An Automated Morphometric Approach for Global Lentic and Lotic Classification of Inland Waters

The distinction of lentic (still) and lotic (flowing) inland waters is fundamental for understanding ecosystem functions, hydrodynamic behavior, nutrient cycling, and biogeochemical exchanges across terrestrial and aquatic interfaces. These systems influence carbon storage, sediment balance, biodiversity support, water residence time, and regional climate regulation, making accurate separation essential for large-scale hydrological assessments. However, existing classification approaches often depend on site-specific information, manual interpretation, or large training datasets, and commonly struggle to classify inland waters smaller than 3 hectares due to resolution limitations and insufficient annotated samples. This work presents an Automated Data Efficient Morphometric Approach (ADEMA) for classifying inland waters down to 0.09 ha (single LANDSAT pixel) using multi-dimensional morphometric interpretations derived using the Global Surface Maximum Extent (GSMW) dataset. The approach was trained and validated using 17,391 expert-labeled samples from 66 geographically diverse locations across multiple climate zones, varied topographies, and hydrological regimes. Further, ADEMA was benchmarked against optimized machine learning, deep learning, and global classification products. Results showed that across all size classes (small: <10 ha, medium:10-1,000 ha, and large: >1,000 ha), ADEMA provided comparable F1 scores (94%) to machine and deep learning models with minimal omission (2%), demonstrating its ability to achieve reliable classification with significantly lower computational and data requirements. A multi-decadal evaluation from 1991 to 2021 showed stable accuracy, highlighting temporal ADEMA’s robustness (F1 score = 92%). When compared to global classification products, ADEMA achieved substantially higher accuracy (average F1 score: 97% vs. 62%), especially for small and medium inland waters that are often underrepresented in global datasets. The method offers a data-efficient and automated solution suitable for regional to global hydrography. However, the framework excludes inland waters >10,000 ha to maintain computational feasibility, limiting coverage of large systems. Single-pixel detections (~0.09 ha) are less reliable due to noise, vegetation, and GSMW uncertainty, with accuracy stabilizing above ~0.5 ha. With further advancements, ADEMA could improve global open-water inventories, guide conservation strategies, and strengthen our understanding of how small inland waters collectively shape hydrology and ecosystem resilience across different environments.