Hard-Constrained Deep Learning for Climate Downscaling
Authors: Paula Harder, Alex Hernandez-Garcia, Venkatesh Ramesh, Qidong Yang, Prasanna Sattegeri, Daniela Szwarcman, Campbell Watson, David Rolnick
JMLR 2023 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We compare different constraining approaches and demonstrate their applicability across different neural architectures as well as a variety of climate and weather data sets. Besides enabling faster and more accurate climate predictions through downscaling, we also show that our novel methodologies can improve super-resolution for satellite data and natural images data sets. |
| Researcher Affiliation | Collaboration | Paula Harder EMAIL Fraunhofer ITWM, Kaiserslautern, Germany Mila Quebec AI Institute, Montreal, Canada Alex Hernandez-Garcia Mila Quebec AI Institute, Montreal, Canada University of Montreal, Montreal, Canada Venkatesh Ramesh Mila Quebec AI Institute, Montreal, Canada University of Montreal, Montreal, Canada Qidong Yang Mila Quebec AI Institute, Montreal, Canada New York University, New York, USA Prasanna Sattegeri IBM Research, New York, USA Daniela Szwarcman IBM Research, Brazil Campbell D. Watson IBM Research, New York, USA David Rolnick Mila Quebec AI Institute, Montreal, Canada Mc Gill University, Montreal, Canada |
| Pseudocode | No | The paper describes its methodologies using mathematical equations and textual explanations, but it does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is available at https://github. com/Rolnick Lab/constrained-downscaling |
| Open Datasets | Yes | Our code is available at https://github. com/Rolnick Lab/constrained-downscaling and our main data set can be found at https: //drive.google.com/file/d/1IENh P1-a TYyq Ok Rcnm CIvx Xkv UW2Qbdx/view. We use climate and weather data sets based on European Center for Medium-Range Weather Forecasts (ECMWF) reanalysis data version 5 (ERA5) (Hersbach et al., 2020), Weather Research and Forecast Model (WRF) data (Auger et al., 2021), and the Norwegian Earth System Model (Nor ESM) (Seland et al., 2020) data |
| Dataset Splits | Yes | To obtain our high-resolution data points we extract a random 128 128 pixel image from each available time step... We randomly sample 40,000 data points for training and 10,000 for each validation and testing. ... OOD data set ... we create a data set with a split in time. Here, we expect patterns to appear in the later time steps that are out-of-distribution of what was previously observed. We train on older data and then test on more recent years: for training, we use the years 1950-2000, for validation 2001-2010, and for final testing 2011-2020. |
| Hardware Specification | Yes | All models were trained with the Adam optimizer, a learning rate of 0.001, and a batch size of 256. We trained for 200 epochs, which took about 3 6 hours on a single NVIDIA A100 Tensor Core GPU, depending on the architecture. |
| Software Dependencies | No | The paper mentions using the Adam optimizer and MSE as a criterion, but it does not specify any software names with version numbers for libraries, frameworks, or programming languages (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | Our models were trained with the Adam optimizer, a learning rate of 0.001, and a batch size of 256. We trained for 200 epochs, which took about 3 6 hours on a single NVIDIA A100 Tensor Core GPU, depending on the architecture. All models use the MSE as their criterion, the GAN additionally uses its discriminator loss term. All the data are normalized between 0 and 1 for training, except for the cases where the Sc Add CL is applied. In the case of this constraint layer we scale the data between -1 and 1 as proposed in Geiss and Hardin (2023). |