INFORMATION TECHNOLOGY

Chinese scientists use artificial intelligence technology to improve the performance of climate models


Schematic diagram of ocean vertical hybrid parametric scheme based on deep learning

(a) Simulation errors in the equatorial eastern Pacific Ocean temperature obtained for the current ocean model using a KPP parametric scheme based on physical empirical relationships (0o, 140oW); (b) Error after applying a new parametric scheme based on neural network (NN); (c) and (d) are the improved effects of the NN method over the KPP method. The results show that the warm error of 60-80 meters is reduced by about 44% of the ocean-supplied map

Recently, the reporter learned from the Institute of Oceanography of the Chinese Academy of Sciences that under the clear physical constraints, the team of Zhang Ronghua, a researcher at the Institute, designed the first ocean vertical hybrid parameterization scheme based on deep learning and turbulence observation data, and applied to the ocean and climate models, and its simulation effect is better than the traditional parameterization method based on the relationship between physical experience, which effectively improves the simulation performance of ocean and climate models. The results were published in the National Science Review.

Since 2021 Nobel Laureate in Physics, Shuro Manabe, and others first established a coupling model covering the global atmosphere and ocean aliquots in 1969, the sea-air coupling model has been an important tool for climate research. Since 1995, the World Climate Research Programme (WCRP) has organized six international coupling model comparison programs (such as the recent CMIP6 product), which have greatly promoted the development and improvement of climate models and have become an important scientific basis for the preparation of the intergovernmental panel on climate change (IPCC) assessment report. However, even the newly released CMIP6 simulations in 2021 still have large systemic differences from observations, and these errors severely constrain the model’s ability to simulate current climate and predict future climate change, and directly affect the credibility of IPCC reports. In view of this, the attribution and elimination of climate pattern errors has always been one of the important contents of climate research.

Among the many reasons why the systematic error of climate patterns comes from, there is a large uncertainty in the ocean vertical hybrid parametric scheme, which is recognized as an important source of error. The current climate model adopts a parametric scheme based on physical empirical relations (such as the KPP scheme based on ocean state such as ocean current shear and stability), and the mixed coefficients estimated by these schemes are quite different from the observation facts, and it is difficult to accurately characterize the observed vertical turbulence heat exchange process of the ocean, which leads to errors in variables such as SST simulations. Especially in the tropical Pacific ocean area, the vertical vortex diffusion coefficient estimated based on the parametric scheme of physical relations is significantly larger and the downward turbulence heat flux is too strong, which is an important reason for the “cold tongue” cold error in the coupling mode.

The key to the uncertainty of the parametric scheme is that the current common schemes are based on pre-assumed physical empirical relationships; Since the current physical understanding of the mixed processes of ocean turbulence is still very limited, the empirical relationship based on these limited understandings naturally produces great uncertainty.

In order to solve this problem, Zhang Ronghua’s team used the turbulence observation records of the tropical Pacific Ocean in the past decade to design the first ocean vertical hybrid parameterization scheme based on deep learning under clear physical constraints, and further applied this parameterization scheme to the ocean circulation and sea-air coupling mode, confirming its ability to characterize the vertical mixing coefficient and vertical heat flux of the upper ocean, thereby effectively improving the temperature simulation results of the tropical Pacific Ocean.

Professor Gustau Camps-Valls, an international authoritative expert in artificial intelligence earth sciences, wrote a special article commenting on the research results, he believes that the research has constructed a more performant and generalized parameterization scheme in a concise and practical way, and ultimately achieved the goal of improving the performance of the climate model.

The first author of the paper is Zhu Yuchao, associate researcher of the Institute of Oceanography, corresponding author is Zhang Ronghua, and the collaborators include Fan Wang, Xiaofeng Li, Delei Li, and James N. Moum, a professor at Oregon State University. The research results have been funded by the Ocean Science Research Center of the Chinese Academy of Sciences, the Qingdao Marine Science and Technology Pilot National Laboratory, the Quaternary Science and Global Change Excellence and Innovation Center of the Chinese Academy of Sciences, the Strategic Pilot Science and Technology Project of the Chinese Academy of Sciences and the National Natural Science Foundation of China. (Source: China Science Daily, Liao Yang, Li Hezhao)

Related paper information:https://doi.org/10.1093/nsr/nwac044



Source link

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button