Towards a Subseasonal Excessive Heat Outlook System

Towards a subseasonal excessive heat
outlook system
Augustin Vintzileos
University of Maryland ESSIC/CICS-MD
and
Jon Gottschalck and Mike Halpert
NOAA/CPC
Why:
• As NOAA's administrator mentioned at the 2015 AMS
general assembly: “…our emergency management
partners tell us that 8-14 day advanced warning of
extreme heat predictions would improve public
preparedness significantly; today’s outlooks are 6-10
days"
• It is expected (IPCC) that heat waves will increase in
frequency, duration and amplitude
• Pushing forecasts to the limits of predictability will
facilitate resilience
Outline:
• Brief review of sources of predictability at Week-2 and beyond.
• Introduce a quantitative definition of heatwaves which takes into
account the challenges of subseasonal forecasting.
• Present a monitoring – verification system based on this metric.
• Discuss the design of the forecasting system.
• Summary and work to follow.
Sources of predictability
Week-1
Atmospheric Initialization,
deterministic forecast of
weather
Week-2
Week-3
Modulation of the statistics of weather by
slow atmospheric modes (e.g. MJO) and
coupling to the ocean and land. Initialization
of the atmospheric, oceanic and land
components. Realistic simulation of slow
atmospheric modes.
Model forecast skill at Week-2
Anomaly correlation between in situ temperature observations and CFS forecasts
Definition of a subseasonal Heat Wave Index
Blending:
Ingredients of a heatwave:
(a) Impacts of heat increase non-linearly
with increasing temperature (and
with humidity when relevant)
Challenges of subseasonal forecasting:
(a) Forecast models present systematic
biases that depend on forecast lead
time (drift).
(b) Duration and timing of a heat event
is very important
(b) Forecast skill is not the same for the
different ingredients of a heatwave
(temperature, humidity etc).
(c) Solar radiation and wind
(c) Subseasonal forecasts are probabilistic.
Definition of a subseasonal Heat Wave Index
Let 𝑨𝑨𝒊𝒏 be the maximum or mean apparent temperature for day n on grid point i – in this work we use the NOAA
Heat Index to define apparent temperature.
𝒊
and let
Let 𝚪𝐧𝐢 be the percentile of 𝑨𝑨𝒊𝒏 defined by the historical record of 𝑨𝑨𝒏=𝟏,…,𝑵
Let 𝒑𝒊𝒏 be the probability of occurrence of 𝑨𝑨𝒊𝒏 . Then if α is a user defined threshold:
n = -N
…
𝒊
<𝜶
Γ−𝟑
Normal day
n = -3
𝒊
≥𝜶
Γ−𝟐
Heat day
𝒊
≥𝜶
Γ−𝟏
Heat day
Γ𝟎𝒊 ≥ 𝜶
Heat day
n = -2
n = -1
Day n = 0
Heat event
The definition of the Heat Wave Index for day n = 0 is Κ 𝑖0 = −𝑙𝑙𝑙 𝒑𝒊−𝟐 ∙ 𝒑𝒊−𝟏 ∙ 𝒑𝒊𝟎
Definition of a subseasonal Heat Wave Index
𝑀
Κ 𝑖𝑛 𝛼, 𝑀𝑚𝑚𝑚 = −log � 𝑝𝑑𝑖
𝑑=1
With 𝑴𝒎𝒎𝒎 ≤ 𝑴 the minimum amount of consecutive days that the apparent temperature must exceed the
𝜶 percentile. The highest the value of the K-index the rarest the heat wave is and thus the more intense.
Advantages:
• This index is based on the historical record of apparent temperatures for a geographical location. Additionally,
historical data from only a short window around a given day can be used for calculating the cumulative
distribution function. Therefore geographical and seasonal dependence is assured.
• Model biases and drifts are automatically taken into consideration by computing cumulative distribution
functions as a function of forecast lead time.
• Results from multiple 𝜶 values can be presented on the same plot.
Disadvantage:
• Requires historical data and reforecasts of adequate length
An excessive heat monitoring system based on the subseasonal Heat Wave Index
In the following examples:
𝑴𝒎𝒎𝒎 = 𝟐 i.e., apparent temperatures must be above the alpha threshold for at least 2 consecutive days
𝜶 ∈ 𝟖𝟖𝟖 𝟗𝟗𝟗 𝟗𝟗𝟗 stacking the three different levels of heatwave severity
Heat wave Index color-map for each severity class
85%
90%
95%
Monitoring Heat Waves based on
NCAR/NCEP Reanalysis
• Each box represents a grid box of
the reanalysis
• Contour colors are constant for
each class
• Interior colors range according to
the intensity of the heat wave for
each class with darker colors
indicated more intense heat
waves
Fine tuning the monitoring system: Heat and abnormal mortality
Mortality data (courtesy S.
Sheridan) were obtained
from the National Center
for Health Statistics for the
period 1975-2010. These
data were summed to
provide a daily total of allcause mortality across each
of the metropolitan areas
(as defined by the US
Census) studied. Data were
partitioned to examine two
groups separately, all age
mortality and those 65 and
older. In this work we use
the later category.
Gaps in mortality
data from Texas.
Fine tuning the monitoring system: Heat and abnormal mortality
Locations with abnormal mortality > 99% percentile
Design of the forecast system (stage 1)
Global Ensemble Forecast System (GEFS)
• 81 ensemble members per day up to
Week-2 (plans Week-3 and 4).
Pre-processing
(bias corrections)
• There is a GEFS reforecast initialized
daily from 1985 to 2014
Verification and
monitoring
system
Subseasonal
Excessive Heat
Outlook System
Design of the forecast system (stage 2)
NAVGEM
CFS
Canadian
JMA
Bias correction
and optimal
consolidation
Subseasonal
Excessive Heat
Outlook System
ECMWF
GEFS
Verification
Summary:
• We introduced a measure for quantifying heatwaves that is adaptable to the
challenges of probabilistic forecasting.
• We are developing and refining the real time monitoring system (to be used for
verification). Testing higher resolution Reanalysis products.
• We are exploring the GEFS reforecast database in order to evaluate its forecast
skill and develop bias correction methods.
Work to follow:
• Developing the experimental realtime outlook and verification system.
• Gradually introducing more models into the forecasting component.
• Evaluating the possibility to extend this product to Week-3 and Week-4.
Questions?
[email protected]
Summer 1995
2003
2003
2003
la journée du 6
juillet est la plus
chaude de l’été
avec 35° à Bourg
St Maurice, 36° à
Clermont Ferrand
et 37°à Paris ainsi
qu’à Reims.
Tropical to extra-tropical interactions
Winter
Hemisphere
Summer
Hemisphere
(from Jin and Hoskins, 1995).
The Equatorial Path
Gill – Matsuno solutions
Distribution of the phase of the MJO for all days
Frequency
of
occurrence
Phase of the RMM index
Distribution of the phase of the MJO for all days satisfying a given criterion
100%
No relation
16.6%
Deterministic
relation
Frequency
of
occurrence
Phase of the RMM index
Phase of the RMM index
Mortality spike > 2xσ AND Apparent temperature percentile > 80%