Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

ABSTRACT: In this study, remotely sensed data and geographic

information system (GIS) tools were used to estimate storm runoff


response for Simms Creek watershed in the Etonia basin in north-
east Florida. Land cover information from digital orthophoto quar-
ter quadrangles (DOQQ), and enhanced thematic mapper plus
(ETM+) were analyzed for the years 1990, 1995, and 2000. The cor-
responding infiltration excess runoff response of the study area was
estimated using the U.S. Department of Agriculture (USDA), Natu-
ral Resources Conservation Service Curve Number (NRCS-CN)
method. A digital elevation model (DEM)/GIS technique was devel-
oped to predict stream response to runoff events based on the travel
time from each grid cell to the watershed outlet. A comparison of
predicted to observed stream response shows that the model pre-
dicts the total runoff volume with an efficiency of 0.98, the peak
flow rate at an efficiency of 0.85, and the full direct runoff hydro-
graph with an average efficiency of 0.65. The DEM/GIS travel time
model can be used to predict the runoff response of ungaged water-
sheds and is useful for predicting runoff hydrographs resulting
from proposed large scale changes in the land use.
(KEY TERMS: surface water hydrology; spatially distributed mod-
eling; GIS; remote sensing; curve number.)
Melesse, Assefa M. and Wendy D. Graham, 2004. Storm Runoff Prediction
Based on a Spatially Distributed Travel Time Method Utilizing Remote Sensing
and GIS. Journal of the American Water Resources Association (JAWRA)
40(4):863-879.
INTRODUCTION
Accurate and up to date information on land cover
and the state of the environment are critical for mod-
eling hydrologic processes such as surface runoff,
evaporation, and infiltration that take place at the
ground/atmosphere interface. Land cover information
is used in hydrologic modeling to estimate the value
of surface roughness or friction as it affects the veloci-
ty of the overland flow of water. It may also be used to
determine the amount of infiltration resulting from
rainfall onto a surface. Land cover data can be
acquired from field survey, aerial photography in the
form of digital orthophoto quad quadrangles (DOQQ)
or satellite imagery. Topography also plays an impor-
tant role in the distribution and flux of water and
energy within the natural landscape. Landscape fea-
tures such as slope, aspect, flow length, drainage
areas, stream network, and flow direction can be
extracted from digital elevation models (DEMs).
The use of the geographic information system (GIS)
to facilitate the estimation of runoff from watersheds
has gained increasing attention in recent years (Oliv-
era and Maidment, 1999). This is mainly due to the
fact that it is often desirable to include the spatial
and geomorphologic variation of the watershed in the
rainfall/runoff model. GIS is a computer based tool
that acquires, displays, stores, analyzes, retrieves,
manages, and generates spatial and nonspatial
(attribute) data. The spatially distributed capabilities
of GIS facilitate the linkage of watershed data to dis-
tributed parameter models (Maidment, 1993). The
tremendous data handling capabilities of GIS and
improved processing offered by GIS modules allow
creation of hydrologic models that can better simulate
spatially distributed hydrology. The increasing avail-
ability of spatially distributed topographic, soils, land
use, land cover, and precipitation data in GIS ready
format provides the framework for the development,
verification, and eventual acceptance of new hydrolog-
ic models capable of taking full advantage of these
1
Paper No. 02150 of the Journal of the American Water Resources Association (JAWRA) (Copyright 2004). Discussions are open until
February 1, 2005.
2
Respectively, Professor, Department of Agricultural and Biological Engineering, University of Florida, 110570 Rogers Hall, Gainesville,
Florida 32611-0570; and Assistant Professor, Earth System Science Institute, Odegard School of Aerospace Sciences, University of North
Dakota, University and Tulane Drive, Grand Forks, North Dakota 58202-9007 (E-Mail/Graham: wgraham@ufl.edu).
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 863 JAWRA
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
AUGUST AMERICAN WATER RESOURCES ASSOCIATION 2004
STORM RUNOFF PREDICTION BASED ON A SPATIALLY DISTRIBUTED
TRAVEL TIME METHOD UTILIZING REMOTE SENSING AND GIS
1
Assefa M. Melesse and Wendy D. Graham
2
new data, while acknowledging the uncertainty inher-
ent in the data.
Maidment (1992a,b, unpublished reports; 1993)
presented a grid based methodology for determining a
spatially distributed unit hydrograph that assumes a
time invariant flow velocity field. The study showed
the velocity time invariance is a requirement for the
existence of a unit hydrograph with a constant time
base and relative shape. The concept emphasizes the
conditions for linearity of a routing system. Maidment
demonstrated that if a constant velocity can be esti-
mated for each grid cell, a flow time grid can be
obtained and subsequently the isochronal curves and
the time area diagram can be determined. A more
elaborate flow simulation process, accounting for both
translation and storage effects in grid cells, was pre-
sented by Maidment et al. (1996), where total water-
shed response is predicted as the sum of the
responses of individual subwatersheds, which are
determined as a combined process of channel flow
(translation process) followed by a linear reservoir
routing (spreading process).
Muzik (1995) applied Maidments procedure to a
229 km
2
watershed in the Canadian Rockies using
GRASS GIS. Muzik estimated the flow velocity
through each grid cell by combining the continuity
equation and Mannings equation. The resulting time
area curve had the expected S-hydrograph shape. To
test the model, a one-hour unit hydrograph was
derived and convolved with a storm event that had an
intensity of 3.7 mm/hr. Based on a visual inspection,
the predicted hydrograph showed good agreement
with the observed hydrograph. Ajward (1996) applied
the same spatially distributed unit hydrograph tech-
nique to a 228 km
2
watershed located on the eastern
slopes of the Canadian Rockies in Alberta, Canada,
comparing the predictions from the spatially dis-
tributed unit hydrograph with the observed hydro-
graph for eight rainfall events. He found that the
model generally gave good predictions of the peak
flow rate and the time to peak.
In this study, an integrated technique employing
satellite imagery to derive Level 1 (Anderson et al.,
1976) land cover classes, and grid GIS to compute
hydrologic parameters from DEMs, is presented that
predicts spatially distributed excess rainfall volume
and routes the water to the watershed outlet. A rout-
ing model based on travel time is developed to gener-
ate the direct runoff hydrograph without relying on
unit hydrograph theory. The methodology is demon-
strated and validated using data measured in the
Simms Creek watershed in northeast Florida. Pre-
dicted runoff hydrographs from the travel time
method is also compared to those obtained using the
traditional Snyder synthetic lumped unit hydrograph
technique (Chow et al., 1988).
METHODOLOGY
Remote Sensing
Remote sensing uses measurements of the electro-
magnetic spectrum to characterize the landscape,
infer its properties, and in some cases, actually mea-
sure hydrologic state variables. The continued devel-
opment of remote sensing technology has added new
techniques that hydrologists can use in a large num-
ber of applications. Most of the hydrologically rele-
vant information is provided indirectly by remote
sensing. One of the advantages of using remote sens-
ing data for hydrological modeling and monitoring is
its ability to generate information over large space
and frequent time scales, which is very useful for suc-
cessful model analysis, prediction, and validation.
Remote sensing can be incorporated into hydrologic
analysis in a variety of ways: to estimate land cover
characteristics and impervious surface area, for pro-
viding initial conditions for flood forecasting, and for
monitoring the extent of flooded areas (Neumann et
al., 1990). Kite and Kouwen (1992) showed that a
semi-distributed model based on Landsat derived sub-
basins performed better than a lumped model. GIS
allows for the combination of remotely sensed data
with other spatial data forms, such as topography soil
maps, and hydrologic variables, such as rainfall dis-
tribution and soil moisture. In a study of the impact
of land use change on the Mosel River Basin, Ott et
al. (1991) and Schultz (1993) defined hydrologically
similar units (HSU) using DEM data, soil maps, and
satellite derived land cover. They also used satellite
data to determine a normalized difference vegetation
index (NDVI) to delineate areas where a subsurface
supply of water is available to vegetation.
Land Cover
A raster layer of land cover was derived for each
study year either from DOQQs (1990 and 1995) or
from image processing of a Landsat Enhanced The-
matic Mapper Plus (ETM+) (2000) image (30 m spa-
tial resolution) to U.S. Geologic Survey (USGS) land
use land cover (LULC) system Level-1 land cover
types (Anderson et al., 1976).
The DOQQ data were converted by the St. Johns
River Water Management District (SJRWMD) into a
vector layer with a nominal scale of 1:24,000 using
an on-screen digitizing approach and stereo
plotter together with ancillary data such as soil, land
parcels, and Federal Emergency Management Agency
floodplains to assist interpretation for land use
classification. The classification accuracy reaches
JAWRA 864 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
MELESSE AND GRAHAM
approximately 90 percent with mapping units of 0.5
ha for wetlands and 2 ha for other land cover types,
which are the minimum mapping units respectively
(SJRWMD, 2002). For the purpose of this study the
vector GIS coverage of the land cover was converted
into 30 m grids using ArcView.
The satellite image was processed using an unsu-
pervised classification technique employing iterative
self organizing data analysis technique (ISODATA)
(Tou and Gonzalez, 1974) to yield 30 spectral classes
(ERDAS, 1999). Using vegetation indices versus sur-
face temperature scatter diagrams, the 30 classes
were recoded into seven Level 1 land cover classes.
The technique used to map land covers using the
above procedure is described in Melesse et al. (2001).
Land cover mapped for Simms Creek using the aug-
mented ISODATA technique employing vegetation
indices and surface temperature was found to be more
than 80 percent accurate in a ground truthing exer-
cise conducted in the Simms Creek watershed
(Melesse et al., 2001).
Runoff Depth
The NRCS-CN runoff equation (USDA, 1986) is
widely used in estimating direct runoff. Since daily
rainfall data were used in the development of the
equation, the time distribution and duration of storms
were not considered. The equation also tends to over-
predict runoff volume for discontinuous storms,
because it does not account for the recovery of soil
storage caused by infiltration during periods of no
rain. The results of studies comparing NRCS method
predictions against measured values have been
mixed. For example, Kumar and Jain (1982) and
Wood and Blackburn (1984) found discrepancies with
measured values. However, Mostaghimi and Mitchell
(1982) have found satisfactory agreement. Dingman
(2001) concluded that the NRCS-CN approach will
continue to be used since (1) it is computationally
simple, (2) it uses readily available watershed infor-
mation, (3) it has been packaged in readily available
tables, graphs, and computer programs, (4) it appears
to give reasonable results under many conditions, and
(5) there are few other practicable methodologies for
obtaining a priori estimates of runoff that are known
to be better.
Land cover classifications and soils maps were
used to determine maps of NRCS-CNs, which were
used to estimate excess precipitation in this study.
For each rainfall event, the total volume of excess pre-
cipitation was calculated for each grid cell in the
watershed using the total volume of rainfall estimat-
ed for each cell, the antecedent moisture condition
(AMC) of the cell, and the cell CN. Antecedent mois-
ture condition is an indicator of watershed wetness
and availability of soil moisture storage prior to a
storm, and can have a significant effect on runoff vol-
ume. Recognizing its significance, NRCS developed a
guide for adjusting the CN according to AMC deter-
mined by the total rainfall in the five-day period pre-
ceding a storm and growing season using data
collected and analyzed by NRCS (USDA, 1985). Three
levels of AMC are used in the CN method: AMC-I for
dry, AMC-II for normal, and AMC-III for wet condi-
tions. Runoff volume computed without adjusting the
CN for AMC will overpredict or underpredict the val-
ues. For instance, for a CN of 78 for AMC II, the cor-
responding CNs for AMC I and AMC III are 39 and
89, respectively. For a storm volume of 40 mm, the
predicted runoff depth will be 0.14, 6.9, and 17.4 mm
for AMC I, II, and III, respectively. For the purpose of
this study, the dormant season was assumed to be
October through May and the growing season was
assumed to be June through September. Evapotran-
spiration is neglected because of the short term, event
based nature of predictions.
Runoff Routing
The runoff hydrograph at the outlet of the water-
shed was developed by routing the spatially distribut-
ed excess precipitation through the watershed using
topographic data. Calculation of travel times from
each cell to the watershed outlet requires computa-
tion of a runoff velocity for each grid cell. Velocity for
each grid cell can be estimated depending on whether
the grid cell represents an area of diffuse overland
flow or more concentrated channel flow. Sensitivity
analysis was conducted to determine the threshold
value for the number of cells that must flow into a
downstream cell before it should be classified as a
concentrated channel flow cell (see Appendix A).
Overland flow travel time in a grid cell was approx-
imated as the time of concentration in the grid cell (or
the time at which the entire grid cell is contributing
flow at the outlet boundary), which was estimated by
combining a steady state kinematic wave approxima-
tion with Mannings equation. Using this methodolo-
gy, the overland flow travel time, t
o
(seconds), is
calculated (Chow et al., 1988; Muzik, 1995),
where i
e
is the average excess rainfall intensity (cal-
culated as the total volume of excess rainfall for the
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 865 JAWRA
STORM RUNOFF PREDICTION BASED ON A SPATIALLY DISTRIBUTED TRAVEL TIME METHOD UTILIZING REMOTE SENSING AND GIS
t
L n
i S
o
e

