The Grid-Based Spatial ARIMA Model: An Innovation for Short-Term Predictions of Ocean Current Patterns with Big HF Radar Data

Abstract

Marine natural disasters have direct impacts on countries as well as their residents living on and near the coast. Warning and monitoring system can aid in reducing the loss of lives in the event of a disaster. HF (high frequency) radar, an IoT-enabled ocean surface current monitoring system, implementation is one of the first attempts towards achieving this goal. Although HF systems can monitor sea current patterns in terms of speed and direction for each of the pixels of the coverage area, it fails to predict future values, which are essential to many applications such as oil-spill trajectory prediction (using the GNOME suite: General NOAA Operational Modeling Environment), water quality control and management, and optimized sea navigation. In this paper, we propose a model, called the grid-based spatial ARIMA (auto-regressive integrated moving average), to estimate the forecast values. As a result, the full potential of the HF systems can be utilized. The method considers not only observations of POI (point of interest), but also its neighboring pixels when predicting future values. The proposed method is implemented and compared with other existing approaches, including baseline, kNN, traditional ARIMA model, and LSTM (long short-term memory) techniques. The experimental results showed that our approach outperformed other methods in V comp prediction (with RMSEs of 6.23265) with a configuration of (2, 0, 1) as (p, d, q) and a historical dataset of 1 day and 7 h prior. This configuration was found to be the best combination.