close

Вход

Забыли?

вход по аккаунту

?

asl.182

код для вставкиСкачать
ATMOSPHERIC SCIENCE LETTERS
Atmos. Sci. Let. 9: 61–66 (2008)
Published online 30 April 2008 in Wiley InterScience
(www.interscience.wiley.com) DOI: 10.1002/asl.182
Seasonal hydrologic predictions of low-flow conditions
over eastern USA during the 2007 drought
Haibin Li,* Lifeng Luo and Eric F. Wood
Environmental Engineering and Water Resources, Department of Civil and Environmental Engineering, Princeton University, USA
*Correspondence to:
Haibin Li, Environmental
Engineering and Water
Resources, Department of Civil
and Environmental Engineering,
Princeton University, Princeton,
NJ, 08544, USA.
E-mail: [email protected]
Received: 7 January 2008
Revised: 14 March 2008
Accepted: 18 March 2008
Abstract
A seasonal streamflow monitoring and forecasting component is implemented into the
Drought Monitoring and Prediction System (Luo and Wood, 2007a), which supplements
the existing soil-moisture-based analysis framework by providing real-time streamflow
monitoring and forecasting up to 6 months lead time. Evaluations were conducted over
four basins in eastern USA to understand the forecast skill of the system for the extensive
hydrologic droughts in 2007. Consistent with the agricultural drought forecasts reported by
Luo and Wood (2007a), the streamflow subsystem can forecast low-flow conditions for up
to three months in advance, with Brier scores ranging from 0.10 to 0.49. Copyright  2008
Royal Meteorological Society
Keywords:
seasonal forecasting; drought; low flow
1. Introduction
Drought is the most costly natural disaster in the USA
(Federal Emergency Management Agency (FEMA),
1995), with economic damages from the 1988 Midwest drought estimated to have been in the billions
of dollars (FEMA, 1995) and for the 1998–2002
southeastern (SE) US drought exceeding $10 billion
(National Climatic Data Center, 2006). Thus, there is a
need for a well-developed prewarning drought system
on which decision makers can respond accordingly
before drought onsets. Such a drought ‘early warning
system’ is central to the vision of the National Integrated Drought Information System (NIDIS) (Western
Governors Association, 2004), which recognizes that
such a forecast system can provide timely and reliable
information about land-surface hydrologic conditions,
particularly soil moisture and streamflow.
Seasonal soil-moisture information is essential for
agricultural activity and wildfire control; streamflow
conditions are crucial to river navigation, hydropower
production, thermal power plant cooling, ecosystem
function and water quality, as well as for recreational
use. Many drought systems provide drought outlook products based solely on meteorological drought
(e.g. Palmer Drought Severity Index), or agricultural drought (usually constructed from soil moisture).
Streamflow forecasting is mainly carried out by the
National Weather Service’s regional River Forecast
Centers, with a focus more on flood forecasting than
low-flow events.
The lack of a systematic low-flow seasonal forecast system motivated the authors to expand their
successful Drought Monitoring and Prediction System
(DMAPS) (Luo and Wood, 2007a) from soil moisture
and agricultural drought prediction to include streamflow monitoring and forecasting. This allows for a
Copyright  2008 Royal Meteorological Society
comprehensive forecast system for the land surface
hydrologic conditions over monthly to seasonal time
scales.
During 2007, hydrologic drought conditions developed in the region of Tennessee, Alabama, and Georgia
in late February through early March, and expanded
west to Mississippi, Louisiana, and to the northeast
(NE) (US Geological Survey (USGS), 2007). Above
normal temperature from late summer through early
fall exacerbated the drought conditions. Extensive
media coverage has been devoted to the immediate
drought impact on social and economic activities. Luo
and Wood (2007a) demonstrated that the DMAPS
exhibited promising skill in predicting the 2007 SE
US agricultural drought. In this article, we present for
the first time the augmented system that provides lowflow forecasts over four eastern regions: the SE, the
NE, Ohio, and Lower Mississippi River basins.
The article is organized as follows: Section 2 gives
a brief introduction of the DMAPS monitoring and
prediction system; Section 3 presents the verification
study of the 2007 low-flow seasonal forecasts, followed by a summary in Section 4.
