Neural Network-Based Approach for Identification of Meteorological Factors Affecting Regional Sea-Level Anomalies

The geographically nonuniform sea-level change has increased the importance of assessing sea-level variability and the factors controlling it on regional scales. This study provides a framework, based on the rules governing an artificial neural network (ANN), to identify an ensemble of large-scale meteorological variables (MVs), which significantly affect long-term monthly water-level anomalies (MWLA) in northern coast of the Persian Gulf (1990-2013). Horizontal spatial grid cells of 10 degrees x10 degrees, bounded between (0-100 degrees E and 0-70 degrees N), create a surface control to address the patterns of six MVs consisting of zonal and meridional wind velocity, total precipitable water, 1,000-500hPa thickness, relative humidity, and air temperature at surfaces of 300 and 700hPa, respectively. Additionally, 14 representative marine regions are also taken into consideration to assess the potential impact of sea surface temperature (SST) and sea-level pressure (SLP) on local sea-level variability. The multicollinearity problem is effectively tackled by principal components analysis, which classified the MVs into the independent categories. A neural network-based pruning algorithm under a statistical hypothesis test is introduced to discern redundant factors, and then estimate the relative importance of each of the significant predictors in simulating the MWLA. The pruning algorithm detected the nine meteorological components, which are able to predict up to 56% of the total variance in the MWLA. Moreover, it is found that more than half of the predicted variability is manifested by zonal wind, SST, and SLP patterns.