11
Drought Monitoring over the United Drought Monitoring over the United
States States
Kingtse Mo Kingtse Mo
Climate Prediction Ct Climate Prediction Ct NCEP/NWS/NOAA NCEP/NWS/NOAA
Our Goals Our Goals
Provide users timely information and analysis Provide users timely information and analysis on drought.
on drought.
Monitor atmospheric and hydrological Monitor atmospheric and hydrological
conditions in support of operational Drought conditions in support of operational Drought
Monitor and Drought Outlook Monitor and Drought Outlook
Develop regional applications in support of the NIDIS pilot project
http://www.cpc.ncep.noaa.gov/products/Drought
33
Drought Drought Briefing Briefing
Kingtse.mo@noaa.gov
Current status Current status
http://www.cpc.ncep.noaa.gov/products/Drought
Current conditions:Current conditions:
Surface conditions ,drought indicesSurface conditions ,drought indices, E, P soil , E, P soil moisture – from the ensemble NLDAS ( VIC, MOSAIC, moisture – from the ensemble NLDAS ( VIC, MOSAIC, SAC and VIC) SAC and VIC)
Atmospheric conditions and budget terms : NARRAtmospheric conditions and budget terms : NARR
Forecasts: Forecasts:
U. Washington (ESP)U. Washington (ESP)
Princeton U-EMC (downscaling from the CFS using Princeton U-EMC (downscaling from the CFS using VIC)VIC)
NSIPP (from NSIPP model)NSIPP (from NSIPP model)
55
Define drought based on the drought Define drought based on the drought
Indices Indices
Meteorological droughtMeteorological drought: Precipitation deficit. : Precipitation deficit.
IndexIndex: : Standardized Precipitation IndexStandardized Precipitation Index
Hydrological droughtHydrological drought: Streamflow or runoff deficit: Streamflow or runoff deficit Index:Index: Standardized runoff indexStandardized runoff index
Agricultural droughtAgricultural drought: Total soil water storage deficit : Total soil water storage deficit
IndexIndex: : SM anomaly percentileSM anomaly percentile
A wet region A wet region
drought drought
6 mo running mean black line 6 mo running mean black line
3 mo running mean (black line) 3 mo running mean (black line)
No smoothing No smoothing
Red line: monthly 75-85W,31-35N
77
SM has much lower freq. over the western region SM has much lower freq. over the western region
A dry region Utah-AZ
Example: Jun 2008 Example: Jun 2008
1. Heavy flooding over Iowa, Mo,&
Illinois in June 2. Dryness over
the Southeast &
Texas and California 3. This pattern
persisted from April
June 2008 AMJ 2008
Kansas : western Kansas : D3 drought Eastern Kansas: Floods
mm/day
99
1. Rainfall over the Mississippi basin influenced all SPI indices.
2. Dry: Southeast, southern Texas and California 3. Wet: Central U.
S. and the upper Missouri basin (RFC 2 area)
Meteorological drought
D3 D2 D1 D0
Hydrological Drought Hydrological Drought
Western Kansas Dry,
1111
Agricultural drought Agricultural drought
University of Washington ensemble:
They all capture basic features:
Drought: SE, southern Texas and California Wetness: great Plains.
But details differ
UW and NCEP both show wet conditions over the great Plains , Dryness over the Southwest and California.
UW percentiles are higher
EMC/NCEP
Don’t worry, be Don’t worry, be
happy!!
happy!!
Different models have different total soil moisture, but anomalies are
similar
Robock et al. 2004;
Dirmeyer et al. 2004, Koster et al. 2008
1313
Contributors Contributors
University of WashingtonUniversity of Washington
Dennis Lettenmaier, Ted Bohn and Dennis Lettenmaier, Ted Bohn and
Shraddhanad ShuklaShraddhanad Shukla
EMCEMC
EMC: Youlong Xia, Ken Mitchell & Mike EkEMC: Youlong Xia, Ken Mitchell & Mike Ek
The EMC NCEP system (NCEP) The EMC NCEP system (NCEP)
Four models: Noah, VIC, Mosaic and SACFour models: Noah, VIC, Mosaic and SAC
From 1979- presentFrom 1979- present
On 0.125 degrees grid (the study was done for On 0.125 degrees grid (the study was done for 0.5 degrees)
0.5 degrees)
P forcing : From the CPC P analysis based on P forcing : From the CPC P analysis based on rain gauges with the PRISM correction.
rain gauges with the PRISM correction.
