• An upgradation of machine learning framework using PCHIP approach has been proposed for improving the wave parameters prediction. • PCHIP based LSTM and RF models were applied to buoy data from the Bay of Bengal region to predict wave parameters and estimate wave power. • Validation shows a notable error reduction of about 70% to 90% for cyclonic, pre cyclonic and post cyclonic periods. • The proposed work is found to be effective under limited data availability during normal and extreme events. This study proposes two hybrid machine learning models employing piecewise cubic hermite interpolation polynomial (PCHIP) technique : PCHIP-LSTM and PCHIP-RF, to improve the wave parameter forecasting. Seven datasets from the Bay of Bengal (13 ∘ 27′09.3″ N and 84 ∘ 08′20.4″ E), consisting cyclonic, pre-cyclonic, and post-cyclonic conditions during short-term periods between 2021 and 2023, were analysed to evaluate significant wave height (SWH) and average wave period (APD) using the proposed models. PCHIP technique is utilised to enhance the temporal resolution of data by iteratively splitting the time durations and accumulating interpolated values from the original datasets. Scatter plot results for SWH and APD shows good predictive performance of the proposed models over the traditional models. Results obtained through these models did not disturb the originality of the datasets and agree well with that of the buoy measurements and reanalysis data. Validation against buoy measurements shows a notable error reduction of about 70% in RMSE, MAE and MAPE. Friedman test and Posthoc pairwise comparison ensures the statistical significance of the proposed models thereby serves as a validation. A key advantage of the proposed approach is its effectiveness under limited data availability where continuous measurements are often sparse or interrupted during severe weather. This work not only mitigates the limitations of traditional ML models for studying small datasets but also enhances the wave power estimation in all oceanic conditions.
Vijay et al. (Fri,) studied this question.