Hydrologic Modeling for the Miller and Walker Creeks Basin Plan
A hydrologic model was developed as part of the analysis for the Miller and Walker Creeks Basin Plan in the early 2000s. The model allowed stream flows to be predicted given assumptions about land cover and rainfall. It allowed the project team to answer "what if" questions - such questions as "What would stream flows be like if this were all forest?" or "How would the stream behave if the cities and the county required a certain uniform level of detention?"
The first step in hydrologic modeling was to develop the model using HSPF (Hydrologic Simulation Program - Fortran). HSPF is a continuous time-series model that is frequently used to simulate hydrologic conditions in streams. The relationship between rainfall and stream flow was determined by assuming certain land cover and geology characteristics. In other words, assumptions about what happened to the rain when it landed on the ground were made - how much of it runs off the land and enters the stream and how much of it infiltrates (sinks into the ground)? These assumptions were made based on the geology of the basin and the extent and type of development that has occurred. Assumptions about the geology were based on geology mapping done by the United States Geological Survey and other local studies. Assumptions about existing development in the basin were based on 1995 satellite mapping of the basin, and relationships between development type, impervious surface coverage, and runoff based on past modeling efforts. In addition, the comprehensive plans for the cities of Burien, Normandy Park, and SeaTac were examined. The Comprehensive Stormwater Management Plan for the Port of Seattle was also consulted. After initial modeling parameters were developed, the model was calibrated using stream flow data collected at flow gauges in the streams and rainfall data collected at precipitation gauges in the watersheds.
In Miller Creek, the model was calibrated with one precipitation gauge (42U) and two flow gauges, one near the confluence with Walker Creek (42A) and one at the outflow of the Lake Reba regional detention facility (42B). In Walker Creek, records were used from one precipitation gauge (42U) and two flow gauges, one near the confluence with Miller Creek (42E) and one just downstream of the headwater wetland (42C). (View latest stream flow gauge information.)
The table below summarizes the gauge information.
Gauge Records Used for Calibration
42U (10/94 - 9/97) At Lake Reba
42U (10/94 - 9/97) At Lake Reba
42A (10/94 - 9/97) Just upstream of confluence with Walker
42B (10/94 - 9/97) Miller Creek Regional Detention Facility outflow
42E (10/94 - 9/97) Just upstream of confluence with Miller
42C (10/94 - 9/97) Just downstream of the headwater wetland at Des Moines Memorial Drive
The calibration procedure consisted of making adjustments to the variables in the model to match the predicted model flows to recorded stream flows. The degree to which the model agreed with the recorded stream flows was a measure of the ability of the model to successfully predict stream flows based on hypothetical rainfall events. This level of agreement was calculated and assigned a numeric value called a correlation coefficient. For Miller Creek, the correlation coefficient for the mean daily flow was 75 percent (0.75) at the Miller Creek Regional Detention Facility (MCRDF). This means that the model can correctly account for 75 percent of the variability observed at that location. While this may not seem as high as desirable, it is a fairly typical degree of a model's predictive power in the natural environment. In the Miller Creek example, it means that the modeled results for mean daily flow may be inaccurate by approximately 25 percent at the MCRDF. The largest source of error in the model is the assumption that rainfall measured at one location, the rainfall gauge, falls uniformly across the basin. The only way to reduce the error associated with this assumption is to install a greater number of precipitation gauges throughout the basin. The degree of variability explained by the models, as represented by their correlation coefficients, is shown in the table below.
Degree of Variability Explained by the Models (based on correlation coefficients)
Mean Daily Flow
75% (MCRDF), 86% (Mouth)*
79% (DMMD), 75% (Mouth)*
Daily Peak-Hour Flow
67% (MCRDF), 75% (Mouth)*
70% (DMMD), 78% (Mouth)*
* Two values are presented because there are 2 stream flow gauges used in the calibration. For Miller Creek the flow gauges are located at the Miller Creek Regional Detention Facility (MCRDF) and at the mouth. For Walker Creek, the flow gauges are located at Des Moines Memorial Drive (DMMD) and at the mouth.
After development and calibration, the program was used to assess the response of the stream to varying amounts of precipitation under various land covers. For each model run, 50 years of rainfall was allowed to "rain" on the basins. The rainfall assumed during model runs was based on precipitation measured at Seattle-Tacoma International Airport from 1949 to 1998. The resulting data were used to generate curves showing peak flows (high flows that occur during storms) and low flows [minimum flows that occur during times of no precipitation when streams rely solely on base flow (i.e., ground water)]. In addition, calculations of erosive work were performed for the Miller Creek basin, as it has experienced historically high degrees of erosion. The same historic rainfall was used for all model runs so that a consistent, long-term basin response to precipitation could be determined.
To assess the impact of future development on the basin, assumptions were made concerning its likelihood of occurrence. Initially, it was assumed that full development would occur consistent with existing zoning. This assumption, however, lead to an enormous increase in impervious surface area that appeared inconsistent with existing development patterns and rates. Instead, an analysis of development potential was done by examining the relationship between land values and improvement values. Parcels for which the improvement value was less than or equal to the land value were identified as being likely to develop in the near future. Those parcels, the "red parcels," were assumed to be developed in subsequent model runs (see Figure MW1). Although there is no way to accurately predict exactly which parcels will be developed and when, this designation of red parcels seems to be a much more realistic approach than assuming that complete re-development of the basin under existing zoning will occur in the next 20 years.
Consultants hired by the Airport Communities Coalition reviewed the development, calibration, and early use of the model. They provided a favorable review and saw no barriers to using the model to more fully assess basin responses to potential future scenarios.
A number of model runs for each of the basins are presented below. Each figure is accompanied by a summary of what the modeling scenario shows. A key describing the assumptions of each run is also included