Using the WRF Model in an Operational Streamflow Forecast

Transcription

Using the WRF Model in an Operational Streamflow Forecast
FEBRUARY 2012
GIVATI ET AL.
285
Using the WRF Model in an Operational Streamflow Forecast System
for the Jordan River
AMIR GIVATI
Israeli Hydrological Service, Israeli Water Authority, Jerusalem, Israel
BARRY LYNN
Weather-It-Is Ltd., Efrat, Israel
YUBAO LIU
National Center for Atmospheric Research,* Boulder, Colorado
ALON RIMMER
The Lake Kinneret Limnological Laboratory, Israel Oceanographic and Limnological Research,
Ltd., Migdal, Israel
(Manuscript received 19 April 2011, in final form 11 September 2011)
ABSTRACT
The Weather Research and Forecasting (WRF) model was employed to provide precipitation forecasts during
the 2008/09 and 2009/10 winters (wet season) for Israel and the surrounding region where complex terrain
dominates. The WRF precipitation prediction has been coupled with the Hydrological Model for Karst Environment (HYMKE) to forecast the upper Jordan River streamflow. The daily WRF precipitation forecasts were
verified against the measurements from a dense network of rain gauges in northern and central Israel, and the
simulation results using the high-resolution WRF indicated good agreement with the actual measurements. The
daily precipitation amount calculated by WRF at rain gauges located in the upper parts of the Jordan River basin
showed good agreement with the actual measurements. Numerical experiments were carried out to test the
impact of the WRF model resolution and WRF microphysical schemes, to determine an optimal model configuration for this application. Because of orographic forcing in the region, it is necessary to run WRF with a
4–1.3-km grid increment and with sophisticated microphysical schemes that consider liquid water, ice, snow, and
graupel to produce quality precipitation predictions. The hydrological modeling system that ingests the highresolution WRF forecast precipitation produced good results and improved upon the operational streamflow
forecast method for the Jordan River that is now in use. The modeling tools presented in this study are used to
support the water-resource-assessment process and studies of seasonal hydroclimatic forecasting in this region.
1. Introduction
Accurate analysis and prediction of precipitation
amounts and their spatial distribution are vital for
regional- and local-scale hydrological applications.
* The National Center for Atmospheric Research is sponsored
by the National Science Foundation.
Corresponding author address: Amir Givati, Israeli Hydrological
Service, Israeli Water Authority, P.O. Box 36118, Jerusalem 91360,
Israel.
E-mail: [email protected]
DOI: 10.1175/JAMC-D-11-082.1
Ó 2012 American Meteorological Society
This is especially true for arid and semiarid regions
such as in the Middle East, where estimation and prediction of the highly variable precipitation during the
rainy season is critical for predicting streamflows and the
recharge of reservoirs. The estimates of the streamflow in
the Jordan River and the water level of Lake Kinneret
play a crucial role in Israeli agricultural and hydrological
planning and in flood control. Hydrological forecasts are
instrumental for decision-support activities at the Israel
Water Authority.
Major operational weather forecast centers provide
relatively coarse (;16–25-km grid increment) precipitation analyses and forecasts, which are incapable of
286
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
resolving the necessary details of the complex precipitation structures that are forced by mesoscale orography,
land surface heterogeneities, and land–water contrasts.
In the eastern Mediterranean region, strong air–sea interaction and orographic forcing produce precipitation
with dramatic gradients that are generally missed by
the coarse-grid operational models. In Israel, the precipitation patterns are particularly complex, and large
precipitation contrasts occur over a relatively small
geographical distance (2–10 km). Large climatological
precipitation gradients in Israel are caused by the preferred tracks of extratropical cyclones, the complex
orography, and the coastline shape. Rich spatial structures are observed in individual precipitation events as
well as in the monthly, seasonal, and annual accumulated precipitation amounts (Saaroni et al. 2009).
Because an objective of this study is to employ a highhorizontal-resolution model to better predict precipitation structures in Israel for streamflow calculations and
prediction, it is important to review previous studies
of the relationship between horizontal resolution and
the predictability of precipitation, In fact, other studies
have shown that reducing the horizontal grid spacing
of the atmospheric model may enhance the model’s
ability to simulate precipitation (Katzfey 1995; Martin
1996; McQueen et al. 1995; Doyle 1997; Colle et al. 1999;
Davis and Carr 2000; Petch et al. 2002; Adlerman and
Droegemeier 2002; Bryan et al. 2003; Xue and Martin
2006; Schwartz at el. 2009; Kain at el. 2006, 2008; Weisman
et al. 2008), and many studies have used high-resolution
numerical weather predictions in hydrologic applications. For example, Hay et al. (2002) used daily precipitation and temperature on a grid with 52-km grid
spacing as input to a distributed hydrologic model for
four basins in the United States. Their results indicated
that the coarse-resolution regional-climate-model output could be used reliably for basin-scale hydrological
modeling after bias correction; even when bias corrected, however, the regional-climate-model output did
not contain the day-to-day variability necessary for hydrologic modeling.
Grell et al. (2000) showed that high-resolution (1-km
grid increment) model simulations might be necessary
when the complex pattern of the precipitation distribution becomes important. Leung and Qian (2003) examined the sensitivity of precipitation and snowpack
simulations in complex terrain to model resolution by
employing a nested regional climate model. They found
that higher horizontal spatial resolution improves the
accuracy of simulated snowpack. Westrick and Mass (2001)
looked at a 7-day rain-on-snow event in the Snowqualmie
River watershed using the nesting capabilities of the fifthgeneration Pennsylvania State University–National Center
VOLUME 51
for Atmospheric Research Mesoscale Model (MM5;
36-, 12-, and 4-km grid spacing) and the Distributed
Hydrology–Soil–Vegetation Model (DHSVM) and found
that the accuracy of the atmospheric fields increased
with higher horizontal resolution, although underforecasting of precipitation was evident for all three
resolutions. The simulated river flows captured 67%
(36 km), 75% (12 km), and 72% (4 km) of the total
flow and 52% (36 km), 58% (12 km), and 62% (4 km)
of the event peak flow.
