Conditional Local Convolution for Spatio-Temporal Meteorological Forecasting

Authors: Haitao Lin, Zhangyang Gao, Yongjie Xu, Lirong Wu, Ling Li, Stan Z. Li7470-7478

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our model is evaluated on realworld weather benchmark datasets, achieving state-of-the-art performance with obvious improvements. We conduct further analysis on local pattern visualization, model s framework choice, advantages of horizon maps and etc.
Researcher Affiliation Academia Haitao Lin,*1 3 Zhangyang Gao,*1 3 Yongjie Xu, 1 3 Lirong Wu, 1 3 Ling Li, 2 Stan Z. Li 1 1 Center of Artificial Intelligence for Research and Innovation, Westlake University 2 Eco-Environmental Research Laboratory, Westlake University 3 Zhejiang University linhaitao, gaozhangyang, xuyongjie, wulirong, liling, stan.zq.li@westlake.edu.cn
Pseudocode No The paper describes the method and architecture in text and figures (e.g., Figure 5) but does not contain a dedicated pseudocode or algorithm block.
Open Source Code Yes The source code is available at https://github.com/BIRD-TAO/CLCRN.
Open Datasets Yes The datasets used for performance evaluation are provided in Weather Bench (Rasp et al. 2020), with 2048 nodes on the earth sphere.
Dataset Splits Yes The hyper-parameters are chosen through a carefully tuning on the validation set (See Appendix D1 for more details).
Hardware Specification No No specific hardware details (e.g., GPU/CPU models, memory specifications) used for experiments were mentioned in the paper.
Software Dependencies No No specific software dependencies with version numbers (e.g., library names with versions) were explicitly mentioned in the paper.
Experiment Setup Yes All the models are trained with target function of MAE and optimized by Adam optimizer for a maximum of 100 epoches. The hyper-parameters are chosen through a carefully tuning on the validation set (See Appendix D1 for more details). The learning rate is set to 0.001 with decayed by 0.7 for every 5 epoches. The batch size is set to 64. The hidden size is 32.