Downscaling Global Climate Model Outputs to Fine Scales over Sri Lanka for Assessing Drought

Impacts

Sanjaya RatnayakeFoundation for Environment Climate and Technology

Sri Lanka

Overview

• Introduction• Problem Definition• Aims and Objectives• My approach• Data and Study Area• Methodology• Results• My Prediction• Conclusion• Acknowledgement

Introduction

• Due to the geographical distribution the island is affected with different kinds of natural disasters.

• The most frequent natural hazards that affect the island are drought, floods, landslides, cyclones and coastal erosion.

• Among all natural disasters, droughts occur most frequently, have the longest duration, and cover the largest area.

Problem Definition

• So far there is no proper drought index over Sri Lanka.

• So far there is no proper drought characteristics predicting mechanism.

• So far drought relief payments made in districts across the country by the government over a 40 years period used as a proxy for drought risk.

Aims and Objectives

• Asses the APCC CLIK tool skill over Sri Lanka• Observer station correlations with NINO

regions• Identify most appropriate downscale

parameters for individual station for different seasons

• Develop drought index over Sri Lanka• Predict Seasonal Droughts

My Approach

• Calculate SPI over Sri Lanka• Obtain downscale results using SPI• Observe skill of forecasted downscale SPI

results• Amplitude the results using variance inflation

methods• Predict SPI using most appropriate method.

Data and Study Area

• 132 station monthly precipitation data covering entire island.

• 26 stations with missing and corrupted data.• SPI 30 years (24 years here)• CLIK most models support : 1984 to 2008

• Finally, 34 stations covering southern and central parts of the Island.

(a)

(b)

(c)

Methodology

• Standard Precipitation Index (SPI)• Climate Information tool Kit 2.0• Preliminary Study• Detailed Downscaling• Variance Inflation• Heidke Skill Score

Meteorological drought index • Standard Precipitation Index requires only

precipitation as input (McKee, et al., 1993)• SPI scale 1, 3, 6 and 12– Appropriate for short term drought studies – Appropriate for seasonal drought studies – 3 months droughts have a drastic impact on the

agriculture – Highly successful with drought data

SPI ClassificationSPI Drought classification

>2.0 Extreme wet1.5 to 1.99 Very wet1.0 to 1.49 Moderately wet

-0.99 to 0.99 Near Normal-1.0 to -1.49 Moderately Dry-1.5 to -1.99 Severe Dry

<-2.0 Extremely Dry

SPI Drought category

0 to -0.99 Mild drought-1.0 to -1.49 Moderately drought-1.5 to -1.99 Severe drought

<-2.0 Extremely drought

Climate Information tool Kit 2.0 (CLIK)

• By APEC Climate Center (APCC) • Statistical Downscaling• Linear Regression• 16 GCM models• 10 predictor variables• 3 months seasonal predictions• Screening Process– First Screening Process (FSC) : Hopeful stations– Second Screening Process (SCP) : Good stations

Preliminary Study

Predictor• Training Period: 1984-2008

• Variables: SLP, SST, U850

• Models: NCEP, NASA, MSC_CANCM4, HMC, CWB, IRIF, GDAPS_F, COLA, MSC_CANCM3, MGO, POAMA, PNU and BCC

• Region : NINO1+2, NINO3, NONO4, NINO3.4

Preliminary Study

Predictand• Season: JFM, AMJ, JAS, OND

• Variables: Precipitation

• Statistical Downscaling: Linear Regression

• Significant Level: 5%

• Minimum pattern score: 0.3

Detailed Downscaling StudySeason JFM AMJ JAS OND

Predictor Model NCEP NCEP NCEP MGO

Predictor Region NINO3 NINO3 NINO3.4 NINO3

Predictor Variable SLP SLP SST U850

No of Stations 34

Predictand Variable SPI

Training Period 1984-2008

Statistical Downscaling Linear Regression

Significant Level 5%

Mini. Pattern score 0.3

Variance Inflation

• Improver magnitude of downscaled forecasted results

Heidke Skill Score

Observed ForecastYes No

Yes A BNo C D

𝐻𝑆𝑆=2 (𝑎𝑑−𝑏𝑐 )

[ (𝑎+𝑐 ) (𝑐+𝑑 )− (𝑎+𝑏) (𝑏+𝑑 ) ]

• For Mild, Moderate, Severe and Extreme droughts HSS was calculated separately.

• JFM and OND seasons were considered separately.

HSS 1: Perfect forecast HSS 0: No skill HSS < 0: Chance forecast is better

Results

• Preliminary Study Results• Downscale results• HSS Results comparison for– Downscale results– VIF1 Results– VIF2 Results

Preliminary Study Results

Preliminary Study Results

• OND showed the best correlations fallowed by JFM

• NINO3 showed higher correlation• JFM : NCEP model with SLP variable• OND : MGO models with U850 variable• Low results for AMJ and JAS

Detailed Downscaling resultsSeason JFM AMJ JAS OND

Only FSP 3 - 2 8

SCP 15 1 7 18

Highest Correlation 0.52 0.26 0.59 0.7

Corr > 0.6 - - - 3

Corr > 0.3 9 - 5 13

JFM AMJ

JAS OND

-WR - -WR - -WR - -WR WR WR WR

JFM

AMJ

JAS

OND

Variance Inflation

• Mild drought– Almost Downscale and VIF2 similar results– VIF2 slight improvement in OND

• Moderate drought : Good improvement for VIF2

(specially in OND)• Severe Droughts : Good improvement from VIF1

• Extreme Droughts : Longer period data are required

Mild Drought

Downscale VIF1 VIF2

JFM

Downscale VIF1 VIF2

OND

Moderate Drought

Downscale VIF1 VIF2 JFM

Downscale VIF1 VIF2 OND

My Prediction for 2012 OND

Conclusion

• Overall CLIK is skillful over Sri Lanka• Variance Inflation methods are important

when drought magnitude increases• OND was high skill season• Central High hills showed higher correlation• NINO3 showed good correlation over all

seasons except AMJ

How to use my study results

• Even though SPI is not a drought predicting tool using downscaled SPI it was possible to forecast drought in Sri Lanka.

• Predicted drought characteristics for better water resource planning in order to reduce agriculture risk

• Use my drought index to proper drought payment relief payment

Acknowledgement• Funders

– Asia-Pacific Economic Corporation (APEC) Climate Center (APCC)

• Supervisors– Dr. Soo-Jin Sohn, Team Leader(Climate Prediction Team), APCC– Dr. Lareef Zubair, Principal Scientist, Foundation for Environment Climate and Technology, Sri Lanka

• Data Providers– Precipitation data : Ministry of Irrigation and Water Resources Management, Sri Lanka– GCM model data providers

• Individuals– Mr. Wimal Ratnayake (Irri Mini) : Station details– Ms. Hye-Jin Park (APCC) : CLIK– Dr. Prasanna Venkatraman (APCC) : ENSO– Ms. Ruvindee Rupasinghe (Uni Peradeniya) : English and Report corrections – Ms. Sooyang Joo (APCC) : Operation support– Mr. Dede Tarmana (The Indonesian Agency for Meteorology Climatology and Geophysics, Indonesia) : Arc GIS

• Free Software– SPI_tool : The National Drought Mitigation Center– Arc Portable : ESRI– Goole earth : Google Inc– Goole Fusion : : Google Inc– CLIK : APCC – Eviews : HIS Inc.

Flashback

• Introduction• Problem Definition• Aims and Objectives• My approach• Data and Study Area• Methodology• Results• Forecast• Conclusion• Acknowledgement

Thank you