0 6 0 6
0 4 0 3
. .
. .
(1)
cell divided by total rainfall duration, with units of
meters/second); L is the length of the overland flow in
the grid cell (meters); n is the Mannings roughness
coefficient; and S is the slope (meters/meter). Man-
nings roughness coefficient for estimating overland
and channel flow velocities was determined using
land cover information. The rainfall intensity (i
e
) in
Equation (1) was determined from the cell excess
rainfall determined by the CN method.
The channel flow velocity, V
c
(meters/second)
(Equation 4), was estimated using Mannings equa-
tion (Equation 2) and the steady state continuity
equation for a wide channel (Equation 3) (Muzik,
1995).
Q = V
c
By
where y is the depth of flow (meters); Q is the cumula-
tive discharge (cubic meters/second) through the cell,
obtained by summing upstream flow contributions
and the contribution from precipitation excess for that
cell; and B is the channel width (meter). It was
assumed that channels had a rectangular cross sec-
tion with depth of flow y and effective width B. The
effective flow width was obtained through sensitivity
analysis (see Appendix A).
The travel time for each channel cell was computed
by dividing travel distance by the cell velocity.
Depending on the direction of flow, the travel distance
was 30 m (grid size) for cells having horizontal or ver-
tical flow directions and 42 m for cells having diago-
nal flow directions. The cumulative travel time of the
storm runoff to the watershed outlet was computed by
summing the travel times along the respective flow
path from each cell following the flow direction.
Once the travel time of each cell to the outlet was
computed, the volumetric flow rate contributed by
that cell (excess rainfall intensity of each cell multi-
plied by the cell area) at that time was noted. The
direct runoff flow was determined by the sum of the
volumetric flow rate from all contributing cells at
each respective travel time. This approach preserves
the effective spatially distributed excess rainfall on
the streamflow hydrograph. Unlike previous
approaches (Maidment,1993; Muzik, 1995; Ajward,
1996), the travel time concept proposed in this study
develops a direct hydrograph for each spatially dis-
tributed rainfall event without relying on developing
a spatially lumped unit hydrograph. The model is an
extension to previous approaches and it utilizes limit-
ed data (rainfall, DEM, soils, and land cover).
APPLICATION TO SIMMS CREEK WATERSHED
The study area, Simms Creek, is located in the Eto-
nia subbasin, Florida (Figure 1). This watershed
drains about 11,433 ha (114 km
2
) of the Etonia sub-
basin. It flows east with Rice and Etonia Creeks to
the St. Johns River, and is mainly covered with forest,
wetlands, agriculture, and urban, built-up areas. Eto-
nia is a moderately sloped watershed with elevation
ranging from 3.6 m to 72 m above mean sea level. The
average annual rainfall is estimated to be 142 cm (56
in) with approximately 60 percent of the rainfall
occurring between June and September. Both frontal
and convective rainfall mechanisms occur in this
region. Frontal precipitation usually occurs when a
cold front from the north results in the lifting of air
masses. Frontal precipitation generates longer peri-
ods of rain and small intensities usually during the
dry season, November through March. In contrast,
convective rainfall during the wet season (May
through September) is characterized by short storm
duration with fairly high rainfall intensities.
Average temperature in the Simms Creek water-
shed is 22C with the minimum and maximum tem-
peratures of 14.8 and 29C occurring in February and
JAWRA 866 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
MELESSE AND GRAHAM
V
S
n
y
c

