Mapping and monitoring of habitats are requirements for protecting biodiversity. In this study, we investigated the benefit of combining airborne (laser scanning, image-based point clouds) and satellite-based (Sentinel 1 and 2) data for habitat classification. We used a two level random forest 10-fold leave-location-out cross-validation workflow to model Natura 2000 forest and grassland habitat types on a 10 m pixel scale at two study sites in Vienna, Austria. We showed that models using combined airborne and satellite-based remote sensing data perform significantly better for forests than airborne or satellite-based data alone. For frequently occurring classes, we reached class accuracies with F1-scores from 0.60 to 0.87. We identified clear difficulties of correctly assigning rare classes with model-based classification. Finally, we demonstrated the potential of the workflow to identify errors in reference data and point to the opportunities for integration in habitat mapping and monitoring.

The frequency and severity of wildfires in the Alpine region will likely increase due to climate change. Most fire danger forecasts currently adopted in this region are based on meteorological data, such as the Canadian Fire Weather Index (FWI). They are typically only available at relatively coarse spatial resolutions (up to ca. 1 km) and, therefore, are of limited use in mountain regions with complex topography. Other factors, such as vegetation type and structure and the role of humans causing ignitions, are typically not considered. We address this gap by presenting a novel, high-resolution, satellite-supported integrated forest fire danger system (IFDS) for Austria. For this purpose, we use radar and optical satellite data from the Copernicus Sentinel-1 and Sentinel-2 missions, airborne laser scanning (ALS), socio-economic data, and topographic properties next to meteorological data. Two independent methods were investigated: (i) an expert-based approach that allows combining various data layers with different weightings assigned by experts and (ii) a machine-learning approach. Here, we focus on the results of the machine learning approach for a study area covering the federal state of Styria in Austria (ca. 16 400 km²). We use several data layers computed within our study as predictors in random forest models. Moisture indicators and tree species maps were derived from satellite data from the Copernicus Earth observation programme. Vegetation structure parameters, solar potential and a digital surface model (DSM) were derived from ALS data. In addition to the remote sensing data, we used meteorological variables, fire weather indices (FWI) and socio-economic data. We trained the model using forest fire events from the Austrian fire database. The cross-validation showed that the best-performing model predicts high fire danger for most fire events (87%). By integrating all the information layers compared to a baseline model using only FWI, the overall accuracy improved from 68% to 87%. The feature importance showed that the vegetation structure parameters, tree species, socio-economic parameters and DSM are essential for the model in addition to the meteorological predictors. Using this data-driven approach allowed us to learn from past fire occurrences and improved the spatial representation of fire ignition drivers, their importance and interactions. Also, this method permitted the identification of areas with higher danger risk, typically located in the vicinity of densely populated settlements. This study has been performed within the CONFIRM project with funding from the Austrian Research Promotion Agency (FFG).

Terrain-induced variations of radar backscatter represent an important limiting factor of many Synthetic Aperture Radar (SAR)-based applications. Radiometric terrain flattening (RTF) is a well-established method that minimizes these variations in SAR imagery. To fully understand the implications of SAR RTF, validation of its impact on the derived products is needed. In this study, we quantified the influence of the RTF on a forest mapping and classification algorithm over Austria, and compared the classification results for the conventional sigma naught and radiometrically terrain-corrected gamma backscatter. The overall accuracy for forest/non-forest mapping and forest type classification improved by 2% and 4%, respectively, over the whole of Austria, with improvements of up to 16% and 20%, respectively, in regions with strong topography.

