Abstract
In this study, a total of 22,371 satellite images acquired between 5 March 2000 and 6 November 2024 from Sentinel-1 synthetic aperture radar (SAR) and the Moderate Resolution Imaging Spectroradiometer (MODIS) sensors onboard the Terra and Aqua satellites were analyzed. From these images, 44,225 internal solitary wave (ISW) events were identified in the Andaman Sea and its surrounding waters. Based on these observations, the spatial distribution and propagation-direction characteristics of ISWs in the Andaman Sea were statistically examined. In addition to the reflection phenomena previously observed near the northern continental slope, similar multi-directional propagation features were also detected in the central and southern parts of the basin. Using the identified ISW dataset, monthly ISW event count time series were constructed for the northern, central, and southern subregions of the Andaman Sea. These datasets were subsequently used to evaluate the predictive performance of the Seasonal Gated Recurrent Unit (SGRU) model through hindcast experiments. Specifically, during the period from November 2020 to October 2024, the one-month lead predictions of monthly ISW event counts show good agreement with observations, with correlation coefficients exceeding 0.85. Sensitivity experiments conducted under different historical window lengths and forecast initialization months indicate that the SGRU model exhibits strong dependence on both initialization timing and historical window length. This suggests that the ISW event count time series in the Andaman Sea contains pronounced seasonal and interannual variability modulated by climatic factors. These results demonstrate the potential of deep-learning approaches for predicting ISW event counts in complex marginal seas and provide new insights into the climate-scale variability of internal wave activity in the Andaman Sea.