Oregon District Active Projects

Estimating Maximum Water Temperature in Small Streams in Western Oregon Using Neural Network Models

BACKGROUND

Elevated water temperature is the single most common water-quality violation for streams in Oregon. Hundreds of stream reaches exceed the State standard, which is 64-degrees Fahrenheit (17.8-degrees Celsius) based on a 7-day moving average maximum daily temperature for most streams with cold-water fisheries. In light of the growing list of threatened and endangered salmon, steelhead, and other species under the Endangered Species Act, the need to address these violations has become critical in the State of Oregon. Because each of these temperature violations must be addressed under the Total Maximum Daily Load provision of the Clean Water Act, and because of the large number of violations, the task is daunting. If agencies rely on individual data-collection and modeling studies of each reach or subbasin to determine the need and degree of remediation, the process will be slow and expensive. Consequently, a different, more regionalized method is needed for evaluating whether water temperatures are significantly elevated in a stream reach and whether remediation efforts are warranted.

The Oregon Department of Environmental Quality (ODEQ), as well as other State and Federal agencies (including EPA, NMFS, USFWS, and USGS), are evaluating a new method to assess whether a stream is considered "healthy." This method compares the biological assemblage in a stream to similar reference sites that are minimally affected by human activities. The concept is called the "Healthy Stream Standard." It is more of a site-specific evaluation of a stream that relies less on universally applied chemical and physical standards for water quality (such as dissolved oxygen, dissolved solids, and toxins) and more on letting the stream biology tell the story. If the biological population in a stream is appropriate for that ecoregion (in terms of density, diversity, and abundance of stress-intolerant species), then the stream will not be targeted for remediation under the TMDL provision of the Clean Water Act. This proposed Healthy Stream Standard recognizes the fact that some streams can be healthy even if its chemical and physical characteristics do not meet existing standards. It also recognizes that when a stream is not biologically healthy, site specific information is needed in order to assess remediation needs. One of the most critical pieces of site specific information is an estimate of maximum water temperature that would be expected at a site given "background" or "undisturbed" conditions.

Currently, the Oregon standard for water temperature is largely based on the biological requirements of cold-water fish rather than on the physical processes that control water temperature. The temperature standard is difficult to apply across a broad terrain (such as the entire Willamette Valley or all of western Oregon) because stream reaches differ markedly in elevation, geographical location, aspect, riparian vegetation species, stream order, etc. Each of these factors affects water temperature by directly influencing the degree of shading or the ambient climatic conditions (air temperature, humidity, and solar radiation). For example, maximum water temperatures would be expected to differ markedly between a wide, low altitude stream in southern Oregon as compared to a narrow, well-shaded mountain stream in northern Oregon. Streams in diverse settings behave very differently, and the temperature standards should reflect those differences.

It has become clear that all the involved agencies, such as ODEQ, the Oregon Watershed Enhancement Board, and Federal agencies (EPA, NMFS, USFWS, USFS, and BLM), need better estimates of maximum water temperatures that reflect "natural" or undisturbed conditions. The agencies are lacking information on what is physically achievable, given undisturbed conditions, for each stream reach. If this information were available, it would help agencies in their tasks of (1) setting reach-specific temperature standards that can be scientifically defended, (2) identifying and prioritizing stream reaches that are grossly out of compliance and most in need of remediation, and (3) establishing attainable temperature-reduction goals for reaches that have naturally high water temperatures.

The USGS will work with the involved State and Federal agencies to develop neural network models to aid in estimating natural, or background, maximum water temperatures in small western Oregon streams. A neural network model is a flexible mathematical structure capable of describing complex nonlinear relationships between input and output data sets that are typically found in natural systems. Neural network models are increasingly being used in environmental sciences, particularly for problems where the characteristics of the processes are difficult to describe using physically based equations.

OBJECTIVES:

1. Develop neural network models for estimating maximum water temperatures in small unregulated streams in western Oregon having either disturbed or undisturbed riparian conditions.

2. Provide error estimates for these maximum water temperatures so regulators and managers know the reliability of the method in the event it is used to establish site-specific standards.

3. Determine the expected frequency distribution of water temperatures at a site during the months June through September. (The frequency distribution data may be useful if a multitiered temperature standard is developed).

4. Work with agencies to use the above water-temperature statistics and methodology to modify or update water temperature standards, particularly as ODEQ continues to develop the Healthy Streams Standard calculating statistics in other unmonitored stream reaches) for unregulated and undisturbed stream reaches in western Oregon.

