HYDROLOGIC MODULE
Unit Model
The hydrologic module of the GEM simulates vertical
water fluxes for a locality that is assumed to have spatially
homogeneous characteristics. GEM takes into account a variety
of hydrologic functions controlled by physical and biotic processes
including the following:
-
Vertical water movement between surface, unsaturated
and saturated storage from percolation, aquifer-stream interactions
and evapotranspiration.
-
Surface water flow rates as a function of dynamically
varying plant biomass, density, and morphology in addition to
surface and water elevation.
-
Spatial climatic forcing based on rainfall, temperature,
humidity and wind condition data.
-
Transpiration fluxes dependent on plant growth,
vegetation type and relative humidity.
Conceptual model of unit hydrology
Taking into account the spatial and temporal scales
of the model, we have adopted the following assumptions to simplify
the unit hydrologic model:
- We assume that, rainfall infiltrates immediately
to the unsaturated layer and only accumulates as surface water
if the unsaturated layer becomes saturated or if the daily infiltration
rate is exceeded. Ice and snow may still accumulate.
- Surface water may be present in cells as rivers,
creeks and ponds. Surface water is removed by horizontal runoff
or evaporation.
- Within the day time step, surface water flux
will also account for the shallow subsurface fluxes that rapidly
bring the water distributed over the landscape into the micro
channels and eventually to the river.. Thus, the surface water
transport takes into account the shallow subsurface flow that
may occur during rainfall, allowing the model to account for the
signficantly different nutrient transport capabilities between
shallow and deep subsurface flow.
Spatial Model
The spatial model combines the dynamics of the unit
model which are calculated at each time step for each cell in
the landscape, and adds the spatial fluxes which control the movement
of water and materials between cells. Each cell generates stock
and flow values which provide input to or accept output from the
spatial flux equations.
In the spatial implementation, a major hypothesis
that we are testing is that overland and channel flow can be modelled
similarly. Given the cell size of the model (200 m or 1 km),
we may assume that in every cell there will be a stream or depression
present where surface water can accumulate. Therefore it makes
sense to consider the whole area as a linked network of channels,
where each cell contains a channel reach which discharges into
a single adjacent channel reach. The channel network is generated
from a link map which connects each cell with its one downstream
neighbor out of the eight possible nearest neighbors.
The horizontal flows are driven by
stage gradients in adjacent cells.
After the water head in each raster cell is modified
by the vertical fluxes controlled in the GEM unit model, the surface
water and its dissolved or suspended components move between cells
based on one of two algorithms being tested. In the simplified
algorithm a certain portion of water is taken out of a cell and
added to the next one downstream. This operation is reiterated
several (10-20) times a day, effectively generating a smaller
time step to allow fast riverflow. The number of iterations needed
for the hydrologic module is calibrated so that the water flow
rates match gage data.
Algorithms of flow routing between cells.
A. Water is routed to the next cell on the Link Map. Reiterated several times in one time step.
B. Water is routed several cells downhill. Number of iterations can be decreased.
C. The number of cells downhill over which the water can travel in one time step
is a function of the stage in the donor cell.
For the saturated water, a modified Darcy equation
was employed. For each cell the flux was determined as a function
of saturated conductivity and water head difference between the
current cell and the average head of the cell and its eight neighbors.
We assume one vertically homogenous aquifer interacts with the
surface water.
Patuxent Calibration
Calibrating and running a hydrologic model of this
level of complexity and resolution requires a multi stage approach.
We first identified two spatial scales at which to run the model
- a 200 m and 1 km cell resolution. The 200 m resolution was
more appropriate to capture some of the ecological processes associated
with landuse change but was too detailed and required too much
computer processor time to perform the numerous model runs required
for calibration and scenario evaluation. The 1 km resolution
reduced the total number of model cells in the watershed from
58,905 to 2,352 cells.
Hierarchy of subwatersheds used to test and run the model
Secondly, we identified a hierarchy of subwatersheds.
The Patuxent watershed has been divided into a series of nested
subwatersheds to perform analysis at three scales. A small (23
km2) subwatershed of Cattail creek in the northern
part of the Patuxent basin was used as a starting point. The
next larger watershed was the upper non-tidal half of the Patuxent
watershed that drained to the USGS gage at Bowie (940 km2).
And finally we examined the whole Patuxent watershed (2,352 km2).
The number of total model cells grew from 566 cells initally,
to 23484 cells for the half watershed, and then to 58905 cells
for the entire study area at the 200 m resolution.
