Machine learning applied to remote sensing in the context of intertidal zone mapping: A literature review

Accurate digital elevation models for intertidal zones are essential for coastal management, yet traditional survey methods and models based on normalized index from remote sensing products often struggle to represent these dynamic environments. Recently, machine learning has emerged as a promising alternative for mapping intertidal morphology. This study systematically reviewed 60 articles using PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines and bibliometric analysis, focusing on machine learning applications for intertidal mapping. Among 19 identified machine learning models: U-Net (convolutional architecture) and Random Forest (decision tree architecture) delivered the most accurate results. Model performance was highly dependent on environmental characteristics; in general, wider tidal ranges reduced accuracy, while substrates with clearer spectral signatures improved it. Machine learning approaches consistently outperformed models without machine learning in representing complex coastal morphology. Most studies used open-source Landsat or Sentinel imagery; however, commercial satellites and high-resolution video cameras, though less common, significantly improved model performance. Validation strategies based on high-resolution data from active sensors such as LiDAR (Light Detection and Ranging) proved essential to assess the accuracy and reliability of predictions, especially in heterogeneous environments. This comprehensive review highlights not only the most effective model architecture but also key research gaps, such as the limited exploration of temporal dynamics, the underrepresentation of macrotidal settings, and the challenges of transferring models across regions. Future research should prioritize convolution-based architectures trained with multi-source, high-resolution datasets, integrate UAV and active sensor data for robust validation, and explore scalable solutions to enhance the generalizability of machine learning models for intertidal zone mapping. • First systematic bibliometric review on intertidal topographic inversion using remote sensing and ML models to generate intertidal DEMs. • Sixty peer-reviewed studies analyzed using PRISMA, with bibliometric mapping and model performance comparison (e.g., RMSE, accuracy). • Random Forest and U-Net models outperformed others especially when applied to high-resolution imagery and validated with high resolution in-situ data. • Environmental and methodological factors - including tidal regime, spatial scale, sensor resolution and bottom type affect model performance. • Provides practical guidance on model selection , dataset design, and validation, while identifying gaps in transferability and uncertainty handling.