Output from the neural network models will include:

• Calibration results at each data-collection site;

• Models for calculating maximum water temperatures in small study area streams having either disturbed or undisturbed riparian conditions, given certain ancillary data;

• A method to calculate the error bounds associated with these maximum water temperatures;

• A method to estimate the frequency distribution (based on 30) years of climatic data) of the maximum water temperatures at the 200+ monitoring sites for the summer months;

• A method to estimate the magnitude and percentage of time a site could be violating (or complying with) the current 64-degree Fahrenheit 7-day moving average daily maximum temperature standard.

APPROACH

Calibration data for the neural network models will be obtained by deploying micro-thermistors in small unregulated streams in western Oregon during June through September 1999. Sites will generally be on first, second, or third order streams. Selection of stream locations will be closely coordinated with ODEQ, who will be deploying about 115 microthermistors in western Oregon during the same months as part of another program. (The two agencies will ensure that good coverage is obtained throughout western Oregon.) In 1999, ODEQ will collect data from coastal streams (including the Klamath Basin) and from the central Cascade Range (see table below). USGS will collect data from the rest of the Willamette River Basin, including its agricultural lowlands, foothills, and some of the eastern flowing streams from the Coast Range. ODEQ's large program outside of the Willamette Basin will allow the USGS to expand the neural network models throughout western Oregon. ODEQ collected data from 63 sites in 1998, 53 in the coastal basins and Klamath River Basin combined, and 10 in the Willamette River Basin and Sandy River Basin combined. Data from these sites will also be used during the calibration of the neural network models.

Of the 180 sites planned for monitoring in 1999, at least 65 will be selected as reference sites. Reference sites are selected from a pool of sites where the amount of human disturbance of the riparian vegetation is small. The remainder of the sites will be randomly selected from a larger pool of candidate sites. These random sites will have a variable degree of human disturbance of the riparian vegetation, ranging from negligible disturbance to sites with all riparian vegetation removed. In 1998, ODEQ's sampling included 18 reference sites and 45 random sites.

The following characteristics will be determined at each temperature monitoring stream site: topographic and vegetative shading, elevation, stream order, drainage area, channel slope, average width, average depth, width-depth ratio, geographical location (or ecoregion), aspect, and streamflow at the time temperature monitoring begins. This information will be collected during extensive field surveys when the microthermistors are deployed. Other information will be taken from GIS coverages, USGS orthophoto-quad maps, and other data bases held by other agencies. For any given stream cross section, riparian shading coefficients (for both banks) will be calculated using the Stream Network Temperature Model (SNTEMP) developed by the USGS Biological Resources Division, Fort Collins, Colorado. For input the model uses stream and vegetative characteristics measured in the field, data from orthophoto- and topographic-quad maps, stream reach aspect, latitude, and Julian date.

The USGS will work with ODEQ to ensure that protocols for temperature monitoring and stream-site characteristics-data collection are compatible between the two agencies. Protocols would cover microthermistor calibration techniques and methods to ensure that the placement of thermistors adequately represents the stream's cross-sectional water temperature.

Calibration of the neural network models will be done using data from the 1998 and 1999 periods. Input data will include the stream-site characteristics mentioned above plus climatic data for this period from all the weather stations in and around western Oregon. Most of the input data are available from National Weather Service and the U.S. Natural Resources Conservation Service's snow survey program. These climatic data include daily maximum and minimum air temperatures, humidity, and precipitation. Hourly short-wave solar radiation data are currently collected at eight or more locations in western Oregon through the Bureau of Reclamation AgriMet program, the University of Oregon, and the H.J. Andrews Experimental Forest. Many municipal airports also collect continuous cloud cover data, which might be used as input data. Monthly and annual air temperature and precipitation GIS coverages, generated by the Parameter-Elevation Regressions on Independent Slopes Model (PRISM) from the Oregon Climate Service, will be used for data interpolation between climate stations and the water-temperature -monitoring field sites.

When the neural network models are calibrated, they will be used with 30 years of climatic data to generate robust water-temperature statistics and frequency distributions for the 200+ sites. The use of 30 years of climatic data will provide a much larger data set with which to evaluate the predicted frequency distribution of daily maximum water temperatures at these sites. The past 30 years of climatic data includes both warmer- and cooler-than-normal summers.

REPORTS

WRIR 02-4218. Estimating Water Temperatures in Small Streams in Western Oregon Using Neural Network Models, by John C. Risley, Edwin A. Roehl, Jr., and Paul A. Conrads. (4/03) Abstract and link to online version