The existing forest inventories have difficulties in providing data with sufficient spatial and temporal resolution to quantify dynamic properties of forests, which are mainly caused by short-term events such as drought, storm, snow damages, or pest infestations. This study aims to explore in a first step the potential of space-borne LiDAR for forest parameters extraction over Alpine forests in Austria. The space-borne LiDAR data investigated in this study is ICESat2 (Ice, Cloud and Land Elevation Satellite-2) and GEDI (Global Ecosystem Dynamics Investigation). GEDI is a full-waveform, multibeam laser altimeter on the International Space Station, with a footprint diameter of about 25m (Dubayah, 2020). As for ICESat-2, it carries a micropulse, multi-beam photon-counting laser altimeter, with a footprint diameter of about 17m (Neuenschwander and Magruder, 2019). Two footprint-level products are used in this study: GEDI L2A and ICESat-2 ATL03. The GEDI L2A product provides footprint-level elevation and height metrics that extract terrain height, canopy height, and relative height metrics from the received waveform. ICESat-2 ATL03 records horizontal coordinates and ellipsoidal heights of all photon data, and the classification labels are extracted from its higher product, ATL08. The DTM with a resolution of 1m and the DSM derived from ALS (Airborne Laser Scanning) point clouds acts as “ground truth” to assess the accuracy of the terrain and canopy height of the two space-borne LiDAR products, respectively. For ICESat-2 ATL03, only photons with a signal confidence flag ranging from medium confidence (sigal_confidence=3) or high confidence (sigal_confidence=4) are included for evaluation. For GEDI L2A, only waveforms with a valid quality flag (quality_flag=1) are included for evaluation. To evaluate the performance of ICESat2 and GEDI for different forest types and topographic conditions, two study sites in Austria are selected: western part of Tyrol and the Vienna Woods. A preliminary results of terrain and canopy height accuracies shows that the terrain height of the two space-borne LiDAR products fits well with the DTM. Compared to GEDI L2A, ICESat2 ATL03 has a better correlation with DTM values. The canopy height accuracy is not as good as the terrain height accuracy. It has been shown that ICESat-2 tend to underestimate the canopy top height as derived from airborne LiDAR. Overall, GEDI has a better canopy height accuracy than ICESat-2. Furthermore, we have investigated the influence of different beam power, data acquisition time and season. In general, the accuracy of both ICESat-2 and GEDI data acquired in nighttime is higher than that of the daytime data. The statistic also shows that for the ICESat-2 data, the terrain height accuracy of weak beam footprints only slightly worse than that of strong beam footprints. For GEDI, footprints of strong beams always perform better than that of weak beams in terms of terrain and canopy height. Regarding the impact of season, for both ICESat-2 and GEDI, the canopy height is more accurate in summer than the collection in winter. For GEDI, the terrain height can be better measured in winter than in summer.

Microwave and optical imaging methods react differently to different land surface parameters and, thus, provide highly complementary information. However, the contribution of individual features from these two domains of the electromagnetic spectrum for tree species classification is still unclear. For large-scale forest assessments, it is moreover important to better understand the domain-specific limitations of the two sensor families, such as the impact of cloudiness and low signal-to-noise-ratio, respectively. In this study, seven deciduous and five coniferous tree species of the Austrian Biosphere Reserve Wienerwald (105,000 ha) were classified using Breiman’s random forest classifier, labeled with help of forest enterprise data. In nine test cases, variations of Sentinel-1 and Sentinel-2 imagery were passed to the classifier to evaluate their respective contributions. By solely using a high number of Sentinel-2 scenes well spread over the growing season, an overall accuracy of 83.2% was achieved. With ample Sentinel-2 scenes available, the additional use of Sentinel-1 data improved the results by 0.5 percentage points. This changed when only a single Sentinel-2 scene was supposedly available. In this case, the full set of Sentinel-1-derived features increased the overall accuracy on average by 4.7 percentage points. The same level of accuracy could be obtained using three Sentinel-2 scenes spread over the vegetation period. On the other hand, the sole use of Sentinel-1 including phenological indicators and additional features derived from the time series did not yield satisfactory overall classification accuracies (55.7%), as only coniferous species were well separated.

Green areas play an important role within urban agglomerations due to their impact on local climate and their recreation function. For detailed monitoring, frameworks like the flora fauna habitat (FFH) classification scheme of the European Union’s Habitat Directive are broadly used. By date, FFH classifications are mostly expert-based. Within this study, a data-driven approach for FFH classification is tested. For two test areas in the municipality of Vienna, ALS point cloud data are used to derive predictor variables like terrain features, vegetation structure and potential insulation as well as reflection properties from full waveform analysis on a 1 m grid. In addition, Sentinel-1 C-Band time series data are used to increase the temporal resolution of the predicting features and to add phenological characteristics. For two 1.3 × 1.3 km test tiles, random forest classifiers are trained using different combinations (ALS, SAR, ALS+SAR) of input features. For all model test runs, the combination of ALS and SAR input features lead to best prediction accuracies when applied on test data.