Other atmospheric forcing: From the NARROther atmospheric forcing: From the NARR
1515
The University of Washington The University of Washington
System (UW) System (UW)
Four models: Noah, VIC, SAC and CLMFour models: Noah, VIC, SAC and CLM
From 1915-presentFrom 1915-present
On 0.5 degrees gridOn 0.5 degrees grid
P, Tsurf and low level winds from P, Tsurf and low level winds from NOAA/NCDC co-op stations
NOAA/NCDC co-op stations
No diurnal cycle in P, solar radiation was No diurnal cycle in P, solar radiation was computed from Tsurf
computed from Tsurf
Possible reasons for differences Possible reasons for differences between the UW and NCEP systems between the UW and NCEP systems
Different base period: Different base period:
NCEP (1979-2006);
NCEP (1979-2006);
UW: 1920-2003 UW: 1920-2003
Different models in the ensembleDifferent models in the ensemble
Different forcingDifferent forcing
1717
Procedures:
Procedures:
Compute monthly mean climatology for the base period Compute monthly mean climatology for the base period
Anomaly: Departure from the monthly mean Anomaly: Departure from the monthly mean climatology for the base period.
climatology for the base period.
Standardized anomaly: Anomaly divided by the Standardized anomaly: Anomaly divided by the monthly mean standard deviation for that month.
monthly mean standard deviation for that month.
SPI, SRI were calculated from monthly mean SPI, SRI were calculated from monthly mean precipitation and runoff data for the base period precipitation and runoff data for the base period
SM anomaly percentiles were calculated from SM anomaly percentiles were calculated from standardized anomalies (Mo 2008)
standardized anomalies (Mo 2008)
Differences due to the base period Differences due to the base period
The UW system was used to study the impact of base The UW system was used to study the impact of base periods because their data cover 1915-2007.
periods because their data cover 1915-2007.
SM %, SRI and SPI were computed for the period SM %, SRI and SPI were computed for the period 1979-2007 using the base period 1920-2007 and 1979-2007 using the base period 1920-2007 and
1979-2007 respectively.
1979-2007 respectively.
RMS difference between SM % ( SRI or SPI) from RMS difference between SM % ( SRI or SPI) from the two base periods was computed for each model the two base periods was computed for each model
Same calculations were performed for the ensemble Same calculations were performed for the ensemble means (equally weighted mean)
means (equally weighted mean)
1919
RMS diff of sm anomaly % for two base periods
Differences are small (less than 10%) except for the CLM model.
Noah
VIC
SAC
CLM
RMS of the Ensemble mean difference RMS of the Ensemble mean difference
SPI3SPI3
SRI3SRI3
SM%SM%
1.1. Ensemble means decrease the Ensemble means decrease the uncertainties of the NLDAS;
uncertainties of the NLDAS;
2.2. For SPI3 and SRI3, the 29-yr For SPI3 and SRI3, the 29-yr base period is sufficient.
base period is sufficient.
3. The differences for SM are less 3. The differences for SM are less
than 10% over the eastern region than 10% over the eastern region
and about 10-15 % over the and about 10-15 % over the
2121
Differences due to base period Differences due to base period
Ensemble mean decreases uncertainties and the RMS Ensemble mean decreases uncertainties and the RMS differences between two periods.
differences between two periods.
For ensemble means, the 30-yr period is sufficient for For ensemble means, the 30-yr period is sufficient for SPI or SRI 3 months or longer.
SPI or SRI 3 months or longer.
For SM %, the differences are less than 10% over the For SM %, the differences are less than 10% over the eastern United States, and 10~ 15% over the western eastern United States, and 10~ 15% over the western
U. S.
U. S.
If sudden changes occur, then outputs need to be If sudden changes occur, then outputs need to be calibrated before and after the change separately.
calibrated before and after the change separately.
2222
mean SM percentiles and
standardized anomalies for area 38-
Black: monthly mean P wrt base (1920- 2006);
Red: 6-mo running mean
The VIC has larger high frequency components. There were enough large positive/negative values before & after 1979 to make the percentiles similar for both base periods.
