Extreme Precipitation in a Changing Climate: Ankara Case Study

Extreme Precipitation in a Changing Climate:

Ankara Case Study

Middle East Technical University

Graduate School of Natural and Applied Sciences

Earth System Science (ESS) Interdisciplinary Program

Sertaç Oruç1, İsmail Yücel 1, 2 and Ayşen Yılmaz1, 3

1 Earth System Science (ESS) Interdisciplinary Program, Middle East Technical University 2 Dept. of Civil Eng., Water Resources Division, Middle East Technical University 3 Institute of Marine Sciences, Middle East Technical University

2

Outline

Motivation

Objectives

Methodology

Observed Precipitation Data Analyses

Projected Precipitation Data Analyses

Summary and Conclusions

3

Motivation: Climate Change Global Precipitation Change (For the last 2 decades and Projected Periods)

Low (RCP 2.6) ensemble average (dark line) and

spread of ensemble members (shaded area). Values

are for the model grid cell containing: 39.912°N

32.84°E

High (RCP 8.5) ensemble average (dark line) and

spread of ensemble members (shaded area). Values

are for the model grid cell containing: 39.912°N

32.84°E.

https://gisclimatechange.ucar.edu/inspector RCP : Representative Concentration Pathways

4

Figure 1.1. Annual count of extreme events in Turkey in the period of 1940-2017 Figure 1.2. Distribution of extreme events and their types in

Turkey in 2017

Motivation: Climate Change and Extreme Events

Annual count of extreme events in Turkey shows an increasing trend in 1940-2017 period (Climate Assessment 2017

Report, February 2018 – State Meteorological Service).

During 2017 most hazardous extreme events were; heavy rain/floods (31%), wind storm (36%), hail (16%), heavy

snow (7%), and lightning (4%)

5

Problem Statement

• To analyze the rainfall extreme value frequencies for stationary and nonstationary conditions in Ankara region,

• To produce Return Levels in stationary and non-stationary conditions with observed data and future projections,.

• To figure out the superiority of nonstationary and stationary models to each other,

• Climate change in Turkey has been evaluated in many different studies with its different aspects. Majority of analysis

performed and the future estimation works were focused on temperature and precipitation changes which are the most

important climate parameters causing the extreme events.

• In the last decades, heavy rainfall and flash flooding caused various damages in Turkey; for example settlements were

damaged, road transportation and vehicles are disrupted, and life was negatively affected in Ankara

Objectives

The methodology of precipitation analysis in this study consists of;

(1) Trend analysis is carried out for observed (1950-2015) and projected data (2015-2098)

(2) Projected data is disaggregated into finer scales (5 min) and then it is aggregated to next analysis time scales (10, 15,

and 30 min, …)

(3) Stationary GEV (St) models are developed, return levels are derived for desired return periods considering single and

multi-time periods for observed and single period for projected data

(4) Non-stationary GEV (NSt) models are developed, return levels are derived for desired return periods for observed and

projected data

(5) Stationary and Non-stationary model results were compared

6

Methodology and Data

• Observed Data for Ankara - 1950-2015 (State Meteorological Services)

• Projected Data; Three global climate models (GCM) are used; namely HadGEM2-ES, MPI-ESM-MR and GFDL-

ESM2M. These models are operated with the RCP 4.5 and RCP 8.5 emission scenarios - 2015-2098 (State

Meteorological Services)

7

Rainfall Data

Observed

Trend Analysis

GEV Models

Stationary

Return Levels for Desired Storm

Durations and Periods

Nonstationary

Mean/Median of Return Levels for Desired Storm Duration

Define Covariates

Obatining Blocks

Projected

Disaggregation

Aggregation Obatining Blocks

GEV Models

Stationary

Return Levels for Desired Storm

Durations and Periods

Nonstationary

Mean/Median of Return Levels for Desired Storm Duration

Define Covariates

Figure 1.3. Rainfall Data Analyses Framework

Methodology

Trends & Change Point

8

Observed Data

Figure 1.4. Sub-Hourly Time Series Trend Figure 1.5. Hourly Time Series Trend

Figure 1.6. Average annual maximum rainfall intensities (mm) for sub-hourly and hourly storm durations