Mass et al. (2002) used MM5 and DHSVM to show that
increasing the spatial resolution improved the forecast in
terms of various meteorological variables. Improvement
of the forecast from an increase in the atmospheric
model resolution was also found by Lin et al. (2002) and
Seuffert et al. (2002). Hay et al. (2006) examined the
accuracy of high-resolution nested mesoscale-model
simulations (MM5 with 5- and 1.7-km grid increments)
as an input to the U.S. Geological Survey’s distributed
hydrologic model (Precipitation Runoff Modeling System). Their results showed a general increase in the accuracy of simulated runoff with an increase in resolution.
Younis et al. (2008) showed the benefit of using highresolution operational weather forecasts for flash-flood
warning in France. Daly et al. (1994) used a statistical–
topographical model for mapping climatological precipitation over mountainous terrain in the western United
States, and Pandey et al. (1999) and Neiman et al. (2002)
related orographic precipitation in California at different
elevations to low-level winds and moisture.
2. Existing methods for calculating streamflow in
the Jordan River
Another type of regional rainfall analysis, which is
especially relevant to the Lake Kinneret watershed, is
the regression method of Rimmer and Salingar (2006),
which was developed especially for the Hydrological
Model of Karst Environment (HYMKE). The model was
applied to the Mount Hermon range (70–2800 m MSL) in
the upper catchments of the Jordan River, an area with
complex karst terrain that contributes most of the water
to the Jordan River. At the time of this study, precipitation measurements were not available from most of the
Mount Hermon range, especially from the upper parts
in Syria and Lebanon. Therefore, precipitation measurements from eight lower-elevation rain gauges in the
north of Israel, with elevations between 100 and 960 m
MSL, were used to create a linear regression model of
precipitation as a function of elevation. The authors extrapolated the precipitation from the lower elevations to
the full Mount Hermon range and created maps of
monthly and annual precipitation.
FEBRUARY 2012
GIVATI ET AL.
Similar regression methods were used in other studies
of rainfall and spring river flow in this nonmonitored
mountainous region (Simpson and Carmi 1983; Gilad
and Bonne 1990; Rosenfeld and Farbstein 1992; Gur
et al. 2003). Using this regression analysis, Rimmer and
Salingar (2006) showed that the linearity of rainfall
versus elevation was slightly weaker and less significant
during the beginning of the winter (October), increased
toward the middle of the winter (January), and gradually decreased toward its end (April). Halfon (2008) and
Saaroni et al. (2009) describe the overall precipitation
forcing in Israel, showing that the Cyprus-low storms are
associated with strong westerly winds, which enhances
the orographic effect, whereas the convective component of precipitation during the autumn and spring is
much higher. This can probably explain the difference in
correlation strength between precipitation and elevation that was described by Rimmer and Salingar (2006).
The main purpose of this paper is to offer a new tool
that will improve the current streamflow calculations
and forecasting. By using the high-resolution WRF with
improved precipitation prediction for every possible
point in the Hermon basin, we can drive the hydrological
simulations in a more physical way by representing well
the precipitation fields in the basin. This system will allow us to produce improved hydrological predictions,
relative to the traditional approach that is based on
rainfall extrapolations for high elevation, using the empirical elevation–precipitation relation. This work can
also help to develop future scenarios for water availability at the Jordan River under climate-change conditions using dynamical downscaling for existing regional
climate models for the twenty-first century.
3. The study areas
The study area in this work includes the Lake Kinneret
watershed, located in the central part of the Jordan Rift
Valley (northern Israel; see Figs. 1a,b). Lake Kinneret is
the most important surface water resource in Israel,
providing approximately 35% of the annual drinking
water to a population of almost 7.5 million inhabitants.
The major water source for the lake is the ;1700-km2
upper catchments of the Jordan River (UCJR), including ;920 km2 in Israel with the remainder located in
Syria and Lebanon. In the north of the UCJR, the higher
elevations of the Mount Hermon range (1200–2800 m)
are the wettest area in the Lake Kinneret watershed,
where the average annual precipitation ranges from 1200
to 1500 mm.
The Mount Hermon basins feed the three major
tributaries of the Jordan River: the Dan with ;250
million m3 (Mm3) annually, the Hermon (also known as
287
Banias) with ;110 Mm3, and the Snir (also known as the
Hazbani River) with ;115 Mm3. The Mount Hermon
area and the surrounding region contribute over 60% of
the water to the major rivers in the Kinneret basin. The
contribution of the Jordan River and its three tributaries
to the national water balance is even higher during dry
years. The other part of the Lake Kinneret basin is the
direct watershed, located in the immediate vicinity of
the lake. The area of the direct watershed is ;965 km2,
577 km2 of which are in the southern part of the Golan
Heights to the east of the lake and the other 388 km2
of which are in the eastern Galilee Mountains to the
west of the lake. The average annual water flowing into
Lake Kinneret over the past decades is 660 Mm3 (for
the period 1979–2007) including the Jordan River
(455 Mm3), direct rainfall (65 Mm3), direct watershed
runoff and artificial diversion (100 Mm3), and springs
flowing directly into the lake (40 Mm3). An average of
230 Mm3 evaporates every year (Mekorot Watershed
Unit 2008) so that the average annual volume of available water is ;410 Mm3. Most of this volume is pumped
out of the lake every year for consumption, while some
water occasionally overflows to the Lower Jordan River
toward the Dead Sea.
