Preprints
https://doi.org/10.5194/gmd-2022-14
https://doi.org/10.5194/gmd-2022-14
Submitted as: development and technical paper
26 Apr 2022
Submitted as: development and technical paper | 26 Apr 2022
Status: this preprint is currently under review for the journal GMD.

Repeatable high-resolution statistical downscaling through deep learning

Dánnell Quesada-Chacón, Klemens Barfus, and Christian Bernhofer Dánnell Quesada-Chacón et al.
  • Institute of Hydrology and Meteorology Technische Universität Dresden, Germany

Abstract. One of the major obstacles for designing solutions against the imminent climate crisis is the scarcity of high spatio-temporal resolution model projections for variables such as precipitation. This kind of information is crucial for impact studies in fields like hydrology, agronomy, ecology and risk management. The currently highest spatial resolution datasets on a daily scale for projected conditions fail to represent complex local variability. We used deep learning (DL) based statistical downscaling (SD) methods to obtain daily 1 km resolution gridded data for precipitation in the Eastern Ore Mountains in Saxony, Germany. We built upon the well established climate4R framework, while adding modifications to its base-code and introducing skip connections based DL architectures, such as U-Net and U-Net++. We also aimed to address the known general reproducibility issues by creating a containerized environment with multi-GPU and TensorFlow’s deterministic operations support. The perfect prognosis approach was applied using the ERA5 reanalysis and the ReKIS (Regional Climate Information System for Saxony, Saxony-Anhalt, and Thuringia) dataset. The results were validated with the VALUE framework. The introduced architectures show a clear performance improvement when compared to previous SD benchmarks. Characteristics of the DL models configurations that promote their suitability for this specific task were identified, tested and argued. Full model repeatability was achieved employing the same physical GPU.

Dánnell Quesada-Chacón et al.

Status: open (until 22 Jun 2022)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse

Dánnell Quesada-Chacón et al.

Data sets

Predictors and predictand for "Repeatable high-resolution statistical downscaling through deep learning" Quesada-Chacón, Dánnell https://doi.org/10.5281/zenodo.5809553

Model code and software

Singularity container for "Repeatable high-resolution statistical downscaling through deep learning" Quesada-Chacón, Dánnell https://doi.org/10.5281/zenodo.5809705

dquesadacr/Rep_SDDL: Submission to GMD Quesada-Chacón, Dánnell https://doi.org/10.5281/zenodo.5856118

Dánnell Quesada-Chacón et al.

Viewed

Total article views: 327 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
244 76 7 327 3 2
  • HTML: 244
  • PDF: 76
  • XML: 7
  • Total: 327
  • BibTeX: 3
  • EndNote: 2
Views and downloads (calculated since 26 Apr 2022)
Cumulative views and downloads (calculated since 26 Apr 2022)

Viewed (geographical distribution)

Total article views: 307 (including HTML, PDF, and XML) Thereof 307 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 20 May 2022
Download
Short summary
We improved the performance of past perfect prognosis statistical downscaling methods while achieving full model repeatability with GPU calculated deep learning models using the TensorFlow, climate4R and VALUE frameworks. We employed the ERA5 reanalysis as predictors and ReKIS (Eastern Ore Mountains, Germany, 1 km resolution) as precipitation predictand, while incorporating modern deep learning architectures. The achieved repeatability is key to accomplish further milestones with deep learning.