Bayesian precipitation inference from streamflow

Knowledge graph centered on Bayesian precipitation inference from streamflow with 8 nodes and 14 connections. Top connected: Crested Butte, Evaluating 3 decades of precipitation in the Upper, Weather Research and Forecasting Model, Multi-dataset precipitation comparison and validation, William Rudisill.

Method synopsis

Bayesian inference method combining temperature index snow accumulation modeling with bucket hydrology to estimate basin mean precipitation from streamflow observations and snow remote sensing data. Uses MCMC sampling to quantify parameter uncertainty.

Synthesized from method descriptions across 1 paper using this protocol.

Procedure from a recent peer-reviewed implementation

Steps below were extracted from the most recent peer-reviewed implementation of this protocol in the corpus — Evaluating 3 decades of precipitation in the Upper Colorado River basin from a high-resolution regional climate model (2023), Geoscientific Model Development. Implementations in other papers (listed below) may differ.

1. Set up Bayesian precipitation inference
   Implement temperature index snow accumulation model (SNOW-17) coupled with bucket hydrologic model based on FUSE framework. Use two state variables for soil moisture in top and bottom buckets (SM1, SM2) with maximum capacities SMmax. Calibrate SNOW-17 parameters for Butte SNOTEL site using precipitation undercatch term and error minimization algorithm.
   Quantities: 2 soil moisture state variables, 1 SNOTEL calibration site (Butte station 380)Duration: Calibration performed for available SNOTEL data periodConditions: Temperature and precipitation inputs from calibrated sources, missing data corrected using adjusted data from nearby sites
   Equipment: SNOW-17 snow model, FUSE hydrologic model framework, Butte SNOTEL data
2. Incorporate snow remote sensing data
   Use Airborne Snow Observatory (ASO) snow water equivalent measurements from water years 2018 and 2019. ASO produces 3m scale estimates of snow water equivalent by taking repeated lidar measurements of snow-covered and snow-free surfaces using energy balance modeling. Average ASO SWE from each elevation band and compare against SNOW-17 run in elevation bands.
   Quantities: 2 water years of ASO data (2018, 2019), 3-meter spatial resolutionDuration: Peak snow accumulation period for water years 2018-2019Conditions: High and low precipitation years respectively (>2000mm in 2019, ~1000mm in 2018)
   Equipment: Airborne Snow Observatory lidar, SNOW-17 model, Energy balance modeling
3. Perform Bayesian parameter inference
   Use PyMC3 library to implement Markov Chain Monte Carlo (MCMC) sampling with Metropolis algorithm. Sample from posterior distribution P(θ|d) = P(d|θ)P(θ) where θ is vector of model parameters and d is data. Use heteroscedastic error model with discharge growth coefficient σt = a1Qt + b1. Perform inference in two parts: time-invariant subsurface parameters, then meteorological parameters independently for each year.
   Quantities: ~15,000 MCMC samples, multiple chains for convergence diagnosticsDuration: MCMC sampling performed until convergence achievedConditions: Bayesian inference conditioned on model structure and streamflow data
   Equipment: PyMC3 library, Python programming environment, MCMC sampling algorithms

Typical Equipment

• SNOW-17 snow model
• FUSE hydrologic model framework
• Airborne Snow Observatory lidar
• PyMC3 library

Output Measurements

• posterior precipitation estimates
• parameter uncertainty
• model skill metrics

Papers Using This Protocol (1)

Places Where Used