4. Methods
a. The WRF model
WRF, version 3.011, was set up with a 90 3 90 point
horizontal-grid outer domain with a 36-km grid interval;
a second 136 3 136 point domain at 12-km grid spacing,
which is nested in domain 1; and a third 181 3 181 point
nested-grid domain at 4-km grid spacing, which is nested
in domain 2. To study the impact of model resolution,
simulations were carried out individually by dropping
the fine-mesh domains in sequence. Also tested is another triply nested model with 27-, 9-, and 3-km grid
spacings. Experiments with a 1.3-km grid domain were
also performed for selected cases. The model domain is
displayed in Figs. 2a,b.
Each simulation was begun at 0000 UTC and had a duration of 30 h. The first 6 h of the forecast were dropped
to allow for model spinup. Hence, the precipitation simulations that are valid between 0600 and 0600 UTC of
the following day were analyzed in this paper. The WRF
simulations were initialized with 0.58 National Oceanic and
Atmospheric Administration National Centers for Environmental Prediction Global Forecast System (NCEP
GFS) model analysis.
Note that this study has been based on ‘‘cold start’’
initiation WRF forecasts. During the last two decades,
several advanced mesoscale data assimilation approaches,
288
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 51
FIG. 1. Maps of the (a) eastern Mediterranean sea and adjacent land and (b) the Lake Kinneret watershed, which
includes the international borders, the UCJR, and the direct watersheds (east and west).
such as the National Center for Atmospheric Research
real-time four-dimensional data assimilation (Liu et al.
2008a,b), have been developed to improve short-term
quantitative precipitation forecasts.
b. Choosing optimal WRF microphysical schemes
and resolution
The first step in the study was to test a variety of WRF
microphysical schemes and different horizontal resolutions for a subset of weather events in Israel. WRF allows several choices of microphysical parameterization
schemes. The microphysics schemes tested in this study
include mp2 and mp6 (WRF single-moment six-class
scheme; Hong and Lim 2006), mp7 (the Goddard microphysics scheme), mp8 (the Thompson scheme), and
mp10 (the Morrison double-moment scheme, which
predicts both mass and number of each category of
cloud/precipitation particles). To choose the optimal
model configuration for Israel, WRF was run for a few
winter storms: 23–25 December 2008, 10–12 February
2009, and 27 February–1 March 2009. The ‘‘best’’ model
configuration was determined by comparing the results
of the model experiments with a cluster of 17 rain gauge
observations in northern and central Israel (see the rain
gauge locations and the different aquifers in Israel in
Fig. 3).
The Kain–Fritsch convective parameterization scheme
was activated in the WRF domains that had grid sizes
larger than 4 km. The ‘‘Noah’’ land surface model and
the 1.5-moment closure TKE-based boundary layer
scheme were used in all model runs. Both schemes are
described in detail in the WRF user’s manual. The GFS
forecasts were used to update the WRF boundary conditions. The model in its final configuration was run
FEBRUARY 2012
GIVATI ET AL.
289
FIG. 2. (a) Map of the four nested grids of WRF simulations in the eastern Mediterranean region (D1 5 36 km,
D2 5 12 km, D3 5 4 km, and D4 5 1.3 km from outer grid to innermost grid, respectively), and (b) the Lake
Kinneret watershed topographical map.
operationally with 36-, 12-, 4-, and 1.3-km grid-increment
domains for the winters of 2008/09 and 2009/10.
c. The hydrological model: HYMKE
HYMKE was developed by Rimmer and Salingar
(2006) and is employed in this study. For this study,
HYMKE was set up to simulate the hydrological processes in the Mount Hermon region where there are
three surface flow catchments and several large karstic
springs originating from groundwater. HYMKE contains
four main modules (see the model schematic description
in Fig. 4): the surface layer (labeled 0), the vadose zone
(labeled 1), groundwater (labeled 2), and surface flow
(labeled 3). Here, the land surface processes of the entire
geographical basin include recharge by precipitation,
drying by evaporation, surface runoff, and percolation to
deeper layers. A more detailed illustration of HYMKE
can be found in Rimmer and Salingar (2006).
The karstic nature of the lands was modeled with
a surface layer (‘‘epikarst’’) composed of both low- and
high-permeability sections that feed the karst network.
The surface layer is drained continuously as a function
of moisture content. Saturation excess is generated
when the surface layer is saturated. Part of the excess
saturation is then transformed into the surface flow
(module 3) while the other part forms a downward
preferential flow component. Therefore, the percolation
into the vadose zone (module 1) includes both a ‘‘slow
flow’’ component through the matrix and a ‘‘quick flow’’
component through the high-permeability section, where
the latter is active mainly during the peak of the wet season. The output from the vadose zone (module 1) feeds the
groundwater reservoir (module 2). The location of the
springs and the differences among the patterns of spring
discharge required the separation of module 2 into four
groundwater linear reservoirs.
Reservoir (or ‘‘tank’’)-type models (Dooge 1973;
Sugawara 1995) are a common way to translate a conceptual model of a groundwater system into mathematical formulations. The linear reservoir concept, in which
discharge is linearly dependent on the reservoir storage,
is especially useful in karst environments, because the
necessary information for physical approaches is usually
not available (Jukic and Denic-Jukic 2009). This has become evident in a large number of recently published
studies (e.g., Fleury et al. 2007; Hartmann et al. 2011;
Jukic and Denic-Jukic 2009; Kessler and Kafri 2007; Le
Moine et al. 2008; Rimmer and Salingar 2006).
In the Mount Hermon region, three linear reservoirs
feed the Dan, Snir, and Hermon baseflow, and one
reservoir contributes to the residual of groundwater to
springs in the eastern part of Mount Hermon within
Syria. The accumulated output from the surface runoff
(module 3) and the baseflow (module 2) for each tributary makes up all of the natural flows in the hydrological model for the region. The sum of all three
tributaries creates the flow in the main stream, the
Jordan River.
