Satellite imagery and fuzzy neural networks enable accurate estimation of key water quality indicators in large reservoirs.
The Complexity of Monitoring Water Quality in Reservoirs
Water quality in large artificial reservoirs represents a complex challenge, where biological, chemical, and environmental factors interact in nonlinear and often hard-to-observe ways. Algal proliferation phenomena, in particular, can compromise both drinking and recreational uses of water resources, making it necessary to adopt monitoring strategies that are more systematic, less invasive, and more predictive than traditional approaches based solely on in situ sampling.
Combining Satellite Remote Sensing and Fuzzy AI
To address this challenge, an approach is developed that combines satellite remote sensing with fuzzy neural network models, aiming to estimate key water quality parameters such as chlorophyll-a, algal concentration, and turbidity. The core idea is to exploit the information contained in high-resolution multispectral satellite images and transform it into a numerical representation suitable for processing by machine learning models.
Wavelet-Based Feature Extraction from Multispectral Images
Satellite images are first selected at locations corresponding to reservoir sampling points and then preprocessed using a wavelet transformation. This step decomposes the spectral information into components that capture local variations and meaningful spatial structures, while simultaneously reducing noise and the dimensionality of the problem. The resulting components are concatenated into a compact feature vector, which serves as input for the predictive models.
Neuro-Fuzzy Models for Nonlinear Environmental Estimation
For the estimation phase, neuro-fuzzy systems and radial basis function networks are adopted, as they are capable of modeling strongly nonlinear relationships while incorporating a form of gradual reasoning typical of fuzzy logic. These models learn the mapping between the spectral response observed by the satellite and the measured physicochemical parameters, achieving generalization even in the presence of seasonal dynamics and pronounced environmental variability.
Validation on a Real Reservoir with Algal Blooms
The approach is validated on a real reservoir affected by recurrent algal bloom events, using data collected over multiple years and across different locations within the lake. The results show good predictive performance for all considered parameters, with particularly strong accuracy in turbidity estimation and stable behavior even for more challenging biological variables.
Toward Scalable and Continuous Water Quality Surveillance
Overall, this work demonstrates how the integration of satellite observation and fuzzy artificial intelligence can provide an effective tool for continuous water quality monitoring. An approach that reduces reliance on costly and discontinuous measurement campaigns, and opens the door to more responsive, scalable environmental surveillance systems capable of supporting informed water resource management decisions.
Autori
H. A. N. Silva, A. Rosato, M. Panella
