Comparison of Statistical and Dynamic Downscaling methods for ...

Comparison of Statistical and Dynamic
Downscaling methods for Hydrologic
Applications in West Central Florida
Wendy Graham and Syewoon Hwang,
University of Florida Water Institute
Alison Adams and Tirusew Assefa
Tampa Bay Water

Goal of the research
• To compare the ability of three statistical downscaling methods and one
dynamic downscaling to generate precipitation fields at hydrologically
relevant space-time scales over the Tampa Bay region of west-central
Florida.
In this research we…
• quantitatively evaluate the skills of existing methods in terms of both temporal and
spatial precipitation patterns.
• develop a new stochastic downscaling technique which improves over the existing
statistical downscaling methods in reproducing observed spatiotemporal variability of
daily precipitation
• evaluate hydrologic implications of differences in downscaled climate scenarios over the
study area using an existing integrated hydrologic model.
2

General circulation models
200~300km resolution
reanalysis data
1 GCM
4 GCMs
RCM (Misra et al)
10 km resolution
BCSD,SDBC,
BCSA
12 km resolution
Evaluation of
climate information
Integrated Hydrologic Model
Applications
The impact assessment
3

Data
GCMs used in this study
Modeling Group, Country
Bjerknes Centre for Climate
Research, Norway
US Dept. of
Commerce/NOAA/Geophy
sical Fluid Dynamics
Laboratory, USA
Canadian Centre for
Climate Modeling &
Analysis, Canada
WCRP
CMIP3*
I.D.
BCCR-
BCM2.0
GFDL-
CM2.0
CGCM3.1
Primary
Reference
Furevik et
al., 2003
Delworth et
al., 2006
Flato and
Boer, 2001
National Center for
Atmospheric Research, USA CCSM Collins et
al., 2006
* WCRP CMIP3: World Climate Research Programme's
Coupled Model Inter-comparison Project phase 3
Latitude
31
30
29
28
27
26
25
BCCR, 2.8 o x 2.8 o
GFDL, 2.0 o x2.5 o
CGCM3, 2.8 o x2.8 o
CCSM, 1.4 o x1.4 o
24
-88 -87 -86 -85 -84 -83 -82 -81 -80
Longitude
Gridded observations
12x12km 2 , 1/8 degree resolution: Maurer et al., 2002
4

Method
Schematic representation
of the methodology
5

Temporal Statistics
Mean daily precipitation
Std Dev daily precipitation
Averaged daily preciptiation
10
8
6
4
2
Gobs
R2+RCM
CCSM+RCM
CCSM+BCSD
CCSM+SDBC
CCSM+BCSA
avg. std. of daily precipitation
18
14
10
6
Gobs
R2+RCM
CCSM+RCM
CCSM+BCSD
CCSM+SDBC
CCSM+BCSA
0
Jan Feb Mar Apr May Jun
Jul Aug Sep Oct Nov Dec
2
Jan Feb Mar Apr May Jun
Jul Aug Sep Oct Nov Dec

Transition Probability
Wet to Wet
Dry to Wet
1
0.8
0.6
Gobs
R2+RCM
CCSM+RCM
CCSM+BCSD
CCSM+SDBC
CCSM+BCSA
1
0.8
0.6
Gobs
R2+RCM
CCSM+RCM
CCSM+BCSD
CCSM+SDBC
CCSM+BCSA
P11
P01
0.4
0.4
0.2
0.2
0
Jan Feb Mar Apr May Jun
Jul Aug Sep Oct Nov Dec
0
Jan Feb Mar Apr May Jun
Jul Aug Sep Oct Nov Dec

Variograms (spatial variability analysis)
Wet season (Jun.‐Sep.)
Dry season (Oct.‐May)
Variogram (mm 2 )
400
300
200
Gobs
R2+RCM
CCSM+RCM
CCSM+BCSD
CCSM+SDBC
CCSM+BCSA
Variogram (mm 2 )
400
300
200
Gobs
R2+RCM
CCSM+RCM
CCSM+BCSD
CCSM+SDBC
CCSM+BCSA
100
100
0
0 100 200 300 400 500
Separation distance (km)
0
0 100 200 300 400 500
Separation distance (km)

