Integrating geospatial attributes into a non-stationary transformer model for improved stream chemistry predictions
Year: 2026
Type: preprint
Abstract: Key Points:• We propose an NST-Geo model architecture for several days ahead stream solute concentration predictions.• The NST-Geo model links spatial heterogeneity to temporal solute dynamics and identifies influential regions within the catchment.• Vegetation dynamics (NDVI) exert the strongest control on predictive skill across solutes, highlighting vegetation-solute coupling effects.
Language: English
Subfield: Environmental Engineering
Field: Environmental Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7156354792
