Multi-scale landscape controls on water quality in high-Andean watersheds: A machine learning approach

Rapid land-use conversion in tropical high-mountain environments threatens downstream water security, yet the scale-dependent mechanisms linking landscape patterns to water quality remain insufficiently resolved. This study integrates multi-scale spatial analysis with Random Forest machine learning to evaluate the influence of land use, topography, and remotely sensed biophysical indicators on physicochemical and microbiological water quality in the high-Andean Tomebamba and Yanuncay watersheds (Ecuador). A dataset of 86 observations from 43 monitoring sites was analyzed across contrasting hydrological seasons and four spatial configurations ranging from local riparian buffers to the full upstream catchment. Exploratory modeling identified scale-dependent patterns in landscape associations. Microbiological contamination showed strongest correlations with proximal anthropogenic activities within the 100-m riparian corridor (R 2 ≈ 0.28), whereas riparian habitat integrity (QBR index) and dissolved solutes responded more strongly to cumulative pressures at the catchment scale (R 2 ≈ 0.32). Conversely, dissolved nutrients exhibited limited predictability (R 2 < 0.15), highlighting a decoupling where localized point-source inputs override diffuse landscape signals. Seasonal analysis indicated dual hydro-climatic influences: wet periods mobilize particulate and microbial contaminants through runoff-driven wash-off processes, while dry periods concentrate dissolved pollutants under reduced dilution capacity. Random Forest screening identified potential non-linear associations and threshold responses, suggesting landscape tipping points that warrant validation through process-based modeling. Remotely sensed spectral indicators provided complementary functional information to static land-cover classifications, enhancing the detection of vegetation and moisture dynamics influencing pollutant transport. Collectively, these findings support a hierarchical watershed governance framework that prioritizes riparian buffer protection alongside catchment-scale land-use planning to sustain hydro-ecological resilience in Andean headwaters.