Challenge of defining normal precipitation with medians
Defining a normal from a climatological distribution of precipitation is not a trivial exercise because precipitation is non-continuous, positively skewed, and often
characterized by alternating periods
of rainy and dry conditions that can either be attributed to noise or physical drivers. A standard practice at CPC is to estimate the median climatology for precipitation as opposed to the mean, which can
be sensitive to outliers.
i) Precipitation is inherently noisy.
ii) Precipitation has non-Gaussian distributions, with medians less than the means.
iii) Raw annual cycles of precipitation climatologies may be non-physical.
iv) Smoothing the raw annual cycles of precipitation risks being arbitrary.
v) The calculation of precipitation medians from reforecasts is not a trivial task.
By raising awareness of above pitfalls, it will lead to the development of robust, meaningful climatologies that are useful to the research and forecasting community.
Geographical separation of seasonal prediction skill between statistical
tool and dynamical model
Statistical tool: Constructed Analog (CA) (van den Dool 1992, 2007)
i) Data: HAD SST (45°S-45°N,
ii) Ensemble size: 24 members
(1-4 seasons data in ICs, 6 EOF cutoffs (35, 40, 45, 50, 55, 60)
Dynamical model: North American Multi-Model Ensemble (NMME) (Kirtman et
Initial condition (IC) season: MAM, JJA, SON, DJF
Forecast lead-time: 1 and 5 months
Skill metrics: Anomaly Correlation (AC)
Assessment time period: CA ~ 1981-2015; NMME ~ 1981-2010
Verification data: NOAA-OI-v2 SST
Puzzle: It was found distinct geographical separation of
seasonal prediction skill with decent skill shown over the tropical western
Pacific and Indian Ocean by CA and that over the tropical central-eastern
Pacific by NMME (e.g.
Fig. 1). A summary of all cases (varied initial seasons and lead times)
is given by
Challenges: The Constructed Analog (CA), a statistical
tool, clearly revealed appreciable predictability over the tropical western
Pacific and Indian Ocean, where dynamical models had little skill; pointing
to possibly missing of important process(es) common to dynamical models,
whose development efforts more focused on improving ENSO forecast
case of week-2 forecast running to the opposite of the observation
Date of initial condition : 11/21/2006
Forecast target period: 11/29-12/5/2006
Predictant: Mean surface temperature anomaly
Model: NCEP GFS
model ensemble forecast was totally out of phase compared with the
examined the near-range forecast and found that the 6-10 day forecast from
11/26 initial condition can correctly capture the 500 hPa height observed
pattern, while that from 11/25 initial condition cannot.
between temperature and precipitation in 2010-2011 seasonal forecasts
winter and spring of late 2010 and 2011 were characterized by a moderate to
strong La Niña across the tropical Pacific Ocean, which shaped CPC’s
seasonal outlooks for those seasons.
forecasts for September – November 2010 through April – June 2011 scored at
least 30% better than a climatological forecast, the longest streak (eight)
of successful forecasts since CPC began issuing seasonal forecasts in 1995
the temperature forecasts during the heart of the winter (November –
January, December – February, and January – March) were not as successful,
with Heidke skill scores near or below zero. What caused the disparity in
skill between the temperature and precipitation forecasts? The answer could
be the seasonally dependent influence of unpredictable factors,
i.e. AO, PNA et al.
Predicting 2011/12 La Nina onset by models
— Where was the early warning?
In CPC Sanity Check of November 2011, the failure of Nino 3.4 SST forecast
from June initial condition by almost all models was brought to attention
with two exceptions of good forecasts by ESSIC Intermediate Coupled Model
(ICM) and by Japan Frontier Research Center for Global Change Coupled
General Circulation Model. Looking at individual model performance, a common
problem of phase delay in prediction of ENSO transition is clearly shown.
What key physical processes were missed or misrepresented?
Difficulties in prediction of ENSO phase changes
and impact on outlooks of 2006 North Atlantic hurricane season & 2006/07 DJF
The predictive ENSO condition is very important information for forecasters
when making seasonal outlooks. Forecasters found that all tools are too
much like persistence, according to which too late phase transition would
also occur in the forecast, the events starting too late and then lasting
beyond the time they should. Following is a case, showing forecasts of
ENSO phase transitions in late 2006 and early 2007 and the influence on the
outlooks of 2006 North Atlantic hurricane season and 2006/07 DJF US
precipitation. The focus is on the NCEP Climate Forecast System (CFS),
though other dynamical and statistical models have similar problem as seen
in the ENSO prediction plume graph produced routinely by the International
Research Institute for Climate and Society (IRI).