Temporal downscaling: a comparison between artificial neural network and autocorrelation techniques over the Amazon Basin in present and future climate change scenarios

Several studies have been devoted to dynamic and statistical downscaling for both climate variability and climate change. This paper introduces an application of temporal neural networks for downscaling global climate model output and autocorrelation functions. This method is proposed for downscaling daily precipitation time series for a region in the Amazon Basin. The downscaling models were developed and validated using IPCC AR4 model output and observed daily precipitation. In this paper, five AOGCMs for the twentieth century (20C3M; 1970–1999) and three SRES scenarios (A2, A1B, and B1) were used. The performance in downscaling of the temporal neural network was compared to that of an autocorrelation statistical downscaling model with emphasis on its ability to reproduce the observed climate variability and tendency for the period 1970–1999. The model test results indicate that the neural network model significantly outperforms the statistical models for the downscaling of daily precipitation variability.     

A cautionary note on automated statistical downscaling methods for climate change

The urge for higher resolution climate change scenarios has been widely recognized, particularly for conducting impact assessment studies. Statistical downscaling methods have shown to be very convenient for this task, mainly because of their lower computational requirements in comparison with nested limited-area regional models or very high resolution Atmosphere–ocean General Circulation Models. Nevertheless, although some of the limitations of statistical downscaling methods are widely known and have been discussed in the literature, in this paper it is argued that the current approach for statistical downscaling does not guard against misspecified statistical models and that the occurrence of spurious results is likely if the assumptions of the underlying probabilistic model are not satisfied. In this case, the physics included in climate change scenarios obtained by general circulation models, could be replaced by spatial patterns and magnitudes produced by statistically inadequate models. Illustrative examples are provided for monthly temperature for a region encompassing Mexico and part of the United States. It is found that the assumptions of the probabilistic models do not hold for about 70 % of the gridpoints, parameter instability and temporal dependence being the most common problems. As our examples reveal, automated statistical downscaling “black-box” models are to be considered as highly prone to produce misleading results. It is shown that the Probabilistic Reduction approach can be incorporated as a complete and internally consistent framework for securing the statistical adequacy of the downscaling models and for guiding the respecification process, in a way that prevents the lack of empirical validity that affects current methods.     

Evaluation of machine learning tools as a statistical downscaling tool: temperatures projections for multi-stations for Thames River Basin, Canada

Many impact studies require climate change information at a finer resolution than that provided by global climate models (GCMs). This paper investigates the performances of existing state-of-the-art rule induction and tree algorithms, namely single conjunctive rule learner, decision table, M5 model tree, and REPTree, and explores the impact of climate change on maximum and minimum temperatures (i.e., predictands) of 14 meteorological stations in the Upper Thames River Basin, Ontario, Canada. The data used for evaluation were large-scale predictor variables, extracted from National Centers for Environmental Prediction/National Center for Atmospheric Research reanalysis dataset and the simulations from third generation Canadian coupled global climate model. Data for four grid points covering the study region were used for developing the downscaling model. M5 model tree algorithm was found to yield better performance among all other learning techniques explored in the present study. Hence, this technique was applied to project predictands generated from GCM using three scenarios (A1B, A2, and B1) for the periods (2046–2065 and 2081–2100). A simple multiplicative shift was used for correcting predictand values. The potential of the downscaling models in simulating predictands was evaluated, and downscaling results reveal that the proposed downscaling model can reproduce local daily predictands from large-scale weather variables. Trend of projected maximum and minimum temperatures was studied for historical as well as downscaled values using GCM and scenario uncertainty. There is likely an increasing trend for T _max and T _min for A1B, A2, and B1 scenarios while decreasing trend has been observed for B1 scenarios during 2081–2100.     