1 2
2 3
/
/
V
S
n
Q
B
c

_
,

1
]
1
1
1 2 2 3
3 5
/ /
/
(2)
(3)
(4)
Figure 1. Study Location Map.
June, respectively. The annual average potential
evapotranspiration is estimated to be 116 cm (45.7
in). The dominant soils of Simms Creek watershed
are Myakka and Tavares, which constitute more than
60 percent of the watershed area. The hydrologic soil
grouping (HSG) of these soils is D, poorly drained
with high runoff potential. Melesse (2002) showed
that infiltration excess rather than saturation excess
was the dominant runoff generation mechanism for
this watershed.
Data
The data collected for this study included land
cover processed from DOQQs and Landsat ETM+
images, HSG from the USDA-NRCS soil maps, topo-
graphic information in the form of 30 m DEMs from
the USGS, and hourly rainfall and runoff discharge
data at Simms Creek obtained from the SJRWMD
and the USGS.
For each study year, a raster layer of land cover
was derived either from DOQQs or from image pro-
cessing of Landsat image (30 m spatial resolution) to
USGS-LULC system Level 1 land cover types (Ander-
son et al., 1976). The soils GIS coverages showing
HSG were obtained from the USDA-States Soils Geo-
graphic (STATSGO) database. Since the STATSGO
database has a scale of 1:250,000 and the soil map
unit identification in the database can have more
than one HSG, hardcopy county level soil survey
maps (1:15, 840) were consulted to improve the accu-
racy of assigning HSG. Soils are classified into four
hydrologic soil groups (A, B, C, and D) according to
their minimum infiltration rate, which is obtained for
a bare soil after prolonged wetting. Soils in Group A
have lowest runoff potential, soils in Group B have
moderately low runoff potential, soils in Group C have
moderately high runoff potential, and Group D soils
have the highest runoff potential.
DEMs were used in ArcView to determine the
hydrologic parameters of the watersheds such as
slope, flow accumulation, flow direction, drainage
area delineation, and stream network. Hourly rainfall
data were obtained from the SJRWMD for 1990, 1995,
1999, and 2000. Using the three rainfall gauges near-
est to Simms Creek, Thiessen polygons were con-
structed using ArcView to estimate the spatial
distribution of rainfall throughout the watershed for
computation of spatially variable excess rainfall. This
provides three zones in Simms Creek with three dif-
ferent rainfall volumes. Historical measurements of
discharge in Simms Creek were obtained for 1990,
1995, 1999, and 2000 (Figure 2).
A total of 16 isolated storms with observable runoff
responses and volumes of not less than 12.75 mm (0.5
in) were selected from the four study years to test the
stream flow prediction model (see Table 1 and Figures
2a-2d). This volume represents the minimum amount
of rainfall recommended for use with the NRCS-CN
method (USDA, 1986). The storms were also selected
to represent a range of rainfall volumes and intensi-
ties, and to span different seasons and years.
The hydrographs were separated into two distinct
components using the straight line base flow method:
a fast intermittent response following heavy rainfall
and a slow continuous base flow response. The fast
response was separated from the slow response by a
horizontal straight line drawn from the point at the
sharp rise of the hydrograph to the recession limb.
Figure 3 illustrates the separation of the hydrograph
components for the April 1, 1999, storm. A compari-
son of rainfall and runoff volumes (Table 1) shows
that for storms in this basin an average of 7.1 percent
of the total rainfall in each event generates the fast
response runoff volume. The minimum runoff
response of 1.2 percent was observed for Storm 15 on
March 30, 2000, and the maximum of 20.4 percent
was observed for Storm 16 on September 29, 2000.
Depending on the intensity and duration of the storm
the observed time to peak for the 16 test storm events
ranged from 13 to 64 hours (Table 1).
AMC levels were determined for each storm based
on five-day antecedent rainfall volume (Table 2), then
NRCS CNs were determined for each storm based on
antecedent moisture, land cover, and hydrologic soil
group. Figure 4 shows the CNs for each study year
based on AMC II. This figure shows that the CN
increased throughout the study period, particularly in
the central part of the domain, due to urbanization of
the region. The spatially distributed runoff depths
were estimated for each storm using spatially vari-
able rainfall (estimated using the Thiessen polygon
method) and spatially distributed CNs. The spatial
distribution of Mannings roughness coefficient was
determined for each study year based on values pub-
lished in literature for the appropriate land cover
(Brater and King, 1976; Montes, 1998) (Figure 5).
The Mannings coefficients also changed over the
study period due to changes in land cover.
Comparison to Snyder Synthetic Lumped Unit
Hydrograph Predictions
Snyder published a method for estimating synthet-
ic unit hydrographs in the 1930s. He analyzed data
from several Appalachian watersheds and concluded
that unit hydrographs can be estimated from three
watershed geometry parameters and two empirical
shape coefficients (i.e., A, watershed area; L,
hydraulic length, flow path length from outlet to
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 867 JAWRA
STORM RUNOFF PREDICTION BASED ON A SPATIALLY DISTRIBUTED TRAVEL TIME METHOD UTILIZING REMOTE SENSING AND GIS
JAWRA 868 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
MELESSE AND GRAHAM
Figure 2. Hourly Rainfall and Discharge Hydrograph for Simms Creek: (a) 1990, (b) 1995, (c) 1999, and (d) 2000.
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 869 JAWRA
STORM RUNOFF PREDICTION BASED ON A SPATIALLY DISTRIBUTED TRAVEL TIME METHOD UTILIZING REMOTE SENSING AND GIS
Figure 2 (contd). Hourly Rainfall and Discharge Hydrograph for Simms Creek: (a) 1990, (b) 1995, (c) 1999, and (d) 2000.
hydraulically farthest point; L
C
, centroid length, flow
path length from outlet to watershed centroid; C
t
,
timing coefficient (dimensionless); and C
p
, peak coeffi-
cient (dimensionless). From these parameters and
coefficients, Snyder developed equations for five char-
acteristics of the synthetic unit hydrograph (Chow et
al., 1988): the peak discharge per unit of watershed
area, q
pR
, the basin lag, t
pR
, the base time, t
b
, and the
widths, W (in time units), of the unit hydrograph at
50 and 75 percent of the peak discharge.
For Simms Creek, C
t
and C
p
were determined
