TemperatureGAN: Generative Modeling of Regional Atmospheric Temperatures

Includes engineered dataset for creating TemperatureGAN and trained models for inference. Stochastic generators are useful for estimating climate impacts on various sectors. Projecting climate risk in various sectors, e.g. energy systems, requires generators that are accurate (statistical resemblance to ground-truth), reliable (do not produce erroneous examples), and efficient. Leveraging data from the North American Land Data Assimilation System, we introduce TemperatureGAN, a Generative Adversarial Network conditioned on months, locations, and time periods, to generate 2m above ground atmospheric temperatures at an hourly resolution. We propose evaluation methods and metrics to measure the quality of generated samples. We show that TemperatureGAN produces high-fidelity examples with good spatial representation and temporal dynamics consistent with known diurnal cycles. If you use the data or models, you must cite our paper (paper in related links). The complete dataset used to train the model is too large to upload here, thus we offer a random split of the original data, which should be sufficient to train a good model to reproduce many of our results. Please refer to TemperatureGAN paper for details on data engineering.本数据集包含用于构建TemperatureGAN的工程化数据集以及用于推理的预训练模型。随机生成器可用于评估气候对各行业的影响。对能源系统等各行业的气候风险进行预测，需要生成器兼具精准性（与基准真值的统计分布一致）、可靠性（不会生成错误样本）与高效性。本研究借助北美陆面数据同化系统（North American Land Data Assimilation System）的数据，提出了TemperatureGAN——一种以月份、地理位置和时间段为条件的生成式对抗网络（Generative Adversarial Network），用于生成地面2米高度的小时级分辨率大气温度数据。我们提出了用于评估生成样本质量的评估方法与指标。实验结果表明，TemperatureGAN可生成高保真度样本，其具备良好的空间表征能力，且时间动态特性与已知的日循环周期相符。若您使用本数据集或模型，请务必引用我们的论文（相关链接中附有论文信息）。用于训练模型的完整数据集体积过大，无法在此上传，因此我们提供了原始数据的随机拆分子集，该子集足以训练出性能优异的模型，复现我们的多项研究结果。有关数据工程的详细信息，请参阅TemperatureGAN相关论文。