Integrating Medium Resolution Satellite Data and CNN-RF Machine Learning for Shoreline Dynamics: Assessing Coastal Erosion and Accretion under Climate Change Scenarios

Sea-level rise (SLR) driven by climate change has exacerbated coastal erosion, posing significant challenges for coastal management. Effective management necessitates robust tools to evaluate shoreline dynamics under varying climate scenarios, facilitating the identification of high-risk areas. However, the pixelated nature of coastlines and the limited scope of large-scale coastal projections under diverse climate conditions hinder comprehensive risk assessment. This study addresses these gaps by utilizing medium-resolution Landsat data integrated with a Convolutional Neural NetworkCNN - Random Forest-RF, enhanced by an activation function and five max-pooling, to process training predictors based on spectral indices MNDWI, NDWI, NDVI, GCVI, and SAVI, shorelines detection and demarcation. The analysis applies Bruun Rule to assess shoreline retreat relative to SLR along Pakistan's coast at five-year intervals from 2020 to 2050. SLR and SST are sourced from multiple satellite sensors, including AVHRR and SLSTR, and computed using CMEMS relative to a 2000–2023 baseline. Climate projections are derived from a multi-model ensemble of CMIP6 General Circulation Models (GCMs), spanning Shared Socioeconomic Pathways SSP1-2.6 to SSP5-8.5. The proposed CNN-RF model demonstrated high accuracy, achieving precision, recall, and F1 scores of 95.01%, 96.16%, and 96.91%. Results from historical regression rates, combined with SLR and SST projections, indicate widespread erosion in Indus Delta, with alarming retreat rates of -80.4 ±1.15 m/year between 2000 and 2010, corresponding to SLR values ranging from 0.015 to 0.085 m/year. From 2010 to 2023, SLR accelerated to 0.087–0.15 m/year, with SST increasing from 297.79 K to 300.3 K. Conversely, the Sandspit coast exhibited accretion, gaining 23.24 km² at rates of up to mean 49.45 ±1.16 m/year. Notable warming trends were observed, with central Arabian Sea SSTs exceeding 302.41 K, correlating strongly with SLR (R² = 0.40 by 2023). Under the high-emission scenario SSP5-8.5, projections for 2020–2025 indicate persistent erosion in the Indus Delta, with retreat rates of -25 to -60 m/year, while Gwadar Port up to 10 to 15 m/year. For 2025–2030 and 2030-2050 erosion in the Indus Delta, retreat rates up to -68 m/year and of -101 to -120 m/year, Sonmiani Aquifer may transition erosion up to mean -55.1 and -110 m/year). SST anomalies exhibit variability (0.3°C–0.8°C) and periodic spikes linked to climatic events, with annual increases of 0.02°C–0.05°C and a coefficient of variation of 12%–25%. Pearson’s correlation (R² = 0.6–0.8) suggests a positive relationship between SST and SLR, but highlighted variability, indicating areas for refinement. The impacts of the intrusion on the local coastal community are also analyzed with trends of communities’ migration. Our analysis revealed that erosion also results from reduced sediment flow linked to water infrastructures. Future policy and action plans should prioritize Integrated Coastal Zone Management frameworks (ICZMF), providing critical insights into erosion dynamics and addressing integrated nature-based solutions.Keywords: Sea-level rise (SLR), Coastal Erosion, CNN-Random Forest (RF), Landsat, CMIP6, Integrated Coastal Zone Management frameworks (ICZMF)