Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Developing a Sequential Deep Learning Pipeline to Model Alaskan Permafrost Thaw Under Climate Change

Addina Rahaman

TL;DR

This study develops an end-to-end sequential deep learning pipeline to model Alaskan active-layer soil temperatures under climate change. By integrating ERA5-Land reanalysis with static lithology data, applying 24-month sliding windows and latitude-band embeddings, and evaluating five DL architectures (TCN, Transformer, Conv1DLSTM, GRU, BiLSTM), it demonstrates robust capture of seasonal, latitudinal, and depth-dependent patterns. A derived scenario signal and quantile mapping enable scenario-aware learning from CMIP5 RCP projections, with SHAP used for interpretability. While GRU often performs best and QM helps reveal sinusoidal trends, limitations arise from CMIP5 data biases and missing snow/zero-curtain dynamics, underscoring the framework's potential and need for richer cryospheric inputs for more divergent scenario forecasts.

Abstract

Changing climate conditions threaten the natural permafrost thaw-freeze cycle, leading to year-round soil temperatures above 0°C. In Alaska, the warming of the topmost permafrost layer, known as the active layer, signals elevated greenhouse gas release due to high carbon storage. Accurate soil temperature prediction is therefore essential for risk mitigation and stability assessment; however, many existing approaches overlook the numerous factors driving soil thermal dynamics. This study presents a proof-of-concept latitude-based deep learning pipeline for modeling yearly soil temperatures across multiple depths. The framework employs dynamic reanalysis feature data from the ERA5-Land dataset, static geologic and lithological features, sliding-window sequences for seasonal context, a derived scenario signal feature for long-term climate forcing, and latitude band embeddings for spatial sensitivity. Five deep learning models were tested: a Temporal Convolutional Network (TCN), a Transformer, a 1-Dimensional Convolutional Long-Short Term Memory (Conv1DLSTM), a Gated-Recurrent Unit (GRU), and a Bidirectional Long-Short Term Memory (BiLSTM). Results showed solid recognition of latitudinal and depth-wise temperature discrepancies, with the GRU performing best in sequential temperature pattern detection. Bias-corrected CMIP5 RCP data enabled recognition of sinusoidal temperature trends, though limited divergence between scenarios were observed. This study establishes an end-to-end framework for adopting deep learning in active layer temperature modeling, offering seasonal, spatial, and vertical temperature context without intrinsic restrictions on feature selection.

Developing a Sequential Deep Learning Pipeline to Model Alaskan Permafrost Thaw Under Climate Change

TL;DR

This study develops an end-to-end sequential deep learning pipeline to model Alaskan active-layer soil temperatures under climate change. By integrating ERA5-Land reanalysis with static lithology data, applying 24-month sliding windows and latitude-band embeddings, and evaluating five DL architectures (TCN, Transformer, Conv1DLSTM, GRU, BiLSTM), it demonstrates robust capture of seasonal, latitudinal, and depth-dependent patterns. A derived scenario signal and quantile mapping enable scenario-aware learning from CMIP5 RCP projections, with SHAP used for interpretability. While GRU often performs best and QM helps reveal sinusoidal trends, limitations arise from CMIP5 data biases and missing snow/zero-curtain dynamics, underscoring the framework's potential and need for richer cryospheric inputs for more divergent scenario forecasts.

Abstract

Changing climate conditions threaten the natural permafrost thaw-freeze cycle, leading to year-round soil temperatures above 0°C. In Alaska, the warming of the topmost permafrost layer, known as the active layer, signals elevated greenhouse gas release due to high carbon storage. Accurate soil temperature prediction is therefore essential for risk mitigation and stability assessment; however, many existing approaches overlook the numerous factors driving soil thermal dynamics. This study presents a proof-of-concept latitude-based deep learning pipeline for modeling yearly soil temperatures across multiple depths. The framework employs dynamic reanalysis feature data from the ERA5-Land dataset, static geologic and lithological features, sliding-window sequences for seasonal context, a derived scenario signal feature for long-term climate forcing, and latitude band embeddings for spatial sensitivity. Five deep learning models were tested: a Temporal Convolutional Network (TCN), a Transformer, a 1-Dimensional Convolutional Long-Short Term Memory (Conv1DLSTM), a Gated-Recurrent Unit (GRU), and a Bidirectional Long-Short Term Memory (BiLSTM). Results showed solid recognition of latitudinal and depth-wise temperature discrepancies, with the GRU performing best in sequential temperature pattern detection. Bias-corrected CMIP5 RCP data enabled recognition of sinusoidal temperature trends, though limited divergence between scenarios were observed. This study establishes an end-to-end framework for adopting deep learning in active layer temperature modeling, offering seasonal, spatial, and vertical temperature context without intrinsic restrictions on feature selection.

Paper Structure

This paper contains 23 sections, 7 equations, 12 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Data and Methods
  3. Study Region and Climate Context
  4. Data Sources and Processing
  5. Data Availability
  6. Data Preparation
  7. Scenario Signal
  8. Temporal Windowing for Sequential Modeling
  9. Additional Preprocessing
  10. Data Pipeline
  11. Deep Learning Models
  12. Convolutional Neural Networks (CNNs)
  13. Recurrent Neural Networks (RNNs)
  14. Transformers
  15. Model Initialization and Setup
  16. ...and 8 more sections

Figures (12)

  • Figure 1: Band map of Alaska used in the analysis.
  • Figure 1: Comparison of model predictions (blue) with ground truth (dashed) for the year of 2023, stratified by latitude band (rows) and soil layer (columns).
  • Figure 2: Diagram of sliding windows for sequence $j$ and subsequent sequences.
  • Figure 2: Comparison between TCN predictions per band-layer pair for each RCP scenario. RCP 2.6 is in green, RCP 4.5 is in blue, RCP 6.0 is in yellow, and RCP 8.5 is in red.
  • Figure 3: Proposed end-to-end deep-learning workflow of model training, testing, and final soil temperature predictions.
  • ...and 7 more figures