290
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 51
FIG. 3. Rain gauge locations and major aquifers in Israel.
d. Jordan River streamflow prediction
We chose to demonstrate the use of the WRF rainfall to
predict daily streamflow in the upper catchments of the
Jordan River, using HYMKE and following the procedure described by Rimmer and Salingar (2006). Here,
the initial conditions of streamflow in each tributary
(Hermon, Snir, and Dan) were determined from the
measured discharge on 1 October 2009 (the first day of the
2009/10 rainy season). At this date the soil moisture content in the surface layer (module 0) is set to the minimal
(residual) value, the storage and discharge of the surface
flow (module 3) and the vadose zone (module 1) are
considered to be zero, and therefore the entire streamflow
is attributed to the groundwater sources (module 2). The
storage of each groundwater reservoir can then be evaluated from the typical discharge–storage relations of each
reservoir. These conditions change throughout the season
with the application of rainfall.
Three sets of HYMKE runs were done: 1) with
measured rainfall, 2) using the WRF-modeled rainfall
output, and 3) using the Rimmer and Salingar (2006)
FEBRUARY 2012
GIVATI ET AL.
FIG. 4. Schematic description of the Mount Hermon hydrological
model. Module 0 is the surface layer, module 1 is the vadose zone,
module 2 consists of four groundwater reservoirs, and module 3
simulates the surface flow. The calculated baseflow and the surfaceflow components of each tributary result in their full natural flow,
and the combined flow creates the Jordan River.
regression-model rainfall output. The measured precipitation input for the latter model was from a rain
gauge located at the Hermon base ski resort (1640 m
MSL) and operated by the Israeli Hydrological Service
since 2009. This location has the highest orographic
effect and contributes most of the water in the basin.
WRF daily precipitation forecasts from the 12-, 4-,
and 1.3-km grid models were extracted for this station
and used as input for HYMKE. In each simulation,
HYMKE produced time series of baseflow, surface flow,
and full natural flow for the Dan, Snir, Hermon, and
Jordan Rivers.
5. Results
a. WRF tests with different microphysical
parameterizations
The five previously noted microphysical schemes
available in WRF were tested for the three major storms
that occurred during 2008/09 in northern and central
Israel. Figure 5 displays the accumulated precipitation
simulated by WRF with the five microphysical schemes
and a comparison with the actual precipitation. Each
point in Fig. 5 represents the accumulated precipitation
on the 4-km grid. Figure 5 indicates that all of the
schemes showed fairly good results but that mp6 produced precipitation with the highest correlation and
smallest bias relative to the observations. The mp6
scheme is a WRF single-moment, six-class-hydrometeor
291
FIG. 5. Accumulated observed precipitation at 17 rain gauges
during three storms (23–25 Dec 2008, 10–12 Feb 2009, and 27 Feb–
1 Mar 2009) vs the forecast precipitation from WRF run at 4-km
grid spacing with different microphysical schemes: mp2, mp6, mp7,
mp8, and mp10.
microphysical scheme. Note that mp6 is an enhanced
version of mp2 (J. Dudhia 2010, personal communication): the advances of mp6 over mp2 appear to be favorable for wintertime precipitation modeling over
mountainous regions where cold-cloud microphysical
processes are dominant.
b. WRF resolution impacts for precipitation
Another important step for optimizing WRF for the
region is to examine the impact of the model resolution.
The preferred microphysical scheme (mp6) was used in
the WRF runs for these resolution experiments. The
correlation between the observed accumulated precipitation and the calculated precipitation with the WRF
runs at different resolutions was examined. Also, the
GFS-forecast rains at three water-resources basins in
Israel were examined: the Kinneret basin, the mountain
aquifer, and the coastal aquifer. This was done using the
17 selected representative rain gauges for eight winter storms that occurred during the 2008/09 season
(23 December 2008–17 April 2009).
Figure 6 displays the correlation between the observed and the modeled precipitation from the WRF
runs with grid increments of 3 (Fig. 6a), 4 (Fig. 6b), 12
(Fig. 6c), and 36 (Fig. 6d) km and from the GFS model
(Fig. 6e). It can be seen clearly that the WRF-predicted
precipitation at the higher resolutions, with 3- and 4-km
grid sizes, correlates very well with the measured rain,
with coefficients r 5 0.92 and 0.96, respectively. The
correlation coefficients for the 12- and 36-km forecasts
292
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
FIG. 6. WRF-forecast total precipitation from 17 stations during eight storms between 23 Dec
2008 and 17 Apr 2009 running at grid spacings of (a) 3, (b) 4, (c) 12, and (d) 36 km, and (e) the
GFS forecast and a comparison with the observed total precipitation for the same period.
VOLUME 51
FEBRUARY 2012
293
GIVATI ET AL.
TABLE 1. Comparison of observed and calculated WRF precipitation for eight storms during 2008/09 at different model resolutions.
