Simulation and Evaluation of Statistical Downscaling of Regional Daily Precipitation over North China Based on Self-Organizing Maps

A statistical downscaling method based on Self-Organizing Maps (SOM), of which the SOM Precipitation Statistical Downscaling Method (SOM-SD) is named, has received increasing attention. Herein, its applicability of downscaling daily precipitation over North China is evaluated. Six indices (total season precipitation, daily precipitation intensity, mean number of precipitation days, percentage of rainfall from events beyond the 95th percentile value of overall precipitation, maximum consecutive wet days, and maximum consecutive dry days) are selected, which represent the statistics of daily precipitation with regards to both precipitation amount and frequency, as well as extreme event. The large-scale predictors were extracted from the National Center for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) daily reanalysis data, while the prediction was the high resolution gridded daily observed precipitation. The results show that the method can establish certain conditional transformation relationships between large-scale atmospheric circulation and local-scale surface precipitation in a relatively simple way. This method exhibited a high skill in reproducing the climatologic statistical properties of the observed precipitation. The simulated daily precipitation probability distribution characteristics can be well matched with the observations. The values of Brier scores are between 0 and 1.5 × 10−4 and the significance scores are between 0.8 and 1 for all stations. The SOM-SD method, which is evaluated with the six selected indicators, shows a strong simulation capability. The deviations of the simulated daily precipitation are as follows: Total season precipitation (−7.4%), daily precipitation intensity (−11.6%), mean number of rainy days (−3.1 days), percentage of rainfall from events beyond the 95th percentile value of overall precipitation (+3.4%), maximum consecutive wet days (−1.1 days), and maximum consecutive dry days (+3.5 days). In addition, the frequency difference of wet-dry nodes is defined in the evaluation. It is confirmed that there was a significant positive correlation between frequency difference and precipitation. The findings of this paper imply that the SOM-SD method has a good ability to simulate the probability distribution of daily precipitation, especially the tail of the probability distribution curve. It is more capable of simulating extreme precipitation fields. Furthermore, it can provide some guidance for future climate projections over North China.

Statistical Downscaling of Precipitation for Mahanadi Basin in India - Prediction of Future Streamflows

Climate change has long-term impacts on precipitation patterns, magnitude, and intensity, affecting regional water resources availability. Besides, understanding the interannual to decadal variations of streamflows in a river basin is paramount for watershed management, primarily extreme events such as floods and droughts. This study investigates impact of climate change in streamflows estimation for four sub-basins of the Mahanadi River, in India. The study includes three major components: (i) Historical trend analysis of hydroclimatic variables, using Mann-Kendall test; (ii) Statistical downscaling of GCM generated precipitation using change factor method and KnnCAD V4 stochastic weather generator; (iii) Dependable flow analysis of future streamflows predicted using Soil Water Assessment Tool (SWAT) model for various future GCM scenarios. It is observed that during the historical period, there is a decrease in number of rainy days and total annual precipitation in all sub-basins. However, the analysis also indicates an increase in flood intensity in two of the sub-basins. The decadal future precipitation has a marginal decrease in precipitation (up to 10%) for future scenarios; however, the precipitation events with high intensities increases. The results indicate that the magnitudes of 5% and 10% dependable flows are higher than the historically observed streamflows, for all future scenarios. This indicates a significant increase in extreme flood events in the basin. Further, only one of the sub-basins has shown an increase in water availability for agriculture and drinking water purposes (75% and 95% dependable flows) in the future. Understanding future flood events in the Mahanadi basin can help decision-makers to implement appropriate mitigation strategies.

­Comparison of Statistical Downscaling of Summer Daily Precipitation Through a Certain Perfect Prognosis and Bias Corrections - A Case Study Across China

Fewer perfect prognosis (PP) based statistical downscaling were applied to future projections produced by global circulation models (GCM), when compared with the method of model output statistics (MOS). This study is a trial to use a multiple variable based PP downscaling for summer daily precipitation at many sites in China and to compare with the MOS. For the PP method (denoted as ‘OGB-PP’), predictors for each site are screened from surface-level variables in ERA-Interim reanalysis by an optimal grid-box method, then the biases in predictors are corrected and fitted to generalized linear models to downscale daily precipitation. The historical and the future simulations under the medium emission scenario (often represented as ‘RCP4.5’), produced by three GCMs (CanESM2, HadGEM2-ES and GFDL-ESM2G) in the coupled model intercomparison project phase five (CMIP5) were used as the downscaling bases. The bias correction based MOS downscaling (denoted as ‘BC-MOS’) were used to compare with the OGB-PP. The OGB-PP generally produced the climatological mean of summer precipitation across China, based on both ERAI and CMIP5 historical simulations. The downscaled spatial patterns of long-term changes are diverse, depending on the different GCMs, different predictor-bias corrections, and the choices on selecting PP and MOS. The annual variations downscaled by OGB-PP have small differences among the choices of different predictor-bias corrections, but have large difference to that downscaled by BC-MOS. The future changes downscaled from each GCM are sensitive to the bias corrections on predictors. The overall change patterns in some OGB-PP results on future projections produced similar trends as those projected by other multiple-model downscaling in CMIP5, while the result of the BC-MOS on the same GCMs did not, implying that PP methods may be promising. OGB-PP produced more significant increasing/decreasing trends and larger spatial variability of trends than the BC-MOS methods did. The reason maybe that in OGB-PP the independent precipitation modeling mechanism and the freely selected grid-box predictors can give rise to more diverse outputs over different sites than that from BC-MOS, which can contribute additional local variability.