With the increasing occurrence of forest fires in the mid-latitudes and the alpine region, fire risk assessments become important in these regions. Fuel assessments involve the collection of information on forest structure as, e.g., the stand height or the stand density. The potential of airborne laser scanning (ALS) to provide accurate forest structure information has been demonstrated in several studies. Yet, flight acquisitions at the state level are carried out in intervals of typically five to ten years in Central Europe, which often makes the information outdated. The Sentinel-1 (S-1) synthetic aperture radar mission provides freely accessible earth observation (EO) data with short revisit times of 6 days. Forest structure information derived from this data source could, therefore, be used to update the respective ALS descriptors. In our study, we investigated the potential of S-1 time series to derive stand height and fractional cover, which is a measure of the stand density, over a temperate deciduous forest in Austria. A random forest (RF) model was used for this task, which was trained using ALS-derived forest structure parameters from 2018. The comparison of the estimated mean stand height from S-1 time series with the ALS derived stand height shows a root mean square error (RMSE) of 4.76 m and a bias of 0.09 m on a 100 m cell size, while fractional cover can be retrieved with an RMSE of 0.08 and a bias of 0.0. However, the predictions reveal a tendency to underestimate stand height and fractional cover for high-growing stands and dense areas, respectively. The stratified selection of the training set, which we investigated in order to achieve a more homogeneous distribution of the metrics for training, mitigates the underestimation tendency to some degree, yet, cannot fully eliminate it. We subsequently applied the trained model to S-1 time series of 2017 and 2019, respectively. The computed difference between the predictions suggests that large decreases in the forest height structure in this two-year interval become apparent from our RF-model, while inter-annual forest growth cannot be measured. The spatial patterns of the predicted forest height, however, are similar for both years (Pearson’s R = 0.89). Therefore, we consider that S-1 time series in combination with machine learning techniques can be applied for the derivation of forest structure information in an operational way.

The constellation of two Sentinel-1 satellites provides an unprecedented coverage of Synthetic Aperture Radar (SAR) data at high spatial (20 m) and temporal (2 to 6 days over Europe) resolution. The availability of dense time series enables the analysis of the SAR temporal signatures and exploitation of these signatures for classification purposes. Frequent backscatter observations allow derivation of temporally filtered time series that reinforce the effect of changes in vegetation phenology by limiting the influence of short-term changes related to environmental conditions. Recent studies have already shown the potential of multitemporal Sentinel-1 data for forest mapping, forest type classification (coniferous or broadleaved forest) as well as for derivation of phenological variables at local to national scales. In the present study, we tested the viability of a recently published multi-temporal SAR classification method for continental scale forest mapping by applying it over Europe and evaluating the derived forest type and tree cover density maps against the European-wide Copernicus High Resolution Layers (HRL) forest datasets and national-scale forest maps from twelve countries. The comparison with the Copernicus HRL datasets revealed high correspondence over the majority of the European continent with overall accuracies of 86.1% and 73.2% for the forest/non-forest and forest type maps, respectively, and a Pearson correlation coefficient of 0.83 for tree cover density map. Moreover, the evaluation of both datasets against the national forest maps showed that the obtained accuracies of Sentinel-1 forest maps are almost within range of the HRL datasets. The Sentinel-1 forest/non-forest and forest type maps obtained average overall accuracies of 88.2% and 82.7%, respectively, as compared to 90.0% and 87.2% obtained by the Copernicus HRL datasets. This result is especially promising due to the facts that these maps can be produced with a high degree of automation and that only a single year of Sentinel-1 data is required as opposed to the Copernicus HRL forest datasets that are updated every three years.