0

5

10

15

20

25

30

35

40

45

501

95

0

195

2

195

4

195

6

195

8

196

0

196

2

196

4

196

6

196

8

197

0

197

2

197

4

197

6

197

8

198

0

198

2

198

4

198

6

198

8

199

0

199

2

199

4

199

6

199

8

200

0

200

2

200

4

200

6

200

8

201

0

201

2

201

4

Pre

cip

ita

tio

n H

eig

ht

(mm

)

Years

5 min 10 min 15 min 30 min

0

10

20

30

40

50

60

70

80

90

195

0

195

2

195

4

195

6

195

8

196

0

196

2

196

4

196

6

196

8

197

0

197

2

197

4

197

6

197

8

198

0

198

2

198

4

198

6

198

8

199

0

199

2

199

4

199

6

199

8

200

0

200

2

200

4

200

6

200

8

201

0

201

2

201

4

Pre

cip

ita

tio

n H

eig

ht

(mm

)

Years

1 hour 2 hours 3 hours 6 hours

0

10

20

30

40

50

60

70

80

90

5 min 10 min 15 min 30 min 1 hour 2 hours 3 hours 6 hours

Pre

cip

ita

tio

n H

eig

ht

mm

Storm Duration

Average 1950-1975

Maximum 1950-1975

Average 1976-2015

Maximum 1976-2015

Average 1950-2015

Maximum 1950-2015

9

Model Location Scale Shape

NStGEV1 𝜇t =𝛽0 +𝛽1t 𝜎 (constant) 𝜉 (constant)

NStGEV2 𝜇t =𝛽0 +𝛽1t 𝜎t =𝛽0 +𝛽1t 𝜉 (constant)

NStGEV3 𝜇 (constant) 𝜎t =𝛽0 +𝛽1t 𝜉 (constant)

NStGEV4 𝜇t =𝛽0 +𝛽1temperature 𝜎 (constant) 𝜉 (constant)

NStGEV5 𝜇t =𝛽0 +𝛽1t 𝜎t =𝛽0 +𝛽1exp(temperature) 𝜉 (constant)

NStGEV6 𝜇t =𝛽0 +𝛽1exp(temperature) 𝜎t =𝛽0 +𝛽1exp(temperature) 𝜉 (constant)

NStGEV7 𝜇t =𝛽0 +𝛽1exp(temperature) 𝜎t = (constant) 𝜉 (constant)

NStGEV8 𝜇 (constant) 𝜎t =𝛽0 +𝛽1temperature 𝜉 (constant)

Table 1.1. Non-stationary models with time and covariate (temperature)

dependent location and scale parameters

Stationary Models (St)

5 Minutes 10

Minutes 15

Minutes 30

Minutes 1 Hour 2 Hours 3 Hours 6 Hours

NonStationary Models (NSt)

5 Minutes 10

Minutes 15

Minutes 30

Minutes 1 Hour 2 Hours 3 Hours 6 Hours

Figure 1.7. Storm Durations Used for Stationary Models

2-year 5-year 10-year 25-year 50-year 100-year 200-year

Mean Value Change

FiveMin -4% -4% -5% -9% -11% -15% -18%

TenMin -14% -13% -12% -9% -7% -5% -3%

FifteenMin -1% -4% -6% -9% -12% -14% -17%

ThirtyMin 0% -3% -6% -10% -13% -16% -19%

OneHour -7% -5% -3% 0% 3% 6% 10%

TwoHours 0% -3% -4% -5% -6% -7% -7%

ThreeHours 0% -3% -5% -8% -11% -13% -16%

SixHours 1% -1% -2% -3% -4% -5% -5%

Median Value Change

FiveMin -3% -2% -4% -7% -9% -12% -15%

TenMin -13% -12% -10% -7% -4% -2% 0%

FifteenMin -1% -2% -4% -7% -9% -12% -14%

ThirtyMin 1% -2% -5% -8% -10% -13% -16%

OneHour -8% -5% -2% 2% 5% 8% 12%

TwoHours -1% -2% -3% -4% -4% -5% -5%

ThreeHours -1% -2% -4% -7% -9% -11% -14%

SixHours 0% -1% -2% -2% -3% -3% -4%

Table 1.2. Nonstationary GEV Best Fit Model Return Levels (mm) - Mean and

Median Value Change with Respect to Stationary GEV Model

Observed Data

10

Observed Data

• The shorter the storm duration the larger the differences between the non-stationary and stationary extremes.