using gauged information from representative storms
in 1990, 1995, and 2000 that had similar volume and
intensity. The storms used to compute these coeffi-
cients were Storm 1 (April 28, 1990), Storm 5 (April
24, 1995), and Storm 10 (February 28, 1999). The val-
ues of L and L
c
were computed from the DEM using
ArcView. The values of C
t
and C
p
(Table 3) are slight-
ly different for the three study years. The C
t
decreased from 6.9 to 4.7 over the study period, lead-
ing to a decrease in travel time due to an increase in
impervious surface areas. The C
p
value decreased
from 0.07 in 1990 to approximately 0.05 in later
years, indicating a decrease in the peak flow. Normal-
ly, peak flow rate, and thus C
p
, would be expected to
increase as travel time, and thus C
t
, decreases due to
an increase in impervious surface area throughout
the watershed. The discrepancy found in this study
could be due to nonuniform storm distributions over
the watershed, which cannot be accounted for in the
lumped Snyder unit hydrograph approach.
For each of the 16 rainfall events, the appropriate
Snyders synthetic unit hydrograph was used with the
excess precipitation determined using the NRCS CN
technique to predict the streamflow response due to
direct runoff at the outlet using the principle of super-
position. The fraction runoff for each storm event
(total runoff divided by total precipitation) was multi-
plied by the total hourly rainfall to compute excess
precipitation for each corresponding hour.
JAWRA 870 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
MELESSE AND GRAHAM
TABLE 1. Summary of Rainfall and Discharge Events.
Rainfall* Runoff
Average Time to
Storm Volume Duration Intensity Volume Volume Peak Peak
No. Storm Date (mm) (hr) (mm/hr) (mm) (percent) (m
3
/s) (hr)
1 April 28, 1990 29.62 3 9.87 1.51 5.1 1.09 20
2 June, 7, 1990 51.05 4 12.76 1.39 2.7 1.22 32
3 June 26, 1990 70.60 22 3.21 2.59 3.7 0.94 64
4 August 17, 1990 21.08 3 7.03 3.19 15.1 1.64 39
5 April 24, 1995 27.43 3 9.14 1.38 5.0 0.76 23
6 May 12, 1995 26.54 8 3.79 3.84 14.5 2.97 30
7 January 3, 1999 24.38 3 8.13 0.62 2.5 0.48 22
8 January 9, 1999 28.44 7 4.06 0.88 3.1 0.49 27
9 January 23, 1999 78.23 10 7.82 8.16 10.4 5.14 32
10 February 28, 1999 30.98 2 15.49 1.46 4.7 0.81 13
11 March, 14, 1999 24.63 7 3.52 2.19 8.9 1.71 27
12 April 1, 1999 33.53 4 8.38 1.37 4.1 0.69 27
13 April 16, 1999 17.52 13 1.35 0.44 2.5 0.25 33
14 June 16, 1999 15.24 2 7.62 1.27 8.3 1.11 24
15 March 30, 2000 58.33 3 19.44 0.69 1.2 0.51 31
16 September 29, 2000 55.50 10 5.55 11.34 20.4 6.08 32
Average 37.07 6.5 7.92 2.65 7.1 1.62 29.75
*Values represent weighted average from the three rain gages.
Figure 3. Hydrograph Analysis Showing Time to Peak
and Recession, and Fast Response Runoff Depth
for Storm 12, April 1 Through 8, 1999.
RESULTS
Runoff hydrograph prediction was conducted using
the spatially distributed direct hydrograph travel
time technique (SDDH) developed here, and Snyders
lumped method, for the 16 selected rainfall events. To
compare the model predictions to measured data, the
mean prediction error and the prediction error stan-
dard deviation were evaluated for total runoff volume,
peak flow rate, and time to peak. In addition, the effi-
ciency (Nash and Sutcliffe, 1970) of the runoff volume,
peak flow rate, and time to peak predictions, as well
the overall model prediction efficiency over the entire
hydrograph, was evaluated using Equations (5), (6),
and (7). Efficiency is a commonly used quantitative
measure of hydrograph prediction performance.
where E is the prediction efficiency, SSE is the sum of
squares of prediction errors, SST is the total sum of
squares of the observations, Observed
i
is the observed
condition (i.e., runoff volume, peak flow rate, time to
peak, stream flow rate) at time i, Observed
i
is the
mean of the observed condition over all times, and
Predicted
i
is the predicted condition at time i. The
index i refers to the storm date for calculating the
prediction efficiencies for runoff volume, peak flow
rate, and time to peak over all storms, and refers to
time during the storm for calculating the efficiency of
a hydrograph for a particular storm.
Table 4 summarizes the results of the model predic-
tions and the prediction error statistics. Note that
Storms 1, 5, and 10, that were used to fit the Snyder
unit hydrograph parameters, were not used to calcu-
late the Snyder prediction error statistics. The SDDH
method predicted the runoff volume and peak flow
rate accurately, with overall prediction efficiencies of
0.98 and 0.85, respectively. The SDDH method was
less accurate predicting the time to peak, with an
overall efficiency of -0.43. The Snyder method also
predicted the runoff volume and the peak flow rate
accurately, with efficiencies of 0.98 and 0.92, respec-
tively. However the Snyder method systematically
underpredicted time to peak and thus had a higher
mean error and lower prediction efficiency (-0.81) for
time to peak than the SDDH method. For the predic-
tion of the full direct runoff hydrograph, the SDDH
method had an average efficiency of 0.65, with
11 of the 16 storms having efficiencies greater than
0.7. Snyders method had an average efficiency of
0.60 for predicting the direct runoff hydrograph, with
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 871 JAWRA
STORM RUNOFF PREDICTION BASED ON A SPATIALLY DISTRIBUTED TRAVEL TIME METHOD UTILIZING REMOTE SENSING AND GIS
TABLE 2. Antecedent Moisture Conditions Estimated From Storm Volumes Five Days Before the Storm Day.
Rainfall (mm)
Storm Days Prior to Storm Event
No. Storm Date 5 4 3 2 1 Total AMC Class*
1 April 28, 1990 0.00 0.00 0.00 0.00 0.00 0.00 I
2 June 7, 1990 0.00 0.00 0.00 0.00 0.00 0.00 I
3 June 26, 1990 6.10 9.91 26.42 7.37 0.00 49.78 II
4 August 17, 1990 0.00 0.00 7.62 19.56 0.25 27.43 I
5 April 24, 1995 0.00 0.51 19.05 0.00 0.00 19.56 II
6 May 12, 1995 0.00 0.00 0.00 0.00 18.29 18.29 II
7 January 3, 1999 0.00 0.00 0.00 0.00 22.86 22.86 II
8 January 9, 1999 0.15 0.00 0.00 0.00 0.00 0.15 I
9 January 23, 1999 4.70 0.10 0.00 0.00 0.08 4.88 I
10 February 28, 1999 0.00 0.00 0.00 0.00 0.20 0.20 I
11 March 14, 1999 0.00 0.00 0.00 0.00 0.00 0.00 I
12 April 1, 1999 0.00 0.00 0.00 0.00 0.00 0.00 I
13 April 16, 1999 0.00 0.00 0.00 0.00 0.00 0.00 I
14 June 16, 1999 0.41 0.15 1.78 0.71 8.74 11.79 I
15 March 30, 2000 0.00 5.89 12.70 0.00 0.15 18.75 II
16 September 29, 2000 10.16 0.63 0.00 0.00 0.00 10.79 II
*For areas classified as urban built-up, water, and wetlands, AMC II values were used.
E
SSE
SST
1
SSE
n
Observed edicted
i i
i
n

