Goals


Streamflow estimation is essential for many monitoring and assessment activities. However, sampling sites are rarely colocated with USGS streamflow gages. There are many techniques, but this novel approach pairs streamflow magnitude (hydrologic condition) with the rate of change. This is an attempt to characterize 1) hydrologic condition, including when anomalous conditions occur (e.g. very low or high flows) and 2) hydrograph variability (e.g. hydrologic conditions may appear ‘normal’, but a sample was taken during a rapidly changing hydrograph). This second example typically takes place when a summer thunderstorm causes rapid increases in flow (and potentially sediment transport), but due to low flows preceding the rain event, the hydrologic condition would not indicate high flow conditions.


This model leverages flow duration curves (FDCs) for 146 active USGS gages with minimally altered hydrology that have a >= 10 year period of record between 1993-2022. Hydrologic condition at an ungaged site is interpolated using 3 geographically closest gages with the most similar drainage sizes. This interpolation approach is intended to reduce the influence of any individual gage, which becomes essential during periods of unevenly distributed of intense precipitation.


A workflow was developed that will allow reproducible hydrologic condition interpolation at ungaged sites of interest. This is currently coded based on a query for samples/sites/projects of interest from the Sample Information System (SIS), but with a few modifications could be used with any dataset where latitude, longitude, date, and time are included.


A quality assurance (QA) exercise was also run to examine accuracy of the model.


Details of the model are contained in subsequent tabs:

  1. Workflow - A graphical overview of the development and application
  2. Development - Technical details of model development
  3. Application - A step-by-step example of model application with sample data
  4. QA - quality assurance (QA) exercise was run to examine the performance of the hydrologic condition interpolation model


Back to top

Workflow


Back to top

Development


Step 1

Identify all USGS streamflow gages in HUC8s that intersect PA

  • Must have ≥ 10 year period of record in 1993 – 2022 time period
  • Must still be active

A total of 465 gages were identified with HUC8s that intersect the PA boundary. Of these, 332 gages had a sufficient period of record and were still active. These gages represent the population of gages that could potentially serve as reference gages to interpolate flow at ungaged locations.


Step 2

Flow duration curves

Flow duration curves (FDCs) were created for each of the 332 gages identified in the first step. FDCs are cumulative frequency curves that shows the percent of time specified discharges were equaled or exceeded during a given period. FDCs combine in one curve the flow characteristics of a stream throughout the range of discharge, without regard to the sequence of occurrence (Searcy 1969).

Both annual and seasonal (May - September) FDCs were created. Due to the inclusion of higher flow months in the annual FDCs, streamflows (in cfs) were higher than seasonal FDCs for the same flow percentile. For example, the 50th percentile (P50) for Penns Creek was 305 cfs for the annual FDC and 210 cfs for the seasonal FDC (Figure 1). This P50 indicates median flow, where flows were equaled or exceeded 50% of the time for each given period. The annual P90 for Penns Creek was 1040 cfs for the annual FDC and 734 cfs for the seasonal FDC. This indicates that streamflow exceeded 1040 cfs 10% of the time annually, and exceeded 734 cfs 10% of the time during May - September (Figure 1).

Figure 1. Annual and seasonal flow duration curves (FDCs) for Penns Creek and the West Branch Lackawaxen River. Streamflow in cubic feet per second (cfs) is on the vertical axis and flow percentile is on the horizontal axis.


Step 3

Identify reference gages with minimally altered hydrology

Many of the gages where streamflow data were available represent altered hydrology due to diversions, dam operations, or other anthropogenic activities. These gages would not be suitable for use to estimate hydrologic condition at ungaged locations, as they do not represent natural hydrology. To determine gages with minimally altered hydrology, the GAGES-II: Geospatial Attributes of Gages for Evaluating Streamflow publication was utilized (Falcone 2011). Two metrics were used: 1) the hydrologic disturbance index score (HYDRO_DISTURB_INDX) and 2) the density of major dams in the upstream watershed (MAJ_DDENS_2009). A filter for HYDRO_DISTURB_INDX was set at <= 20 and MAJ_DDENS_2009 <= 0.1. These filters removed 186 gages due to altered hydrology and elevated density of dams upstream, and retained 146 gages with minimally altered hydrology.

