SHC Hourly Flow Product Calibration and Validation

 Calibration and validation are performed by splitting the available observed data into two datasets: one for calibration, and another for validation. 

Currently,  USGS flow data  is used for calibration and validation of the SHC HFP.  Two types of flow data are available.  Instantaneous data is from stream flow measurements (discharge) carried out in the stream using flow meters, while continuous flow data is obtained from instream sensors for which a depth to flow relationship has been established. Continuous flow data is available at USGS 01472157 and 01473169 (red hexagons), while instantaneous data is available at many locations (brown hexagons). Going forward, flow data from EnviroDYI stations and stream sampling will extend model validation into second and third order tributaries.

Calibration for hourly flow is carried out with  SWAT Calibration and Uncertainty Program  (SWATCUP). In simple terms,  the calibration of hydrological models involves adjusting parameters to more accurately reflect physical conditions.   SHC HFM was calibrated against hourly flow data from the USGS 01472157 station at Hoffecker Road in the French Creek for the year 2018.

4 parameters were mildly adjusted for calibration. Channel roughness is decreased to align hourly peaks and curve number increased to align peak volume. Alpha baseflow is obtained from analysis at USGS 01472157. A google sheet with the calibration data and the list of parameters adjusted is  here .

To illustrate an example of the calibration process, consider how the roughness of a stream channel affects stream flow. A smooth stream channel with few or no rocks will have less resistance to stream flow while a channel with more roughness will have more resistance. During storms, as the peak flow moves down stream, the smoother channel with deliver the peak earlier than a rough channel. The model has default values for channel roughness that can be adjusted in order to align the simulated peak flow to the observed peak flow--this is model calibration and it is necessary because the model does not have information about the actual channel roughness. Peak alignment calibration is a priority at an hourly time step--an hour or two misalignment between observed peaks and simulated peaks is a failure at an hourly time step, but would be unnoticeable at a daily time step.

 Validation is the process where simulations produced by the calibrated model are compared to observed data using a statistical analysis . The SHC HFP was validated against continuous flow data from the USGS 01472157 (French Creek) and USGS 01473169 (Valley Creek)(ground water withdrawals of 20% are added to the Valley Creek simulations based on published data) with satisfactory results. The SHC HFP is also being validated against the many instantaneous flow measurements made by USGS at locations throughout the SHC; the figure shows validation at USGS 01472190, from 2002 to 2023.

SHC HFP validation will be ongoing with the USGS sites where data is available.

SHC HFP inputs can be divided into two categories. The first are static GIS layers that do not vary during the simulation. The second are weather inputs which do vary through simulation. There are other inputs which can also vary--for example data for groundwater withdrawals.

Static GIS layers are currently the National Land Cover Dataset (NLCD2016), Soils Survey Geographic Database, and 3 meter digital elevation model (DEM). Other landcover datasets can be used. Several high resolution hybrid land cover products have been developed and tested with the goal of producing high resolution impervious cover inputs.

Weather inputs are precipitation, temperature, relative humidity, wind speed and solar radiation. Precipitation is obtained from  NEXRAD , which collects precipitation data at 5 minute intervals with radar stations located across the US. This data is quality controlled and then aggregated at the national level to create the NCEP/EMC 4KM Gridded Data Stage IV Data. Precipitation is available at 1 hour, 6 hour and 24 hour intervals from 2002 forward. The other weather data inputs are currently obtained from  gridMET .

The model is simulating the movement of water from its beginning as accumulating precipitation through to its departure from the watershed. By quantitatively following the progress of water from initial radar observations to stream flow observations we are connecting actual precipitation with actual stream flow through a physical model. This connection leverages both types of data to improve our understanding of stormwater events--and we need better understanding of these events. However, there are uncertainties at every point in this process. There are uncertainties in flow measurements, radar measurements, agricultural practices, stream channel dimensions, underlying geology, forest biomass and so on. These uncertainties combine and propagate resulting in differences between the modeled flow and the observed flow. In 2023, the differences are acceptable--the simulations are performing well and we can start using them. Going forward there are many places where these uncertainties can be evaluated and reduced. Examples are reducing the uncertainties in precipitation inputs, reducing the uncertainties in the measured peak flows, and additional changes to model inputs to more accurately reflect actual conditions such as forest biomass, stormwater facilities, agricultural practices, and groundwater withdrawals/exports.