2. DMAPS – the Princeton hydrologic
nowcast and forecast system
Luo and Wood (2007a,b) provide a detailed description
about the DMAPS; readers are directed to these
references for further details. The core component of
the system is the Variable Infiltration Capacity (VIC)
land surface model (Liang et al., 1996; Nijssen et al.,
1997) that was calibrated for diverse climate regimes
and has been widely used in land-surface modeling and
climatic change related research (e.g. Nijssen et al.,
1997; Cherkauer and Lettenmaier, 1999; Nijssen et al.,
62
2001; Maurer et al., 2002; Sheffield et al., 2004). The
US nowcast subsystem uses real-time precipitation and
temperature forcing fields from the North American
Land Data Assimilation System (NLDAS) (Mitchell
et al., 2004). The NLDAS hourly forcing is aggregated
temporally to a daily resolution to drive the VIC
model at a 1/8th degree spatial resolution. The nowcast
component not only provides near-real-time estimates
for the land-surface hydrologic conditions but also
creates realistic initial conditions that can be used by
the forecasting subsystem.
For the forecast portion of the system, the precipitation and temperature forecasts from the National
Centers for Environmental Prediction (NCEP)’s Climate Forecast System (CFS) (Saha et al., 2006) are
used after preprocessing. Specifically, the forecasts are
combined with observed climatology via a Bayesian
merging approach developed by Luo et al. (2007)
to (1) remove forcing biases, (2) statistically downscale the coarse spatial resolution forcing of the climate models to finer resolution for hydrologic applications, and (3) produce forcing ensembles with sufficient spread to better represent the forcing uncertainties. The merging framework is very effective in
achieving these purposes (see Luo et al., 2007 for
details).
With respect to streamflow, the gridded runoff outputs are routed to river gauging stations with a linear routing model (Lohmann et al., 1996, 1998). To
further reduce possible biases due to inadequate calibration for the hydrologic and routing models, postprocessing is conducted on the modeled streamflow
through an equal-quantile mapping procedure that corrects model simulation bias by utilizing the empirical
probability distributions for observed and simulated
flows (Wood et al., 2002; Hashino et al., 2007). This
equal-quantile mapping method is only applied to the
monthly streamflow forecasts, and not to the daily
nowcasts.
The Princeton DMAPS (http://hydrology.princeton.
edu/forecast) was initiated in August 2005 with a
focus on soil moisture. The generation of streamflow
nowcasts started in August 2007, and has since been
updated weekly in conjunction with the soil-moisture
fields. As streamflow exhibits large day-to-day variation, the 7-day cumulative streamflow statistics are calculated for over 400 stations across the eastern USA.
The 7-day cumulative streamflow climatology is based
on offline simulations with the VIC model for the
period 1949–2004, driven by gridded observational
records (Maurer et al., 2002; Cosgrove et al., 2003),
and extended to the present as part of the NLDAS
project. As the nowcast is compared to simulated climatology from the model itself, rather than observations, the need for postprocessing is eliminated. The
first two years were not used in the analysis to ensure
enough spin-up.
A snapshot of the streamflow nowcast conditions
for the week 26 October 2007 to 1 November 2007 is
provided in Figure 1. The conditions are expressed in
Copyright  2008 Royal Meteorological Society
H. Li, L. Luo and E. F. Wood
terms of percentile classes, based on the cumulative
probability calculated using the Gringorten plotting
position method (Gringorten, 1963). The flows are
classified into seven classes: low (less than historical observed minimum), <10%, 10–25%, 25–75%,
75–90%, >90%, high (larger than the historical maximum). For the week shown in Figure 1, rivers across
North and South Carolina recorded historically high
flows while some rivers around the area of Tennessee, Georgia, and Alabama reported much below
normal flow conditions (less than 10% of their climatological distribution). The spatial characteristics
resemble that of the USGS real-time monitoring
[http://water.usgs.gov/waterwatch/?m=real&r=us&
w=real%2Canimation] which is based on gauge measurements, though the USGS site offers far more station points than we currently monitor.
