The Hydrologic Sampler Plugin provides three hydrologic sampling methods (Table 1). The user must specify only one of these methods per hydrologic sampling alternative (review Create a Hydrologic Sampling Alternative). Table 1 provides descriptions and data requirements for each hydrologic sampling method that might be helpful in selecting the appropriate sampling method.

The following sections provide an overview of each available sampling option in the Hydrologic Sampler. 

Flow – Correlated Flow Frequency Curves

The Correlated Flow Frequency Curves method generates randomly sampled flood events based on a user-provided flow-frequency curve for each inflow location, spatial correlations between locations (frequency curves), hydrograph shapes, and a flood season distribution. This method generates random flood magnitudes at hydrograph locations (with specified linear correlations) that defines the spatial distribution of flow throughout a watershed. Though more flexible in creating possible flood events than the Bootstrapping Historical/Synthetic Basin-wide Events sampling method, the set of frequency curves and linear spatial relationships can be a simplistic representation of extreme event hydrology for a complex watershed.

The user provides location-specific flow-frequency curves to capture the probability distributions of event magnitudes ranging from small to extremely large. These flow-frequency curves have generally been estimated from analysis of a historical record of annual maximum (or coincident) flows, as are the spatial correlations between locations within the watershed. The user provides sets of scalable hydrograph shapes sets to capture examples of possible event timing, both for a given location and in combination with other locations. Finally, the user provides a distribution of flood dates (month and day) based on analysis of historical flood dates.

In this method, for each random flood event the Hydrologic Sampler generates a hydrograph for each specified inflow location using the following steps:

1. randomly samples a peak flow magnitude for each primary location with correlated sampling from the flow-frequency curves,
2. randomly samples a shape set from the available basin-wide sets and scales each hydrograph shape to match the corresponding sampled flow magnitude, and
3. places the event on a randomly sampled date.

The resultant set of flood hydrographs for the watershed is thus a combination of date, magnitude and timing of the event arrival. At any location, the user may choose to sample channel stage rather than flow.

When flooding in the watershed results from different flooding mechanisms that cannot be captured adequately by a single season with associated frequency curves and shapes, the user may define two seasons, each with complete user-inputs as detailed above, including frequency curves, shape sets and flood season distribution. The user may choose to sample an event in each season for each year of the lifecycle, or randomly choose the season from which to sample one event each year of the lifecycle.

Flow – Bootstrapping Historical/Synthetic Basin-wide Events

The Bootstrapping Historical/Synthetic Basin-wide Events (bootstrapping) flow sampling method re-samples a provided historical record (plus optional derived synthetic events), as watershed-wide flood events. The method is limited to a discrete set of known flood events that are reasonable, rather than a continuous range of possibilities, but a well-constructed set can capture a more complex spatial description of flood event probability in the watershed than the Correlated Flow Frequency Curves method.

The user can include as many events in the historical record as desired, but the record must contain data at every location for each event. The user may also choose to supplement the historical record with synthetic flood events that are greater or less than measured flows, or have a different spatial distribution of flow volumes, to ensure that a good characterization of feasible extreme events is included. For each event, the user defines a flood hydrograph for each inflow location which encompasses the peak and as much of the hydrograph recession limb as they deem necessary, with all locations spanning the same event windows. The hydrographs can vary in length and shape from one event to the next (though a common use-case includes the full-year for each). The user-defined set of historical and synthetic hydrographs should collectively capture the probabilistic relationships of magnitude and spatial distribution of floods in the watershed.

In this method, the Hydrologic Sampler starts with a "bucket" of flood events filled with the user-defined, basin-wide historical record and additional synthetic flood events. The sampling method randomly samples an event from the flood bucket, and the flood hydrographs of the sampled event at all basin inflow locations are used in the event compute. This method maintains flood event frequency by sampling events at the frequency they occurred or would occur, meaning an equal likelihood for each historical event and user-assigned frequency for synthetic events. The spatial correlation of flood magnitudes between locations is implicitly maintained , because the definition of each historical year and synthetic event is for the entire watershed. By default, flood events sampled for each year are independent of the previous year, but the user may specify serial (temporal) correlation between the volume of annual events that is maintained for a key location.

In addition to flow, any parameter needed by the subsequent models in the HEC-WAT model sequence may be defined and provided at any specified location. A sampled flood event will include hydrographs of flow and those of other parameters for event.

With this sampling method, hydrologic forecasts may also be sampled, randomly deflecting the volume of an event over the defined forecast period based on user-defined forecast error statistics.

Precipitation – Basin Average Frequency Curve

The Basin Average Frequency Curve method samples precipitation from a precipitation-frequency curve, and in this way is similar to the Correlated Flow Frequency Curves method for sampling flows. The method uses a basin average precipitation-depth frequency curve provided by the user, as well as various hyetograph shape sets and a flood season distribution. Precipitation hyetographs are generated for each flood event by scaling randomly chosen shape sets to match randomly sampled basin average precipitation depths. The primary difference compared to the flow sampling method, however, is that one precipitation depth is sampled from an area-weighted basin average frequency curve, then "disaggregated" by subbasin area and relative depth at various subbasin locations, rather than sampling values for each subbasin location from separate frequency curves.

The user provides a basin average precipitation-frequency curve to capture the likelihood of event magnitudes, ranging from small to extremely large. This curve has generally been estimated from historical precipitation records from gages across the watershed. The user also provides sets of location-specific scalable hyetograph shapes (shape sets) that together capture the array of possible event timing and spatial distribution of rainfall throughout the watershed, with each set usually based on a large event from the historical record. Finally, the user provides a distribution of possible flood dates based on analysis of the historical flood dates.

In this method, for each random storm event the Hydrologic Sampler generates a hyetograph for each subbasin location using the following steps:

1. randomly samples an annual maximum precipitation depth from an aggregated basin average precipitation depth frequency curve,
2. randomly selects a shape set,
3. disaggregates the basin average precipitation depth among the subbasin locations based on subbasin area and shape set depth,
4. scales every hyetograph in the shape set to the subbasin depth, and
5. places the scaled hyetographs on a peak date randomly sampled from the flood season distribution.

The resultant set of precipitation hyetographs for the watershed is thus a combination of date, magnitude, spatial distribution and timing of the event arrival.

The user may specify multiple watersheds in the Hydrologic Sampler, each with subbasin locations and a basin-average frequency curve used to sample precipitation magnitudes for that watershed. Spatial correlations are specified between the frequency curves. Each separate watershed is intended to correspond either to a separate rainfall/runoff model in the WAT model sequence, or simply a basin that can be reasonably averaged by area. A flood event is generated by sampling a precipitation depth from each basin-average frequency curve (maintaining correlation with the other watersheds), and scaling its locations in the selected shape set to reproduce that basin-average depth. The storm event thus produced has hyetographs for all locations in all defined watersheds. The most common case, however, will be use of a single watershed and basin-average frequency curve.

When flooding in the region results from different storm mechanisms that cannot be captured adequately by a single season and associated basin-average frequency curve, the user may define two seasons, each with complete user-inputs as detailed above, including basin-average frequency curve, shape set hyetographs and flood season distribution (and multiple watersheds, if used). The user may choose to sample an event in each season for each year of the lifecycle, or randomly choose the season from which to sample one event each year of the lifecycle.