Page 21 - APEC CLIMATE CENTER 2025 Annual Report

APEC CLIMATE CENTER 2025 ANNUAL REPORT

Highlighted 4. Initiative for Developing an East Asia Climate

Achievements Extremes Dataset!

in 2025 ㉖ Dr. Yun-Young Lee (yyalee@apcc21.org) Dr. Miae Kim (miaekim@apcc21.org)

Dr. Uran Chung (uchung@apcc21.org)

The most critical and fundamental step in developing AI models for predicting climate ex-
tremes is the systematic construction of training data. However, defining climate extremes
is challenging in the initial stages of data construction because definitions vary by mete-
orological element, and criteria differ depending on research objectives or regions. Con-
1)
sequently, a systematically organized and classified inventory of major climate extreme
phenomena is now needed. Establishing such an inventory is essential for enhancing the
reproducibility and utility of future AI-based prediction research. Furthermore, it holds
significant meaning as it promotes the qualitative and quantitative expansion of East
Asian climate extreme research by lowering the barrier to data access for researchers in
academia and related organizations.

The primary achievement of this research is the establishment of a sharing system that
systematizes climate extreme data and analysis codes for the East Asia region (21–48°N,
114–141°E) via the GitHub repository ‘EastAsiaClimateExtremes’ (https://github.com/yy-
2)
alexlee/EastAsiaClimateExtremes). The main components are as follows:

Fig 15 Detailed metadata provided by the EastAsiaClimateExtreme GitHub repository
3)
Fig 16 Sample Jupyter Notebook scripts supporting statistical analysis, visualization, and data storage

◎ Construction of Long-term East Asian Climate Extremes Data: Utilizing observa-
tion-based reanalysis data such as ERA5 and OISST, along with ECMWF-hindcast dy-
namical model data, major extreme phenomena including Anomalously High Tem-
peratures (AHT), Heavy Rainfall (HR), and Marine Heatwaves (MHW) were quantified on
a long-term, grid basis.
◎ Inclusion of Grid-based Detailed Extreme Indices: Daily and weekly climatological nor-
mals and percentile thresholds (90th/95th) were calculated. Based on these, detailed
profiles including occurrence frequency, duration, mean and max intensity, and im-
pact factors, as well as weekly extremeness metrics, were generated.
◎ Provision of Analysis Tools and Flexibility: Through Python-based Jupyter Notebooks,
the system provides codes that allow for the reproduction of the entire workflow, from
data loading to extreme event calculation, storage, and visualization using time-series
graphs or heatmaps. Additionally, it is designed with a flexible structure that allows us-
Glossary ers to easily modify the research domain, variables, and time periods to suit their needs.

1) Climate Extremes Inventory: This system is expected to substantially contribute to the analysis of mechanisms behind
Refers to a ‘comprehensive informa- East Asian climate extremes and the improvement of prediction accuracy.
tion system’ established by identifying ◎ Acceleration of AI Research: The constructed inventory can be immediately utilized
extreme weather events within a spe- as training and labeling data as well as validation datasets for AI-based climate pre-
cific scope (spatial and temporal) and diction models, thereby increasing research efficiency.
quantifying the data according to stan- ◎ Ease of Model Evaluation: By providing both observation-based data and Global
Glossary
dardized methods.
Circulation Model (GCM) data, this system facilitates performance comparisons and
3) Jupyter Notebook: evaluations between AI models and existing dynamical models.
2) GitHub Repository An all-in-one analysis tool that com- ◎ Provision of Research Collaboration Infrastructure: The distribution via a public re-
Fig 14 Screenshot of EastAsiaClimateExtreme GitHub repository page (https://github.com/
A web-based storage space used to yyalexlee/EastAsiaClimateExtremes) bines code, execution results (graphs), pository and the provision of reproducible code will create an environment where
store and share a project's source code and documentation in a single docu- researchers can collaborate on a single platform based on a ‘common baseline
and version history. ment. dataset.’

20
20 21 21