To understand how well our hydrologic model
simulates the streamflow for the basins selected in
this article, the Nash–Sutcliffe efficiency coefficient
(NS) was computed (Nash and Sutcliffe, 1970). On a
monthly scale, the calculated NS is generally above
0.8 for most gauges in the Ohio River basin and the
NE, while the coefficient is relatively low (less than
0.25) for stations located in the SE. On a daily scale,
the NS is smaller, but still comparable for gauges in
the Ohio River basin. For the SE, the model tends
to overestimate river discharge in general. As both
the distributed (larger basins) and lumped (smaller
basins) routing models produce similar results, the
Figure 1. Probability streamflow nowcast for the week 26
October 2007 to 1 November 2007, expressed as percentile
class. The classes are based on a 7-day cumulative streamflow
(the nowcast period) compared to the same period of off-line
simulations of 1951–2004.
Atmos. Sci. Let. 9: 61–66 (2008)
DOI: 10.1002/asl
Seasonal hydrologic predictions over eastern USA
63
original VIC model parameter calibration may be
inadequate for this particular region. This was also
pointed out by Mitchell et al. (2004). Troy et al.
(2008) are recalibrating the model over the USA
using a grid-based calibration scheme. The results will
be reported elsewhere when the work is completed.
Nonetheless, the nowcast system is still capable of
capturing the general spatial patterns of the observed
categorical probability, mainly because of the fact that
overestimations exist in both background climatology
simulations and the nowcast simulations so that it has
little effect on the calculated statistics.
As soil moisture is a time-integrated variable in the
hydrologic system, the success of the prediction for
the latest drought event across the USA makes us
believe that the DMAPS is also capable of skillfully
forecasting streamflow conditions. To evaluate this, we
further examined streamflow forecasts for over 300
gauging stations over the study area. Specifically, we
want to test whether the system is able to predict lowflow events from one to up to six months in advance.
Here, we selected the 25th quantile of the observed
historical monthly distribution as the threshold for
low-flow events. This level is sufficiently low so that
water managers must consider adaptive procedures
for managing water demands under these conditions.
The forecast is expressed in terms of exceedence
probability and thus, can be compared to the binary
event for the observations. Explicitly, for a given
month, if the observed monthly-mean streamflow is
greater or equal to the threshold, the event is defined
as 1; otherwise, 0 is assigned. For the forecast, the
smaller the exceedence probability, the higher the
chance of a low-flow event happening. Figure 3 gives
the probabilistic streamflow forecasts similar to that of
Figure 2. The first row represents the binary event for
streamflow observations (either above or below the
25th quantile threshold.) Similar to the precipitation
map, persistent low-flow conditions are observed for
much of the SE. The forecast system predicted well the
observed low-flow conditions, especially for May and
July 2007 when hydrologic droughts extended further
northward and eastward toward the Ohio River basin
and the NE. At the 3-month lead time, skill still exists,
although the probability gets closer to 50%. For some
months, such as May, even at a 5-month lead time,
skill is still detectable. This may also be attributable,
to some extent, to the initial streamflow conditions in
addition to the quality of preprocessed precipitation
forecasts used as VIC forcing. An unresolved issue is
the relative impact from initial condition uncertainty
versus precipitation forecast uncertainty in streamflow
forecast skill. We are conducting further research work
on this subject and will present the results when it is
completed.
From a conservative perspective, we may want to
accept that drought is likely to occur if the exceedence
probability for predicted low-flows is less than 50%.
Under such circumstances, the ‘hit rate’ for 1-month
lead time forecast is about 0.5 for forecasts initialized
in May, July, and September. For 3-month lead time,
the hit rate drops to about 0.3. It is close to 0 for
longer lead times. We also calculated the BS for the
streamflow forecasts. Similar to precipitation, small
BS values (more skillful forecasts) can always be
found for short lead time forecasts and forecasts for
cooler seasons. The regional BS value for a given time
is calculated by a basin area-weighted average under
the assumption that the forecast skill is basin-size
dependent. Intuitively, one supposes that predictions
for large basins are better compared to small basins
because of the longer time lags in large basins and the
3. Analyses of the 2007 hydrologic drought
forecasts for Eastern USA
Low-flow conditions are usually accompanied by
below-average soil moisture values (Luo and Wood,
2007a). A series of 6-month streamflow forecasts
were made for every month from November 2006 to
September 2007 using NCEP CFS precipitation and
temperature forecasts as input forcings into VIC after
preprocessing them using the Bayesian merging technique of Luo et al. (2007). The forecasts for months
1, 3, and 5, starting in January 2007, are presented.
