Parallel Computing Implementation of Marine Heat Waves Detection

Marine heat waves (MHW) are anomaly increases in sea surface temperature that can negatively impact marine life, such as the mass death of marine creatures. Therefore, MHW detection is required to determine areas whose high potential risk for MHW. Extreme event detection, such as MHW, is a computing challenge as it is costly and needs huge memory. Basically, MHW detection is calculated individually for each location, which is inefficient for large domain. This study aims to accelerate MHW detection by implementing a parallel computation technique. The parallel computation of MHW detection combines the MHW Python library proposed by Oliver et al., 201S and the joblib library based on pipelining in Python. We propose the parallelization in a single and nested loop and evaluate the performance of the parallelization in single and multi-nodes. We found that the proposed parallelization can efficiently accelerate the computation time in a single node up to 10 times. Moreover, the parallelization in the nested loop performs more efficiently than the single loop.

I. Introduction

Prolonged discrete warm water anomalies have had a significant impact on marine ecosystems. Marine heat waves (MHW) are events characterized by anomalously warm water occurring within a specific time frame, described in terms of duration, intensity, evolution rate, and spatial extent. Warm water events at the sea surface can be classified as MHW if they persist for at least five consecutive days, with temperatures exceeding the 90th percentile of historical baseline temperatures over a minimum period of 30 years. These heat waves can be attributed to oceanographic and atmospheric processes [1].