An Automated Data Efficient Morphometric Approach to Define Global Lentic and Lotic Inland Waters

Abstract Defining lentic and lotic system types is critical for understanding hydrological, ecological, and biochemical processes. Traditional classification methods rely on non‐generalizable site‐specific parameters such as visual characteristics, historical inventory, and residence time. While machine learning and deep learning models address these challenges to some extent, they are limited by high data requirements, unverified training data sets, computational demands, and the inability to accurately detect inland waters smaller than 3 ha. To address this gap, this study introduces a novel Automated Data Efficient Morphometric Approach (ADEMA) that classifies inland waters into lentic and lotic system types globally up to 0.09 ha (33 times smaller than previous studies) using multi‐dimensional morphometric interpretations. ADEMA was developed and validated using 17,391 expert‐labeled inland waters spanning 66 globally diverse locations and compared against state‐of‐the‐art, comprehensively optimized machine learning, deep learning, and global models. Results show ADEMA equivalently performed to the machine learning and deep learning models, achieving F 1 scores of 92%, 95%, and 71% in small, medium, and large inland waters, respectively. Across 17,391 expert‐labeled samples, ADEMA maintained a high performance with a precision of 89%, a recall of 99%, and an F 1 score of 94%. Analysis across four decadal intervals (1991–2021) demonstrated ADEMA's temporal invariance, with consistently high F 1 scores (90%–93%) and negligible omission errors (0%–2%). Further, ADEMA surpassed global classification products (average F 1 score: 97% vs. 62%). These findings emphasize ADEMA's potential for accurately classifying global inland waters into lentic and lotic system types.