Figure 2 shows the bias corrected and downscaled precipitation forecasts, presented in probabilistic space,
with the observations being the first row. The forecasts (rows 2–6) are shown relative to the corresponding monthly climatological observations. For a given
threshold (e.g. historical monthly mean precipitation),
the probability for observation is either 1 (larger or
equal to the threshold) or 0 and thus, essentially a
binary event. With respect to the forecasts, the probability is the percentage of ensemble members in the
total ensemble that is larger or equal to the given
threshold. For all selected months in 2007, observations indicate widespread below-normal rainfall conditions, which are particularly evident for the SE. In late
spring and early summer, below-normal conditions
spread to nearly the entire study area. The forecast
system successfully predicted persistent below-normal
precipitation, which is very important for hydrologic
forecasting, as precipitation is the primary driver for
land surface hydrology. The Brier score (BS) (Brier,
1950) was computed to quantify the forecast skill of
the system for precipitation; the scores are shown in
Figure 2. As expected, the mean BS is, in general,
smaller for the cold season than that for the warm
season as the latter is characterized by higher climatic
variability that is harder to forecast. With respect to
lead time, forecasts for the 1-month lead time tend to
have smaller BS values than longer lead time forecasts.
The success of the system may be attributed to two
things: first, CFS itself possesses some skill for this
event, and secondly, the Bayesian bias correction and
downscaling technique creates more realistic postprocessed precipitation and temperature forcings, which is
the unique advantage of the forecast subsystem (Luo
et al., 2007).
Copyright  2008 Royal Meteorological Society
Atmos. Sci. Let. 9: 61–66 (2008)
DOI: 10.1002/asl
64
H. Li, L. Luo and E. F. Wood
Figure 2. Probabilistic forecasts for monthly mean precipitation (below the historical monthly mean). The first row is the
observed precipitation, expressed as a binary event, against historical means (green below the mean, white above the mean). Row
2–6 represents forecasts starting from January to September 2007, every other month respectively. Only forecasts for 1-, 3-, and
5-month lead times are plotted to match with observations. The inset table, representing rows 2–6, gives the average of Brier
score for all grid points.
sensitivity to spatial variability in the forcing fields
in small basins. To assess whether skill is basin-size
dependent, the BS values were averaged and plotted
according to the basin sizes and different lead times
(Figure 4). The basin categories are defined such that
the smallest 50 basins in terms of the basin area are
grouped together to form the ‘small basin’ group and
50 basins with intermediate size are treated as the
medium group and the largest 50 basins formed the
‘large basin’ category.
The streamflow forecast skill, quantified by the BS,
decreases with the increase of the lead time in cold
seasons, which is consistent with our aforementioned
discussion. This is directly linked to the precipitation forecast – cold season forecasts are, in general,
more skillful than the counterpart in warm seasons. In
terms of the scale dependence, forecast skill tends to
increase in cold seasons with increase of basin size.
This does not hold, however, for warm seasons, e.g.
forecasts initiated in July 2007, when the precipitation
variability is strong and forecast skill is more limited.
Copyright  2008 Royal Meteorological Society
However, it is still interesting to note that the variability in July is well picked up by the 3-month lead time
forecast initiated in May. Another factor, which also
contributes to lower-forecast skill in warm season, is
the intense water management in summertime especially for large basins. The basins used in our analysis
were not screened for water management intensity, so
the forecasts include a blend on management effects
that need further analysis.
