During the BorealScat radar tower experiment (2016 to 2021) in southern Sweden, a single forest stand was observed using a multi-polarization tomographic radar for several years. A variety of measurement sequences at multiple frequency bands were used to study the effects of soil moisture, weather and seasonal changes on radar signals at time scales ranging from less than a second to several years. The experiment was partly an ESA campaign for studying weather-induced radar measurement disturbances in support of present and future synthetic aperture (SAR) radar missions with the goal of estimating above-ground biomass, tree height and forest structure. In addition to characterising the temporal variation of boreal forest radar reflectivity, effects of tree water dynamics, drought-induced stress and insect damage were observed during the experiment.
The experiment consisted of a 50-m high tower with 30 antennas mounted at the top. The tower overlooked a mature stand of Norway spruce. The radar system allowed tomographic imaging whereby images could be produced representing the vertical cross section of the forest. This allowed backscatter contributions from different parts of the forest (e.g., ground and canopy) to be separated. Radar measurements were conducted at P-band (436 MHz), L-band (1270 MHz) and C-band (5400 MHz), offering different degrees of forest penetration. These frequency bands are supported by the main current and future SAR missions for forest radar observations such as ESA’s BIOMASS, ROSE-L and Sentinel-1 missions, NASA/ISRO’s NISAR mission and JAXA’s ALOS-2/4 missions. BorealScat observations were also used for studying short-term decorrelation in preparation for the ESA’s candidate Earth Explorer mission Hydroterra. Tomographic images were acquired at 5-minute intervals while Doppler observations were carried out over 10-s periods to study sub-second variations.
A weather station was installed for correlating radar time series measurements to weather variables. Freeze-thaw cycles caused the largest fluctuations in radar time series observations as moisture in the forest turned to ice in sub-zero temperatures and thawed again in warmer conditions. Soil moisture variations caused large fluctuations in ground-level backscatter, while canopy backscatter was relatively stable during non-frozen conditions, supporting P-band forest parameter estimation methods based on tomography and interferometric ground notching. During the hot summer of 2018, diurnal cycles were observed in radar time series, which were suspected to be connected to diurnal tree water content fluctuations. Water stored in tree tissues contribute to transpired water during hot days to sustain high transpiration rates. This effect was accentuated by strong winds that accelerate the rate of transpiration under certain conditions, showing clear drops in backscatter during the day. In sustained hot and dry conditions, this diurnal cycle would decrease in magnitude, indicating water stress. Digital point dendrometers were installed on tree stems to observe fluctuations in tree water content. Sap flow sensors, measuring the rate of water transport in stems, were also installed to complement the dendrometers. Forest damage due to bark beetles were detected in 2019, causing progressive deaths of trees in the forest stand. These effects could be observed in radar backscatter time series as a gradual drop in reflectivity. Early onset of damage was most clearly observed in L-band canopy attenuation observations, acquired by placing a large reflector within the forest. A similar effect was observed at C-band as trees died from bark beetle attacks.
BorealScat observations were also used to assess temporal coherence without the effects of spatial baselines and volume decorrelation that are always present in airborne and spaceborne SAR data. Temporal coherence cycles at diurnal timescales were observed during warm periods, offering the possibility of repeat-pass L- and C-band interferometry. Correlation tomography, a tomographic imaging method for overcoming temporal decorrelation, was investigated at L- and C-band. Vertical backscatter distributions from correlation tomography were not significantly affected by dynamic weather conditions. These results strongly motivate the implementation of a single-pass bistatic interferometer mission at L- or C-band for forest applications.
The BorealScat experiment provided forest radar observations that reveal a close connection between vegetation water transport mechanisms and tree vitality. The dataset has been made publicly available. The long overpass times of satellites do currently not facilitate the observation of some of these effects from space. To continue this study of the relationship between forest radar observations, tree water dynamics, forest-atmosphere water exchange and tree vitality, a new long-term experiment has been initiated in northern Sweden: BorealScat-2. A better understanding of how tree evapotranspiration, tree water content, stress and tree vitality can be sensed using radar will pave the way for new data exploitation opportunities and new earth observation missions.
Fast, repeatable, synoptic, and cost-effective monitoring of forest ecosystems over large areas can be achieved by means of remote sensing. In particular, data acquisitions with hyperspectral remote sensing systems can capture detailed vegetation information from hundreds of contiguous narrow spectral bands, which improves the observation of ecosystem functions and services.
The DLR Earth Sensing Imaging Spectrometer (DESIS) is a hyperspectral instrument integrated in the Multi-User-System for Earth Sensing (MUSES) platform installed on the International Space Station (ISS) acquiring data since 2019. DESIS measures objects in the spectral range from 400 and 1000 nm with a spectral sampling distance of 2.55 nm. With its ability to provide accurate spectral measurement in a moderately fine spatial resolution of 30 m, DESIS offers the opportunity for spatio-temporal quantification of vegetation parameters at a larger scale. To test and evaluate these potentials, we analysed several DESIS images of the Bavarian Forest National Park (BFNP).
