Estimating Daily Streamlow at Ungaged Sites in NH: A Modeling Update Neil M. Fennessey HYSR/Hydrologic Services HYSRConsult@gmail.com NH Department of Environmental Services Concord, NH March 29, 2018
Estimating Daily Flows at Ungaged Sites NH DES needs to determine PISFs at ungaged sites in presently and future designated rivers NHDES needs a reliable way to estimate daily streamflow time series at these ungaged sites The QPPQ Transform method is one approach Conceived and developed by Fennessey (1994) Has been used in the northeast and central US, Canada and South Africa
Other Ways to Estimate Daily Flows at Ungaged Sites: Pluses & Minuses Watershed Area Ratio (WAR) method Very simple to use Assumes that all watersheds are exactly alike except for the size of the watershed Rainfall-Runoff Models (HSPF, Sacramento Soil Moisture Accounting Model) Physically based but parameter-intensive Used by NOAA for flood forecasting Requires long stream gage period-of-record for site calibration and validation
Does USGS Like QPPQ Transform? The QPPQ Transform method has been adopted or is being considered by several US Geological Survey district offices: Massachusetts and Rhode Island USGS Pennsylvania USGS Iowa USGS Minnesota USGS New York USGS New Hampshire and Vermont USGS
Has the QPPQ Been Extensively Tested? The QPPQ Transform method was extensively tested by the USGS and bested 17 alternative methods for estimating daily flow time series at 182 gaged watershed sites located in West Virginia Tennessee North Carolina South Carolina Georgia Alabama Florida
Have States or Agencies Adopted the QPPQ? The QPPQ Transform method has been adopted or is being considered by several state agencies: Massachusetts EOEEA Massachusetts Water Management Act program Sustainable Water Management Initiative The Delaware River Basin Authority Maryland Dept. of the Environment New Hampshire Vermont
Examples of Local Applications The QPPQ Transform method has been used for Many surface water reservoir system assessments in Massachusetts and Connecticut River basin and watershed studies in Massachusetts and Connecticut Systems analysis study of the Connecticut River watershed Complex litigation in Connecticut
The QPPQ Transform Method
Four Steps to the QPPQ Transform Method 1. Chose a USGS Streamgage Site for the time series: Q gage (t) 2. Use that time series to construct a streamflow duration curve (FDC): Q gage (p) 3. Construct a FDC at the ungaged site using its watershed’s soil characteristics, climate, and topography and a regional FDC model: Q ungaged (p) 4. Generate a streamflow time series at the ungaged site: Q ungaged (t)
Minnesota USGS 2016
Use Q gage (t) to Construct ‘observed’ FDC Q gage (p ) Rank sort Q gage (t) “n” days from largest to smallest Q (1) ≥ Q (2) ≥ Q (3) … Q (n-2) ≥ Q (n-1) ≥ Q (n) “n” is number of days in the period of record Use “plotting position” formula to estimate exceedance probability p=P[ Q≥q ] “i” is the rank-order i=1,n POR 30 years x 365 = 10,950 days P[Q≥Q (n) ]x100 =10,950/10951 x 100 = 99.991
Step 1: Q gage (t) and Step 2: Q gage (p) Common factor between the flow date, time, and the FDC probability, p, is Q = a 3-D relationship Knowing Q gage (t) & p(Q gage ) => p(t)
Update Step 3: Regional FDC Model HYSR study focused on update of the Fennessey (1994) regional FDC model: Q ungaged (p) Constructed a streamgage network of USGS gaged watersheds found in both the HCDN (1992) and GAGES-II (2010) national streamgage networks All gages have at least 20 years POR in 1950- 1990 base-period HYSR prelim. network of 142 gaged watersheds in ME, NH, VT, MA, RI, CT, NH, PA, and NJ
Step 3: Regional FDC Model Need a FDC for the ungaged site without historic daily streamflow data
Assess Alternative Probability Function Candidates for Regional FDC Model Statistically analyze each of the 142 gage site POR to estimate the L-moments (linear probability weighted moments) Construct a L-kurtosis Diagram to compare alternative probability functions Conduct Discordancy analysis to determine the “region”; eliminate sites “outside” the core region 8 NJ gage sites and 1 PA site near NJ eliminated Final HYSR network: 133 USGS stream gage sites, 20-year POR in 1950-1990 base period
Final Probability Function Candidates Three-parameter Generalized Pareto (GPA) Three-parameter Log Normal (LN3) Fennessey (1994) conducted extensive “goodness of fit” tests on both and chose the GPA. Use L-moments to calculate the 3 parameters: ζ , α , κ for each of the 133 HYSR network sites Test the “fitted” FDC against the “observed” FDC
Baker River, near Rumney, NH
Construct Regional FDC Model Gather HCDN and GAGES-II network watershed specific soil, climate, and topography data for each of the 133 USGS gage sites in the HYSR network Use each of the 3 GPA parameters statistically determined for each of 133 HYSR watersheds from the streamgage data as dependent variables Use OLS multivariate regression and each watershed’s specific soil, climate, and topography data as candidate independent variables Keep only those particular soil, climate, and topographic variables that are “statistically significant” (pass various tests)
To construct a FDC at the ungaged site, the analyst must determine ten soil, climate, and topography characteristics for that watershed: Watershed area (mi 2 ) Avg annual precipitation (in/yr) Avg annual temperature ( o F) Potential maximum soil moisture retention (in) Percent of watershed area covered by HSG C soil (%) Avg watershed elevation (ft) Avg watershed aspect (degrees) 0-360 O Percent of watershed area covered by lakes, ponds and reservoirs (%) Percent of watershed area covered by impervious surface (%) Slope of main stream channel (ft/mile)
Baker River, near Rumney, NH
Compare FDCs and Assess “Goodness-of-Fit” Construct “observed,” “fitted,” and regional “model” FDCs at eleven NH USGS gages sites, watershed areas range: 12.1 mi 2 – 622 mi 2 Graph all three FDCs for visual comparison Estimate the Bias and RMSE between the “observed” FDC and the “fitted” FDC to see how suitable the GPA is as basis for the Reginal FDC Model Estimate the Bias and RMSE between the “observed” FDC and the regional “model” FDC to see how well it compared with the “fitted” FDC.
Regional FDC Model Development Conclusions GPA is a good choice for regional FDC model “Fitted” FDC is good compared to “obs.” FDC Parsimonious: only 3 parameters to estimate Regional FDC model regression equations Ten watershed variables are estimated from GIS data layers and USGS topographic maps “Model” compares well to “fitted” FDC Regional FDC model is continuous and monotonic Because E[P] and E[T] are independent variables, can do assessment of potential impacts on streamflow due to climate change
QPPQ Transform Steps 1 - 3
Step 4: Generate Q ungaged (t) Assume for Steps 2 and 3: p(Q ungaged ) = p(Q gaged ) Recall 3-D relationship between t, Q and p => p(t) Use regional FDC model with p(t) instead of just p
Summary The QPPQ Transform method was conceived and developed by Fennessey (1994) The method has been widely adopted The Step 3 regional FDC model was updated in the present HYSR study The goals of the QPPQ Transform remain the same: Provide method to generate a quality time series of daily streamflow data at ungaged sites The analyst can choose to do whatever they want with the generated time series, Q ungaged (t)
Questions and Comments ? Neil M. Fennessey Hydrologic Services 49 School Street South Dartmouth, MA 02748 HYSRConsult@gmail.com
Recommend
More recommend
