Stream Temperature: Model & metrics for planning & management

Daniel Hocking, Ben Letcher, Kyle O'Neil
April 7, 2015

Conte Anadromous Fish Research Center, USGS

Modeling Daily Stream Temperature

Objective Model daily stream temperature over broad spatial and temporal extents from daily climate, land-use, and landscape data

Rationale

  • Most biota in streams are temperature dependent including trout & salamanders
  • Stream chemistry is also influenced by temperature
  • Policy and management decisions are often on the basis of temperature as a surrogate for biotic and abiotic concerns
  • Summary metrics of interest (mean summer temp) can be derived from daily temperature models

Stream Temperature Data

  • State and Federal Agencies, Academic Researchers, and NGOs
  • Current: New England States
  • Upload/Access Source: www.ecosheds.org
  • Original Temporal Resolution: 5-60 minutes
  • QA/QC

Climate Data

Daymet: www.daymet.ornl.gov

Model daily climate variables:

  • Tmax: Maximum daily temperature
  • Tmin: Minimum daily temperature
  • Prcp: Total daily precipitation
  • RH: Relative humidity
  • Swe: Snow-water equivalent
  • Srad: Shortwave solar radiation
  • Dayl: Day length

Climate Data

Daymet: www.daymet.ornl.gov

Resolution

  • Temporal: Daily
  • Duration: 1980-2013
  • Spatial: 1 km x 1 km
  • Extent: US, Mexico, Canada (south of 52 degrees North)

Landuse & Landscape Data

  • National Hydrography Dataset
    • High Resolution Flowlines (1:24,000-scale or better)
    • Clipped to 0.75 sq km
  • USGS Watershed Boundary Dataset (HUCs)
  • DEM from NALCC
  • USDA NRCS: SSURGO (soil and geology)
  • National Land Cover Database (NALCD)
  • US FWS National Wetlands Inventory

Problem with Ice

plot of chunk unnamed-chunk-2

plot of chunk unnamed-chunk-3

Solution: Air-Water Synchronization

\[ Index_{sync}=\frac{(T_{w}-T_{a})}{(T_{w}+0.00001)} \]

alt text

Solution: Air-Water Synchronization

\[ Index_{sync}=\frac{(T_{w}-T_{a})}{(T_{w}+0.00001)} \]

plot of chunk unnamed-chunk-4

Model Fitting

Validation: 30% of stream reaches held out at random

Model fitting (70% calibration/training data)

  • Bayesian Inference
  • MCMC (Gibbs Sampler) using JAGS
  • Vague Priors
  • Allow for correlation among coefficients

Results

            parameter    mean     sd sig
1           Intercept 17.7035 0.2486   *
2                AirT  2.1721 0.1472   *
3          7-day AirT  1.5792 0.1362   *
4         Development  0.1709 0.0559   *
5         Agriculture -0.0583 0.0665    
6    Impoundment Area  0.3678 0.0660   *
7  AirT x Impoundment -0.0288 0.0229    
8       AirT x Forest -0.0176 0.0265    
9   AirT x Prcp2 x DA -0.0036 0.0016   *
10 AirT x prcp30 x DA -0.0020 0.1666    
11                Day  0.0506 0.1070    
12              Day^2 -0.5141 0.0887   *
13              Day^3 -0.0834 0.0778    
14                AR1  0.7696 0.0073   *

Results

plot of chunk unnamed-chunk-6

Model Fit

  • Good fit to calibration data
  • No apparent bias or autocorrelation
  • Root mean squared error (RMSE) = 0.60

plot of chunk unnamed-chunk-7

Model Validation

  • Excellent predictive power
  • RMSE = 0.74 for validation data

plot of chunk unnamed-chunk-8

Model Validation

  • Excellent predictive power
  • RMSE = 0.74 for validation data

plot of chunk unnamed-chunk-9

Identifying Problems

Dam

Predictions

Long_Prediction.png

Predictions

Short Prediction

Derived Metrics

Jennifer Jacobs et al. http://www.epa.gov/region1/neaeb2010/pdfs/6A-SummerWaterTemperaturesNHMA.pdf

Thermal_Regime

Derived Metrics for Decision Support

www.EcoSHEDS.org

  • Mean July Temp
  • Mean Summer Temp
  • Mean Days per year > 18 C

SHEDS

July Mean Temperature: Northeast

Mean July Temp NENY

July Mean Temperature: Deerfield River

Mean July Temp Deerfield

Benefits of Approach

Hierarchical (Mixed Effects) Model

  • Accounts for spatial correlation
  • Accounts for temporal autocorrelation
  • Able to use time series of widely varying lengths
  • Easily incorporate disjunct time series from a location
  • Partitioning of uncertainty from variation in time and space

Benefits of Approach

Bayesian Estimation with JAGS

  • Ease of coding and estimating complicated models
  • Naturally quantify variability and uncertainty across all parameters

Future Directions

  • Other variables or Interactions (Day Length, Solar Radiation)
  • Riparian Forest Cover
  • Derived Metrics: State Cold Water Regulatory Classifications
  • Climate Sensitivity
  • Climate Change Projections

Questions

Thanks