4. Summary
In this article, a streamflow monitoring and forecasting
component, as an integrated part of Princeton University’s DMAPS, was introduced. The usefulness of
the component was verified against the 2007 hydrologic drought observed across four river basins over
the eastern USA. The system is shown to be capable
Atmos. Sci. Let. 9: 61–66 (2008)
DOI: 10.1002/asl
Seasonal hydrologic predictions over eastern USA
65
Figure 3. Similar to Figure 2 but for probabilistic streamflow forecast for monthly river flow exceeding the 25th percentile of
monthly climatology. The first row is the binary event for observed monthly mean streamflow (blue above the 25th quantile, red
below). Row 2–6 represents the exceedence probability for forecasts initialized from January to September 2007 for 1-, 3, and
5-month lead times, respectively. The inset table representing rows 2–6 is the basin area-weighted average Brier score for the
forecasts.
Jan–2007
Mar–2007
0.7
0.7
0.6
0.5
1–month lead time
3–month lead time
5–month lead time
0.6
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0
Small
Medium
Large
0
Small
May–2007
0.6
0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
Small
Medium
Large
Jul–2007
0.6
0
Medium
Large
0
Small
Medium
Large
Figure 4. Brier score for different basin size categories and lead times. Black, red, and green lines are for 1-, 3-, and 5-month lead
times, respectively. The spread across the basins is represented by one standard deviation. Basin size categories are defined in the
text.
Copyright  2008 Royal Meteorological Society
Atmos. Sci. Let. 9: 61–66 (2008)
DOI: 10.1002/asl
66
of skillfully forecasting such hydrologic low flows for
up to three months in advance. The results are consistent with and complementary to recent analysis of
soil moisture by Luo and Wood (2007a). Thus, in conjunction with soil-moisture-based drought monitoring
and forecasting, the DMAPS provides a comprehensive picture of land surface hydrologic conditions.
The ultimate goal of a drought monitoring and
forecasting system is effective drought management,
as called for in the NIDIS system (Western Governors
Conference, 2004). Such a management can result in
huge economic benefits (Steinemann, 2006). How to
translate the drought forecast products into information
easily understandable to water management sections
requires close collaboration between different sectors
and remains a challenge to be addressed within NIDIS.
Acknowledgements
We thank two anonymous reviewers for their constructive
comments. This research was supported through NOAA’s
Climate Prediction Program for the Americas (CPPA) by Grant
NA17RJ2612. This support is gratefully acknowledged.
References
Brier GW. 1950. Verification of forecasts expressed in terms of
probability. Monthly Weather Review 75: 1–3.
Cherkauer KA, Lettenmaier DP. 1999. Hydrologic effects of frozen
soils in the upper Mississippi River basin. Journal of Geophysical
Research 104(D16): 19599–19610, DOI: 10.1029/1999JD900337.
Cosgrove BA, Lohmann D, Mitchell KE, Houser PR, Wood EF,
Schaake JC, Robock A, Sheffield J, Duan Q, Luo Li, Higgins RW,
Pinker RT, Tarpley JD. 2003. Land surface model spinup
behavior in the North American Land Data Assimilation System
(NLDAS). Journal of Geophysical Research 108(D22): 8845, DOI:
10.1029/2002JD003119.
Federal Emergency Management Agency (FEMA). 1995. National
Mitigation Strategy; Partnerships for Building Safer Communities.
Mitigation Directorate. Federal Emergency Management Agency:
Washington, DC; 2.
Gringorten II. 1963. A plotting rule for extreme probability paper.
Journal of Geophysical Research 68: 813–814.
Hashino T, Bradley AA, Schwartz SS. 2007. Evaluation of biascorrection methods for ensemble streamflow volume forecasts.
Hydrology and Earth System Sciences 11: 939–950.
Liang X, Wood EF, Lettenmaier DP. 1996. Surface soil moisture
parameterization of the VIC-2L model: evaluation and modifications.
Global and Planetary Change 13: 195–206.
Lohmann D, Nolte-Holube R, Raschke E. 1996. A large-scale
horizontal routing model to be coupled to land surface parameterization schemes. Tellus 48A: 708–721.
Lohmann D, Raschke E, Nijssen B, Lettenmaier DP. 1998. Regional
scale hydrology: I. Formulation of the VIC-2L model coupled to a
routing model. Hydrological Sciences Journal 43: 131–141.
