ESA CCI Research Fellowship: Alexandra Runge

Project: Permafrost Vulnerability from multiple Essential Climate Variables (PVE)

Fellowship summary

Permafrost, which is frozen ground covering vast areas of the Earth, serves as a key indicator of global climate change. It's tightly linked to the atmosphere, biosphere, and geosphere, making it sensitive to changes in various environmental and climatic factors. Recent studies reveal a global warming trend in permafrost based on in situ observation and modelling studies, yet the specific drivers at large-scale behind this remain unclear.

To address this knowledge gap, we conducted a comprehensive analysis by combining Earth Observation (EO) time series data sets representing key environmental and climatic variables influencing permafrost ground temperature. These variables include land surface temperature, soil moisture, land cover, snow cover, albedo, fire, air temperature, and precipitation, all defined Essential Climate Variables (ECV). Additionally, we integrated in situ measurements and modelled permafrost ground temperature time series data into our study. The goal was to assess the impact of EO-based variables on permafrost, identify their combined influence, and develop a framework to understand permafrost vulnerability.

Our approach involved spatiotemporal assessments of the individual variables, correlation analyses, principal component analyses (PCA), and setting up a Machine Learning (ML) approach. The ML aspect aims to predict changes in permafrost ground temperature based on the environmental and climatic variables and identify the primary drivers, using explainable Artificial Intelligence for improved interpretability. We conducted the analysis on two assessment scales: 1) utilizing the data from the Mean Annual Ground Temperature (MAGT) borehole time series data of observed permafrost ground temperature, which is rather sparsely distributed across the circumpolar Arctic, and 2) employing a randomly stratified sampling based on the Boreal-Arctic Wetland and Lake Dataset (BAWLD POI1).

The spatiotemporal variability assessments revealed regional differences in the statistically significant trends of the individual ECVs. Following this, we sampled the ECV trend products with for the MAGT and BAWLD POI1 data set. Overall, the correlation between ECVs was low for both MAGT and BAWLD POI1 data sets. The PCA for MAGT highlighted that trends in precipitation, soil moisture, and fire predominantly account for variance in permafrost ground temperature changes. Meanwhile, the PCA for BAWLD POI1 showed that trends in land surface temperatures, precipitation, and air temperature explain most of the variance in permafrost ground temperature changes. Overall, these results fit well to previous local and regional studies, that described air temperature and/or precipitation to be the main drivers of permafrost ground temperature warming. This indicates that especially climate variables drive permafrost warming and changes in environmental variables are less prominent in this context.

The Machine Learning and explainable Artificial Intelligence work is still ongoing and expected to provide further insights into the permafrost-climate dynamics at large-scale. However, challenges persist due to varying data quality in EO-based data sets in northern high latitudes.

In summary, our results indicate that the assessment scale is crucial, with BAWLD POI1 appearing to be a better fit than the unevenly distributed MAGT data. Nevertheless, the correlation analyses and PCAs for MAGT and BAWLD POI1 yielded similar outcomes, reinforcing the robustness of our findings. This project demonstrates the value of large-scale EO-based assessments and the strength in consistent long-term globally available products.

Summary slides: PVE_ARunge_final_meeting

Research Fellow: Alexandra Runge

Host Institution: Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research (AWI Potsdam)