Storm
Obs precipitation (mm)
23 Dec 2008
16 Feb 2009
3 Mar 2009
10–12 Feb 2009
16–17 Apr 2009
19–21 Feb 2009
23–25 Mar 2009
27 Feb–1 Mar 2009
Tot precipitation
65
26
27
76
36
94
41
70
435
23 Dec 2008
16 Feb 2009
3 Mar 2009
10–12 Feb 2009
16–17 Apr 2009
19–21 Feb 2009
23–25 Mar 2009
27 Feb–1 Mar 2009
Tot precipitation
3-km WRF
12-km WRF
36-km WRF
50-km GFS
Kinnert basin
68
60
21
24
31
29
62
76
64
28
92
93
48
45
72
70
459
425
56
12
11
36
42
57
18
48
280
33
6
11
36
30
45
18
39
218
50
13
15
40
18
57
27
49
269
79
11
7
33
4
60
32
121
347
Yarqon–Taninim
94
100
5
10
6
6
15
22
5
5
61
38
21
38
86
108
293
327
30
6
7
25
18
52
4
77
220
18
6
5
25
6
14
6
48
129
35
7
3
28
9
44
19
89
235
23 Dec 2008
16 Feb 2009
3 Mar 2009
10–12 Feb 2009
16–17 Apr 2009
19–21 Feb 2009
23–25 Mar 2009
27 Feb–1 Mar 2009
Tot precipitation
42
18
10
21
20
58
23
82
272
Coastal aquifer
38
41
19
12
5
11
4
8
15
2
56
32
12
42
52
62
201
211
35
15
7
24
23
55
6
66
231
26
8
5
25
10
29
8
51
161
29
14
7
20
13
47
16
63
209
23 Dec 2008
16 Feb 2009
3 Mar 2009
10–12 Feb 2009
16–17 Apr 2009
19–21 Feb 2009
23–25 Mar 2009
27 Feb–1 Mar 2009
Tot precipitation
76
26
25
37
19
76
33
116
408
Western Galilee
48
41
16
13
13
10
28
41
19
16
55
49
26
42
110
102
315
314
50
9
17
39
37
50
17
78
296
34
9
16
39
29
55
19
69
270
46
13
13
40
22
54
30
77
294
were poorer (r 5 0.79 and 0.69). The GFS model, although it has coarser resolution (at around 50-km grid
size), showed better results than the WRF forecasts at
36 km. More complete results for WRF model-resolution
impact are given in Table 1. These findings suggest the
importance of using a high-resolution model for operational precipitation forecasts in regions with complex
terrain.
Figure 7 displays the correlation of modeled and observed daily precipitation for four rain gauges in the
upper part of the Kinneret basin for the full 2009/10
rainy season, in the Hermon range and its foothills. The
rain gauges are Hermon base, Majdal Shams, El Rom, and
Phicman, with locations shown in Fig. 8. By comparing
4-km WRF
WRF forecasts at 1.3- and 4-km grid spacing, the high
agreement between the calculated and the observed
precipitation at both resolutions can be seen again, and
the 1.3-km solution shows better results. Figure 9 summarizes the WRF runs for the full 2009/10 rainy season
at three different grid increments (12, 4, and 1.3 km) for
the cluster of those four rain gauges versus the observed
accumulated precipitation. Note that the difference between the observed amount and the forecast precipitation is small, especially for WRF running at 1.3-km grid
spacing. The differences between the observed precipitation amounts and the calculated values in Fig. 9 are
only ;3% for the 1.3-km WRF runs and 5% at for the 4-km
WRF runs.
294
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 51
FIG. 7. The correlation between daily precipitation during the 2009/10 rainy season for the four rain gauges in the
upper Jordan River basin (Hermon, Majdal Shams, El Rom, and Phicman) and the WRF precipitation forecasts at
(a) 1.3- and (b) 4-km grid spacing.
c. WRF versus the linear-regression-calculated
precipitation
Rimmer and Salingar (2006) calculated the precipitation for the Hermon range using regression between
elevation and precipitation at the lower elevations in the
north of Israel, as was described in the first paragraph of
section 2. Figure 10 displays the annual precipitation
amount that was measured at the Hermon base’s Israeli
Hydrological Service rain gauge versus the WRF 1.3and 4-km precipitation forecasts and versus that calculated for this rain gauge station using the original
regression-model equations. It can be seen that the
WRF simulations did much better than the regression
calculation. The actual precipitation in this gauge was
1306 mm, the WRF 1.3-km simulation calculated 1336 mm,
the WRF 4-km simulation calculated 1278 mm, and the
regression equation produced just 1022 mm (12%, 22%,
and 222% relative to the observed, respectively). Such an
underestimation by the regression model would lead to
underestimation of the streamflow volumes in the river
and its tributaries.
measured and predicted flows for the Hermon stream
(Fig. 11a) resulted in r 2 5 0.74, as compared with r 2 5
0.89 in a 20-yr model calibration with measured rainfall
(Rimmer and Salingar 2006). Similar results were found
d. Simulations of daily flow in the Jordan River and
its tributaries
The actual measured precipitation at Mount Hermon,
the amount of precipitation computed by the WRF 1.3-km
model, and that computed from the regression model
were used to drive the calculations of daily flow volumes
in the three major tributaries and at the Jordan River
using HYMKE. Figure 11 displays simultaneously the
measured and the calculated daily streamflows derived
from the WRF 1.3-km rainfall model. Figure 11 indicates that the correlation between the time series of
FIG. 8. The locations of the four rain gauges that were used in the
HYMKE hydrological model: Hermon base, Majdal Shams, El
Rom, and Phicman.
FEBRUARY 2012
GIVATI ET AL.
295
FIG. 9. Observed accumulated precipitation during 2009/2010 for the four rain gauges in the
upper part of the Kinneret basin (Hermon base, Majdal Shams, El Rom, and Phichman), and
the WRF-forecast precipitation with 1.3-, 4-, and 12-km grid spacing. Note that the difference
between the observed amount and the forecast precipitation is small, especially for WRF at the
1.3-km grid spacing.
for the Snir (Fig. 11b; r 2 5 0.70, as compared with r 2 5
0.84), the Dan (Fig. 11c; r 2 5 0.79, as compared with
r 2 5 0.77), and the Jordan River itself (Fig. 11d; r 2 5 0.61,
as compared with r 2 5 0.94) at Sede Nehehemia (hydrometric station of the Israeli Hydrological Service).
The hydrological model during 2009/10 predicted better
the baseflows in this karstic region, whereas the model
system needs to be improved for forecasting the very
extreme peaks of the streamflows.