• Among the storm durations, only one hour time series exhibit larger values for its nonstationary model return level

values, however this is not valid for shorter return periods such as 5 years or 20 years

• Sub-hourly storm durations indicate larger difference than hourly storm durations and non-stationary estimates are

smaller than their corresponding stationary values

Figure 1.8. Stationary and Best Fit Nonstationary Model Return Level (mm) Comparison - Return Period vs. Return Level

0

5

10

15

20

2 5 10 20 25 50 100 200

Ret

urn

Lev

el m

m

Return Period (Years)

StFiveMin

NStFiveMin

0

5

10

15

20

25

30

2 5 10 20 25 50 100 200

Ret

un

Lev

el m

m


StTenMin

NStTenMin

0

10

20

30

40

2 5 10 20 25 50 100 200

Ret

urn

Lev

el m

m


StFifteenMin

NStFifteenMin

0

10

20

30

40

50

60

70

2 5 10 20 25 50 100 200

Ret

urn

Lev

el m

m


StThirtyMin

NStThirtyMin

0

20

40

60

80

100

2 5 10 20 25 50 100 200

Ret

urn

Lev

el m

m


StOneHour

NStOneHour

0

20

40

60

80

100

2 5 10 20 25 50 100 200

Ret

urn

Lev

el m

m


StTwoHours

NStTwoHours

0

20

40

60

80

100

2 5 10 20 25 50 100 200

Ret

urn

Lev

el m

m


StThreeHours

NStThreeHours

0

20

40

60

80

100

2 5 10 20 25 50 100 200

Ret

urn

Lev

el m

m


StSixHours

NStSixHours

11

(Non) Stationary ((N)St)

10 Minutes

MPI-ESM-MR (MPI)

RCP 4.5

RCP 8.5

HADGEM2-ES (HG)

RCP 4.5

RCP 8.5

GFDL-ESM2M (GFDL)

RCP 4.5

RCP 8.5

15 Minutes

MPI

RCP 4.5

RCP 8.5

HG

RCP 4.5

RCP 8.5

GFDL

RCP 4.5

RCP 8.5

1 Hour

MPI

RCP 4.5

RCP 8.5

HG

RCP 4.5

RCP 8.5

GFDL

RCP 4.5

RCP 8.5

6 Hours

MPI

RCP 4.5

RCP 8.5

HG

RCP 4.5

RCP 8.5

GFDL

RCP 4.5

RCP 8.5

Figure 1.9. Projected Storm Durations Used for Stationary Models for 2015-2098 period

Projected Data

12

Projected Data: Trends

Figure 1.10. Projected 10-15 Minutes (a,b) and 1-6 Hours (c,d) Annual Maximum Time Series for 2015-2098

(a) (b)

(c) (d)