Only one gage with minimally altered hydrology exceeded 1,000 mi2 in drainage area (West Branch Susquehanna River at Renovo - 2,975 mi2). The rest of the minimally altered gages ranged in size from 1.5 - 718 mi2. Gages with altered hydrology were more prevalent in larger drainages, with drainage areas ranging from 2.5 - 18,300 mi2 (Figure 2). As a result, this method is most accurately applied at ungaged sites with drainage areas <1,000mi2.

The 146 gages with minimally altered hydrology have relatively good spatial coverage, with noticeable gaps in western and northeastern PA (Figure 3).

Figure 2. Histogram of drainage area of USGS gages with altered (top) and minimally altered (bottom) hydrology.


Figure 3. Interactive map of USGS gages with minimally altered hydrology (n=146). Points are symbolized by drainage area (mi2). Clicking points opens a pop-up box with additional information on each gage.


Step 4

Create a lookup table based on USGS flow regime definitions

The final development step was to create a lookup table to reflect USGS flow regime definitions. This lookup table assigned flow regime definitions to flow percentile ranges described below, and visible which align with the USGS waterwatch website.


Table 1. Categorical USGS flow regime definitions. Background colors and definitions align with the USGS waterwatch website.
Flow Regime Definition Flow Percentile
Much Below Normal 0 - 9
Below Normal 10 - 24
Normal 25 - 75
Above Normal 76 - 90
Much Above Normal 91 - 100


Literature Cited

Falcone, J. A. 2011. GAGES-II: Geospatial Attributes of Gages for Evaluating Streamflow. U.S. Geological Survey. Reston, VA.

Searcy, J. K. 1969. Manual of Hydrology: Part 2. Low-Flow Techniques. U.S. Geological Survey. Washington, D.C.


Back to top

Application

After gages with minimally altered hydrology were identified and FDCs were created, it is now possible to interpolate hydrologic condition at ungaged locations. R code has been developed that makes model application reproducible with only minor edits needed.


Step 1

Pull data where hydrologic condition is needed

  • DEP users should pull samples/monitoring points/projects of interest from SIS
  • Pull data from SIS:
    • Samples from DWH_DBA:DWH_ALL_SAMPLE_RESULTS
    • NHD data from DWH_DBA:DWH_NHD
    • Join the results and make a dataframe of distinct combination of COMID, location (lat, long required), date, and time
    • Samples must be snapped to the NHD flowline


Step 2

Pull in drainage area of sites

  • Make a vector of unique COMIDs from the above distinct dataframe
  • Pull drainage area for each COMID from the NHDplus dataset
    • Streams in eastern/northeastern PA often do not join to the DEP GDC.NHDFlowline
    • This requires a manual process because drainage area is required
    • See examples on how to manually populate in the R code


Step 3

Calculate distance from each sampling location to each reference gage

  • Keep only the 5 closest gages to each sampling location
  • Calculate a weight for each gage based on similarity in drainage area (0-1)
  • Keep the gages with the top 3 weights
  • Recalculate weights for the top 3 gages (0-1 scale, sums to 1)

A visual representation of gage selection is shown below in Figure 4.


Figure 4. Results of proximal gage selection. The final three reference gages (circles) are shown, symbolized by weight (0-1, based on similarity in drainage area) for each sampling location (squares) where hydrologic condition is needed.


Step 4

Pull instantaneous flow data from the gages with the highest weights

  • Instantaneous gage data is available from dataRetrieval::readNWISuv
    • parameter_code = ‘0600’ | tz = “EST”
    • start_date and end_date should correspond to earliest and most recent dates in the sample dataframe
    • This part of the code may take a while, depending on the number of gages and length of time requested. Be cognizant of what your asking for before submitting your query!
    • For reference, with a high-speed internet connection it took 21 minutes for M. Shank to query instantaneous flow data for 18 gages for 3.5 years (resulted in >2 million records)


Step 5

Hydrologic Condition Interpolation

  • Join instantaneous flow to the sample dataset, keeping only the measurement closest in time to each sample
    • Since we’re pulling instantaneous flow data that is usually collected every 15 minutes, the temporal difference should be very slight
  • Join FDCs, determine the percentile of the instantaneous reading
  • We now have instantaneous flow at 3 gages paired to each sample
    • Calculate percentile values from each gage using weights
    • Sum the percentiles to get aggregated hydrologic condition
  • Join the USGS explanation to get categorical hydrologic condition!

