RAINFALL PREDICTION USING STATISTICAL MULTI MODEL ENSEMBLE OVER SELECTED REGION IN INDONESIA

1
RAINFALL PREDICTION USING STATISTICAL MULTI MODEL
ENSEMBLE OVER SELECTED REGION IN INDONESIA

• INTERNATIONAL WORKSHOP ON
• IMPLEMENTATION OF DIGITIZATION HISTORICAL DATA
  AND SACAD / ICAD AND CLIMATE ANALYSIS IN THE
  REGIONAL ASEAN
•    02 05 APRIL 2012
• JAKARTA / BOGOR, INDONESIA

Fierra Setyawan R D of BMKG fierra.setyawan_at_bmkg
.go.id
2
Outline

• Background
• Data and Methods
• Objective
• Result
• Conclusion
• Introduction ClimaTools
• Future Plans

3
Background
4
Bmkg as the provider climate information

• The behaviour of climate (rainfall) ? high
  variability , such as shifting and changing of
  wet/dry season, climate extrem issues recently
• Users need climate information regulary, accurate
  and localized
• BMKG has been challenged to provide climate
  information
• The limitation of human resources and tools to
  provide climate information in high resolution
• Dynamical Climate Model is high technologies
  computation requirements ? expensive resources
• Statistical model as a solution to fullfill
  forecaster needs in local scale

5
Spatial Planning
Crops
Statistical Models
Water resources
EOF
AR
ANFIS
HyBMG ClimaTools
Filter Kalman
Plantation
Wave- let
Non- Linier
Ensemble
High Res. Weather Climate Forecasts
Multi- regr.
Dissemination
PCA
CCA
Fishery
Statistical Downscaling
Energy Industry
AO- GCM
RCM
Dynamical Downscaling
Hidromet. Disaster Management
Numerical/Dynamical Models
Tourism
MM5, DARLAM, PRECIS, RegCM4, CCAM
6
Why we need ensemble forecast ?

• To antcipate and to reduce the entity of climate
  itself (chaotic)
• Ensemble forecast is a collection of several
  different climate models ? forcaster no need to
  worry which one of model that fitted for one
  particular location especially for his location
• Various ensemble methods have been introduced
  such as a lagged ensemble forecasting method
  (Hoffman and Kalnay, 1983), breeding techniques
  (Toth and Kalnay, 1993), multimodel superensemble
  forecasts (Krishnamurti et al. 1999).
• Dynamic models, because each different model has
  its own variability generated by internal
  dynamics (Straus and Shukla 2000) as a result,
  performance of a multi-model ensemble is
  generally more reliable/ skillful than that of a
  single model (Wandishin et al, 2001, Bright and
  Mullen 2001).

7
Data and Methods
8
Data

• Rainfall Data from 12 location (Lampung, Java,
  South Kalimantan and South Sulawesi)
• Period
• 1981 2009

9
Methods

• Prediction Techniques
• Univariate Statistical Method
• most common statistical (ARIMA),
• Hybrid (ANFIS, Wavelet Transform)
• Multivariate Statistical Method Kalman Filter

10
Methods contd.

• Multi Model Ensemble
• Simple Composite Method ? Simple composite of
  individual forecast with equal weighting

11
skill

• Using Taylor Diagram
• Correlation Coefficient
• Root Mean Square Error
• Standard Deviation

Hasanudin 2006
12
objectives

• To investigate statistical model univariate and
  multivariate in selected location (12 location)
• To provide tools for local forcaster to improve
  quality and accuracy of climate information
  especially in local scale

13
Results
14
Correlation Coefficient

• Multivariate Technique

• Univariate Technique

15
Correlation Coefficient contd.

• Univariate

• Multivariate

16
All Years
17
All Years
18
siNGLE YEAR

• Hasanudin 2007

• Hasanudin 2006

19
conclusion

• The function of Multi model ensemble is a single
  model and it has a better skill
• Correlation value is significant rising, marching
  to eastern part Indonesia, from Lampung, West
  Java, Central Java, East Java, South Kalimantan
  and South Sulawesi
• MME improves accuracy of climate prediction
• Multivariate Statistic technique is not always
  has a better prediction than univariate technique
  

20
Introduction ClimaTools v1.0
21
About ClimaTools v1.0 Software

• The ClimaTools Software is an application for
  processing climate data using statistical tools
  whether univariate or multivariate techniques. It
  contains tools for data processing, analysis,
  prediction and verification.
• The ClimaTools version 1.0 Software includes the
  following statistical packages
• Data analysis single wavelet power spectrum and
  empirical orthogonal function (EOF).
• Prediction Techniques Kalman Filter technique
  and Canonical Correlation Analysis (CCA).
• Verification Methods Taylor Diagram and
  Receiver Operating Characteristic (ROC).

22
(No Transcript)
23
Future plans

• Spatial Climate Prediction embedded in ClimaTools
  
• Integration Statistical Model HyBMG into
  ClimaTools
• Optimalization of output multimodel ensemble by
  adjustment using BMA (Bayesian Model Averaging)
  (koreksi)

24
Thank You
Visit Us http//172.19.1.191 Contact puslitbang_at_bm
kg.go.id