0

2

4

6

8

10

12

14

16

201

5

201

8

202

1

202

4

202

7

203

0

203

3

203

6

203

9

204

2

204

5

204

8

205

1

205

4

205

7

206

0

206

3

206

6

206

9

207

2

207

5

207

8

208

1

208

4

208

7

209

0

209

3

209

6

Hei

gh

t m

m

MPI 4.5 MPI 8.5 GFDL 4.5 GFDL 8.5 HG 4.5 HG 8.5

0

5

10

15

20

25

201

5

201

8

202

1

202

4

202

7

203

0

203

3

203

6

203

9

204

2

204

5

204

8

205

1

205

4

205

7

206

0

206

3

206

6

206

9

207

2

207

5

207

8

208

1

208

4

208

7

209

0

209

3

209

6

Hei

hg

t m

m

MPI 4.5 MPI 8.5 GFDL 4.5 GFDL 8.5 HG10 4.5 HG10 8.5

0

5

10

15

20

25

30

35

40

45

50

201

5

201

8

202

1

202

4

202

7

203

0

203

3

203

6

203

9

204

2

204

5

204

8

205

1

205

4

205

7

206

0

206

3

206

6

206

9

207

2

207

5

207

8

208

1

208

4

208

7

209

0

209

3

209

6

Hei

gh

t m

m


0

10

20

30

40

50

60

70

80

201

5

201

8

202

1

202

4

202

7

203

0

203

3

203

6

203

9

204

2

204

5

204

8

205

1

205

4

205

7

206

0

206

3

206

6

206

9

207

2

207

5

207

8

208

1

208

4

208

7

209

0

209

3

209

6

Hei

gh

t m

m


13

Return Period

-Years 2 5 10 25 50 100 200 2 5 10 25 50 100 200

Model Mean Value Change Median Value Change

MPI45 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%

MPI85 0% -1% -1% -2% -2% -2% -2% 1% 0% -1% -1% -2% -2% -2%

GFDL45 -1% -1% 0% 0% 1% 1% 2% 1% 1% 1% 1% 2% 2% 3%

GFDL85 1% 0% 0% -1% -1% -2% -2% 2% 2% 2% 2% 2% 2% 1%

HG45 0% -1% 0% 1% 2% 2% 4% 0% -1% 0% 1% 2% 2% 4%

HG85 0% 0% 0% 0% 0% 0% 1% -1% -1% -1% -1% 0% 0% 0%

Mean Value Change Median Value Change

MPI45 0% 0% 0% 0% 1% 2% 3% 0% 0% 0% 1% 1% 2% 3%

MPI85 0% -1% -1% -2% -2% -2% -3% 1% 0% -1% -1% -2% -2% -2%

GFDL45 -1% -1% -1% 0% 0% 0% 1% -1% -1% -1% 0% 0% 0% 1%

GFDL85 0% 0% -1% -1% -1% -1% -1% 0% 0% 0% 0% -1% -1% -1%

HG45 -1% -1% 0% 2% 3% 5% 6% -1% -1% 0% 2% 3% 5% 6%

HG85 0% 0% 0% 0% 0% 0% 1% -2% -1% -1% -1% -1% 0% 0%


MPI45 -1% -1% -1% -1% -2% -2% -2% -1% -1% -1% -1% -2% -2% -2%

MPI85 0% -2% -2% -1% -1% 0% 1% 1% -1% -2% -1% 0% 1% 2%

GFDL45 -1% 0% 0% 1% 2% 3% 4% -1% 0% 0% 1% 2% 3% 4%

GFDL85 0% 0% -1% -2% -3% -3% -4% 2% 2% 1% 1% 0% 0% -1%

HG45 0% -1% -1% 0% 0% 0% 1% 0% -1% -1% 0% 0% 0% 1%

HG85 0% -1% -1% -1% -1% -1% -1% -3% -3% -2% -2% -2% -2% -2%


MPI45 0% -1% -1% -2% -2% -3% -4% 0% -1% -1% -2% -2% -3% -4%

MPI85 2% 2% 1% 0% 0% -1% -2% 2% 3% 3% 3% 3% 2% 2%

GFDL45 1% 0% -2% -5% -7% -10% -13% 1% -1% -3% -6% -9% -12% -15%

GFDL85 0% -2% -3% -5% -7% -9% -12% 1% 0% -1% -3% -4% -5% -7%

HG45 0% -1% -3% -5% -7% -9% -11% 2% 2% 1% -1% -2% -4% -6%

HG85 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 1%

Table 1.3. Nonstationary Mean and Median Value Change with Respect to Stationary Model - Projected Data

