Multi-branch Spatio-Temporal Graph Neural Network For Efficient Ice Layer Thickness Prediction

Understanding spatio-temporal patterns in polar ice layers is essential for tracking changes in ice sheet balance and assessing ice dynamics. While convolutional neural networks are widely used in learning ice layer patterns from raw echogram images captured by airborne snow radar sensors, noise in the echogram images prevents researchers from getting high-quality results. Instead, we focus on geometric deep learning using graph neural networks, aiming to build a spatio-temporal graph neural network that learns from thickness information of the top ice layers and predicts for deeper layers. In this paper, we developed a novel multi-branch spatio-temporal graph neural network that used the GraphSAGE framework for spatio features learning and a temporal convolution operation to capture temporal changes, enabling different branches of the network to be more specialized and focusing on a single learning task. We found that our proposed multi-branch network can consistently outperform the current fused spatio-temporal graph neural network in both accuracy and efficiency.

Graph Neural Network as Computationally Efficient Emulator of Ice-sheet and Sea-level System Model (ISSM)

The Ice-sheet and Sea-level System Model (ISSM) provides solutions for Stokes equations relevant to ice sheet dynamics by employing finite element and fine mesh adaption. However, since its finite element method is compatible only with Central Processing Units (CPU), the ISSM has limits on further economizing computational time. Thus, by taking advantage of Graphics Processing Units (GPUs), we design a graph convolutional network (GCN) as a fast emulator for ISSM. The GCN is trained and tested using the 20-year transient ISSM simulations in the Pine Island Glacier (PIG). The GCN reproduces ice thickness and velocity with a correlation coefficient greater than 0.998, outperforming the traditional convolutional neural network (CNN). Additionally, GCN shows 34 times faster computational speed than the CPU-based ISSM modeling. The GPU-based GCN emulator allows us to predict how the PIG will change in the future under different melting rate scenarios with high fidelity and much faster computational time.

Graph Neural Networks for Emulation of Finite-Element Ice Dynamics in Greenland and Antarctic Ice Sheets

Although numerical models provide accurate solutions for ice sheet dynamics based on physics laws, they accompany intensified computational demands to solve partial differential equations. In recent years, convolutional neural networks (CNNs) have been widely used as statistical emulators for those numerical models. However, since CNNs operate on regular grids, they cannot represent the refined meshes and computational efficiency of finite-element numerical models. Therefore, instead of CNNs, this study adopts an equivariant graph convolutional network (EGCN) as an emulator for the ice sheet dynamics modeling. EGCN reproduces ice thickness and velocity changes in the Helheim Glacier, Greenland, and Pine Island Glacier, Antarctica, with 260 times and 44 times faster computation time, respectively. Compared to the traditional CNN and graph convolutional network, EGCN shows outstanding accuracy in thickness prediction near fast ice streams by preserving the equivariance to the translation and rotation of graphs.

Learning Spatio-Temporal Patterns of Polar Ice Layers With Physics-Informed Graph Neural Network

Learning spatio-temporal patterns of polar ice layers is crucial for monitoring the change in ice sheet balance and evaluating ice dynamic processes. While a few researchers focus on learning ice layer patterns from echogram images captured by airborne snow radar sensors via different convolutional neural networks, the noise in the echogram images proves to be a major obstacle. Instead, we focus on geometric deep learning based on graph neural networks to learn the spatio-temporal patterns from thickness information of shallow ice layers and make predictions for deep layers. In this paper, we propose a physics-informed hybrid graph neural network that combines the GraphSAGE framework for graph feature learning with the long short-term memory (LSTM) structure for learning temporal changes, and introduce measurements of physical ice properties from Model Atmospheric Regional (MAR) weather model as physical node features. We found that our proposed network can consistently outperform the current non-inductive or non-physical model in predicting deep ice layer thickness.

Physics-Informed Machine Learning On Polar Ice: A Survey

The mass loss of the polar ice sheets contributes considerably to ongoing sea-level rise and changing ocean circulation, leading to coastal flooding and risking the homes and livelihoods of tens of millions of people globally. To address the complex problem of ice behavior, physical models and data-driven models have been proposed in the literature. Although traditional physical models can guarantee physically meaningful results, they have limitations in producing high-resolution results. On the other hand, data-driven approaches require large amounts of high-quality and labeled data, which is rarely available in the polar regions. Hence, as a promising framework that leverages the advantages of physical models and data-driven methods, physics-informed machine learning (PIML) has been widely studied in recent years. In this paper, we review the existing algorithms of PIML, provide our own taxonomy based on the methods of combining physics and data-driven approaches, and analyze the advantages of PIML in the aspects of accuracy and efficiency. Further, our survey discusses some current challenges and highlights future opportunities, including PIML on sea ice studies, PIML with different combination methods and backbone networks, and neural operator methods.