Animation of Cattail creek flow over
one month and several rainfall events
|
|
We staged a set of experiments with the small Cattail
creek subwatershed to test the sensitivity of the surface water
flux. Three crucial parameters controlled surface flow in the
model: infiltration rate, horizontal conductivity and number
of iterations per time step of the unit model. Riverflow peak
height was strongly controlled by the infiltration rate. The
conductivity determined river levels between storms and the number
of iterations modified the width of the storm peaks.
Surface water flow was calibrated against the
13 USGS gaging stations in the area that have
data concurrent with the climatic data series (1980-90). After the first
200 days over which the model adjusts its initial conditions,
results are in fairly good agreement with
the gage data. The model is able to consider antecedent moisture
and runoff-generating areas in a spatially realistic manner based
on topography, land use and soil type, giving the simulation a
fairly high degree of precision. The general hydrologic trends
seem to be well captured by the model.
We have then performed some additional calibration for the 5 year period of 1986-1990
to match the calibration period of the HSPF model previously implemented for the
Patuxent watershed. The rainfall pattern over the 9 years between 1984 and 1992
shows that during 1986-1992 we had a good sample of both wet and dry years. The
baseflow is perfectly correlated with rainfall, whereas the peak flow (represented
by the max 10% of flow) is not necessarily matching the patterns of total annual
rainfall. As seen from the graph below there was less rainfall in 1988 than in
1987, nevertheless the peak flow as well as the total runoff in 1988 exceeded that
of 1987.
We may also observe how runoff was driven by precipitation during this period of time.
It was interesting to compare the rainfall dynamics with the variations in total
amount of water in saturated and unsaturated storage.
The comparison with the corresponding HSPF statistics shows
that our model performs somewhat better for this watershed.
In addition to Cattail Creek we were investigating the hydrological and
water quality dynamics in the Upper Patuxent subwatershed. This was mostly
stipulated by the fact that the USGS gaging station at Unity is recording the
water quality parameters in addition to the water flow characteristics.
Therefore the Unity subwatershed became essential for the water quality
calibrations.
For both subwatersheds we have been using
climatic data registered at the Damascus and Clarksville stations. Within the
model, data from various stations are interpolated spatially to provide a smooth coverage
for all the study area. The approximations used for climatic data, especially the
precipitation data (spatial - interpolated; temporal - daily averaged) seem to be
one of the important factors of errors in predicting some flow events, peak flows, in
particular. In the temporal scale we miss some important characteristics of the
rainfall events, such as duration and intensity, that very much affect the infiltration
patterns. Spatially we may miss some rainfall events entirely if they were not at a
location of a meteorologic station, or we may significantly overestimate the effect of
certain local rainfalls by interpolating them over a larger watershed portion.
Once the model was calibrated for the smaller Cattail and Unity subwatersheds, we moved
to the next level in the spatial hierarchy, which is the upper part of the
whole Patuxent watershed, draining at the USGS gaging station at Bowie.
The general pattern of simulated flow was in good agreement with the measurements
(although the
height of some flood peaks was over or under estimated). Model estimated
cummulative flow data fell within 3% of integrative statistics such as minimum
or maximum flow and total annual flow. Note that the observed high peak on
day 220 is not associated with any rainfall event and is probably caused by
discharge from a reservoir upstream (a factor which is not yet in the model).
Calibration results for USGS gaging data at Bowie
(half watershed scale)
Long-term runs over the 1986-1990 period showed that the model performs well enough with
no additional adjustments of parameters. Therefore this output can be viewed as a
model verification experiment.
Comparisons with the HSPF statistics shows
that our model still performs at least as well.
to view animation of surface water flow for Half Watershed (719Kb to download).
Another important finding was that the model performance at 1 km resolution was
very well correlated with the 200 m model. The difference between the resulting runoff
turned out to be negligible, while the gain in performance was considerable.
Hydrologic Scenarios
The next step was to start running some scenarios in order to estimate
model sensitivity to various processes and changes in the environment.
We here show how hydrologic flow reacted to land use transformation. River flow for
the watershed in all forested land use (i.e. before European settlement)
was compared to the currently existing mix of land uses. The peak storm flows
increased, while the minimum base flow decreased, reflecting changes in the
infiltration and evapotranspiration rates and their spatial distribution.
An example of a scenario run.
Comparison of hydrologic output for present landuse vs. all forested
 Back to other model components
E-mail to Alexey Voinov
|
to view animation of surface water flow for Half Watershed (719Kb to download).