Machine learning-based analysis reveals alarming sea surface warming in Türkiye’s semi-enclosed marine basins (Mediterranean, Aegean, Marmara, and Black Seas) with the Mediterranean warming at nearly twice the global ocean average (+1.98 °C since 1993 vs. +0.88 °C in IPCC AR6). This study integrates 30 years (1993–2023) of high-resolution in-situ coastal observations with satellite-derived CMEMS reanalysis data for robust trend analysis and forecasting. Linear trend estimates indicate accelerated regional warming rates of +0.03–0.05 °C/year. A novel Long Short-Term Memory (LSTM) model enhanced with ENSO and NAO indices and trained on bias-corrected data significantly improves projection accuracy (RMSE: 0.42–0.65 °C; R2 >0.90) and reduces prediction uncertainty by over 15% compared to classical methods. Projections under SSP2-4.5 and SSP5-8.5 scenarios suggest that SSTs could exceed 27 °C in the Mediterranean and reach 21–23 °C in the Black Sea by 2100, intensifying ecological risks such as tropicalization by invasive species, oxygen depletion, and marine heatwaves. Türkiye’s seas are warming 1.5–2 times faster than CMIP6 global model projections, highlighting their status as climate hotspots. This hybrid approach; merging satellite data, in-situ observations, and machine learning; offers a scalable framework for semi-enclosed sea monitoring and climate-informed policy design. Findings emphasize the urgent need to integrate machine learning-based projections into coastal adaptation strategies, fisheries governance, and regional marine resilience planning in alignment with SDGs 13 (Climate Action) and 14 (Life Below Water).