A deep-learning-based consistency test approach for Earth system models on HPC systems

Evaluating climate consistency is a critically important step in the development and optimization of Earth system models (ESMs) on the high-performance computing (HPC) systems. We have developed an Earth system model deep-learning consistency test, referred to as ESM-DCT. The ESM-DCT is based on the unsupervised bidirectional gate recurrent unit-autoencoder (BGRU-AE) model to study the features from the ESM simulation ensembles and adopts the reconstruction errors to evaluate the consistency. We use the Community Earth System Model (CESM) on the new Sunway heterogeneous system to evaluate the ESM-DCT. The results show that the ESM-DCT can determine whether or not the new model simulations are statistically distinguishable from the original trusted ensembles in the case of the heterogeneous computing environment, compiling optimization option changes, and model parameter changes. Our ESM-DCT tool provides an efficient and objective approach for verifying the reliability of the development and optimization of ESMs on HPC systems.