_
,

1
1
1
2
Pr
SST
n
Observed Observed
i
i
n

_
,

1
1
1
2
(5)
(6)
(7)
6 of the 13 storms that were not used to fit
the Snyder parameters having efficiencies
greater than 0.7. Observed and predicted
hydrographs for all storms are shown in
Figures 6a through 6p. The SDDH method
out-performed the Snyder method for 8 of
13 storms not used to fit the Snyder
parameters, performed comparably for
three of 13 storms (i.e., Storms 2, 12, and
16) and performed worse on 2 of 13 storms
(i.e., Storms 9 and 13). A few representa-
tive storms are discussed in more detail
below.
Figure 6d shows the predicted and
observed hydrographs for Storm 4 (August
17, 1990), a well fit small storm, and Fig-
ure 6l shows the predicted and observed
hydrographs for Storm 13 (April 16, 1999),
a poorly fit small storm. Both storms had
approximately the same total rainfall vol-
ume. However, Storm 4 lasted three hours
with an average intensity of approximate-
ly 7 mm/hr, whereas Storm 13 lasted 13
hours with an average rainfall intensity of
approximately 1.35 mm/hr. Furthermore,
although both storms were classified as
AMC I, 27.43 mm of rainfall had occurred
in the five days prior to Storm 4, while no
rainfall had occurred in the five days prior
to Storm 13. As a result, Storm 13 had a
much smaller observed peak flow rate and
observed runoff volume than Storm 4, a
characteristic that was well predicted by
the SDDH model. The SDDH method pre-
dicted the shape of the Storm 4 hydro-
graph extremely well, accurately
reproducing the peak flow rate, time to
peak, and total runoff volume. However for
the longer more complex Storm 13, the
SDDH predicted a much more diffuse,
complex response than observed. The first
peak predicted by the SDDH method for
Storm 13 corresponds fairly well with that
observed hydrograph peak. However the
predicted larger second peak, correspond-
ing to the second large rainfall peak, was
not observed in the actual data.
Figure 6f shows the predicted and
observed hydrographs for Storm 6 (May
12, 1995), a well fit medium storm, and
Figure 6e shows the predicted and
observed hydrographs for Storm 5 (April
24, 1995), an adequately fit medium
storm. These storms had comparable rain-
fall volumes and comparable AMCs. How-
ever, Storm 5 was a shorter duration, more
JAWRA 872 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
MELESSE AND GRAHAM
F
i
g
u
r
e

4
.

