Multispectral CNN-RF Approach for Shoreline Extraction and Climate-Change Driven Coastal Erosion Risk Assessment

Sea-level rise (SLR) driven by climate change is accelerating coastal erosion, which necessitates comprehensive assessment of shoreline dynamics to identify areas for targeted mitigation and adaptive planning. However, extracting coastlines at the sub-pixel level from medium spatial data with the limited annotated large-scale information availability hinder comprehensive risk assessment under climate change projections. This study utilizes geospatial techniques and machine learning based multispectral indices approach combined with Convolutional Neural Network (CNN) and Random Forest (RF) to quantify the erosion and accretion along the 1,365 km coastline of Pakistan with historic and climate change scenarios. Shoreline trends were analyzed, and future projections were derived from CMIP6 General Circulation Models (GCMs) under Shared Socioeconomic Pathways (SSPs) for 2020 to 2050. Results reveal pronounced erosion along the Indus Delta, with retreat rates reaching cumulative − 150.4 ± 1.02 m with increase in SLR (0.015–0.15 m) from 2000 to 2020, accompanied by increases in sea surface temperature (297–301 K). Sandspit coast showed up to 23.24 km² of accretion at 89.45 ± 0.23 m, while Gwadar Port experienced accretion rates up to 50 m with annual temperature increases of 0.02°C-0.05°C. Under the high emission SSP5-8.5 scenario, persistent erosion of -90 ± 1.35 m is expected in the Indus Delta by 2050. Projections across Global Warming Levels (GWL) with 1.5-5°C suggest SLR may reach 0.23 m at 5°C warming. Our findings underscore the importance of Integrated Coastal Zone Management (ICZM) and reinforce the IPCC’s call to limit global warming to safeguard coastal ecosystems and achieve SDG 13 and 15 targets by 2030.