The assessment of forest fire risk has recently gained interest in countries of Central Europe and the alpine region since the occurrence of forest fires is expected to increase with a changing climate. Information on forest fuel structure, which is related to forest structure, is a key component in such assessments. Forest structure information can be derived from airborne laser scanning (ALS) data, whose value for the derivation of respective metrics at a high accuracy level has been demonstrated in numerous studies over the last years. Yet, the temporal resolution of ALS data is low as flight missions are typically carried out in time intervals of five to ten years in Central Europe. ALS-derived forest structure descriptors for fire risk assessments, therefore, are often outdated. Open access earth observation data offer the potential to fill these information gaps. Data provided by synthetic aperture radar (SAR) sensors, in particular, are of interest in this context since this technology has a known sensitivity to the vegetation structure and acquires data independent of weather or daylight conditions. In our study, we investigate the potential to derive forest structure descriptors from time series of Sentinel-1 (S-1) SAR data for a deciduous forest site in the Eastern part of Austria. We focus on forest stand height and fractional cover, which is a measure for forest density, as both of these components impact forest fire propagation and ignition. The two structure metrics are estimated using a random forest (RF) model, which takes a total of 36 predictors as input, which we compute from the S-1 time series. The model is trained using ALS-derived structure metrics acquired during the same year as the S-1 data. We estimated stand height with a root mean square error (RMSE) of 4.76 m and a bias of 0.09 m at 100 m resolution, while the RMSE for the fractional cover estimation is 0.08 with a bias of zero at the same resolution. The spatial comparison of the structure predictions with the ALS reference further shows that the general structure is well reproduced. Yet, fine scale variations cannot be completely reproduced by the S1-derived structure products, and the height of tall stands and very dense canopy parts are underestimated. Due to the high correlation of the predicted values to the reference (Pearson’s R of 0.88 and 0.94 for the stand height and the fractional cover, respectively), we consider S-1 time series in combination with ALS data with low temporal resolution and machine learning techniques to be a reliable data source and workflow for regularly (e.g. < yearly) updating ALS structure information in an operational way.

The Sentinel-1 satellites provide the formerly unprecedented combination of high spatial and temporal resolution of dual polarization synthetic aperture radar data. The availability of dense time series enables the derivation and analysis of temporally filtered annual backscatter signals. The study concentrates on the use of Sentinel-1 seasonal backscatter signatures for forest area estimation and forest type classification. A classification method based on time series similarity measures is introduced and tested in three test areas covered by various forest types including broadleaf temperate, boreal and montane forests. The results are compared with two European-wide Copernicus high resolution layers, namely forest type and tree cover density (TCD). The correspondence of forest/non-forest maps and TCD is high in all test areas, with overall accuracies for forest/non-forest classification between 86% and 91% and Pearson correlation coefficients for TCD between 0.68 and 0.74. The forest type classification (non-forest, coniferous and broadleaf forest classes) provides best results in temperate forests with an overall accuracy of 85%; in boreal forest, the accuracy decreases to only 65%. Generally, the method provides reliable results for forest area estimation, including regions where methods based on static parameters are often problematic (mountainous areas), and it enables forest type classification in temperate forests.

In this paper, we show first experimental results of big data processing for geophysical parameters retrieval from Synthetic Aperture Radar (SAR) data, carried out at the Earth Observation Data Center (EODC) for Water Resources Monitoring. The EODC is utilizing the high-performance computing platform provided by the third generation of the Vienna Scientific Cluster (VSC-3). Different Level-1 SAR datasets including data acquisitions from Sentinel-1 in four case studies were processed at the VSC-3 using the new SAR Geophysical Retrieval Toolbox (SGRT) developed at the Vienna University of Technology. SGRT produces a series of different geophysical parameters through time series analysis of satellite SAR data. This study presents the recent improvements and new features of the SGRT together with the findings obtained from the case studies performed at the EODC high-performance data processing environment.

The recently launched Sentinel-1A provides the high resolution Synthetic Aperture Radar (SAR) data with very high temporal coverage over large parts of European continent. Short revisit time and dual polarization availability supports its usability for forestry applications. The following study presents an analysis of the potential of the multi-temporal dual-polarization Sentinel-1A data for the forest area derivation using the standard methods based on Otsu thresholding and K-means clustering. Sentinel-1 data collected in winter season 2014-2015 over a test area in eastern Austria were used to derive forest area mask with spatial resolution of 10m and minimum mapping unit of 500 m2. The validation with reference forest mask derived from airborne full-waveform laser scanning data revealed overall accuracy of 92 % and kappa statistics of 0.81. Even better results can be achieved when using external mask for urban areas, which might be misclassified as forests when using the introduced approach based on SAR data only. The Sentinel-1 data and the described methods are well suited for forest change detection between consecutive years.