Figure 12 compares the daily Jordan River flows
simulated by the HYMKE model with 1) measured
rainfall and 2) the WRF 1.3-km forecast rainfall. The
comparison between flows predicted by the two cases
results in r 2 5 0.93. The precipitation amount during the
rainy season of 2009/10 was about the average climatemean value. Table 2 summarizes the monthly flow volumes forecast by the hydrological model driven by the
rainfall observations, the rainfall forecast by the WRF
FIG. 10. Observed annual precipitation for 2009/10 at the Hermon base rain gauge vs calculated precipitation from WRF at 1.3-km grid increment, from WRF at 4-km grid increment,
and from the regression model used by Rimmer and Salingar (2006).
296
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 51
FIG. 11. Comparison of daily streamflow forecasts at the three major tributaries of the Jordan River, the (a)
Hermon, (b) Snir, and (c) Dan Rivers, and (d) the streamflow in the Jordan River, using the observed precipitation
from the Hermon, and the WRF-forecast precipitation from the 1.3-km model.
1.3-km run, and that computed on the basis of the
Rimmer and Salingar (2006) regression approach during the 2009/10 rainy season (from 1 October 2009 to 30
April 2010). The total measured flow volume at the
gauge for this period was 354 Mm3: 323 Mm3 at the
gauge and another 31 Mm3 consumption upstream that
was not able to get into the river. The HYMKE-simulated
flow using the actual Hermon precipitation was 362 Mm3,
that using the WRF precipitation was 371 Mm3, and that
using the regression precipitation was just 294 Mm3. It
can be seen in Fig. 12 and Table 2 that using simulated
precipitation from WRF is actually almost like using such
input from a rain gauge network. The findings presented
here have important value for hydrologists, especially in
areas where a good network of rain gauges is not available and there is a need for a precipitation forecast such
as WRF can provide.
for all of the consumers. This fact is why reliable
streamflow forecasting is required throughout the year.
Steep mountains in northern Israel cause complex
structures in the spatial distribution of precipitation,
especially in the Mount Herman region. Such precipitation features make hydrological forecasting challenging, especially for the Jordan River. In this paper,
the use of an advanced mesoscale weather model, the
community WRF, has been explored for driving hydrological model forecasts for the Jordan River. Sensitivity experiments were conducted to study the impact
of WRF microphysical parameterization schemes and
model grid resolution, both of which are crucial for
6. Discussion and conclusions
Improving streamflow forecasts for the Jordan River
and its tributaries is a highly important task, especially
during dry years. This area is disconnected from the
national water system (the ‘‘National Water Carrier’’),
and it has to rely on local water sources: direct pumping,
mostly from the Dan and the Banias Rivers. During
severe drought years like 2000/01 or 2007/08, water
shortages occurred and there was scarcely enough water
FIG. 12. Comparison of the modeled daily Jordan River flow
simulated by HKMKE using measured rainfall and rain forecast
with the WRF 1.3-km grid spacing. The differences of the streamflows
computed with the two approaches are also plotted.
FEBRUARY 2012
297
GIVATI ET AL.
TABLE 2. Measured streamflow at the Jordan River for the rainy season of 2009/10 vs the calculated streamflow using HYMKE with
1) actual precipitation in the Hermon region, 2) calculated precipitation from the 1.3-km WRF run, and 3) calculated precipitation from
the regression model used by Rimmer and Salingar (2006).
Month
Oct 2009
Nov 2009
Dec 2009
Jan 2010
Feb 2010
Mar 2010
Apr 2010
1 Oct 2009–
30 Apr 2010
Measured
streamflow
(Mm3)
Calculated streamflow with
HYMKE using actual
precipitation (Mm3)
Calculated streamflow
with HYMKE using WRF
1.3-km precipitation (Mm3)
Calculated streamflow with HYMKE
using precipitation calculated
with the regression model (Mm3)
24.3
23.3
45.3
69.2
64.4
60.6
36.6
354
25.3
28.1
55.4
64.4
70.7
66.9
50.9
362
25.3
23.1
50.9
72.7
69.6
76.8
52.3
371
25.3
23.0
41.2
54.3
54.6
54.5
40.6
294
wintertime precipitation modeling in regions with steep
mountains. The modeling results suggest that WRF
is capable of forecasting precipitation amounts and
structures in northern Israel reasonably well when highresolution (4–1.3 km) grids are used.
High-resolution WRF-forecast precipitation is used to
drive HYMKE to simulate the Jordan River flow. The
verification of the streamflow forecasts for the upper
Jordan River, using the WRF rainfall, showed improvement relative to the current hydrological simulations
that make use of rainfall from a regression method.
WRF is of special value in areas where there are poor
surface meteorological observations, such as in the Jordan
River basin. The results in Table 2 clearly show that the
Jordan River flows forecast using the WRF 1.3-km model
precipitation are more accurate than those that are based
on the regression approach.
Hydrological modeling using the WRF-simulated precipitation allows us to represent better the important
hydrological processes that occur in the northern part
of the Jordan basin located in Lebanon, such as daily
precipitation–runoff ratios, and the recharge for springs
like the Dan and the Banias. Such improved upstream
information leads to better predictions for downstream
regions.
It can be argued that the high-resolution WRF combined with HYMKE can be used to support hydrometeorological and hydrogeological applications with
dynamical downscaling of global seasonal precipitation
forecasts like those from the NCEP GFS model. This
would facilitate seasonal streamflow forecasting in the
Jordan River. This approach can improve the current
method, which uses statistical downscaling for seasonal
precipitation forecasting for the Kinneret basin (Wu
et al. 2011, manuscript submitted to Int. J. Climatol.).