A comparison of downscaling techniques in the projection of local climate change and wheat yields

This study aims to evaluate the performance of two mainstream downscaling techniques: statistical and dynamical downscaling and to compare the differences in their projection of future climate change and the resultant impact on wheat crop yields for three locations across New South Wales, Australia. Bureau of Meteorology statistically- and CSIRO dynamically-downscaled climate, derived or driven by the CSIRO Mk 3.5 coupled general circulation model, were firstly evaluated against observed climate data for the period 1980–1999. Future climate projections derived from the two downscaling approaches for the period centred on 2055 were then compared. A stochastic weather generator, LARS-WG, was used in this study to derive monthly climate changes and to construct climate change scenarios. The Agricultural Production System sIMulator-Wheat model was then combined with the constructed climate change scenarios to quantify the impact of climate change on wheat grain yield. Statistical results show that (1) in terms of reproducing the past climate, statistical downscaling performed better over dynamical downscaling in most of the cases including climate variables, their mean, variance and distribution, and study locations, (2) there is significant difference between the two downscaling techniques in projected future climate change except the mean value of rainfall across the three locations for most of the months; and (3) there is significant difference in projected wheat grain yields between the two downscaling techniques at two of the three locations.     

两种统计降尺度方法在秦岭山地的适用性

基于ASD（automated statistical downscaling）统计降尺度模型提供的多元线性回归和岭回归两种统计降尺度方法,采用RCP4.5（representative concentration pathways 4.5）和RCP8.5情景下全球气候模式MPI-ESM-LR输出的预报因子数据、NCEP/NCAR再分析数据和秦岭山地周边10个气象站观测数据,评估两种统计降尺度方法在秦岭山地的适用性及预估秦岭山地未来3个时期（2006-2040年、2041-2070年和2071-2100年）的平均气温和降水。结果表明:率定期和验证期内,两种统计降尺度方法均可以较好地模拟研究区域的平均气温和降水的变化特征,且多元线性回归的模拟效果优于岭回归。在未来气候情景下,两种统计降尺度方法预估的研究区域平均气温均呈明显上升趋势,气温增幅随辐射强迫增加而增大。降水方面,21世纪未来3个时期降水均呈不明显减少趋势,但季节分配发生变化。综合考虑两种统计降尺度方法在秦岭山地对平均气温和降水的模拟效果和情景预估结果,认为多元线性回归降尺度方法更适用于秦岭山地气候变化的降尺度预估研究。     

Joint variable spatial downscaling

Joint Variable Spatial Downscaling (JVSD), a new statistical technique for downscaling gridded climatic variables, is developed to generate high resolution gridded datasets for regional watershed modeling and assessments. The proposed approach differs from previous statistical downscaling methods in that multiple climatic variables are downscaled simultaneously and consistently to produce realistic climate projections. In the bias correction step, JVSD uses a differencing process to create stationary joint cumulative frequency statistics of the variables being downscaled. The functional relationship between these statistics and those of the historical observation period is subsequently used to remove GCM bias. The original variables are recovered through summation of bias corrected differenced sequences. In the spatial disaggregation step, JVSD uses a historical analogue approach, with historical analogues identified simultaneously for all atmospheric fields and over all areas of the basin under study. Analysis and comparisons are performed for 20th Century Climate in Coupled Models (20C3M), broadly available for most GCMs. The results show that the proposed downscaling method is able to reproduce the sub-grid climatic features as well as their temporal/spatial variability in the historical periods. Comparisons are also performed for precipitation and temperature with other statistical and dynamic downscaling methods over the southeastern US and show that JVSD performs favorably. The downscaled sequences are used to assess the implications of GCM scenarios for the Apalachicola-Chattahoochee-Flint river basin as part of a comprehensive climate change impact assessment.     

Estimating hydrological characteristics in the Amu Darya River basin under climate change conditions

Under consideration are results of solving the problem of the river water content estimation under conditions of uncertainties of climate change forecasts and the catchment state with a reference to the Amu Darya River basin. When constructing regional climate models, one selected a multimodel approach using the results of several global models and a statistical downscaling method that made the climate scenarios more detailed. The estimates demonstrated that in the medium- and long-term perspective, the Amu Darya River runoff is expected to decrease. As a result of the Bayesian ideology application, using the calculations got with a total probability formula, a prognostic probability curve of an annual river runoff supply of the basin rivers was derived based on different weights given to the estimates of a mean value for different climate scenarios. Prognostic characteristics of the annual runoff for the Amu Darya basin rivers are estimated in a form acceptable for hydrologic and hydroeconomic application.     

