Transcript Statistical modelling of thunderstorms in the present and future climate

Statistical modelling of thunderstorms in the
present and future climates
2nd Workshop on Severe Convection and Climate
March 9-10, 2016, Columbia University, New York, NY
Anja Westermayer
Tomas Pucik
Pieter Groenemeijer
Eberhard Faust
Robert Sausen
Motivation
Goals:
 How well can we model thunderstorm
climatology in Europe using
reanalysis data?
 What changes can we expect in
future climates?
Lightning- and Reanalysis Data
 cloud-to-ground (CG) lightning data
(comparable to NLDN)
 resolution: ∆𝑡 = 1ℎ, ∆𝑥 = ∆𝑦 = 0.25°
Era-Interim global atmospheric reanalysis data**
 resolution: ∆𝑡 = 6ℎ, ∆𝑥 = ∆𝑦 = 0.75°, 𝑧 = 37 𝑙𝑒𝑣𝑒𝑙𝑠
Instability-, moisture- and shear related parameters
 for the time period 2008-2013
Thunderstorm definition:
 at least two CG detections within a gridbox
* DIENDORFER, G., et al., 2010: EUCLID – State of the Art Lightning Detection. 1–21.
**DEE, D., et al. , 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. – Q. J. R. Meteorol. Soc. 137(656), 553–597
Lighting cases between 2008-2013
European Cooperation of Lightning Detection – EUCLID*
Probability of lightning in CAPE – RH parameter space
Dry midlevel air strongly suppresses CI
observations
linear
logistic
regression
relative frequency of lightning 𝑓𝐿
4
Probability of lightning in CAPE – RH parameter space
observations
additive
logistic
regression (2D)
relative frequency of lightning 𝑓𝐿
5
Lightning observations (EUCLID)
2008-2013
south
north
all
north
• Annual cycle of thunderstorms
• July is peak season
• Hotspot in northeast Italy
south
Additive logistic regression (3D: LI, RH and DLS)
ERA-Interim 2008-2013
south
north
• Annual cycle fairly well reproduced
• Overestimation of thunderstorms in winter
• Hotspot in northeast Italy
north
south
Additive logistic regression (3D: LI, RH and DLS)
ERA-Interim 1979-2013
south
north
north
south
• Discontinuity around 2002 for south domain
• Hotspot in northeast Italy
EuroCORDEX regional climate simulations
High-resolution regional climate change ensemble for Europe within the World Climate Research Program
Coordinated Regional Downscaling Experiment (EURO-CORDEX) initiative*.
 resolution: ∆𝑡 = 6ℎ, ∆𝑥 = ∆𝑦 = 0.44° (approx. 50 km)
 periods: 1971 – 2000, 2071 – 2100
 Greenhouse gas emission scenarios (RCPs):
 RCP4.5: radiative forcing after the 21st century at 4.5 W/m²
 RCP8.5: radiative forcing > 8.5 W/m² at the end of 21st century
 Preliminary results from the first model SMHI MPI-ESM-LR (GCM)
Acknowledgment:
Rowan Fealy (NUIM), Claas Teichman (HZG), Erik van Meijgaard (KNMI),
Grigory Nikulin (SMHI), Andreas Prein (WEGC)
* Jacob, D., et al. "EURO-CORDEX: new high-resolution climate change projections for European impact research." Regional Environmental Change 14.2 (2014): 563-578.
Additive logistic regression (3D: LI, RH and DLS)
EuroCORDEX smhi MPI-ESM-LR 2071 – 2100
Distribution of historical thunderstorms
modelled average of thundery
6h periods per year
Change in thunderstorm probability for rcp8.5
1971 - 2000
Underestimation of
instability on North Italy
Differences between 2071-2100 to 1971-2000
EuroCORDEX SMHI MPI-ESM-LR
differences between 2071-2100 to 1971-2000
Change in 1st percentile of LI
rcp 4.5
rcp 8.5
(K)
EuroCORDEX SMHI MPI-ESM-LR
differences between 2071-2100 to 1971-2000
Change in 1st percentile of RH
rcp 4.5
rcp 8.5
(%)
EuroCORDEX SMHI MPI-ESM-LR
differences between 2071-2100 to 1971-2000
Change in 1st percentile of DLS
rcp 4.5
rcp 8.5
(m/s)
Conclusions
How well can we model thunderstorm climatology in Europe using
reanalysis data?
Fairly well with an additive logistic regression (3D: LI, RH and DLS)
with some limitations in winter
What changes can we expect in future climates?
small increases of the number of thunderstorms in most of Central Europe
small decrease in the number of thunderstorms in the South of Europe
Thank you for your attention!
Changes in DLS small throughout the
domain
Decrease in relative humidity
dominating
change in percent
for rcp8.5 2071 – 2100
Conclusions
EuroCORDEX ensemble (14 members)
differences between 2071-2100 to 1971-2000
Change in 1st percentile of LI
rcp 4.5
rcp 8.5
(m/s)
EuroCORDEX ensemble (14 members)
differences between 2071-2100 to 1971-2000
Change in 1st percentile of RH
rcp 4.5
rcp 8.5
(%)
EuroCORDEX ensemble (14 members)
differences between 2071-2100 to 1971-2000
Change in 1st percentile of DLS
rcp 4.5
rcp 8.5
(m/s)
Additive logistic regression (3D: LI, RH and DLS)
EuroCORDEX smhi MPI-ESM-LR 2071 – 2100
Distribution of historical thunderstorms
modelled
average
ofofthundery
modelled areal
average
thundery
periods per
6h6hperiods
peryear
year
Change in thunderstorm probability for rcp4.5
1971 - 2000
Underestimation of
instability on North Italy
Differences between 2071-2100 to 1971-2000
Generalized linear models
(GLM)
Generalized additive models
(GAM)
𝑌 = 𝛽0 + 𝛽1 ∗ 𝑋1 + 𝛽2 ∗ 𝑋2 … + ϵ
𝑌 = 𝒔(𝑋1 , 𝑋2 , … ) + ϵ
constant
response variable
coefficient
error
explanatory or
predictor variables
 Term „linear“ refers to the combination of parameters,
not the shape of the distribution
𝜇
)
1−𝜇
smoothing function
•
Y and X don‘t have to be linearly related
•
No theory or mechanistic model assumption needed
•
Y and X are not linked by coefficient but by a
smoothing function
•
Smoothing function is estimated from available data
 Logit link: 𝑔 𝜇 = log(
•
•
models log of odds of one outcome
(e.g. presence)
Linear Logistic Regression
𝜇
• Logit link: 𝑔 𝜇 = log(1−𝜇)
• Additive logistic regression
Atmospheric parameter:
CAPE, CIN, relative humidity and deep-layer bulk shear
relative
humitity
deep-layer
bulk shear
DLS