R.W. Higgins1, V. Silva2 , W. Shi1 , and V. E. Kousky

1
Comparison of Daily Precipitation Statistics for
the US in Observations and in the NCEP CFS
by R.W. Higgins1, V. Silva2 , W. Shi1 , and V.
E. Kousky 1Climate Prediction Center, Camp
Springs, MD 2RS Information Systems, Mclean, VA
2
Overview

• Compare statistics of daily precipitation within
  seasonal climate over the U.S. using gridded
  station data (1948-2006), NCEP CFS re-forecasts
  (1981-2005) and CFS CMIP 100-yr simulations
• Current Goals
• Identify the regional seasonal dependence of
  the bias in CFS re-forecasts
• Examine differences by ENSO phase
• Examine differences in the frequency of wet and
  dry spells
• Future Goal
• Develop more reliable ensemble-based
  probabilistic forecasts in real-time at
• weeks 2-4 (e.g. risks of heavy rain events)

3
Data

• US Unified Raingauge Database (Higgins et al.
  2000)
• - multi-year (1948-2006) daily analysis
  (12Z-12Z).
• - horizontal resolution (lat,
  lon)(0.25x0.25)
• - domain (140oW-60oW,10oN-60oN)
• - Cressman (1959)
• - advanced QC (duplicate, buddy, std dev, radar)
  
• NCEP CFS retrospective forecasts (re-forecasts)
• - fully coupled O-A-L prediction system T62L64
  (Saha et al. 2006).
• - atmosphere (GFS 2003), ocean (GFDL MOM3)
• - 1981-2005 15 fcsts per calendar month out to
  nine months
• - atmospheric ICs (NCEP/DOE Reanalysis 2
  Kanamitsu et al. 2002)
• - ocean ICs (NCEP GODAS Behringer 2005).
• NCEP CFS CMIP Simulations
• - 100 years T126L64
• - CMIP 1 (init Jan 1, 2002) and CMIP 2 (init Jan
  1, 1984)

4
Average Observed Number of
Days per Season with p gt 1 mm
(1981-2005)
Average Number of Days per Season with p gt 1 mm
(CFS Re-forecasts OBS)

• - Observed annual cycle shows highest number of
  wet days per season in PNW (fall and winter), SE
  (summer), NE (spring, summer) and SW (summer)
• Difference patterns show considerable evolution
  depending on season and lead
• JFM positive bias along northern tier at all
  leads
• AMJ positive bias in west negative bias GP, GC
• JAS negative bias in SW GC
• OND positive bias in NP

5
Avg. Number of Days per Season with p gt 1 mm
CMIP OBS (1981-2005)
Differences between the CFS coupled simulations
observations yield similar patterns.
6
Average Number of Days-per-Season with p gt 1 mm
(CFS Re-forecasts CMIP 1)
(CFS Re-forecasts CMIP 2)
Differences between the CFS re-forecasts and the
CMIP simulations were examined to see if spin-up
might be affecting the CFS re-forecast results at
shorter leads. There are obvious differences
between day 1 and day 10 patterns that appear to
be due to the affects of spin-up, while the day
10 and day 100 patterns are similar.
7
Regional Characteristics of the Bias

• Bias in the number of days with precipitation in
  CFS has considerable spatial and temporal
  variability through the annual cycle.
• To examine regional characteristics of the bias,
  4 regions were selected based on areas of large
  bias in the spatial difference (CFS-OBS). Care
  was taken to choose areas with bias of one sign
  or the other.
• Four regions
• Interior Pacific Northwest (PNW) -
  (42.5oN-47.5oN, 115oW-120oW) North-Central (NC)
  - (45oN-47.5oN, 95oW-100oW)
• Southwest (SW) - (32.5oN-37.5oN, 107.5oW-112.5oW)
  Southeast (SE) - (30oN-35oN, 82.5oW-87.5oW)

8
Average Number of Days per Season with
Precipitation by Categorical Amount
OBS
CFS Day 1
CFS Day 10
CFS Day 100

• PNW Too many precipitation events (Pgt 4
  mm/day) during boreal winter bias increases with
  forecast lead suggesting spin-up effects.
  Orographic influences too far downstream in CFS?
  Effect of model resolution?
• NC Too many precipitation events for all
  precipitation intensities during fall and winter.
  Overactive shallow convection in CFS?
• SE and SW compare well during boreal fall and
  winter

9
Average Number of Days per Season with
Precipitation by Categorical Amount
OBS
CFS Day 1
CFS Day 10
CFS Day 100

• NC Too few light precipitation events during
  boreal summer bias decreases at longer leads.
• SW Too few light precipitation events during
  boreal summer bias decreases at longer leads.
  Weak monsoon?
• SE Negative bias for light events and positive
  bias for heavy events during JAS. Diurnal cycle
  too strong?

10
ENSO Composite Analysis Observations

• El Niño La Niña episodes are identified using
  the Oceanic Niño Index (ONI) (Kousky and Higgins
  2007).
• The ONI is computed from 3-mrm
  values of SSTA in the Niño 3.4 region
  (5N-5S,120W-170W) using a set of homogeneous
  historical SST analyses (ERSST.v2 of Smith and
  Reynolds 2003).
• El Niño La Niña episodes are defined as 5 (or
  more) consecutive 3-month seasons during the
  period 1950-2006.

El Niño
neutral
La Niña
11
ENSO Composite Analysis CMIP
Simulations
CMIP 1

• A similar procedure is used to identify ENSO
  episodes in the CFS CMIP simulations (i.e. define
  an ONI identify El Niño and La Niña episodes)
• The results are similar whether we use 59 years
  (to match the observations) or 100 years, so we
  use 100 years to improve the statistics.
• The Table shows the number of warm (El Niño),
  neutral, and cold (La Niña) episodes in the CMIP1
  and CMIP 2 simulations.

