Demonstrating the Generation and Application of Analysis Ready Data of Optical CubeSat Images at a Rewetted Peatland Site
Dr. Zhan Li | Helmholtz Center for Environmental Research - UFZ | Germany
Peatlands are areas with naturally accumulated thick layers of dead organic materials. While peatlands cover about 3% of the world’s land area, their carbon storage is estimated equivalent to ~30% of all soil carbon, ~75% of all atmospheric carbon, and as much carbon as all terrestrial biomass. Drained peatlands due to past human uses can emit carbon and be a key source of greenhouse gases while rewetted peatlands usually have significantly reduced CO2 emissions and can even become a carbon sink. However, quantifying the potential and limitations of reducing emissions by peatland rewetting is challenging. Carbon fluxes on peatlands are both spatially complex and temporally dynamic owing to their microtopography, changing water levels and associated vegetation status. Here we demonstrated the generation of temporally consistent ~biweekly 5-m images over 8 years (2013-2020) at visible and near infrared bands (VNIR) to track the temporal trajectories of vegetation and surface water and estimate cover-specific carbon fluxes at a rewetted peatland site in northeastern Germany (Figure 1).
To ensure temporally-consistent multispectral images for the subsequent analyses of vegetation/water covers, we set up a two-stage normalization procedure that normalized the images from RapidEye (SmallSats) and PlanetScope (CubeSats) to rigorously calibrated multispectral sensors onboard large satellites (Landsat-7/8 and Sentinel-2). The two-stage normalization procedure produced two levels of image normalization that allows for downstream applications to balance between the quality and the quantity of available normalized CubeSat images in a time series. A quantitative evaluation approach using daily MODIS images as bridging benchmark data revealed that the temporal consistency in CubeSat images was comparable to that in Landsat and Sentinel-2 images, which confirmed the efficacy of the normalization procedure.
The temporal information in the stack of normalized 5-m images helped us estimate the vegetation types and the changes in vegetation/surface water covers throughout the 8 years. The within-year time series of CubeSat images at the three visible and one near-infrared bands showed discernible differences among vegetation types at this peatland sites, which promises systemic mapping of vegetation compositions in peatlands using very-high-resolution CubeSat imagery time series over heterogeneous peatlands. We aggregated vegetation and surface water covers within each year into three condition categories at the peatland site, always emergent vegetation, always surface water, and alternating between vegetation and water. The estimated areas of the three condition categories closely covary with the measured water table depths at the site (Figure 2). The substantial areal expansion of always emergent vegetation at the site, that are captured by the CubeSat imagery time series, aligns well with the timing of three drought events (2016, 2018 and 2019) in this region.
These surface covers and conditions at both high temporal and spatial resolutions from CubeSat images allow us to disaggregate ecosystem-scale measurements of CO2 and CH4 fluxes by the eddy covariance (EC) tower at the site into cover-specific fluxes. We attribute CO¬2 and CH4 fluxes measured by EC over 8 years to the three surface condition categories through a nonparametric approach to flux decomposition using annual maps of surface condition categories and half-hourly EC-measurement footprints. The disaggregated carbon fluxes improve our upscaled estimates of carbon emission/sequestration over rewetted peatland sites. Such spatial-temporally-resolved carbon fluxes in dynamic and heterogeneous peatlands will contribute to better informed restoration and protection of peatlands.
Use of Sentinel-1 for peatland hydrological condition monitoring in near-natural, damaged, and restored northern peatlands.
Linda Toca | Department of Meteorology, University of Reading, Reading, UK; The James Hutton Institute, Aberdeen, UK | United Kingdom
Water table depth along with soil moisture are predominant factors driving biogeochemical processes in peatlands and excessive lowering of water level can lead to peat subsidence, oxidation, and large amounts of carbon being released into the atmosphere. With the increase of northern peatland protection and restoration projects, the assessment of peatland ecosystem hydrological condition is highly necessary, but remains challenging, especially over larger areas in remote locations.
Radar is sensitive to both geometric and water (dielectric) content of vegetation and soil, can cover large areas, and has a high temporal revisit time over northern peatlands; therefore, Sentinel-1 synthetic aperture radar (SAR) data can be expected to offer valuable information on peatland ecosystem condition. This study compared radar backscatter sensitivity to different hydrological conditions in peatlands, first, by undertaking a controlled laboratory-environment experiment and second, by analysing Sentinel-1 time series and hydrological data from degraded, restored, and near-natural peatland sites across Scotland. The unique laboratory experiment used a fully polarimetric C-band SAR system operating in tomographic profiling mode within a 4-8 GHz frequency band to monitor a forced drought of peat and associated peatbog vegetation.
