identify leading determinants influencing LSP archetypes. Our results highlight the advantages of LSP metrics derived from dense Landsat time series for the analysis of heterogeneous ecosystems over large spatial extents. The proposed concept of LSP archetypes can support a range of applications such as carbon quantification, the assessment of biodiversity or vegetation-climate models.

Savannah ecosystems comprise some mixture of woodland and grassland, and variability in proportion and distribution of these two classes can severely limit the detail and accuracy of land cover mapping. Moreover, savannahs are highly dynamic systems, changing over time in response to natural and anthropogenic drivers, but also subject to marked seasonal variations with the appearance of vegetation classes varying significantly throughout the year.

Recent work to monitor woody encroachment in Kruger National Park (KNP), South Africa showed that medium spatial resolution Landsat imagery contained similar information content to very high resolution (VHR) WorldView-2, QuickBird and IKONOS imagery, though it was necessary to generalise thematic class definitions at the coarser scale of observation. However, uncertainty arose regarding the seasonal vegetation signal and how this influenced classification accuracy. In particular, separate wet and dry season land cover maps of a common study area, supported by common reference (field survey) data, yielded similar classification accuracies yet appeared markedly different in visual terms. A further difficulty for monitoring seasonal and other change in savannah environments is that of cloud cover which can reduce the temporal coverage of optical imagery substantially.

In this paper, benchmark work conducted using Landsat and VHR imagery to map savannah land cover at three sites in KNP is compared against a range of spectral and temporal combinations of Sentinel-1 and Sentinel-2 imagery acquired in 2015 and 2016. Specifically, single-date wet and dry, and multitemporal wet/dry, season Sentinel-2 images were tested. Previous research has suggested that dry season optical images often outperform wet season images in savannah environments. Separately, 26 multipolarization Sentinel-1 SAR images were acquired over a one-year period and classified as a single stacked time-series. Also, though, this time-series was broken down into 13 consecutive-date image pairs to enable analysis over an annual phenological cycle, indicating exactly when individual classes are most distinct. Finally, each of the three Sentinel-2 image combinations were integrated with the full Sentinel-1 time-series image.

Research results demonstrate several benefits of the Sentinels mission. Spectrally, the radar capabilities of Sentinel-1 increase the temporal coverage of imagery substantially compared to optical sensors, and this can provide valuable seasonal and phenological information. Spatially, the finer resolution provided by Sentinel-2 compared to Landsat is significant, enabling more detailed thematic classes to be mapped. Specifically, in previous work Landsat imagery could only accurately distinguish vegetation classes on the basis of woodland and grassland proportions, while VHR imagery enabled classification of separate tree, shrub, long grass and sparse grass classes. Here, Sentinel-2 imagery enabled accurate mapping of the separate vegetation classes suggesting its information content may be more closely aligned to VHR imagery than that of Landsat. Overall, the most accurate classification of savannah land cover is provided by combined spectral (optical and SAR) and temporal (wet/dry season Sentinel-2 and time-series Sentinel-1) imagery , demonstrating the value of the Sentinel mission’s multi-platform and multi-sensor configuration.