S
p
a
t
i
a
l

D
i
s
t
r
i
b
u
t
i
o
n

o
f

R
u
n
o
f
f

C
u
r
v
e

N
u
m
b
e
r
s
:

(
a
)

1
9
9
0
,

(
b
)

1
9
9
5
,

a
n
d

(
c
)

2
0
0
0
intense storm. The SDDH method
predicted the shape of the Storm 6
hydrograph very well, accurately
reproducing the peak flow rate, time to
peak, and total runoff volume. Howev-
er, for higher intensity Storm 5, the
SDDH method overpredicted the time
to peak and overpredicted the peak
flow rate.
Figure 6c shows the predicted and
observed hydrographs for Storm 3
(June 26, 1990), the best fit large
storm, and Figure 6b shows the pre-
dicted and observed hydrographs for
Storm 2 (June 7, 1990), the worst fit
large storm. Storm 3 was the largest
storm studied with a total depth of
70.6 mm, an intensity of 3.21 mm/hr,
and had 49.78 mm of antecedent rain-
fall. Storm 2 was a more intense storm
with a total depth of 51.42 mm, an
intensity of 12.86 mm/hr, and no
antecedent rainfall. The SDDH
method predicted the shape of the
Storm 3 hydrograph very well, accu-
rately reproducing the peak flow rate,
time to peak, and total runoff volume,
and achieving an efficiency of 0.88. For
Storm 2, the SDDH method performed
very poorly, overpredicting the peak
flow rate significantly for this very
intense storm.
SUMMARY AND CONCLUSIONS
This study contributes a SDDH
technique to predict streamflow based
on runoff travel time within the water-
shed using spatially distributed data
(rainfall, roughness, and land cover).
The method developed uses only
DEMs, rainfall, and land cover, and
thus can be used in a GIS environment
to predict streamflow in large,
ungaged watersheds where predicting
the effects of alternative land use sce-
narios is of interest. When reliable
spatially distributed rainfall data are
available, the SDDH approach may be
preferable to the time area method
(D.R. Maidment 1992, unpublished
report) because it can directly use spa-
tially distributed rainfall excess,
whereas the time area method is a
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 873 JAWRA
STORM RUNOFF PREDICTION BASED ON A SPATIALLY DISTRIBUTED TRAVEL TIME METHOD UTILIZING REMOTE SENSING AND GIS
F
i
g
u
r
e

5
.

S
p
a
t
i
a
l

D
i
s
t
r
i
b
u
t
i
o
n

o
f

M
a
n
n
i
n
g

s

R
o
u
g
h
n
e
s
s

C
o
e
f
f
i
c
i
e
n
t
:

(
a
)

1
9
9
0
,

(
b
)

1
9
9
5
,

a
n
d

(
c
)