Ten

Minutes

Fifteen

Minutes

One

Hour Six Hours

2-year 1% -5% 0% 1%

5-year 0% -7% -1% 0%

10-year -1% -9% -1% -1%

25-year -2% -10% -1% -2%

50-year -3% -11% 0% -4%

100-year -3% -12% 0% -5%

200-year -4% -12% 1% -6%

Table 1.4. Nonstationary Model-Stationary

Comparison for Projected Data - Average

Values

Projected Data

14 Figure 1.11. Stationary Model Results for Projected Time Series

Projected Data

0

5

10

15

20

25

30

Ret

urn

Lev

el m

m

Return Period

StTenMinMPI45

StTenMinMPI85

StTenMinGFDL45

StTenMinGFDL85

StTenMinHG45

StTenMinHG85

Average

Max

0

5

10

15

20

25

30

35

Ret

urn

Lev

el m

m

Return Period

StFifteenMinMPI45

StFifteenMinMPI85

StFifteenMinGFDL45

StFifteenMinGFDL85

StFifteenMinHG45

StFifteenMinHG85

Average

Max

05

101520253035404550

Ret

urn

Lev

el m

m

Return Period

StOneHourMPI45

StOneHourMPI85

StOneHourGFDL45

StOneHourGFDL85

StOneHourHG45

StOneHourHG85

Average

Max

0102030405060708090

100

Ret

urn

Lev

el m

m

Return Period

StSixHoursMPI45

StSixHoursMPI85

StSixHoursGFDL45

StSixHoursGFDL85

StSixHoursHG45

StSixHoursHG85

Average

Max

15

On average nonstationary models produce mostly lower return levels for mid and longer return periods for all durations and

similar results for short (2 and 5 years) return periods except one hour storm duration.

Projected Data

Figure 1.12. 10-15 Minutes and 1-6 Hours Ensemble Model Comparison for Projected Data

0

5

10

15

20

25

30

2-year 5-year 10-year 20-year 25-year 50-year 100-year200-year

Ret

urn

Lev

el m

m

NStTenMinAvrg

NStTenMinMax

StTenMinAvrg

StTenMinMax

0

5

10

15

20

25

30

35

40

Ret

urn

Lev

el m

m

NStFifteenMinAvrg

NStFifteenMinMax

StFifteenMinAvrg

StFifteenMinMax

0

5

10

15

20

25

30

35

40

45

50

Ret

urn

Lev

el m

m

NStOneHourAvrg

NStOneHourMax

StOneHourAvrg

StOneHourMax

0

20

40

60

80

100

120

Ret

urn

Lev

el m

m

NStSixHoursAvrg

NStSixHoursMax

StSixHoursAvrg

StSixHoursMax

16

St

St

St

St

St

St

St

SMS

SMS

SMS

SMS

SMS

SMS

SMS

NStMean

NStMean

NStMean

NStMean

NStMean

NStMean

NStMean

NstMedian

NstMedian

NstMedian

NstMedian

NstMedian

NstMedian

NstMedian

-

5,00

10,00

15,00

20,00

25,00

30,00


St SMS NStMean NstMedian StMPI45 StMPI85 StGFDL45 StGFDL85

StHG45 StHG85 NStMPI45Mean NStMPI45Median NStMPI85Mean NStMPI85Median NStGFDL45Mean NStGFDL45Median

NStGFDL85Mean NStGFDL85Median NStHG45Mean NStHG45Median NStHG85Mean NStHG85Median StProjectedAvg NStProjectedAvg

Projected Data

Figure 1.15. Ten Minutes Data Model Comparison - Best Fit Nst and St for Observed and Projected Data and SMS (State Meteorological Service) Data

17

Projected Data

Figure 1.16. Fifteen Minutes Data Model Comparison - Best Fit Nst and St for Observed and Projected Data and SMS (State Meteorological Service) Data

St

St

St

St

St

St

St

SMS

SMS

SMS

SMS

SMS

SMS

SMS

NStMean

NStMean

NStMean

NStMean

NStMean

NStMean

NStMean

NstMedian

NstMedian

NstMedian

NstMedian

NstMedian

NstMedian

NstMedian

-

5,00

10,00

15,00

20,00

25,00

30,00

35,00

40,00





18

St

St

St

St

St

St

St

SMS

SMS

SMS

SMS

SMS

SMS

SMS

NStMean

NStMean

NStMean

NStMean

NStMean

NStMean

NStMean

NstMedian

NstMedian

NstMedian

NstMedian

NstMedian

NstMedian

NstMedian

-

10,00

20,00

30,00

40,00

50,00

60,00

70,00

80,00

90,00





Projected Data

Figure 1.17. One Hour Data Model Comparison - Best Fit Nst and St for Observed and Projected Data and SMS (State Meteorological Service) Data

