Climate change in Arctic regions poses a threat to permafrost stability, potentially leading to a cascade of issues, including infrastructure damage, altered ecosystems, and increased greenhouse gas emissions. The challenge of monitoring permafrost temperatures and identifying critical environmental factors is accentuated by the scarcity of subsurface thermal data in remote Arctic locales. To alleviate this lack of soil thermal data and to account for seasonal variations, MERRA-2 reanalysis data empowered by Machine learning are employed in this paper for season-specific subsurface soil temperature prediction in Alaska. First, the most useful reanalysis features are selected based on field experts and statistical analysis. Coupled with time-related features, reanalysis features are preprocessed and then fed to the machine learning prediction model. Different experiments and performance benchmarks are conducted to investigate and optimize the proposed approach, including the best-performing machine learning model, the optimal look-back period of inputs, and the optimal training set size. In particular, multiple prediction models are considered, namely, six conventional machine learning models (e.g., GBDT, RF, SVR) and five statistical baselines (e.g., ARIMA using in-situ air temperature data). The proposed method was assessed using in-situ soil temperature data at multiple depths (about 1m depth) from site fields at Deadhorse and Toolik lake, Alaska, spanning 16+ years. All prediction performances are assessed using Root mean squared error and mean absolute error. Results show that locally trained machine learning models can estimate shallow soil temperatures with an average error of $RMSE=1.308^{\circ}$C.
moredataset / soiltemperatureprediction Goto Github PK
View Code? Open in Web Editor NEWLicense: Apache License 2.0