The laboratory-environment research confirmed a firm linear relationship existing between radar backscatter and peat hydrological characteristics with R2>0.9 when other factors influencing radar backscattering were controlled for. The Sentinel-1 time series analysis demonstrated radar backscatter sensitivity to peatland hydrological condition, but the relationships differed between peatland condition classes. The Sentinel-1 time series were able to show particularly well the impact of extreme weather events on peatland hydrological condition, such as the 2018 summer drought.
Taken together, our data demonstrate close sensitivity of backscatter strength to hydrological patterns in the peatbog ecosystem and demonstrate how Sentinel-1 data can support peatland condition monitoring over northern unforested peatlands. However, to further improve peatland hydrological condition estimates using satellite SAR data, more precise modelling of other elements influencing radar backscatter is necessary.
Peat monitoring in Indonesia based on Sentinel-1 time series
Prof. Dr. Florian Siegert | RSS - Remote Sensing Solutions GmbH | Germany
Peatland ecosystems are known as the largest terrestrial carbon sinks. In Indonesia, extensive often forest covered peatlands form convex-shaped peat domes up to 20 m thick and up to 100 km wide. The total peatland area is estimated at 20-25 Million ha storing approximately 14 – 58 Gt carbon. This makes Indonesian peatlands one of the largest carbon sinks in the world. This ecosystem became under intense pressure in the past two decades. Recurrent fires during the dry season burn the forest vegetation at partially the peat layer, which leads to the release of huge amounts of carbon. It is estimated that emissions from South-east Asian peatlands account for 5 – 8 % of the total global CO2 emissions. Large peatlands are cleared from forests for the plantation business. Additionally, this leads to further damage: peat dome hydrology is disturbed by the loss of vegetation and the construction of drainage channels. In pristine peatlands acidic water prevents the organic substance from being decomposed by micro-biological processes and thus organic matter accumulates. In drained peatlands the organic matter becomes quickly decomposed and CO2 and other greenhouse gases oxidation are released. Drained peatlands are extremely vulnerable to fire, which is frequently used by farmers or also the oil palm industry to clear forested land. Consequently, carbon, previously retrained in the soil, is released into the atmosphere. Over the last 20 years large areas of drained peatlands were burned by uncontrolled wildfires and drained for conversion to oil palm plantations. To reduce CO2 emissions rewetting is the most effective way to prevent peat decomposition, soil subsidence and recurrent fires. To achieve this, blocking of drainage canals by dam building is necessary in order to raise the water level of the surrounding peat. The resulting reduced water flow in the canals allows sedimentation of organic and mineral material upstream of the dam, which in turn facilitates the regrowing of vegetation.
Successful restoration of disturbed tropical peatlands requires a detailed understanding of the hydrological situation. Accurate information is needed about the extent and the volume of the peat layers as well as on the drainage network of canals. The 3D surface topography is crucial to understand the hydrological conditions in the case of convex-shaped ombrogenous peat domes. Mapping the surface topography by collecting in situ GPS data is almost impossible, since the peatlands are often remote, frequently flooded, and usually covered by dense vegetation.
In this study, we used ICESat-2/ATLAS satellite LiDAR data to map the surface topography of extensive carbon-rich ombrogenous peatlands. We compared spaceborne ICESat-2 LiDAR data and correlated it with highly accurate field-validated digital terrain models (DTM) generated from airborne LiDAR. Compared to the airborne DTM, the ICESat-2 LiDAR data produced an R2 of 0.89 and an RMSE of 0.83 m. From ICESat-2 LiDAR footprint transects we set up a continuous 3D peat surface model, which can be used for restoration and rewetting activities. In the second part of the study, we investigated if long-term variations in hydrological characteristics of peatlands can be observed in dense Sentinel-1 backscatter time series in non-forested peatlands which have been burned in the past or which have been converted into oil palm plantation recently (where the palms are still small). Sentinel-1 satellites provide temporally dense and high spatial resolution synthetic aperture radar (SAR) imagery. Since the backscatter coefficient (σ0) on bare soil is sensitive to moisture, the application of Sentinel-1 satellite data can support the monitoring of these climate-relevant soils at high spatial resolution. To reduce the effect of outliers or noise, we applied a time series analysis in non-vegetated areas with a size of 100 ha each. Furthermore, the areas are investigated regarding precipitation and MODIS Fire Hotspots in order to rule out changes in backscatter caused by rainfall or land degradation. First results show a decrease from -12.7 – -13.5 dB in a period from December 2015 – December 2018 in areas that have been drained by the construction of drainage canals, as seen in oil palm plantations that have been recently established. On the other hand, we found an increase of -13.5 – -13.0 dB in a period from January 2019 – October 2021 in an area that is currently being rewetted, indicating an increase in soil moisture. We believe that satellite based information from the ESA Sentinel and other missions will help to monitor the success of restoration activities and thus can help to better manage and protect peatland areas in Indonesia and other countries in the tropics.