JFM AMJ JAS OND
El Nino 21 25 24 23
ENSO-neutral 54 51 52 52
La Nina 25 24 24 25
CMIP 2
JFM AMJ JAS OND
El Nino 23 27 27 25
ENSO-neutral 52 50 52 54
La Nina 25 23 21 21
12
Departures from Average Number of Wet Days (p gt 1
mm) by Season and by ENSO Phase
(OBS)
(CFS CMIP)
(Based on ENSO events during
1948-2006) Base period is 1948-2006
(Based on ENSO events from 100 yr
simulation) Base period is 100 yr

• CFS reproduces many of the classical features of
  the ENSO precipitation anomaly patterns, but
  there are some systematic differences.

13
Average Number of Wet Days (p gt 1 mm) by Season
and by ENSO Phase
(CFS CMIP OBS)
OBS Based on ENSO events during 1948-2006 CMIP
Based on ENSO events from 100 yr
simulation Composite differences are based on
the full fields

• CFS has a positive bias in the average number of
  wet days per season independent of ENSO phase,
  except in the Southwest during the warm season.
  
• The CFS has as a positive bias of up to 30-40
  additional wet days per season, which implies
  that it rains nearly every day at some locations.
  

14
Total Number of Wet Spells for Conterminous US
JFM
OBS (1948-2006) and last 59 yrs of CMIP 1
OND
OND and JFM - CFS has a positive bias in the
number of wet spells in the PNW and along the
northern tier-of-states and a negative bias in
the SE. - Results suggest a systematic northward
shift of the jetstream and stormtrack in CFS, and
a tendency for the flow to be too zonal in CFS
relative to OBS.
15
Total Number of Wet Spells for Conterminous US
JAS
OBS (1948-2006) and last 59 yrs of CMIP 1
AMJ
AMJ and JAS - CFS has positive bias in PNW and
SE negative bias in SW (JAS only), Great Lakes
and Gulf Coast. - Summer monsoon induced
precipitation is too weak in CFS relative to OBS.
- Land-sea breeze induced precipitation is too
weak along the Gulf Coast in CFS (especially
JAS) - These analyses motivate additional
synoptic studies aimed at improving the linkage
between daily precipitation and related
circulation features in CFS
16
Circulation Differences CMIP1-CDAS

• Average 200-hPa wind was examined for JFM and JJA
  over the last 50 years of the CMIP1 free run.
• The differences between those fields and the
  1971-2000 CDAS1 mean fields were used to evaluate
  the performance of the CFS (CMIP1 run).
• Differences are considered to be biases in the
  model climatology.

17
CMIP1 vs- CDAS1 NH flow is more zonally
symmetric. NH westerlies are shifted
poleward. Eastern Pacific low-latitude troughs
are too weak.Consistent with positive bias in
number of wet spells in CFS along northern
tier-of-states
18
CMIP1 vs- CDAS1 SH flow is more zonally
symmetric. NH westerlies are too strong over NH
subtropics. Eastern Pacific low-latitude troughs
are too weak.
19
CMIP1 vs- CDAS1 Upper-level ridge too weak in
the SW (weak monsoon) - Anomalous convergence
(divergence) in SW (PNW) - Consistent with
negative (positive) bias in the number of wet
spells in SW (PNW) Westerlies are too strong
over the hurricane development regions in the
Gulf of Mexico, off the Southeast US coast and
over the western Caribbean.
20
Total Number of Dry Spells for Conterminous US
OBS (1948-2006) and last 59 yrs of CMIP 1
JFM
OND
OND and JFM - CFS has a negative bias in the
number of dry spells at most locations during
boreal fall winter. - There are very few dry
spells longer than about 20 days in the CMIP runs
(difference patterns are similar to observed
patterns and of opposite sign).
21
Total Number of Dry Spells for Conterminous US
OBS (1948-2006) and last 59 yrs of CMIP 1
JAS
AMJ

• AMJ and JAS
• During boreal spring the CFS has a negative bias
  in the number of dry spells of all durations at
  most locations except in the southwestern states.
  
• During boreal summer the CFS has a positive bias
  in the Southwest (weak monsoon induced
  precipitation) and along the Gulf Coast (weak
  land-sea breeze induced precipitation).

22
Statistical Adjustment of CFS Operational
Forecasts

• The re-forecasts and their corresponding
  verifications can be used to calibrate
  ensemble-based probabilistic forecasts.
• The value of this approach has been demonstrated
  for both weather (short term) climate
  predictions (e.g. Hamill et al. 2004 2006).
• Day-to-day weather cannot be predicted at leads
  of 2-6 weeks, but shifts in time averages (e.g.
  average weather over the period) may still be
  predicted skillfully. (Signal-vs-noise).
• -e.g. likelihood of extreme events (e.g.
  risk of heavy rain event s, flash floods, flash
  droughts)
• Bias corrected forecasts would be immediately
  useful in CPC forecast operations for preparation
  of the day 6-10, day 8-14 forecasts and Hazards
  Assessments (US, Africa, global tropics).

23
Summary

• The statistics of daily precipitation within
  seasonal climate over the U.S. from gridded
  station data (1948-2006) from NCEP CFS
  re-forecasts (1981-2005) and 100-yr CMIP
  simulations were intercompared.
• Current Results
• Quantified the regional seasonal dependence of
  the bias in CFS re-forecasts
• Examined differences by ENSO phase
• Examined differences in the frequency of wet
  and dry spells.
• Future Goal
• Develop reliable ensemble-based probabilistic
  forecasts in real-time at
• weeks 2-4 (e.g. risks of heavy rain events,
  flash floods or flash drought)
• Compare current results to those from next
  generation CFS.