19

St

St

St

St

St

St

St

SMS

SMS

SMS

SMS

SMS

SMS

SMS

StMPI45

StMPI45

StMPI45

StMPI45

StMPI45

StMPI45

StMPI45

NStMPI45Mean

NStMPI45Mean

NStMPI45Mean

NStMPI45Mean

NStMPI45Mean

NStMPI45Mean

NStMPI45Mean

NStMPI45Median

NStMPI45Median

NStMPI45Median

NStMPI45Median

NStMPI45Median

NStMPI45Median

NStMPI45Median

-

10,00

20,00

30,00

40,00

50,00

60,00

70,00

80,00

90,00

100,00





Projected Data

Figure 1.18. Six Hours Data Model Comparison - Best Fit Nst and St for Observed and Projected Data and SMS (State Meteorological Service) Data

20

Summary and Conclusions:

• Stationary GEV models were capable of fitting extreme rainfall data for all durations but the developed non-stationary

GEV models showed advantage over the stationary models

• The differences in design rainfall estimates between two time slice, entire period and nonstationary assumption models

support the need to update the current information, with the most recent data and approaches.

• The differences also reveal the need to conduct analysis using future climate data.

• Nonstationary model results are in general exhibited smaller return level values with respect to stationary model results of

each storm duration for the observed data driven model results.

• On average nonstationary models produce mostly lower return levels for mid and longer return periods for all durations

and similar results for short (2 and 5 years) return periods except one hour storm duration for the projected data.

• Almost all the nonstationary model maximum return level results are significantly higher than stationary model maximum

return level results for all storm durations and return periods for the projected data driven model results.

21

References 1. Cheng, L. (2014). Frameworks for univariate and multivariate non-stationary analysis of climatic extremes, PhD. Dissertation, UC Irvine.

2. Cheng, L., & AghaKouchak, A. (2014). Nonstationary precipitation intensity-duration-frequency curves for infrastructure design in a changing climate. Sci.

Rep. 4, 7093. doi:10.1038/srep07093

3. Coles, S. (2001). An Introduction to Statistical Modeling of Extreme Values, Springer, London.

4. Coles, S. G., & Sparks, R. S. J. (2006). Extreme value methods for modelling historical series of large volcanic magnitudes. Chapter 5, Statistics in

Volcanology.

5. Efstratiadis, A., and D. Koutsoyiannis, An evolutionary annealing-simplex algorithm for global optimisation of water resource systems, (2002). Proceedings

of the Fifth International Conference on Hydroinformatics, Cardiff, UK, 1423-1428, International Water Association, (http: //itia.ntua.gr/el/docinfo/524/)

6. Gilleland, E., & Katz, R. (2016). extRemes 2.0: An Extreme Value Analysis Package in R. Journal of Statistical Software, 72(8), 1-39. doi:

10.18637/jss.v072.i08

7. Gilleland, E. (2016). Extreme Value Analysis, Package ‘extRemes’

8. Kossieris, P., Makropoulos, C., Onof, C., & Koutsoyiannis, D. (2016a). A rainfall disaggregation scheme for sub-hourly time scales: Coupling a Bartlett-

Lewis based model with adjusting procedures, Journal of Hydrology, 556, 980-992.

9. Kossieris, P., Makropoulos, C., Onof, C., & Koutsoyiannis, D. (2016b). HyetosMinute, A package for temporal stochastic simulation of rainfall at fine time

scales, Version 2.0.

10. SMS, (2018). Republic of Turkey, the ministry of forestry and water affairs, state meteorological service, 2017 Annual Climate Assessment Report, February

2018

Extreme Precipitation in a Changing Climate: Ankara Case Study

Documents

Transcript of Extreme Precipitation in a Changing Climate: Ankara Case Study