An example of hydrologic condition interpolation output at Birch Run, a 3.5 square mile headwater stream in the South Mountain area of Adams County is shown in the plots below. Samples were collected between August 2020 and November of 2021. The hydrologic condition during the sampling period shows low flows in the fall of 2020 and summer of 2021. Seasonal flows occurred over-winter in 2020-2021, with higher flows during a snowmelt event in March 2021 (Figure 5). Birch Run is impacted by atmospheric definition. Sonde deployment at the site demonstrated that high flows correspond to decreased pH and increased dissolved aluminum concentrations. The correlation of hydrologic condition with each of these variables was consistent with these trends using this model (Figure 6).


Figure 5. Hydrologic condition interpolation output at Birch Run. Points are symbolized by USGS flow regime definitions.


Figure 6. Relationship of hydrologic condition interpolation output with chemical parameters at Birch Run. Points are symbolized by USGS flow regime definitions.


Step 6

Rate of Change Interpolation

  • Join instantaneous flow to the sample dataset, keeping only the measurement closest in time to each sample AND the measurement 6 hours prior to each sample
  • Calculate the percent change of streamflow of the sample compared to 6 hours prior
  • Calculate percent change values from each gage using weights
  • Sum the percent change to get aggregated rate of change

Due to the recommendations that this model be applied in streams <1000 mi2, a 6-hour temporal window was selected to judge the rate of change of the sampling location.

An example is presented using the high flow event on 2023-12-18. Samples were collected at WQN0290 (Sherman Creek, 2.2 mi2 in Wayne County) and WQN0291 (Faulkner Brook, 2.2 mi2 in Wayne County). Streamflow during sampling and 6 hours prior were extracted from the instantaneous flow dataset for each of the three gages that were selected as references based on spatial proximity and similarity in drainage area. WQN0290 was sampled at 10:30, which corresponded to near peak flow conditions at two out of 3 reference gages. WQN0290 was sampled at 12:30, when flow at reference gages was on the descending limb. The resulting rate of change interpolation output, which is an aggregation of the rate of change across reference gages based on weights, was 49% at WQN0290, indicating sampling was conducted during a rapidly increasing hydrograph. Conversely,the rate of change interpolation output was -34% for WQN0291, indicating sampling was conducted on the descending limb (Figure 7).

This rate of change output can provide important information on the status of hydrologic conditions. Intense thunderstorms during summer low flow conditions might result in ‘normal’ (P25 - P75) hydrologic conditions, but a rapidly increasing hydrograph may result in flows with elevated concentrations of suspended sediment. This rate of change metric will assist the analyst to interpret results.


Figure 7. Hydrographs for three gages used for rate of change interpolation at WQN0290 and WQN0291. Points represent the time of sampling and 6 hours prior. Lines represent the slope of the rate of change, and are labeled with percentages.


Back to top

QA


A quality assurance (QA) exercise was run to examine the performance of the hydrologic condition interpolation model. Three gages were randomly selected from the population of gages with minimally altered hydrology. These gages were removed from consideration, and the application code was run to select the most representative gages to estimate hydrologic condition. In effect, these three gages with minimally altered hydrology where FDCs were available were treated as ungaged locations, just as sampling locations were treated in the application workflow

The QA exercise involved comparing hydrologic condition interpolation output with the actual measured instantaneous flow for a hypothetical dataset where samples were collected monthly (first day of each month at noon) for three years, from 2020-2022, resulting in 36 samples.

The hydrologic condition interpolation output show a high degree of concordance with measured flow data for each of the three randomly selected gages (Figures 8 and 9). The annual comparison (Figure 8) was more accurate than the seasonal (May - September) comparison (Figure 9).


Figure 8. Annual hydrologic condition interpolation estimate (horizontal axis) compared to annual flow percentile of measured streamflow (vertical axis). Dashed red line indicates a 1:1 relationship. Blue line is linear relationship of the dataset.


Figure 9. Seasonal hydrologic condition interpolation estimate (horizontal axis) compared to seasonal flow percentile of measured streamflow (vertical axis). Dashed red line indicates a 1:1 relationship. Blue line is linear relationship of the dataset.


Back to top