Copyright  2008 Royal Meteorological Society
H. Li, L. Luo and E. F. Wood
Luo L, Wood EF. 2007a. Monitoring and predicting the 2007 U.S.
drought. Geophysical Research Letters 34: L22702, DOI:10.1029/
2007GL031673.
Luo L, Wood EF. 2007b. Seasonal hydrologic prediction with VIC
model for the Eastern U.S. Journal of Hydrometeorology (submitted).
Luo L, Wood EF, Pan M. 2007. Bayesian merging of multiple climate
model forecasts for seasonal hydrological predictions. Journal of
Geophysical Research 112: D10102, DOI:10.1029/2006JD007655.
Maurer EP, Wood AW, Adam JC, Lettenmaier DP, Nijssen B. 2002. A
long-term hydrologically-based data set of land surface fluxes and
states for the conterminous United States. Journal of Climate 15:
3237–3251.
Mitchell KE, Lohmann D, Houser PR, Wood EF, Schaake JC,
Robock A, Cosgrove BA, Sheffield J, Duan Q, Luo L, Higgins RW,
Pinker RT, Tarpley JD, Lettenmaier DP, Marshall CH, Entin JK,
Pan M, Shi W, Koren V, Meng J, Ramsay BH, Bailey AA. 2004.
The Multi-institution North American Land Data Assimilation
System (NLDAS): Utilizing multiple GCIP products and partners in
a continental distributed hydrological modeling system. Journal of
Geophysical Research 109: D07S90, Doi: 10.1029/2003JD003823.
Nash JE, Sutcliffe JV. 1970. River flow forecasting through conceptual
models part I – A discussion of principles. Journal of Hydrology
10(3): 282–290.
National Climatic Data Center, cited. 2006. Billion dollar U.S. weather
disasters, [Available online at http://lwf.ncdc.noaa.gov/oa/reports/
billionz.html.].
Nijssen B, Schnur R, Lettenmaier DP. 2001. Global retrospective
estimation of soil moisture using the variable infiltration capacity
land surface model, 1980–93. Journal of Climate 14: 1790–1808.
Nijssen B, Lettenmaier DP, Liang X, Wetzel SW, Wood EF. 1997.
Streamflow simulation for continental-scale river basins. Water
Resources Research 33(4): 711–724.
Saha S, Nadiga S, Thiaw C, Wang J, Wang W, Zhang Q, van den
Dool HM, Pan H-L, Moorthi S, Behringer D, Stokes D, Pena M,
Lord S, White G, Ebisuzaki W, Peng P, Xie P. 2006. The NCEP
climate forecast system. Journal of Climate 19: 3483–3517.
Sheffield J, Goteti G, Wen F, Wood EF. 2004. A simulated soil
moisture based drought analysis for the United States. Journal of
Geophysical Research 109: D24108, DOI:10.1029/2004JD005182.
Steinemann CA. 2006. Using climate forecasts for drought water
management. Journal of Applied Meteorology and Climatology 45:
1353–1361.
Troy T, Wood EF, Sheffield J. 2008. The effect of temporal and spatial
resolutions on land surface hydrological model calibration. Water
Resources Research (submitted).
U.S. Geological Survey, cited. 2007. USGS Real-Time Water Data for
the Nation, [http://water.usgs.gov/waterwatch/?m=real&r=us&w=
real%2Canimation].
Western Governors Association. 2004. Creating a drought early
warning system for the 21st Century: the National Integrated Drought Information System (NIDIS), 16, Available at
http://www.westgov.org/wga/publicat/nidis.pdf.
Wood AW, Maurer EP, Kumar A, Lettenmaier DP. 2002. Long
range experimental hydrologic forecasting for the eastern
U.S. Journal of Geophysical Research 107(D20): 4429, DOI:
10.1029/2001JD00065.
Atmos. Sci. Let. 9: 61–66 (2008)
DOI: 10.1002/asl
Документ
Категория
Без категории
Просмотров
2
Размер файла
4 287 Кб
Теги
asl, 182
1/--страниц
Пожаловаться на содержимое документа