|Paper title||Using Unmanned Aerial Systems to unveil the impacts of woody encroachment in subarctic tundra wetlands.|
|Form of presentation||Poster|
Shrubification of arctic tundra wetlands alongside with changes in the coverage and volume of lichens are two well-documented processes in the Fennoscandian tundra. A rapidly warming climate and changes in reindeer grazing patterns are driving shifts in the carbon feedbacks and altering local microclimate conditions. The growth in arctic deciduous shrubs has been documented, and its effects on ecosystem function and structure may range from a greater release of soil carbon to alterations in the local ecohydrology. It is therefore of upmost importance to closely monitor these changes in order to gain a complete understanding of their dynamics and improve the adaptive capacity of the regions under study. In this regard, earth observation data has played a key monitoring role during past decades. However, the fine scale of these processes often renders them invisible or hazy under the eye of satellite sensors. On the other hand, the rapid growth of Unmanned Aerial Systems and sensor capabilities opens new opportunities for mapping and monitoring.
Here, we present a toolset of Unmanned Aerial Systems and Machine Learning algorithms that enables highly accurate monitoring of landcover change dynamics in the sub-arctic tundra. The study area is located in the Fennoscandian oroarctic tundra zone, between the Finnish-Norwegian border. In the mid 1950s, a reindeer fence was built along the border, thus separating two different reindeer grazing strategies. While reindeer graze only during winter in the Norwegian side, grazing occurs all year round in the Finnish side, with reindeer feeding on the new shoots of willows (Salix spp.) and therefore containing the shrubification process.
In order to study the long-term impacts of differential grazing on willow extent and growth, we surveyed the study area with a Sensefly Ebee and a DJI Matrice 200 equipped with a Parrot Sequoia 1.2 megapixel monochromatic multi-spectral sensor, a senseFly rgb S.O.D.A and a FLIR Thermal Imaging kit respectively. We combined multispectral, photogrammetric and thermal data with an ensemble of machine learning algorithms to map the extent of woody shrubs and quantify their above-ground biomass at two wetlands across the Finnish-Norwegian border. Furthermore, we used the same toolset to map topsoil moisture and water table depth, two parameters strongly influenced by the encroachment of willow bushes in subarctic wetlands. The set of algorithms under scrutiny were a pixel-based Random Forest and the more recent XGBoost. The ensemble of algorithms was trained with a comprehensive set of in-situ data collected at the study sites, including plant species composition, above ground biomass, topsoil moisture, water table depth and depth of the peat layer. The validation of results showed a high degree of accuracy, with R2 > 0.85 for biomass prediction and overall accuracy > 80% for plant community distribution maps. The results show a clear expansion of willows in the Norwegian side of the border, alongside a strong increase in the above ground biomass.
The high degree of accuracy obtained in the results unfolds new research prospects, such as the combination of fine-scale remote sensing with chamber and Eddy Covariance measurements to quantify the impact of land cover on the carbon and energy balance. The use of Unmanned Aerial Systems could also help unveil the complexity of greening and browning patterns in the arctic.