2
0
0
0
unit hydrograph approach, which requires spatially
constant precipitation. The overland travel time
method is based on physical concepts and does a good
job estimating travel time distribution of runoff to the
outlet of the watershed. However, calibration of the
channel width and cell threshold casts some doubt on
the physical basis for these parameters. These cali-
brated parameters must be verified before applying
the method to other watersheds.
The spatially distributed NRCS CNs determined
from the soil maps, antecedent precipitation, and land
cover derived from DOQQs and satellite imagery pre-
dicted the precipitation excess leading to direct runoff
very well, with an overall efficiency of 0.98. Direct
runoff hydrographs predicted by the SDDH travel
time approach accurately predicted the observed
hydrographs, with an average efficiency for the 16
storm events of 0.65. In general, the SDDH method
performed better for storms that had higher AMCs,
were simple in structure, and moderate in volume
and intensity. The method reproduced the hydro-
graphs for Storms 1, 3, 4, 6, 7, 12, and 14 extremely
well, with efficiencies over 0.80. The method per-
formed most poorly for Storms 2 and 9, which were
both high volume storms with low AMCs. The SDDH
method also performed relatively poorly for Storm 13,
which was a long duration, low intensity, low
antecedent moisture storm, and for Storm 10, which
was a short duration, high intensity, low antecedent
moisture storm. Errors in direct runoff hydrograph
predictions may be due to a variety of sources includ-
ing inaccurate estimation of spatially distributed
rainfall, inaccurate estimation of antecedent soil
moisture, and violations of the assumptions used to
derive the overland flow velocities.
JAWRA 874 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
MELESSE AND GRAHAM
TABLE 4. Model Prediction and Prediction Error Statistics.
Observed SDDH Snyder
Time Time Time
to to to
Storm Peak Peak Volume Peak Peak Volume Peak Peak Volume
No. Storm Date (m
3
/s) (hr) (mm) (m
3
/s) (hr) (mm) Efficiency (m
3
/s) (hr) (mm) Efficiency
1 April 28, 1990 1.09 20 1.51 1.10 31 1.53 0.88 1.12 20 1.53 0.99
2 June 7, 1990 1.22 32 1.39 2.11 30 2.53 -0.64 2.11 20 2.53 -0.10
3 June 26, 1990 0.94 64 2.59 0.90 65 2.09 0.85 1.30 35 2.09 0.09
4 August 17, 1990 1.64 39 3.19 1.61 50 3.09 0.93 2.26 22 3.10 0.66
5 April 24, 1995 0.76 23 1.38 0.98 36 1.38 0.63 0.76 17 1.35 0.94
6 May 12, 1995 2.97 30 3.84 3.07 34 4.01 0.99 2.11 23 3.94 0.74
7 January 3, 1999 0.48 22 0.62 0.50 31 0.67 0.85 0.37 13 0.66 0.76
8 January 9, 1999 0.49 27 0.88 0.52 25 0.54 0.77 0.30 16 0.53 0.60
9 January 23, 1999 5.14 32 8.16 7.34 27 8.41 0.38 4.50 18 8.33 0.81
10 February 28, 1999 0.81 13 1.46 1.37 20 1.23 0.34 0.67 13 1.23 0.93
11 March 14, 1999 1.71 27 2.19 2.30 21 1.68 0.79 0.91 17 1.66 0.54
12 April 1, 1999 0.69 27 1.37 0.72 24 1.15 0.86 0.84 21 1.16 0.90
13 April 16, 1999 0.25 33 0.44 0.18 76 0.46 0.40 0.25 21 0.46 0.79
14 June 16, 1999 1.11 24 1.27 0.82 33 1.30 0.83 0.72 14 1.29 0.60
15 March 30, 2000 0.51 31 0.69 0.58 36 0.78 0.71 0.43 14 0.77 0.49
16 September 29, 2000 6.08 32 11.34 5.61 35 10.47 0.77 5.74 20 10.36 0.90
Average Error -0.24 -6.13 0.05 0.75 12.77 0.08
Error Standard Deviation 0.62 11.54 0.42 0.65 5.88 0.49
Efficiency 0.85 -0.43 0.98 0.65 0.92 -0.81 0.98 0.60
Note: Storms 1, 5, and 10 are not included in error statistics for the Snyder method.
TABLE 3. Snyder Synthetic Unit Hydrograph
Parameters for Simms Creek.
Parameters 1990 1995 1999/2000
A (km
2
) 114
L (km) 17.3
L
c
(km) 5.8
C
t
6.5 4.9 4.7
C
p
0.07 0.05 0.052
t
r
(hr) 3 3 2
t
p
(hr) 20.5 15.5 11.5
5.5t
r
(hr) 16.5 16.5 11
t
b
(hr) 88 115 114
The Snyder synthetic unit hydrograph, a lumped
approach to predicting stream flow hydrographs, also
performed quite well with an average efficiency of
0.60. The disadvantage of this method, however, is
that the parameters for the Snyder unit hydrograph
must be fit from existing rainfall/runoff data. Thus, it
cannot be used to predict runoff response for proposed
land use changes. In addition, the method cannot
account for nonuniform, spatially distributed rainfall.
On the other hand, the Snyder technique uses fewer
parameters than the SDDH method, which also
requires calibration of the channel threshold and
channel width.
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 875 JAWRA
STORM RUNOFF PREDICTION BASED ON A SPATIALLY DISTRIBUTED TRAVEL TIME METHOD UTILIZING REMOTE SENSING AND GIS
Figure 6. Predicted Versus Observed Runoff Hydrographs for 16 Storms
(note that Storms 1, 5, and 10 were used to fit the Snyder parameters).
Although the method predicted total runoff volume
accurately for most storms, the shape of the hydro-
graph was predicted less accurately for low AMC.
Since the AMC determined from antecedent rainfall is
not accurate and soil moisture data are very tran-
sient, the use of remotely sensed spatial moisture
data might improve the antecedent moisture estima-
tion, and hence runoff prediction. Accurate rainfall
data for the watershed are also very critical for
streamflow prediction. Therefore, the use of spatially
distributed rainfall from radar may provide improve-
ment over the Thiessen method used in this study
since radar sensed rainfall has better spatial accuracy
than a sparse network of rain gage stations.
JAWRA 876 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
MELESSE AND GRAHAM
Figure 6 (contd). Predicted Versus Observed Runoff Hydrographs for 16 Storms
(note that Storms 1, 5, and 10 were used to fit the Snyder parameters).
To better assess the performance of the runoff rout-
ing technique, it is recommended that the model be
tested on different watersheds of varying topography
and land use. The optimal effective channel flow
width of 1 m, and the optimal channel threshold of
one contributing upstream cell, as found for the
Simms Creek watershed, need to be evaluated for
other watersheds of different topography and geome-
try.
The routing technique developed here does not con-
sider storage effects of wetlands, lakes, and depres-
sions. Including the effect of storage could
significantly affect the time to peak and peak flow of
predicted hydrographs, and might improve the
streamflow prediction for low antecedent moisture
storms. Since the kinematic wave model used to
develop the overland flow travel times assumes fric-
tion and bed slope are identical, its applicability for
flat topography is limited and using diffusive wave
assumptions to estimate overland flow velocities
should be explored for these situations.
APPENDIX A
SENSITIVITY ANALYSIS
Sensitivity analysis was conducted to assess the
change in the shape, peak flow, and time to peak of
the hydrographs for changes in model parameters. In
this study, sensitivity analysis was carried out for the
threshold value for stream network delineation (i.e.,
classification of overland versus channel cells) and
effective width of concentrated channel flow.
Sensitivity analyses were conducted for three levels
of channel threshold (1, 5, and 10 cells) and three
channel widths (0.2 m, 1.0 m, and 3.0 m). Table A-1
summarizes the hydrograph prediction efficiencies for
each storm using these values. These data indicate
that, by a significant margin, the best prediction effi-
ciency was found for the threshold value of 1 cell and
an effective channel width of 1 m. Larger channel
thresholds resulted in slower travel times that signifi-