Large scale detection of economically important palm trees using UAVs and Deep learning in peat swamps in the Peruvian Amazon
Ximena Tagle | Wageningen University & Research | Netherlands
Peatlands are well known for storing more carbon below ground than all the rest of vegetation in the world combined, and they provide other ecosystem services as well. Peruvian Amazonia hosts the most extensive peatland palm swamp in South America, that covers more than 4% of the Peruvian territory (an area larger than the full extension of Denmark). This peatland ecosystem is dominanted by the arborescent palm Mauritia flexuosa (aguaje) and also hosts other arborescent palm species like Oenocarpus bataua (ungurahui) and Euterpe precatoria (huasai). These palm species are ecologically, culturally, and economically important. They provide fruits considered as “superfood” due to their high nutritional values, supporting fauna and local communities.
However, these peatlands are threatened from new infrastructure and increasing demand for agricultural land.In order to avoid degradation and deforestation, it is important to use sustainable fruit harvesting techniques while generating income for local communities. The general limitation to expanding sustainable management of palms in intact forests has been the difficulty of mapping resource abundance and distribution at large scales. Traditional ground-based surveys sample small areas, while management decisions require precise information at larger scales. In recent years, Unmanned Aerial Vehicles (UAVs) have become an important tool for mapping forest areas as some are cheap and easy to transport, and they provide high spatial resolution imagery of remote and difficult-to-access areas.
This study combined field data, RGB UAV imagery and deep convolutional neural networks (CNNs) to automatically detect three economically important palm tree species in the peatland palm swamps of Peru. We surveyed 55 sites and ground-referenced 5,170 palm trees using small multirotor UAVs and permanent forest plots during the dry season of 2017-2019 in the Loreto Region. The developed CNN model accounted for differences in flying heights and weather conditions, having a good accuracy for identifying Mauritia flexuosa (86% Precision at 87% recall, 0.87 F1-score) and lower accuracy for Oenocarpus bataua (53% Precision at 64% recall, 0.54 F1-score) and Euterpe precatoria (83% Precision at 69% recall, 0.76 F1-score).
We show that the combination of the use of UAVs with CNNs allows large-scale mapping in Peruvian Amazonia, providing the basis for expanding sustainable management in intact peatlands, especially in regions where the cloud cover limits the use of satellite imagery, and where the large areas and low accessibility restrict ground-based surveys.
Development of a digital terrain model of the Congo Basin by combining TanDEM-X with other data sources
Dr. Ian Davenport | University of Edinburgh | United Kingdom
The world’s most extensive tropical peatlands occur in the Cuvette Centrale depression in the Congo Basin, which stores around 30 petagrams of carbon. Improving our understanding of the genesis, development and functioning of these under-studied peatlands requires knowledge of their topography and, in particular, whether the peat surface is domed, as this implies a rain-fed system. With airborne LiDAR, we have identified two 5m-high peat domes, and we are extending this study to the whole basin.
Measuring the ground elevation under the dense forest canopy presents substantial obstacles, as the canopy scatters most incident radiation back, preventing optical and radar techniques. ESA’s TanDEM-X has generated a high-resolution digital surface model (DSM), estimating the elevation of the canopy-top, but another source of data is required to determine how high the canopy is above ground level, and turn this into a Digital Terrain Model (DTM). Airborne laser altimetry can provide some of this information, as the high pulse rate with a small laser footprint can occasionally retrieve a return from ground level, and we have acquired 33 swathes in the region with a piloted light aircraft, and two with an autonomous drone. This only covers a small fraction of the basin, so a more extensive source of information is also needed. ICESat-2 uses a satellite-based LiDAR, and while unsuitable in regions with this density of tree cover to reliably estimate ground level, we have used it to estimate tree height. This information is sparse, however, with gaps up to 3km between tracks, so a classification has also been used to provide more information in interpreting the ICESat-2 returns.
By comparing the ICESat-2 canopy estimates to those acquired by airborne LiDAR, we evaluate and calibrate them and subtract from the forested areas of TanDEM-X 90m DSM over the Basin, and estimate uncertainty using ground measurements from airborne LiDAR.