Spatially Regularized Graph Attention Autoencoder Framework for Detecting Rainfall Extremes

TL;DR

This work introduces a novel Graph Attention Autoencoder with spatial regularization to address the challenge of scalable anomaly detection in spatiotemporal rainfall data across India from 1990 to 2015, and paves the way for sophisticated spatiotemporal anomaly detection methodologies in climate science.

Abstract

We introduce a novel Graph Attention Autoencoder (GAE) with spatial regularization to address the challenge of scalable anomaly detection in spatiotemporal rainfall data across India from 1990 to 2015. Our model leverages a Graph Attention Network (GAT) to capture spatial dependencies and temporal dynamics in the data, further enhanced by a spatial regularization term ensuring geographic coherence. We construct two graph datasets employing rainfall, pressure, and temperature attributes from the Indian Meteorological Department and ERA5 Reanalysis on Single Levels, respectively. Our network operates on graph representations of the data, where nodes represent geographic locations, and edges, inferred through event synchronization, denote significant co-occurrences of rainfall events. Through extensive experiments, we demonstrate that our GAE effectively identifies anomalous rainfall patterns across the Indian landscape. Our work paves the way for sophisticated spatiotemporal anomaly detection methodologies in climate science, contributing to better climate change preparedness and response strategies.

Figures (5)

• Figure 1: SRGAttAE Model Architecture. The data procurement and preprocessing are performed using publicly available IMD and ERA5 Reanalysis data.
• Figure 2: Heatmaps generated from the anomaly scores, for three years: 1991, 2005, and 2015. Uniformly red regions indicate $90^{th}$ percentile ($TOP$) and $95^{th}$ percentile ($BOTTOM$) anomaly scores. Heatmaps from 1991 through 2015 have been provided in the Appendix.
• Figure 3: Experimentation Results on the IMD (top) and ERA5 (bottom) datasets; the MSE penalizes anomalies harsher than the MAE. As we can see, our GAT Autoencoder with SCR reconstructs the graph with the least or near-least error across both metrics.
• Figure 4: Variation of Node Anomalies over 1990-2015; statistical analysis using the Mann-Kendall trend test revealed no detectable trend. The temporal stability of rainfall events can be attributed to the enduring stability of the underlying graph structures, as shown in Tantary_2023.
• Figure 5: A schematic of the dataset construction process.