ClimateBench
收藏资源简介:
ClimateBench是由麻省理工学院地球、大气和行星科学系等机构创建的气候模拟数据集,用于评估气候模型的性能。该数据集包含753条数据,涵盖了多种气候变量和排放情景。数据集的创建过程包括使用多个气候模型进行模拟,并对结果进行平均处理以减少内部变异性。ClimateBench主要应用于气候模拟和预测领域,旨在提高气候模型预测的准确性。
ClimateBench is a climate simulation dataset developed by institutions including the Department of Earth, Atmospheric and Planetary Sciences at the Massachusetts Institute of Technology (MIT) and other relevant organizations, used to evaluate the performance of climate models. This dataset contains 753 data records, covering multiple climate variables and emission scenarios. The dataset is constructed by running simulations with multiple climate models and averaging their outputs to reduce internal variability. ClimateBench is primarily applied in the fields of climate simulation and prediction, aiming to improve the accuracy of climate model predictions.
气候模拟器数据集概述
数据集描述
该数据集用于比较线性模式缩放(LPS)与CNN-LSTM在气候模拟中的应用。数据来源于ClimateBench和MPI-ESM1.2-LR模型的新数据摘要。
数据下载
Em-MPI数据摘要(<10GB)
bash export DATA_DIR=/path/to/data/dir mkdir -p $DATA_DIR python download_emmpi.py --data_dir $DATA_DIR
ClimateBench NorESM2-LM目标和input4mips排放输入(<2GB)
bash export PATH_CLIMATEBENCH_DATA=$DATA_DIR/data/raw/climatebench/ mkdir -p $PATH_CLIMATEBENCH_DATA wget https://zenodo.org/record/7064308/files/train_val.tar.gz -P $PATH_CLIMATEBENCH_DATA tar -xvf "$PATH_CLIMATEBENCH_DATA/train_val.tar.gz" -C $PATH_CLIMATEBENCH_DATA rm $PATH_CLIMATEBENCH_DATA/train_val.tar.gz wget https://zenodo.org/record/7064308/files/test.tar.gz -P $PATH_CLIMATEBENCH_DATA tar -xvf "$PATH_CLIMATEBENCH_DATA/test.tar.gz" -C $PATH_CLIMATEBENCH_DATA rm $PATH_CLIMATEBENCH_DATA/test.tar.gz
数据处理
线性模式缩放(LPS)结果重现
bash jupyter notebook notebooks/calculate_climatebench_metrics.ipynb
内部变异性实验重现
第一代码测试:在50个成员集合平均Em-MPI数据上训练和评估CNN-LSTM
bash wandb login export TF_GPU_ALLOCATOR=cuda_malloc_async export KERAS_BACKEND=torch vim runs/cnn_lstm/mpi-esm1-2-lr/default/config/config.yaml python emcli2/models/cnn_lstm/train.py --cfg_path runs/cnn_lstm/mpi-esm1-2-lr/default/config/config.yaml --data_var pr --verbose
第二代码测试:在1,2,...,50个成员的子集上训练LPS和CNN-LSTM,然后绘制RMSE随实现次数的变化
bash python emcli2/models/cnn_lstm/train.py --train_m_member_subsets --cfg_path runs/cnn_lstm/mpi-esm1-2-lr/m_member_subsets_with_m50_eval_on_all_spcmp_dwp_manyr/config/config.yaml --data_var pr python emcli2/models/pattern_scaling/model.py --train_m_member_subsets --cfg_path runs/pattern_scaling/mpi-esm1-2-lr/m_member_subsets_with_m50_replace_False_eval_on_all_manyr/config/config.yaml --data_var pr python emcli2/utils/plotting.py --plot_m_member_subsets_experiment --data_var pr
其他图表重现
bash jupyter notebook notebooks/explore_linear_relationships.ipynb jupyter notebook notebooks/explore_local_internal_variability.ipynb
参考文献
如果该数据集对您的分析有用,请考虑引用:
@article{lutjens24internalvar, title={A Cautionary Tale about Deep Learning-based Climate Emulators and Internal Variability}, year={2024}, }

- 1The impact of internal variability on benchmarking deep learning climate emulators麻省理工学院地球、大气和行星科学系,斯克里普斯海洋研究所和哈利克奥卢数据科学研究所,加州大学圣地亚哥分校,麻省理工学院数据、系统和社会研究所 · 2024年