The importance of savannahs worldwide has been widely acknowledged and documented. They are important ecosystems that are found on almost half of the African continent and a fifth of the Earth’s surface. They consist of varying densities of grasses and woody vegetation and, due to their complexity, they have been centre stage for a number of land system science debates, such as equilibrium dynamics, land degradation and desertification, or their contribution to the global carbon cycle, to name but a few. Their significance is directly linked to the numerous recent efforts to map and monitor their land cover as accurately as possible, most commonly by employing Earth observation technologies.
A recent case study by Symeonakis et al. (2018), carried out in a southern African savannah covering an area of ~44,000km2, tested the performance of different combinations of Landsat- and PALSAR-based metrics from the dry and wet seasons. They concluded that the combination of multi-sensor and multi-season data provides the best results when mapping the main land cover types of woody vegetation, grasses, crops and urban/bare. Here, we take this work further by testing the performance of 15 similar models, this time based on a combination of Sentinel-1 and Sentinel-2 metrics from the dry and wet season, using the same study area, training and validation samples (extracted from 0.5 m RGB aerial photography). Our results corroborate the findings of the previous study: the combination of the Sentinel optical and C-band radar data from both seasons yielded the most accurate land cover Random Forests classifications: overall accuracy of 87%, overall k: 0.83. Very high user’s and producer’s accuracies were also found, especially for the woody class (User’s: 93%; Producer’s: 93%), which is of primary concern in these savannah environments, due to its relevance to the land degradation processes that are dominant in the region: bush encroachment, overexploitation for fuelwood and deforestation. Similar to the Landsat/PALSAR study, the models based on SAR data only, were less accurate than the optical models. As anticipated, the L-band PALSAR data used in the previous study performed better in mapping the woody cover than their C-band counterparts, since L-band radiation is able to penetrate through the canopy layer and reach the woody stems and branches more efficiently. However, and interestingly, the Landsat-PALSAR study generally outperformed the Sentinel1/2 study, e.g. reporting the same omission but a 5% lower commission error for the woody class (estimated from the all-parameter model).
Additionally, we tested a novel land cover configuration mapping accuracy approach using PlanetScope 3m-pixel data, with the view to assessing the quality of the mapping of landscape configuration (e.g. number of crop paddocks) and not only its composition (e.g. crop vs other covers). For five test areas of 170 km2 each, we performed object-based classifications of the PlanetScope imagery (two images per test area, one from the dry and one from the wet season). The results were less promising than those achieved for land cover composition: as an example, in one of our areas of focus, we determined 30 crop paddocks, 28 of which were of the circular centre pivot type ranging from 0.06 to 0.75 km2. For the same area, we mapped a significantly larger number of paddocks with the Sentinel 2 data, most of which were of the rectangular-shaped type. This discrepancy was attributed to the spectral similarity between the crop and grassland cover types, as well as the averaging effect of the lower spatial resolution of the Sentinel.
Our findings, however, suggesting a multi-sensor and multi-seasonal approach, are an important addition to the emerging literature and can be used to guide efforts for achieving a highly accurate savannah land cover characterisation. Future work will investigate the use of the PlanetScope object-based classifications as training of the coarser-resolution models (e.g. Sentinel 1- and 2-based metrics), not only for improving the mapping of land cover configuration but also for mapping fractional cover, e.g. % woody vegetation.

Tropical dry forests and savannas cover about 20% of the Earth’s surface, harbor high biodiversity and provide many important ecosystem services. However, these ecosystems have been under severe land-use pressure during the past 30 years, and monitoring woodland condition is therefore important. This poses major challenges given heterogeneous nature of these ecosystems, characterized by complex canopy structures, continuous gradients of tree and shrub cover, and patchy landscapes.

Here, we present a new approach to map, separately, continuous fields of tree and shrub cover across the 1.1 million km² South American Dry Chaco for the year 2015. We made use of the entire Sentinel-1 (SAR) and Landsat-8 archives, gathered a large manually digitized training data set, and used a gradient-boosting framework. Our regression results of 85.5% for tree cover and 68.5% for shrub cover showed that it is possible to reliably map continuous fields of tree and shrub cover across large areas, and that the combined use of SAR and optical data performed substantially better than models using single-sensor data only.
The datasets we generate allow for a much more nuanced characterization of vegetation and a deeper understanding of the factors that shape it. For example, combining tree and shrub layers in a spatial clustering algorithm allowed identifying distinct vegetation formations, including shrub-dominated systems, woodlands with large trees, and tree-dominated wet savannas. Similarly, our shrub and tree layers uncovered considerable edge effects away from croplands (extending up to 2km) and roads and railways (1-1.5km). In addition, our new maps revealed the substantial influence that smallholders have on forest cover in the Chaco through forest grazing and fuelwood extraction (up to about 1.2km from the center). Considering the large number of smallholder ranches in the Chaco (~30,000) this adds to the other existing pressures on the Chaco.
All of these patterns cannot be detected using discrete classifications. Combining SAR and optical data in this context was crucial, further underpinning the potentials that arise through the consolidation of the archives and combining them within analyses to monitor the condition of the world’s remaining dry forests and savannas.