Research notes
Climate-Hydrology Multilevel Analysis
Mapping how aridity, seasonality, and biome context modulate streamflow across hundreds of North American watersheds using multilevel and Bayesian techniques.
Objective
Build a stacked dataset of annual streamflow metrics (Qmean, Q05, Q95) and climate covariates, then disentangle shared vs. local drivers of hydrologic variability.
- Quantify long-run trends in discharge and seasonality at the basin level.
- Separate fixed biome effects from random intercept/slope variation.
- Probe interactions between aridity index, temperature seasonality, and flow.
Analytic design
For every HYDAT watershed I assembled a 20-year panel that pairs gridded precipitation/temperature surfaces with gauge readings. Data QC flagged outliers, filled short gaps, and harmonized coordinate systems.
- Random-intercept + random-slope models (lme4) to capture basin heterogeneity.
- Panel regressions with fixed effects (plm) as a robustness check.
- Hierarchical clustering on random-effects to surface shared risk profiles.
Modeling workflow
Fit Bayesian multilevel models (brms) for posterior distributions on slopes, then stress-tested with generalized additive models (mgcv) to capture residual nonlinearity.
- Decomposition of variance components using performance::variance_decomp.
- Posterior predictive checks to verify overlapping credible intervals.
- Scenario explorer that perturbs AI/SI inputs to score vulnerability.
Applications
Outputs inform water-resource agencies prioritizing storage upgrades and flow maintenance, while also serving as teaching material for multilevel modeling in environmental contexts.
- Targeted monitoring plans for basins with high slope variance.
- Decision briefs for conservation districts balancing ecological flow.
- Reusable code templates for future ecohydrology cohorts at UBC.