TinyVQA: Compact Multimodal Deep Neural Network for Visual Question Answering on Resource-Constrained Devices

Traditional machine learning models often require powerful hardware, making them unsuitable for deployment on resource-limited devices. Tiny Machine Learning (tinyML) has emerged as a promising approach for running machine learning models on these devices, but integrating multiple data modalities into tinyML models still remains a challenge due to increased complexity, latency, and power consumption. This paper proposes TinyVQA, a novel multimodal deep neural network for visual question answering tasks that can be deployed on resource-constrained tinyML hardware. TinyVQA leverages a supervised attention-based model to learn how to answer questions about images using both vision and language modalities. Distilled knowledge from the supervised attention-based VQA model trains the memory aware compact TinyVQA model and low bit-width quantization technique is employed to further compress the model for deployment on tinyML devices. The TinyVQA model was evaluated on the FloodNet dataset, which is used for post-disaster damage assessment. The compact model achieved an accuracy of 79.5%, demonstrating the effectiveness of TinyVQA for real-world applications. Additionally, the model was deployed on a Crazyflie 2.0 drone, equipped with an AI deck and GAP8 microprocessor. The TinyVQA model achieved low latencies of 56 ms and consumes 693 mW power while deployed on the tiny drone, showcasing its suitability for resource-constrained embedded systems.

Graph Neural Networks as Fast and High-fidelity Emulators for Finite-Element Ice Sheet Modeling

Although the finite element approach of the Ice-sheet and Sea-level System Model (ISSM) solves ice dynamics problems governed by Stokes equations quickly and accurately, such numerical modeling requires intensive computation on central processing units (CPU). In this study, we develop graph neural networks (GNN) as fast surrogate models to preserve the finite element structure of ISSM. Using the 20-year transient simulations in the Pine Island Glacier (PIG), we train and test three GNNs: graph convolutional network (GCN), graph attention network (GAT), and equivariant graph convolutional network (EGCN). These GNNs reproduce ice thickness and velocity with better accuracy than the classic convolutional neural network (CNN) and multi-layer perception (MLP). In particular, GNNs successfully capture the ice mass loss and acceleration induced by higher basal melting rates in the PIG. When our GNN emulators are implemented on graphic processing units (GPUs), they show up to 50 times faster computational time than the CPU-based ISSM simulation.

Multi-task Deep Convolutional Network to Predict Sea Ice Concentration and Drift in the Arctic Ocean

Forecasting sea ice concentration (SIC) and sea ice drift (SID) in the Arctic Ocean is of great significance as the Arctic environment has been changed by the recent warming climate. Given that physical sea ice models require high computational costs with complex parameterization, deep learning techniques can effectively replace the physical model and improve the performance of sea ice prediction. This study proposes a novel multi-task fully conventional network architecture named hierarchical information-sharing U-net (HIS-Unet) to predict daily SIC and SID. Instead of learning SIC and SID separately at each branch, we allow the SIC and SID layers to share their information and assist each other's prediction through the weighting attention modules (WAMs). Consequently, our HIS-Unet outperforms other statistical approaches, sea ice physical models, and neural networks without such information-sharing units. The improvement of HIS-Unet is obvious both for SIC and SID prediction when and where sea ice conditions change seasonally, which implies that the information sharing through WAMs allows the model to learn the sudden changes of SIC and SID. The weight values of the WAMs imply that SIC information plays a more critical role in SID prediction, compared to that of SID information in SIC prediction, and information sharing is more active in sea ice edges (seasonal sea ice) than in the central Arctic (multi-year sea ice).

Skip-WaveNet: A Wavelet based Multi-scale Architecture to Trace Firn Layers in Radar Echograms

Echograms created from airborne radar sensors capture the profile of firn layers present on top of an ice sheet. Accurate tracking of these layers is essential to calculate the snow accumulation rates, which are required to investigate the contribution of polar ice cap melt to sea level rise. However, automatically processing the radar echograms to detect the underlying firn layers is a challenging problem. In our work, we develop wavelet-based multi-scale deep learning architectures for these radar echograms to improve firn layer detection. We show that wavelet based architectures improve the optimal dataset scale (ODS) and optimal image scale (OIS) F-scores by 3.99% and 3.7%, respectively, over the non-wavelet architecture. Further, our proposed Skip-WaveNet architecture generates new wavelets in each iteration, achieves higher generalizability as compared to state-of-the-art firn layer detection networks, and estimates layer depths with a mean absolute error of 3.31 pixels and 94.3% average precision. Such a network can be used by scientists to trace firn layers, calculate the annual snow accumulation rates, estimate the resulting surface mass balance of the ice sheet, and help project global sea level rise.

Prediction of Annual Snow Accumulation Using a Recurrent Graph Convolutional Approach

The precise tracking and prediction of polar ice layers can unveil historic trends in snow accumulation. In recent years, airborne radar sensors, such as the Snow Radar, have been shown to be able to measure these internal ice layers over large areas with a fine vertical resolution. In our previous work, we found that temporal graph convolutional networks perform reasonably well in predicting future snow accumulation when given temporal graphs containing deep ice layer thickness. In this work, we experiment with a graph attention network-based model and used it to predict more annual snow accumulation data points with fewer input data points on a larger dataset. We found that these large changes only very slightly negatively impacted performance.