Hierarchical stress sensitive neuro graph learning for use specific surface water quality assessment in Baitarani Brahmani River

ABSTRACT

The figure shows the graphical abstract of the proposed approach with the flow of the document represented in a pictorial format.

Surface water assessment is key to ecosystem protection, irrigation applicability, and appropriate management, yet traditional water quality index (WQI) techniques miss nonlinear interactions, spatial variation, and temporal effects. Hence, a hierarchical stress-sensitive neuro graph learning framework (HSS-NGF) is proposed for early, interpretative, and use-specific water quality evaluation. Initially, scalar indices average various physicochemical variables and conceal sub-lethal effects of contaminants, allowing localized degradation to persist, resulting in a latent discovery of progressive degradation. Thus, a feature-ordinal variational multi-task neuro-fuzzy (FOV-MTNF) is presented in the first hidden layer of the graph neural network (GNN) to maintain parameter-level stress sensitivity, learn the trajectories of progressive degradation, and produce stress embeddings in the latent feature. Moreover, tributary inflows establish mixing areas with contrasting water quality, and localized hotspots to synergically interacting stressors lead to more gradual erosion of the dual-use water utility. So, graph-weighted regression convolutional Gaussian tree is introduced in the second hidden layer of the GNN to learn spatial dependencies, to obtain upstream–downstream propagation, anomalous nodes, and interpretable water quality rules to protect the environment and irrigation suitability. The effectiveness of the HSS-NGF framework has been shown through the experimental results with better accuracy, precision, and explainability.