Understanding the spatiotemporal dynamics of turbidity in large inland lakes is essential for sustainable water resource management and ecosystem conservation. This study investigates turbidity variability in Lake Tana by integrating MODIS-derived Forel–Ule Index (FUI), Diversity-II turbidity observations, and reanalysis-based physical drivers within a hybrid empirical and process-guided modeling framework. First, empirical univariate regression models (EURMs) were developed to estimate turbidity from FUI over the period 2000–2021. Among the tested formulations, the exponential model exhibited the best performance, capturing both moderate and extreme turbidity conditions with satisfactory predictive skill ( R 2 > 0.76). Validation against independent Diversity-II observations confirmed robust performance across multiple statistical metrics, including MAE, RMSE, NSE, IOA, and WI. However, comparison with the process-guided Physics-Informed Neural Network (PINN) revealed a systematic positive bias in the regression model, indicating structural limitations associated with single-predictor formulations and omitted physical controls. Spatiotemporal analyses show pronounced heterogeneity in turbidity distribution, with consistently higher values in the southern and western regions of Lake Tana, primarily driven by major tributary inflows and wind-induced resuspension, while the northern basin remains comparatively clearer. Seasonal patterns indicate elevated turbidity during the rainy season (mid-summer) and reduced levels during winter, reflecting strong hydro-meteorological control. Composite analyses further demonstrate that turbidity responds nonlinearly to physical forcing. Surface pressure influences turbidity primarily through atmospheric instability and runoff-driven inputs, while wind speed regulates spatial redistribution of suspended sediments through mixing and dispersion processes. Temperature effects are vertically structured, with surface warming promoting stratification and reduced turbidity, whereas bottom and total water column warming enhance turbidity through reduced decay rates and intensified internal nutrient cycling. The PINN results indicate that turbidity variability is governed by the combined influence of catchment inputs, atmospheric forcing, and internal lake processes, with wind speed and temperature stratification playing dominant roles, while surface pressure contributes marginally (2–3%). Overall, regression models capture first-order relationships with optical proxies, whereas the PINN framework provides a more physically consistent and interpretable representation of turbidity dynamics. This study highlights the value of integrating satellite-derived indices with physics-informed machine learning to improve long-term monitoring and understanding of inland water quality under increasing anthropogenic and climatic pressures.
Abegaz et al. (Wed,) studied this question.