The CLM picks up the low freq.
component
2323
Wang et al. 2009 JHM VIC: 3 soil
layers. Soil depth: 0.8-3 m
Noah: 4 soil layers. Soil depth 2m
SAC: 5 soil layers. Soil depth 600mm
CLM: 10 soil layers. Soil depth 3.43m
Measure the differences among models Measure the differences among models
Rm for a group of models
m :the mean intermodel variance (or
spread)
int (m): interannual variance of the ensemble mean)
int
( m R
m m
Similar formula was used by Dirmeyer et al (2004). to assess
2525
Differences due to forcing Differences due to forcing
Differences are measured by R which is the Differences are measured by R which is the ratio between spread and interannual
ratio between spread and interannual variability
variability
R was calculated for the NCEP (4 models), the R was calculated for the NCEP (4 models), the UW (4 models) separately and two systems (8 UW (4 models) separately and two systems (8
models) pooled together for 1979-2007.
models) pooled together for 1979-2007.
We computed R for SMWe computed R for SM
R values for SM % R values for SM %
1.1.The spread among the The spread among the members from the same members from the same system (UW or NCEP) is system (UW or NCEP) is small. It is less than 0.4.
small. It is less than 0.4.
(Fig. a and b) (Fig. a and b)
2. R values with all UW and 2. R values with all UW and
NCEP members together is NCEP members together is
much larger (Fig.c).
much larger (Fig.c).
This implies that the mean This implies that the mean differences between two differences between two
systems are large systems are large
2727
1.1.The RMS difference (Fig.d) The RMS difference (Fig.d) between the ncep and the between the ncep and the
UW ensemble SM means UW ensemble SM means
are large over the western are large over the western
U. S. (> 20%).
U. S. (> 20%).
2.2. Largest differences occur Largest differences occur after 2001 as indicated by after 2001 as indicated by
the mean differences for the mean differences for
two periods (Fig. f and g) two periods (Fig. f and g)
•Differences between two systems are
larger than the spread among members of
the same system
•The differences are not The differences are not caused by one model.
caused by one model.
They are caused by They are caused by
forcing.
forcing.
• In general, extreme In general, extreme values from the UW values from the UW
(Green) are larger than (Green) are larger than
from the NCEP (red) from the NCEP (red)
standardized SM anomalies for area 38-42N,110-115W
2929
P and Tsurf differences are larger after 2002 P and Tsurf differences are larger after 2002
when systems went to operation in real time when systems went to operation in real time
There were less There were less station data after station data after
2002 when systems 2002 when systems
went to real time went to real time
operation.
operation.
There are larger There are larger uncertainties in uncertainties in
forcing forcing
The NCEP has The NCEP has larger P variances larger P variances
than the UW
than the UW , so
extreme values have smaller %.
Uncertainties from different systems Uncertainties from different systems
Differences between members of the same system Differences between members of the same system (ncep or uw) are small. (runoff shows the same thing) (ncep or uw) are small. (runoff shows the same thing)
There are large differences between the NCEP and There are large differences between the NCEP and the UW systems. They are caused by forcing.
the UW systems. They are caused by forcing.
Differences for historical period (before 2001 or Differences for historical period (before 2001 or
2002) were small. After systems went to operation in 2002) were small. After systems went to operation in
near real time, station obs dropped and large near real time, station obs dropped and large
differences occurred.
differences occurred.
Currently, both systems will declare drought (wet Currently, both systems will declare drought (wet spells) , but they are likely to give different D
spells) , but they are likely to give different D categories
categories
3131
Model differences Model differences
Characteristic time To– time scale of SM.Characteristic time To– time scale of SM.
Correlations between different variables Correlations between different variables For the common period 1979-2006 on 0.5 For the common period 1979-2006 on 0.5
resolution grid resolution grid
Characteristic time :To Characteristic time :To
Ni
i R N
i T
1
0
1 2 ( 1 / ) ( )
R(i) : auto correlation at time lag i, n=30
• To is calculated for each model and for the ensemble means.
• We also ask the question whether the spectral for SM is red. R(1) was computed for the NCEP
ensemble mean. To was calculated for red noise
Trenberth 1984
3333
Characteristic time T Characteristic time To o to measure the SM persistence
Overall, To is larger over Overall, To is larger over the interior western states the interior western states (24-32 months) than the (24-32 months) than the eastern U. S. (6 months) eastern U. S. (6 months)
To is related to the total To is related to the total water storage of the
water storage of the
model. CLM has less high model. CLM has less high frequency and more water frequency and more water storage so To is large.
storage so To is large.