Future climate change scenario in hot semi-arid climate of Saurashtra, Gujarat by using statistical downscaling by LARS-WG model

Climate change is considered to be the greatest challenge faced by mankind in the twenty first century which can lead to severe impacts on different major sectors of the world such as water resources, agriculture, energy and tourism and are likely to alter trends and timing of precipitation and other weather drivers. Analyses and prediction of change in critical climatic variables like rainfall and temperature are, therefore, extremely important. Keeping this in mind, this study aims to verify the skills of LARS-WG (Long Ashton Research - Weather Generator), a statistical downscaling model, in simulating weather data in hot semi-arid climate of Saurashtra and analyze the future changes of temperature (maximum and minimum) and precipitation downscaled by LARS-WG based on IPCC SRA2 scenario generated by seven GCMs' projections for the near (2011-2030), medium (2046-2065) and far (2080-2099) future periods. Rajkot (22.3° N, 70.78° E) observatory of IMD, representing hot semi-arid climate of Saurashtra, Gujarat state was chosen for this purpose. Daily rainfall, maximum and minimum temperature data for the period of 1969-2013 have been utilized. LARS-WG is found to show reasonably good skill in downscaling daily rainfall and excellent skill in downscaling maximum and minimum temperature. The downscaled rainfall indicated no coherent change trends among various GCMs’ projections of rainfall during near, medium and far future periods. Contrary to rainfall projections, simulations from the seven GCMs have coherent results for both the maximum and minimum temperatures. Based on the ensemble mean of seven GCMs, projected rainfall at Rajkot in monsoon season (JJAS) showed an increase in near future, i.e., 2011-2030, medium future (2046-2065) and far future (2080-2099) periods to the tune of 2, 11 and 14% respectively compared to the baseline value. Model studies indicating tropospheric warming leading to enhancement of atmospheric moisture content could be the reason for this increasing trend. Further, at the study site summer (MAM) maximum temperature is projected to increase by 0.5, 1.7 and 3.3°C during 2011-2030, 2046-2065 and 2080-2099 respectively and winter (DJF) minimum temperature is projected to increase by 0.8, 2.2 and 4.5 °C during 2011-2030, 2046-2065 and 2080-2099 respectively.

Is time a variable like the others in multivariate statistical downscaling and bias correction?

Bias correction and statistical downscaling are now regularly applied to climate simulations to make then more usable for impact models and studies. Over the last few years, various methods were developed to account for multivariate – inter-site or inter-variable – properties in addition to more usual univariate ones. Among such methods, temporal properties are either neglected or specifically accounted for, i.e. differently from the other properties. In this study, we propose a new multivariate approach called “time-shifted multivariate bias correction” (TSMBC), which aims to correct the temporal dependency in addition to the other marginal and multivariate aspects. TSMBC relies on considering the initial variables at various times (i.e. lags) as additional variables to be corrected. Hence, temporal dependencies (e.g. auto-correlations) to be corrected are viewed as inter-variable dependencies to be adjusted and an existing multivariate bias correction (MBC) method can then be used to answer this need. This approach is first applied and evaluated on synthetic data from a vector auto-regressive (VAR) process. In a second evaluation, we work in a “perfect model” context where a regional climate model (RCM) plays the role of the (pseudo-)observations, and where its forcing global climate model (GCM) is the model to be downscaled or bias corrected. For both evaluations, the results show a large reduction of the biases in the temporal properties, while inter-variable and spatial dependence structures are still correctly adjusted. However, increasing the number of lags too much does not necessarily improve the temporal properties, and an overly strong increase in the number of dimensions of the dataset to be corrected can even imply some potential instability in the adjusted and/or downscaled results, calling for a reasoned use of this approach for large datasets.

MSG-GAN-SD: A Multi-Scale Gradients GAN for Statistical Downscaling of 2-Meter Temperature over the EURO-CORDEX Domain

One of the most important open challenges in climate science is downscaling. It is a procedure that allows making predictions at local scales, starting from climatic field information available at large scale. Recent advances in deep learning provide new insights and modeling solutions to tackle downscaling-related tasks by automatically learning the coarse-to-fine grained resolution mapping. In particular, deep learning models designed for super-resolution problems in computer vision can be exploited because of the similarity between images and climatic fields maps. For this reason, a new architecture tailored for statistical downscaling (SD), named MSG-GAN-SD, has been developed, allowing interpretability and good stability during training, due to multi-scale gradient information. The proposed architecture, based on a Generative Adversarial Network (GAN), was applied to downscale ERA-Interim 2-m temperature fields, from 83.25 to 13.87 km resolution, covering the EURO-CORDEX domain within the 1979–2018 period. The training process involves seasonal and monthly dataset arrangements, in addition to different training strategies, leading to several models. Furthermore, a model selection framework is introduced in order to mathematically select the best models during the training. The selected models were then tested on the 2015–2018 period using several metrics to identify the best training strategy and dataset arrangement, which finally produced several evaluation maps. This work is the first attempt to use the MSG-GAN architecture for statistical downscaling. The achieved results demonstrate that the models trained on seasonal datasets performed better than those trained on monthly datasets. This study presents an accurate and cost-effective solution that is able to perform downscaling of 2 m temperature climatic maps.