Using WRF for dynamical downscaling also improves
the precipitation input for regional climate models for
the upper Jordan River and especially for the Hermon
range [as was done for California by Pan et al. (2010)]. In
contrast, previous studies used statistical downscaling to
evaluate future (2015–35) streamflow volumes and regimes in the Jordan River (Samuels et al. 2010; Rimmer
et al. 2011). Thus, the need for interseasonal streamflow
simulations is clear.
Further studies to improve the modeling system should
be undertaken. A next step will be to extend the WRF
domains to cover the full Jordan River region, especially
the highest elevations in the Syrian mountains (2800 m),
to feed HYMKE with more-complete meteorological
input from those areas for basin-flow forecasting.
Acknowledgments. The authors thank Dr. T. T.
Warner for reviewing the manuscript and for valuable
discussions. The authors also thank Guy Kelman from
Weather-It-Is for helping to maintain the WRF operational system at the Israeli Hydrological Service.
REFERENCES
Adlerman, E. J., and K. K. Droegemeier, 2002: The sensitivity of
numerically simulated cyclic mesocyclogenesis to variations in
model physical and computational parameters. Mon. Wea.
Rev., 130, 2671–2691.
Bryan, G. H., J. C. Wyngaard, and J. M. Fritsch, 2003: Resolution
requirements for the simulation of deep moist convection.
Mon. Wea. Rev., 131, 2394–2416.
Colle, B. A., K. J. Westrick, and C. F. Mass, 1999: Evaluation of
MM5 and Eta-10 precipitation forecasts over the Pacific Northwest during the cool season. Wea. Forecasting, 14, 137–154.
Daly, C., R. P. Neilson, and D. L. Phillips, 1994: A statistical–
topographic model for mapping climatological precipitation
over mountainous terrain. J. Appl. Meteor., 33, 140–158.
Davis, C., and F. Carr, 2000: Summary of the 1998 workshop on
mesoscale model verification. Bull. Amer. Meteor. Soc., 81,
809–819.
Dooge, J. C. I., 1973: Linear theory of hydrologic systems. U.S.
Dept. of Agriculture Tech. Bull. 1468, 267–293.
298
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
Doyle, J. D., 1997: The influence of mesoscale orography on
a coastal jet and rainband. Mon. Wea. Rev., 125, 1465–1488.
Fleury, P., V. Plagnes, and M. Bakalowicz, 2007: Modelling of the
functioning of karst aquifers with a reservoir model: Application to Fontaine de Vaucluse (South of France). J. Hydrol.,
345, 38–49.
Gilad, D., and J. Bonne, 1990: Snowmelt of Mt. Hermon and its
contribution to the sources of the Jordan River. J. Hydrol.,
114, 1–15.
Grell, G. A., L. Schade, R. Knoche, A. Pfeiffer, and J. Egger, 2000:
Nonhydrostatic climate simulations of precipitation over
complex terrain. J. Geophys. Res., 105, 29 595–29 608.
Gur, D., M. Bar-Matthews, and E. Sass, 2003: Hydrochemistry of
the main Jordan River sources: Dan, Banias, and Kezinim
springs, north Hula valley. Israel. Isr. J. Earth Sci., 52, 155–178.
Halfon, N., 2008: Spatial patterns of precipitation in Israel and
their synoptic characteristics. Ph.D. thesis, University of Haifa,
185 pp.
Hartmann, A., M. Kralik, F. Humer, J. Lange, and M. Weiler,
2011: Identification of a karst system’s intrinsic hydrodynamic
parameters: Upscaling from single springs to the whole aquifer. Environ. Earth Sci., doi:10.1007/s12665-011-1033-9, in press.
Hay, L. E., M. P. Clark, R. L. Wilby, W. J. Gutowski, G. H.
Leavesley, Z. Pan, R. W. Arritt, and E. S. Takle, 2002: Use
of regional climate model output for hydrologic simulations.
J. Hydrometeor., 3, 571–590.
——, ——, M. Pagowski, G. H. Leavesley, and W. J. Gutowski,
2006: One-way coupling of an atmospheric and a hydrologic
model in Colorado. J. Hydrometeor., 7, 569–589.
Hong, S.-Y., and J.-O. J. Lim, 2006: The WRF single-moment
6-class microphysics scheme (WSM6). J. Kor. Meteor. Soc., 42,
129–151.
——, K.-S. S. Lim, Y.-H. Lee, J.-C. Ha, H.-W. Kim, S.-J. Ham,
and J. Dudhia, 2010: Evaluation of the WRF double-moment
6-class microphysics scheme for precipitating convection. Adv.
Meteor., 2010, 707253, doi:10.1155/2010/707253.
Jukic, D., and V. Denic-Jukic, 2009: Groundwater balance estimation in karst by using a conceptual rainfall-runoff model.
J. Hydrol., 373, 302–315.
Kain, J. S., S. J. Weiss, J. J. Levit, M. E. Baldwin, and D. R. Bright,
2006: Examination of convection-allowing configurations
of the WRF model for the prediction of severe convective
weather: The SPC/NSSL Spring Program 2004. Wea. Forecasting, 21, 167–181.
——, and Coauthors, 2008: Some practical considerations regarding horizontal resolution in the first generation of operational convection-allowing NWP. Wea. Forecasting, 23,
931–952.
Katzfey, J. J., 1995: Simulation of extreme New Zealand precipitation events. Part I: Sensitivity to orography and resolution. Mon. Wea. Rev., 123, 737–754.
Kessler, A., and U. Kafri, 2007: Application of a cell model for
operational management of the Na’aman groundwater basin,
Israel. Isr. J. Earth Sci., 56, 29–46.
Le Moine, N., V. Andréassian, and T. Mathevet, 2008: Confronting
surface- and groundwater balances on the La RochefoucauldTouvre karstic system (Charente, France). Water Resour. Res.,
44, W03403, doi:10.1029/2007WR005984.
