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-based statistical downscaling 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 deep learning 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 (graphic processing unit) 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 robust VALUE framework. The introduced architectures show a clear performance improvement when compared to previous statistical downscaling benchmarks. The best performing architecture had a small increase in total number of parameters, in contrast with the benchmark, and a training time of less than 6 min with one NVIDIA A-100 GPU.
Characteristics of the deep learning 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, which is key to build trust in deep learning applications. The EURO-CORDEX dataset is meant to be coupled with the trained models to generate a high-resolution ensemble, which can serve as input to multi-purpose impact models.
Despite intense research on climate change (CC), regional studies for Central America, which is considered a CC hot spot, remain scarce. The information provided by general circulation models (GCMs) is too coarse to accurately reproduce local-scale climatic features, which are needed for impact assessment. Thus, downscaling techniques are employed to address this scale mismatch. Costa Rica is the present case study, for which suitable predictors were tailored for downscaling related to regional climatic characteristics, such as the Inter-Tropical Convergence Zone, El Niño Southern Oscillation, the Caribbean Low-Level Jet, and the MidSummer Drought. Statistical downscaling models were calibrated for precipitation, maximum and minimum temperature, using the perfect prognosis methodology by means of station data, ERA-INTERIM reanalysis and artificial neural networks, yielding satisfactory results. As found in several studies, the temperature models replicated more accurately the statistics of the observed datasets. However, here, through the implemented approach and the tailored predictors, the precipitation models conveyed an improvement compared to standard methods. Projected daily climate was obtained employing CORDEX data under the RCP8.5 scenario for the central region of the country. Overall, the changes in climate estimated by the end of the 21st century agree with coarser-scale projections. Finally, projected climate extremes indices were calculated and rendered further details on the intensity of future CC by the end of the century.
As a contribution to an Integrated Water Resources Management (IWRM) project in Distrito Federal, Brazil, we address several aspects for a credible downscaling of near‐surface air temperature and precipitation using the Statistical DownScaling Model (SDSM4.2). For instance, we apply a detailed screening of predictors, consider the end user needs in the validation procedure, assess the added value of the downscaling model and include several sources of uncertainties until the downscaling step. Results suggest that the interpolation of large‐scale predictors to the target site is a reasonable alternative to predictors derived from grid‐boxes. The validation metrics, measures (i.e. bias, root‐mean‐square error, and Pearson's correlation coefficient) and quantile–quantile plots reveal that model tends to underestimate near‐surface temperature and precipitation; whereas extreme values are subject of considerable uncertainties. Single‐site projections at daily scale are derived from 27 climate models from the Coupled Model Intercomparison Project phase 5 (CMIP5) forced by Representative Concentration Pathways (i.e. RCP2.6, RCP4.5, RCP6.0 and RCP8.5) scenarios. The downscaling model adds substantial value in terms of amplitude of variability when compared to the host coarse‐resolution projections. Its performance is higher than a quantile‐mapping bias correction technique, particularly in reproducing observed trends. In spite of the elevated level of uncertainties in the magnitude of change, most of the downscaled projections agree on positive changes in near‐surface temperature and precipitation for the period of 2036–2055 when compared to the reference period (i.e. 1986–2005). The massive amount of downscaled projections is of limited application in hydrological studies and, therefore, we suggest a summarized group of projections which are representative to the central tendency and spread of the ensemble.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.