General circulation models
200~300km resolution
reanalysis data
1 GCM
4 GCMs
RCM (Misra et al)
10 km resolution
BCSD,SDBC,
BCSA
12 km resolution
Evaluation of
climate information
Integrated Hydrologic Model
Applications
The impact assessment
9

Method
Integrated Hydrologic Model
• TBW and SWFWMD commissioned the development and application of an integrated surface
water/groundwater model for the Tampa Bay Region.
• The Integrated Hydrologic Model (IHM) was developed which integrates the EPA Hydrologic
Simulation Program-Fortran for surface-water modeling with the US Geological Survey
MODFLOW96 for groundwater modeling.
Ross et al., 2004 (IHM theory manual)
10

Method
Study domain
11

Monthly average streamflow
12
0
5
10
15
20
25
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
Streamflow (m3/s)
BCSD_GCMs
Obs.
Cal.
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
SDBC_GCMs
Obs.
Cal.
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
BCSA_GCMs
Obs.
Cal.
0
2
4
6
8
10
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
Streamflow (m3/s)
BCSD_GCMs
Obs.
Cal.
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
SDBC_GCMs
Obs.
Cal.
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
BCSA_GCMs
Obs.
Cal.
Alafia River
Cypress Creek
BCSD SDBC BCSA
BCSD SDBC BCSA

Mean of errors for monthly average streamflow
Mean of errors (simulated-calibrated) for monthly average
streamflow over four GCM results for each target station
8
Alafia River
BCSD
4
Cypress Creek
BCSD
SDBC
SDBC
4
BCSA
2
BCSA
mean error (m 3 /s)
0
mean error (m 3 /s)
0
‐4
‐2
‐8
‐4

Mean errors for daily streamflow standard deviation
Mean of errors (simulated-calibrated) for daily streamflow
standard deviation over four GCM results for each target
station
20
Alafia River
BCSD
4
Cypress Creek
BCSD
SDBC
SDBC
10
BCSA
2
BCSA
mean error (m 3 /s)
0
mean error (m 3 /s)
0
‐10
‐2
‐20
‐4
14

Frequency of daily streamflow events
# of events per year
# of events per year
1000
100
10
1
0.1
0.01
1000
100
10
1
0.1
0.01
Streamflow (m3/s)
Alafia River
Obs.
Cal.
BCSD_GCMs
SDBC_GCMs
BCSA_GCMs
0 100 200 300 400 500 600
Streamflow (m3/s)
Cypress Creek
Obs.
Cal.
BCSD_GCMs
SDBC_GCMs
BCSA_GCMs
0 10 20 30 40 50 60 70
• The frequency of daily streamflow was not reproduced by SDBC as closely as for the BCSA results.
• Comparing the , it is evident that the under/overestimations of extreme events is canceled out when
evaluating only the monthly mean streamflow.

Conclusions
• Commonly used BCSD method underestimates spatial and temporal variability of
rainfall, and over estimates number of low precipitation days. These errors
propagate into hydrologic predictions producing higher ET and lower
streamflow predictions in summer months
• SDBC method improves estimates of day-to-day temporal variability at fine gridscale,
but underestimates spatial variability of rainfall and thus overestimates
temporal variability of spatially averaged rainfall. As a result SDBC successfully
reproduces mean monthly streamflow, but significantly overestimates
magnitude of peak streamflow events.
• BCSA method reproduces spatial and temporal variability of rainfall accurately
and successfully reproduces mean monthly streamflow. However BCSA slightly
overestimates magnitude of peak streamflow events for the larger river.
16

Conclusions
• Dynamically downscaled results show promise, especially
when driven by reanalysis data. Hydrologic implications of
higher than observed spatial variability and temporal
variability of local rainfall will be investigated.
• Dynamically downscaled retrospective GCM predictions are
way off! Does it make sense to both dynamically downscale
AND bias-correct these results?
17

Future work
• Evaluate the utility of using various downscaling methods with
seasonal GCM predictions and long-term projections of
climate change scenarios in the Tampa Bay region.
• Assess the potential to use downscaled seasonal predictions to
reduce risk for water management operations in the Tampa
Bay region.
• Assess potential future climate change impacts on the
hydrologic system in west-central Florida.
18