Leung, L. R., and Y. Qian, 2003: The sensitivity of precipitation
and snowpack simulations to model resolution via nesting in
regions of complex terrain. J. Hydrometeor., 4, 1025–1043.
Lin, C. A., L. Wen, M. Beland, and D. Chaumont, 2002: A coupled
atmospheric–hydrological modeling study of the 1996 Ha! Ha!
VOLUME 51
River basin flash flood in Québec, Canada. Geophys. Res.
Lett., 29, 1026, doi:10.1029/2001GL013827.
Liu, Y., and Coauthors, 2008a: The operational mesogamma-scale
analysis and forecast system of the U.S. Army Test and
Evaluation Command. Part I: Overview of the modeling system, the forecast products, and how the products are used.
J. Appl. Meteor. Climatol., 47, 1077–1092.
——, and Coauthors, 2008b: The operational mesogamma-scale
analysis and forecast system of the U.S. Army Test and
Evaluation Command. Part II: Inter-range comparison of the
accuracy of model analyses and forecasts. J. Appl. Meteor.
Climatol., 47, 1093–1104.
Martin, G., 1996: A dramatic example of the importance of detailed
model terrain in producing accurate quantitative precipitation
forecasts for southern California. National Weather Service
Western Region Tech. Attachment 96-07, 9 pp. [Available
from Western Regional Climate Center, Desert Research
Institute, 2215 Raggio Parkway, Reno, NV 89512.]
Mass, C. F., D. Ovens, K. Westrick, and B. A. Colle, 2002: Does
increasing horizontal resolution produce more skillful forecasts? Bull. Amer. Meteor. Soc., 83, 407–430.
McQueen, J. T., R. R. Draxler, and G. D. Rolph, 1995: Influence of
grid size and terrain resolution on wind field predictions from
an operational mesoscale model. J. Appl. Meteor., 34, 2166–
2181.
Mekorot Watershed Unit, 2008: The water, solutes and heat balances of Lake Kinneret (in Hebrew). Mekorot Water Supply
Co. Annual Rep., Sapir Site, Israel, 41 pp.
Neiman, J. N., P. J. Ralph, A. B. White, D. E. Kingsmill, and P. O.
G. Persson, 2002: The statistical relationship between upslope
flow and rainfall in California’s coastal mountains: Observations during CALJET. Mon. Wea. Rev., 130, 1468–1492.
Pan, L. L., S.-H. Chen, D. Cayan, M.-Y. Lin, Q. Hart, M.-H. Zhang,
Y. Liu, and J. Wang, 2010: Influences of climate change on
California and Nevada regions revealed by a high-resolution
dynamical downscaling study. Climate Dyn., 37, 2005–2020.
Pandey, G. R., D. R. Cayan, and K. P. Georgaakakos, 1999: Precipitation structure in the Sierra Nevada of California during
winter. J. Geophys. Res., 104, 12 019–12 030.
Petch, J. C., A. R. Brown, and M. E. B. Gray, 2002: The impact of
horizontal resolution on the simulations of convective development over land. Quart. J. Roy. Meteor. Soc., 128, 2031–2044.
Rimmer, A., and Y. Salingar, 2006: Modeling precipitationstreamflow processes in karst basin: The case of the Jordan
River sources, Israel. J. Hydrol., 331, 524–542.
——, A. Givati, R. Smauels, and P. Alpert, 2011: Using ensemble of
climate models to evaluate future water and solutes budgets in
Lake Kinneret, Israel. J. Hydrol., 410, 248–259.
Rosenfeld, D., and H. Farbstein, 1992: Possible influence of desert
dust on seedability of clouds in Israel. J. Appl. Meteor., 31,
722–731.
Saaroni, H., N. Halfon, B. Ziv, P. Alpert, and H. Kutiel, 2009: Links
between the rainfall regime in Israel and location and intensity
of Cyprus lows. Int. J. Climatol., 30, 1014–1025.
Samuels, R., A. Rimmer, A. Hartman, S. Krichak, and P. Alpert,
2010: Climate change impact on Jordan River flow: Downscaling application from a regional climate model. J. Hydrometeor., 11, 860–879.
Schwartz, C. S., and Coauthors, 2009: Next-day convectionallowing WRF model guidance: A second look at 2-km
versus 4-km grid spacing. Mon. Wea. Rev., 137, 3351–3372.
Seuffert, G., P. Gross, C. Simmer, and E. F. Wood, 2002: The influence of hydrologic modeling on the predicted local weather:
FEBRUARY 2012
GIVATI ET AL.
Two-way coupling of a mesoscale land surface model and
a land surface hydrologic model. J. Hydrometeor., 3, 505–523.
Simpson, B., and I. Carmi, 1983: The hydrology of the Jordan River
and its tributaries: Hydrographic and isotopic investigation.
J. Hydrol., 62, 225–242.
Sugawara, M., 1995: Tank model. Computer Models of Watershed
Hydrology, V. P. Singh, Ed., Water Resources Publications,
165–214.
Weisman, M., C. Davis, W. Wang, K. Manning, and J. Klemp, 2008:
Experiences with 0–36-h explicit convective forecasts with the
WRF-ARW model. Wea. Forecasting, 23, 407–437.
299
Westrick, K. J., and C. F. Mass, 2001: An evaluation of a highresolution hydrometeorological modeling system for prediction
of a cool-season flood event, in a coastal mountainous watershed.
J. Hydrometeor., 2, 161–180.
Xue, M., and W. J. Martin, 2006: A high-resolution modeling study
of the 24 May 2002 case during IHOP. Part I: Numerical
simulation and general evolution of the dryline and convection. Mon. Wea. Rev., 134, 149–171.
Younis, J., S. Anquetin, and J. Thielen, 2008: The benefit of highresolution operational weather forecasts for flash flood
warning. Hydrol. Earth Syst. Sci., 12, 1039–1051.