One of the major advantages of the Sentinel-1 data is its capability to provide very high spatio-temporal coverage allowing the mapping of large areas as well as creation of dense time-series of the Sentinel-1 acquisitions. The SGRT software developed at TU Wien aims at automated processing of Sentinel-1 data for global and regional products. The first step of the processing consists of the Sentinel-1 data geocoding with the help of S1TBX software and their resampling to a common grid. These resampled images serve as an input for the product derivation. Thus, it is very important to select the most reliable processing settings and assess the geocoding uncertainty for both backscatter and projected local incidence angle images. Within this study, selection of Sentinel-1 acquisitions over 3 test areas in Europe were processed manually in the S1TBX software, testing multiple software versions, processing settings and digital elevation models (DEM) and the accuracy of the resulting geocoded images were assessed. Secondly, all available Sentinel-1 data over the areas were processed using selected settings and detailed quality check was performed. Overall, strong influence of the used DEM on the geocoding quality was confirmed with differences up to 80 meters in areas with higher terrain variations. In flat areas, the geocoding accuracy of backscatter images was overall good, with observed shifts between 0 and 30m. Larger systematic shifts were identified in case of projected local incidence angle images. These results encourage the automated processing of large volumes of Sentinel-1 data.

The performance of the advanced synthetic aperture radar (ASAR) global mode (GM) surface soil moisture (SSM) data was studied over Australia by means of two widely used bivariate measures, the root-mean-square error (RMSE) and the Pearson correlation coefficient (R). By computing RMSE and at multiple spatial scales and for different data combinations, we assessed how, and at which scales, the spatial sampling error, noise, and the choice of the reference data impact on RMSE and . The results reveal large changes in RMSE and with continental average values of 8% and 18% for the RMSE of relative soil moisture saturation and between 0.4 and 0.7 for depending on the spatial scale of aggregation and the choice of reference data. The combined effect of noise and spatial sampling error accounted for a 79% RMSE increase at 1 km and predominated over the error due to the choise of the reference data also at 5 km scale. The effect of noise on RMSE strongly diminished at spatial scales ≥2 km. By contrast, the impact of uncertainties in the reference data was larger on than on RMSE. This highlights the better potential of to estimate the benefit of observations prior to data assimilation. Based on our results, it is further suggested that a potential way for an improved ASAR GM SSM error assessment is to: 1) aggregate the data to ≥2 km resolution to minimize the noise; 2) subtract the spatial sampling error within the coarse resolution footprint; and 3) remove the reference uncertainty using advanced techniques such as triple collocation.

A surface soil moisture (SSM) product at a 1-km spatial resolution derived from the Envisat Advanced Synthetic Aperture Radar (ASAR) Global Monitoring (GM) mode data was evaluated over the entire African continent using coarse spatial resolution SSM acquisitions from the Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E) and the Noah land surface model from the Global Land Data Assimilation System (GLDAS-NOAH). The evaluation was performed in terms of relative soil moisture values (%), as well as anomalies from the seasonal cycle. Considering the high radiometric noise of the ASAR GM data, the SSM product exhibits a good ability (Pearson correlation coefficient (R) = ~0.6 for relative soil moisture values and root mean square difference (RMSD) = 11% when averaged to 5-km resolution) to monitor temporal soil moisture variability in regions with low to medium density vegetation and yearly rainfall >250 mm. The findings agree with previous evaluation studies performed over Australia and further strengthen the understanding of the quality of the ASAR GM SSM product and its potential for data assimilation. Problems identified in the ASAR GM algorithm over arid regions were explained by azimuthal effects. Diverse backscatter behavior over different soil types was identified. The insights gained about the quality of the data were used to establish a reliable masking of the existing ASAR GM SSM product and the identification of areas where further research is needed for the future Sentinel-1-derived SSM products.