BFNP is located in the south-eastern part of Germany, close to the border of the Czech Republic, and is Germany’s first designated national park (founded in 1970) covering an area of around 24,250 ha. The forest area in BFNP has been affected by the proliferation of the spruce bark beetles since the mid-1990s. Until today, the bark beetle infestation had resulted in a massive dieback of more than 7500 ha of spruce stands. In recent years, forest managers have made efforts to develop regenerated forests in the deadwood areas. DESIS data is applied to emphasize annual changes of forest condition in a more detailed manner from a spectral perspective.
DESIS data has been continuously acquired over BFNP since June 2019. In order to reduce noise and account for minor spectral de-calibration issues, a 4 times software binning has been applied to the DESIS L2A data resulting in a spectral resolution of ~10 nm. After detailed data quality checking and subsequent pre-selection of DESIS data, we used different narrow band vegetation indices (VIs) in order to estimate bio-physical variables. Temporal changes of these indices were examined more closely regarding spatial patterns related to stressed regions (i.e. bark beetle infested areas in particular). Narrow band indices used in this study are sensitive to different biophysical and biochemical variables such as canopy structure, chlorophyll content or leaf pigments. Anomalies in the narrow band indices that incorporate the red edge region (e.g. MCARI, the Modified Chlorophyll Absorption in Reflectance Index) strongly correspond to the affected areas identified in the field.
This presentation will give an overview of the potentials and limitations of using DESIS data for forest health monitoring purposes. It summarizes which spatial patterns of vegetation change can be observed in the BFNP when focusing on spectral information at a 30 m resolution.
This work was conducted in the frame of the Data Pool Initiative for the Bohemian Forest Ecosystem.
A recent increase in deforestation rates and fire activity has sparked widespread concerns about Amazon forest conservation. However, the inability to rapidly separate satellite fire detections by fire type hampers fire suppression and assessments of ecosystem and air quality impacts. Here we report on the development of a near-real time approach that integrates data of the Visible Infrared Imaging Radiometer Suite (VIIRS) and Sea and Land Surface Temperature Radiometer (SLSTR) for tracking individual wildfires and their behavior across the southern Amazon. Our approach combines fire characteristics with land cover information to identify the contributions from deforestation, forest, agricultural, and savanna fires to burned area and emissions. Ongoing efforts to integrate data of Sentinel-2 and Sentinel-5p will help to further constrain burned area and emissions, respectively.
Initial results for the fire season of 2019, reveal that 19,700 deforestation fire events accounted for 39% of all satellite active fire detections and the majority of fire carbon emissions (63%; 69 Tg C) across the southern Amazon. However, direct emissions from forests fires (86 Tg) almost equaled those from deforestation fires (99 Tg) in southern hemisphere South America, including extensive forest fires in the tropical dry forest of Bolivia. The multi-sensor approach resulted in an unprecedented capability to detect low-intensity forest fires, even under dense canopy of tropical forests that remain largely undetected by most existing burned area products. During 2020 our data highlighted how a small number of exceptionally large fires consumed most woodlands of the Pantanal, the world’s largest tropical wetland. Combined, our results highlight the increasing importance of forest fires for regional forest degradation and carbon emissions.
Multi-day fires accounted for about 80% of burned area and 90% of carbon emissions from the Amazon, with many forest fires burning uncontrolled for weeks. Most fire detections from deforestation fires were correctly identified within two days (65%), highlighting the potential to improve situational awareness and improve management outcomes during future fire emergencies.
Spectral indices including for example the differenced Normalized Burn Ratio (dNBR) are often applied to characterize the spatial variability of wildfire severity after a fire has occurred. Mapping wildfire severity is important to quantify ecological and economic damages and plan mitigation measures. Despite the frequent and even operational use of dNBR after fire events, it is not fully understood which exact changes occurring on the land surface and induced during the course of a fire drive the dNBR signal and hence the spatial variability of dNBR observed by multispectral satellites.
One reason for this lack of understanding is that only in rare cases, detailed and spatially highly-resolved pre-fire information about vegetation structure and composition is available. In this study, we applied very high resolution unmanned aerial system (UAS) orthoimages collected briefly before and after the massive wildfires burning in Central Chile in the fire season 2016/2017. More specifically, we derived a set of variables describing the pre- and postfire landscape composition and structure and also included information on the distribution of cast- and terrain shadows. The latter were discussed as potentially influential factor for the dNBR signal but have rarely been examined systematically.
Using Generalized Additive Models (GAM) we were able to explain more than 80% and 75% of the variability in dNBR and RdNBR values as observed by Sentinel-2 images using our set of predictors. Expectedly, a large fraction of the dNBR variability was explained by the fraction of green canopy cover consumed by the fire. However, we also found that cast-shadows shed by snags and dead trees with an intact crown structure also have a notable influence on the observed dNBR values. Contrarily, pre-fire vegetation composition played only a minor role.