SAC has smaller To SAC has smaller To
because of smaller water because of smaller water storage
storage
Ensemble mean To Ensemble mean To
•UW has larger To than the NCEP over the western region (6 month difference)
•Red noise model may be a good approximation for the western region
3535
Runoff P
dt E
dw
W= total Soil moisture W= total Soil moisture
PE
E
Soil moisture balance Eq Soil moisture balance Eq
PE-potential evaporation= E when the PE-potential evaporation= E when the soil is sufficiently saturated
soil is sufficiently saturated Beta= evaporation efficiency Beta= evaporation efficiency
Runoff P
dt PE
dw ( )
Beta is roughly proportional to W. If P-runoff is small,, then we can use red noise model
Persistence of SM Measured by To Persistence of SM Measured by To
To is larger over the interior western region To is larger over the interior western region (about 24-32 months) and smaller over the (about 24-32 months) and smaller over the
eastern region (6 months). Therefore, drought eastern region (6 months). Therefore, drought
over the western region tends to persist.
over the western region tends to persist.
To from the UW is larger than the NCEP. To from the UW is larger than the NCEP.
To is related to the water storage capacity of To is related to the water storage capacity of the model.
the model.
Red noise model fits well for SM over the Red noise model fits well for SM over the western region when P.-runoff is small
western region when P.-runoff is small
3737
Relationships among variables Relationships among variables
Tsurf and P are forcing for the NLAS. How do Tsurf and P are forcing for the NLAS. How do they relate to SM?
they relate to SM?
Are there any direct (linear) relation between Are there any direct (linear) relation between them?
them?
We study them using correlations. We study them using correlations.
Correlations (Tsurf and SM) Correlations (Tsurf and SM)
for ensemble means for ensemble means
No significant correlations for JFM, OND
The UW has
weaker correlation between Tsurf and SM over the
northern plains.
3939
Possible mechanisms Possible mechanisms
NCEP ensemble mean(Noah, VIC and Mosaic)
SM influence T2m SM influence T2m through ET
through ET Mechanism:
Mechanism:
more SM more SM
more ETmore ET
less sensible heat if no less sensible heat if no large radiation changes large radiation changes
cooler T2m cooler T2m
Possible mechanisms Possible mechanisms
E is small in winter JFM and OND so no correlation between SM and Tsurf. E is small in winter JFM and OND so no correlation between SM and Tsurf.
For AMJ, the Northeast is cold. For AMJ, the Northeast is cold.
Warmer T2m => more vegetationWarmer T2m => more vegetation
more transpirationmore transpiration
so T2m and E are correlated. SM does not play a role.so T2m and E are correlated. SM does not play a role.
For JAS, SM feedbacks to E so correlations are highFor JAS, SM feedbacks to E so correlations are high
The UW has lower correlations over the northern plains. The UW has lower correlations over the northern plains.
4141
Monthly mean
correlation between psq and SM for a given
season.
NCEP UW
Runoff P
dt E
dw
Correlations are small except over the Southeast in JFM and AMJ.
Delayed response?
Correlation
between monthly mean Psq(t) and SM(t+1) for a given month.
The relationship is
regionally and seasonally dependent.
Correlations are large over the wet area
4343
P & SM P & SM
For the wet region, P-runoff term is large.
They are comparable with the beta PE term.
For dry region, P-runoff term is small. PE does not depend on SM, so beta has larger influence
Runoff P
dt PE
dw ( )
Current drought conditions &
Current drought conditions &
outlook
outlook
4545
Correlation Correlation
between spi6 between spi6
and sm%
and sm%
They are higher They are higher
than the old than the old
versions of versions of
models. (Mo models. (Mo
2008) 2008)
Again, CLM is Again, CLM is different from different from
the rest the rest
1. correlations between different indices are
much higher than the old NCEP Noah and VIC systems (Mo 2008) 2. SRI6 and SM has lower correlations over the east region.
3. The ncep has stronger relationship between SPI6 and SM than the
Correlations between
ensemble
mean indices
4747
Indices Indices
Correlations among indices are larger than the Correlations among indices are larger than the previous version of models.
previous version of models.
The NCEP has stronger correlation between The NCEP has stronger correlation between SPI6 and SM than the UW.
SPI6 and SM than the UW.
Correlations between SM and SRI are lower Correlations between SM and SRI are lower over the wet region. (runoff takes water out over the wet region. (runoff takes water out from the soil?)from the soil?)
