Deep Learning for multi-year ENSO forecasts Yoo-Geun Ham - - PowerPoint PPT Presentation

Deep Learning for multi-year ENSO forecasts

Yoo-Geun Ham

Department of Oceanography, Chonnam National University

• 2019. 11. 07

*Collaborators : Jeong-Hwan Kim (CNU), Jing-Jia Luo (NUIST)

3

• El Nino is defined when Sea Surface Temperature (SST) over the equatorial Eastern

Pacific is higher than the normal year.

• Definition of El Nino index (Nino3.4 index) : Equatorial Eastern Pacific area-averaged

SST.

What is El Nino?

Area for NIno3.4 index

4

El Nino’s global impact

Heatwave Drought Wildfire Heavy rainfall

Deep Learning Algorithm

Deep Neural Network (Deep Learning)

Convolutional Neural Network (CNN) Recurrent Neural Network (RNN)

Use for the detection of the object in the image

Convolutional Neural Network (CNN)

Autonomous Car (self-driving car) Flower Search

Basin concept of statistical forecast of the El Nino event

Time

1-year before the El Nino occurs 6-months before Peak phase Increased temperature over the equatorial western Pacific propagates to east to induce El Nino.

Input layer Output layer

Amplitude of Nino3.4 index Dog

Conventional usage of CNN Application to ENSO forecast

Application of the CNN to climate forecasts

Input layer Output layer

Structure of CNN for ENSO Prediction

9

• Enough numbers of samples to train the CNN model is crucial (e.g., number
• f samples of image for MNIST set is 100,000)
• The oceanic temperature is observed after late 19th (less than 150 years).
• It is very challenging to train the CNN model only with the observations due

to the lack of samples.

Model simulations from Coupled Model Intercomparison Project is utilized to train the CNN.

Difficulties of applying CNN to climate forecasts

CMIP5 simulations

El Nino precipitation Global Temperature in time in CMIP5 models

11

• 1st training : CMIP5 archives (number of samples : about 2,700)
• 2nd training : Reanalysis from 1871 to 1973
• Initial weighting for 2nd training is final weighting of 1st training

Transfer learning

Large number

• f samples
• f Pear

Small number

• f samples
• f Apples

Model identifying Pear Model identifying Apple

12

All-season correlation Skill of 3-months-averaged Nino3.4

*Validation period : 1984-2017 CNN

13

Sensitivity Exp. : Number of samples & Transfer learning

• Required number of samples : at least 1,000.
• Transfer learning has advantages on ENSO forecast.

14

CNN model Heatmap Off-equatorial WP positive HC (Jin, 1997; Anderson, 2007)

Heat map for the 18-month forecast of 97/98 El Nino event

Southwest IO cooling à nIOD (Izumo et

• al. 2010)

WHWP cooling (Park et al., 2017) x-dim. y-dim.

• Conv. dim.
• Conv. dim.

15

The global impact between EP and CP El Ninos are dramatically different. It is crucial to predict the El Nino type as well as the occurrence of the El Nino.

Eastern Pacific (EP) vs Central Pacific (CP) El Nino

CP El Nino EP El Nino

DJF Precipitation anomalies DJF Precipitation anomalies

Skill of El Nino-type (CP, EP, mixed) forecasts

Visit our website http://ocl.jnu.ac.kr

Summary and Conclusion

• A statistical forecast model employing a deep learning approach produces

skillful ENSO forecast for lead times of up to one and a half years.

• To circumvent the limited amount of observation data, we use transfer

learning to train a Convolutional Neural Network (CNN) first on the Coupled Model Intercomparison Project phase 5 (CMIP5) historical simulations and subsequently on reanalysis from 1871 to 1973.

• During the validation period from 1984 to 2017, the all-season correlation

skill of the Nino3.4 index of the CNN model is significantly higher than those of current state-of-the-art dynamical forecast systems.

• Also, the CNN model is better at predicting the detailed zonal distribution
• f sea surface temperatures, overcoming a weakness of dynamical forecast

models.

• A heatmap analysis indicates that the CNN model predicts ENSO events

based on physically reasonable precursors.

• The CNN model can be a powerful tool for both the prediction of ENSO

events and the analysis of their associated complex mechanisms.

E-mail) ygham@jnu.ac.kr Office number) 062-530-3461 Homepage) http://ocl.jnu.ac.kr/

Thank you for your attention

CMIP5 model output

The CMIP5 models have abilities to simulate realistic ENSO to some extent.

• Eq. SST during the El Nino (Kug et al. 2012)

ENSO-related feedback amplitude (Kim et al., 2013)

22

• Class activation map을 이용하여 CNN 모형이 각 카테고리를 어떻게 판단했는지,

입력의 어떤 부분이 이를 판단하게 만들었는지 찾을 수 있음

• Global average pooling을 통해 파라미터의 수를 획기적으로 감소시킬 수 있음

(CNN에서 fully connected layer가 가장 많은 파라미터를 차지함)

Heat map analysis

23

El Nino forecast results for the period of 1984-2017

*EXP1 : Use 2nd Ensemble of the model integration if possible *EXP2 : Ensemble member is randomly selected *EXP3 : Model realization is randomly selected among all integrations *EXP4 : Model realization is randomly selected for each year

3

18-months lead forecast for DJF Nino3.4 index

25

Forecast skill of Feed-forward Neural Network (FFNN) model

*Correlation skill of the CNN model : 0.64

The FFNN model with any combination of EOF PCs cannot beat the CNN model Correlation skill of DJF Nino3.4 index for 18-months lead

26

Forecast skill of Feed-forward Neural Network (FFNN) model

27

• Negative IOD --> (14-months) El Nino (Izumo et al., 2010)
• Negative IO Basin-wide Warming (IOBW) --> (12-months) El Nino (Kang and Kug, 2006)
• WHWP cooling --> (17-months) El Nino (Park et al., 2018)

SST anomalies from MJJ 1996 from FMA 1997

Negative IOD WHWP cooling NTA cooling Negative IOBW

CP- & EP-El Nino precursor patterns

Through the heatmap analysis, the precursors for CP-El Nino over Indian Ocean and Southern Pacific is revealed.

Lag regression for IO and SP precursor for CP-El Nino

IO SP Regressed SST anomalies after 1-year

Visit our website http://ocl.jnu.ac.kr