cantly delayed the time to peak and time to recession
compared to the observed data. Similarly, channel
widths smaller than 1 m increased the channel flow
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 877 JAWRA
STORM RUNOFF PREDICTION BASED ON A SPATIALLY DISTRIBUTED TRAVEL TIME METHOD UTILIZING REMOTE SENSING AND GIS
Figure 6 (contd). Predicted Versus Observed Runoff Hydrographs for 16 Storms
(note that Storms 1, 5, and 10 were used to fit the Snyder parameters).
velocity resulting in shorter travel times and higher
peak flows, and larger channel widths reduced the
channel velocity and resulted in longer travel times
and lower peak values than the observed data.
It should be noted that the values of channel
threshold and channel width, as recommended and
selected in this study, may not be the recommended
values for other watersheds of different geomorpholo-
gy and dominant land cover.
ACKNOWLEDGMENTS
This study was made possible by the financial support from the
Tsao-Jiin Memorial Foundation of the Kaohsiung Irrigation Associ-
ation, and the University of Florida Institute of Food and Agricul-
tural Sciences Center for Remote Sensing. The authors extend their
appreciation to the St. Johns River Water Management District for
providing rainfall and stream flow data. This is Florida Agricultur-
al Experiment Station Journal Series No. R-09238.
LITERATURE CITED
Ajward, M.H., 1996. A Spatially Distributed Unit Hydrograph
Model Using a Geographical Information System. Ph.D. Disser-
tation, Civil Engineering Dept., University of Calgary, Calgary.
Anderson, J.R., E.E. Hardy, T.J. Roach, and R.E. Witmer, 1976.
A Land Use and Land Cover Classification System for Use With
Remote Sensor Data. U.S. Geological Survey Prof. Paper 964,
U.S. Government Printing Office, Washington. D.C.
Brater, E.F. and H.W. King, 1976. Handbook of Hydraulics for the
Solution of Hydraulic Engineering Problems. McGraw-Hill Book
Company, New York, New York.
Chow, V.T., D.R Maidment, and L.W. Mays, 1988. Applied Hydrolo-
gy. McGraw-Hill Inc., New York, New York.
Dingman, S.L., 2001. Physical Hydrology (Second Edition). Pearson
Education, pp. 656.
ERDAS (Earth Resources Data Analysis System), 1999. ERDAS
Field Guide (Fifth Edition). ERDAS, Inc, Atlanta, Georgia, 671
pp.
Kite, G.W. and N. Kouwen, 1992. Watershed Modeling Using Land
Classifications. Water Resources Research 28(12):3193-3200.
Kumar, S. and S.C. Jain, 1982. Application of SCS Infiltration
Model. Water Resources Bulletin 18:503-507.
Maidment, D.R., 1992a. Grid-Based Computation of Runoff: A Pre-
liminary Assessment. Contract DACW04-92-P-1983, Hydrologic
Engineering Center, U.S. Army Corps of Engineers, Davis, Cali-
fornia (unpublished).
Maidment, D.R., 1992b. A Grid-Network Procedure for Hydrologic
Modeling. Contract DACW04-92-P-1983, Hydrologic Engineer-
ing Center, U.S. Army Corps of Engineers, Davis, California
(unpublished).
Maidment, D.R., 1993. GIS and Hydrologic Modeling. In: Environ-
mental Modeling With GIS, M.F. Goodchild, B.O. Parks,
and L. Steyaert (Editors). Oxford University Press, New York,
New York.
Maidment, D.R., J.F. Olivera, A. Calver, A. Eatherral, and
W. Fraczek, 1996. A Unit Hydrograph Derived From a Spatially
Distributed Velocity Field. Hydrologic Processes 10:831-844.
Melesse, A.M., 2002. Spatially Distributed Storm Runoff Modeling
Using Remote Sensing and Geographic Information Systems.
Ph.D. Dissertation, Agri. and Bio. Engineering Dept., Universi-
ty of Florida, Gainesville, Florida.
Melesse, A.M., J.D Jordan, and W.D Graham, 2001. Enhancing
Land Cover Mapping Using Landsat Derived Surface Tempera-
ture and NDVI. World Water and Environmental Resources
Congress, American Society of Civil Engineers, CD Proceed-
ings).
Montes, S., 1998. Hydraulics of Open Channel Flows. ASCE Press,
Reston, Virginia, 697 pp.
JAWRA 878 JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
MELESSE AND GRAHAM
TABLE A-1. Prediction Efficiencies for Channel Width and Channel Threshold Value Sensitivity Analysis
Storm Channel Width Threshold Value
No. Storm Date 0.2 m 1 m 3 m 1 Cell 5 Cells 10 Cells
1 April 28, 1990 -1.80 0.88 0.11 0.88 0.61 0.22
2 June 7, 1990 -2.34 -0.64 -1.05 -0.64 -0.62 -1.24
3 June 26, 1990 -1.56 0.85 0.71 0.85 0.71 0.36
4 August 17, 1990 -1.22 0.93 0.48 0.93 0.58 0.11
5 April 24, 1995 -1.23 0.63 0.24 0.63 0.39 -0.02
6 May 12, 1995 0.50 0.99 0.26 0.99 0.45 -0.13
7 January 3, 1999 -0.65 0.85 -0.46 0.85 -0.13 -0.58
8 January 9, 1999 0.03 0.77 -0.30 0.77 0.18 -0.21
9 January 23, 1999 -2.75 0.38 0.33 0.38 0.66 0.16
10 February 28, 1999 -0.14 0.34 0.34 0.34 0.39 0.00
11 March 14, 1999 -0.17 0.79 -1.11 0.79 -0.34 -0.55
12 April 1, 1999 -2.11 0.86 0.54 0.86 0.74 0.43
13 April 16, 1999 0.70 0.40 -0.73 0.40 -0.40 -0.95
14 June 16, 1999 -1.13 0.83 -0.82 0.83 -0.18 -0.77
15 March 30, 2000 -4.77 0.71 0.40 0.71 0.69 -0.25
16 September 29, 2000 -0.50 0.77 -0.02 0.77 -0.02 -0.56
Average Efficiency -1.20 0.65 -0.07 0.65 0.23 -0.25
Note: Bold data indicate best fit values.
Mostaghimi, S. and J.K. Mitchell, 1982. Peak Runoff Model Com-
parison on Central Illinois Watersheds. Water Resources Bul-
letin 18(1):9-18.
Muzik I., 1995. GIS Derived Distributed Unit Hydrograph, A New
Tool for Flood Modeling. In: Developments in Computer Aided
Design and Modeling for Civil Engineering, B.H.V. Topping
(Editor). Civil-Comp Press, Edinburg, United Kingdom, pp. 243-
247.
Nash, J.E. and J.V. Sutcliffe, 1970. River Flow Forecasting Through
Conceptual Models. A Discussion of Principles. Journal of
Hydrology 10:282-290.
Neumann, P., W. Fett and G.A. Schultz, 1990. A Geographic Infor-
mation System as Database for Distributed Hydrological Mod-
els. Proc., Int. Symp., Remote Sensing and Water Resources,
Enschede, The Netherlands, pp. 781-791.
Olivera, J.F. and D.R. Maidment, 1999. Geographic Information
Systems (GIS)-Based Spatially Distributed Model for Runoff
Routing. Water Resour. Res. 35(4):1155-1164.
Ott, M., Z. Su, A.H. Schumann, and G.A. Schultz, 1991. Develop-
ment of a Distributed Hydrological Model for Flood Forecasting
and Impact Assessment of Land-Use Change in the Internation-
al Mosel River Basin. Hydrology for the Water Management of
Large River Basins, Proc. of the Vienna Symposium, IAHS Pub.
No. 201.
Schultz, G.A., 1993. Hydrological Modeling Based on Remote Sens-
ing Information. Adv. Space Res., 13(5):149-166.
SJRWMD (St. Johns River Water Management District), 2002. Spa-
tial Data Lineage Report. Available at http://www.sjrwmd.
com/programs/data.html Accessed on May 23, 2002.
Tou, J.T. and R.C. Gonzalez, 1974. ISODATA Algorithm in Pattern
Classification by Distance Functions. In: Pattern Recognition
Principles. Addison-Wesley, Reading, Massachusetts, pp. 97-104.
USDA (U.S. Department of Agriculture), 1985. National Engineer-
ing Handbook Section 4: Hydrology. Soil Conservation Service,
U.S. Government Printing Office, Washington, D.C.
USDA (U.S. Department of Agriculture), 1986. Urban Hydrology
for Small Watersheds. Soil Conservation Service, Technical
Release 55, U.S. Government Printing Office, Washington, D.C.
Wood, M.K and W.H. Blackburn, 1984. An Evaluation of the Hydro-
logic Soil Groups as Used in the SCS Runoff Method on Range-
lands. Water Res. Bulletin 20:379-389.
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION 879 JAWRA
STORM RUNOFF PREDICTION BASED ON A SPATIALLY DISTRIBUTED TRAVEL TIME METHOD UTILIZING REMOTE SENSING AND GIS

You might also like