Our study shows that spatially highly resolved information collected by UAS can be a valuable source of information to improve our understanding of the meaning of common spectral index products which are often used operationally for environmental monitoring while their exact meaning may in many cases not be fully understood.
Regarding the massive drought-induced forest damages in Central Europe, the research project FirSt 2.0 (Forest damage inventory based on rapid Satellite technologies) is investigating forest vitality with multiple sensors and across scales to derive a continuous forest vitality and forest damage information.
In a first stage of the investigation, four areas in Germany were repeatedly covered within the vegetation period from March to September of 2021 with UAS imagery of a hyperspectral recording system (Headwall Nano - 400 to 1000 nm). At the same time, intensive field surveys of damage evaluation (leaf loss) and LAI measurements took place. The areas are mainly covered by either spruce or beech stands, which are the main tree types in Central Europe and both are heavily affected by the drought period in 2018/19/20.
Simultaneously, for the same regions Sentinel-1 and Sentinel-2 images were acquired for the years 2016 to 2021 and compared to the damage evaluations from bark beetle infestations (spruce) and drought related degradation (beech). The images were pre-processed with the FORCE algorithm within the CODE-DE platform and analyzed with a machine learning regression method.
The preliminary results show that remote sensing derived long-term as well as short-term damages of the two tree types in all areas clearly correspond with the field-based observations. While the higher spatial and spectral resolution of the UAS images is more suited for early detection and local management of damages, the satellite imagery is able to detect interrelations of different degradations and subsequently damage chains (e.g. from drought-related degradation to insect infestation to storm event loss). . In combining both data-types, the Sentinel data with additional external geo-data such as German drought monitor as well as terrain and soil data can provide a reliable estimation of the risk of drought stress, while the hyperspectral data can be seen as a magnifier for specific spatial and temporal development stages of forest degradation. The potential combination of these advantages with hyperspectral satellite data, such as PRISMA and EnMap will be the focus of future investigations.
Forest monitoring is a prime objective for Earth observation (EO) satellites. In the tropics, it is widely accepted that, among all the various available remote sensing technologies, low-frequency SAR is the most powerful tool for a consistent and coherent all-year round forest monitoring. While optical sensors are already incapacitated by the slightest cloud cover, high-frequency SARs (X- and C-band) are also regularly hampered by the massive cloud formations and intense rainfalls within the intertropical convergence zone. Moreover, these systems lack the penetration capability into forest canopies rendering impossible to obtain any direct information from stems, understory and ground. Contrariwise, L-band radar can well provide information about the entire vertical forest strata from the top canopy down to the forest ground surface under almost all weather conditions. However, in spite of tremendous accomplishments in the last 3 decades, the reliability of major applications like forest cover mapping, forest classification, above-ground biomass (ABG) estimation, and deforestation detection is still greatly affected by ill-defined spatial-temporal backscatter dynamics. The key to overcome these knowledge gaps and allow breakthrough improvements are sophisticated long-term measurements covering all possible environmental conditions in all tropical regions.
At the time of launch in 2014, ALOS-2 was the first EO satellite in orbit to feature a coherent on-receive dual-polarization ScanSAR mode opening all new opportunities for seamless global scale forest monitoring. Since then ALOS-2/PALSAR-2 has acquired a revolutionary forest data archive which provides unprecedented long-term time-series for the entire equatorial forest belt. Thanks to JAXA’s basic observation scenario (BOS) ALOS-2 has systematically monitored the entire tropical belt with a minimum repeat cycle of 42 days over the last 6 years. In many areas the temporal resolution is even better resulting in a maximum time-series length of more than 75 cycles in areas of the Amazon basin. Moreover, due to the 350-km wide swath coverage with large overlap areas for neighboring paths, approximately 90% of the target area is covered by 2 paths resulting in dual-polarization/dual-incidence angle time-series. Hence, despite the coarse resolution of 50 m and only partial polarimetry, the ScanSAR long-term pantropical forest monitoring archive is arguably one of the most valuable scientific contributions in forest remote sensing to date.
In this study, we highlight some of the groundbreaking achievements of this long-term tropical forest monitoring mission which for the first-time allows to overcome all uncertainties related to seasonal variations, weather, and flooding effects. We discuss the state-of-the-art in operational deforestation detection with JJ-FAST focusing on the latest time-series based false-alarm suppression techniques. We demonstrate the unique potential of PALSAR-2 time-series to create all-new highly-detailed flooded forest maps, and we introduce novel time-series based forest/non-forest maps with unmatched reliability.
Our findings show that ALOS-2/PALSAR-2 is the pioneer for all coming low-frequency forest missions. With the opening of the PALSAR-2 ScanSAR archive in early 2022, JAXA will make the data freely available allowing every researcher to fully exploit the formidable massive big data time-series. This will undoubtedly lead to tremendous improvements in our understanding of tropical forests and their monitoring. All future forest missions including the upcoming ALOS-4, NISAR, and BIOMASS, as well as the planned ALOS-6 and ROSE-L, will build upon the game changing achievements of ALOS-2.