Using Statistical Downscaling to Quantify the GCM-Related Uncertainty in Regional Climate Change Scenarios： A Case Study of Swedish Precipitation

There are a number of sources of uncertainty in regional climate change scenarios. When statistical downscaling is used to obtain regional climate change scenarios, the uncertainty may originate from the uncertainties in the global climate models used, the skill of the statistical model, and the forcing scenarios applied to the global climate model. The uncertainty associated with global climate models can be evaluated by examining the differences in the predictors and in the downscaled climate change scenarios based on a set of different global climate models. When standardized global climate model simulations such as the second phase of the Coupled Model Intercomparison Project （CMIP2） are used, the difference in the downscaled variables mainly reflects differences in the climate models and the natural variability in the simulated climates. It is proposed that the spread of the estimates can be taken as a measure of the uncertainty associated with global climate models. The proposed method is applied to the estimation of global-climate-model-related uncertainty in regional precipitation change scenarios in Sweden. Results from statistical downscaling based on 17 global climate models show that there is an overall increase in annual precipitation all over Sweden although a considerable spread of the changes in the precipitation exists. The general increase can be attributed to the increased large-scale precipitation and the enhanced westerly wind. The estimated uncertainty is nearly independent of region. However, there is a seasonal dependence. The estimates for winter show the highest level of confidence, while the estimates for summer show the least.     

Comparison of regional climate model and statistical downscaling simulations of different winter precipitation change scenarios over Romania

Summary Regional climate model and statistical downscaling procedures are used to generate winter precipitation changes over Romania for the period 2071–2100 (compared to 1961–1990), under the IPCC A2 and B2 emission scenarios. For this purpose, the ICTP regional climate model RegCM is nested within the Hadley Centre global atmospheric model HadAM3H. The statistical downscaling method is based on the use of canonical correlation analysis (CCA) to construct climate change scenarios for winter precipitation over Romania from two predictors, sea level pressure and specific humidity (either used individually or together). A technique to select the most skillful model separately for each station is proposed to optimise the statistical downscaling signal. Climate fields from the A2 and B2 scenario simulations with the HadAM3H and RegCM models are used as input to the statistical downscaling model. First, the capability of the climate models to reproduce the observed link between winter precipitation over Romania and atmospheric circulation at the European scale is analysed, showing that the RegCM is more accurate than HadAM3H in the simulation of Romanian precipitation variability and its connection with large-scale circulations. Both models overestimate winter precipitation in the eastern regions of Romania due to an overestimation of the intensity and frequency of cyclonic systems over Europe. Climate changes derived directly from the RegCM and HadAM3H show an increase of precipitation during the 2071–2100 period compared to 1961–1990, especially over northwest and northeast Romania. Similar climate change patterns are obtained through the statistical downscaling method when the technique of optimum model selected separately for each station is used. This adds confidence to the simulated climate change signal over this region. The uncertainty of results is higher for the eastern and southeastern regions of Romania due to the lower HadAM3H and RegCM performance in simulating winter precipitation variability there as well as the reduced skill of the statistical downscaling model.     

Daily relative humidity projections in an Indian river basin for IPCC SRES scenarios

A two-stage methodology is developed to obtain future projections of daily relative humidity in a river basin for climate change scenarios. In the first stage, Support Vector Machine (SVM) models are developed to downscale nine sets of predictor variables (large-scale atmospheric variables) for Intergovernmental Panel on Climate Change Special Report on Emissions Scenarios (SRES) (A1B, A2, B1, and COMMIT) to R _H in a river basin at monthly scale. Uncertainty in the future projections of R _H is studied for combinations of SRES scenarios, and predictors selected. Subsequently, in the second stage, the monthly sequences of R _H are disaggregated to daily scale using k-nearest neighbor method. The effectiveness of the developed methodology is demonstrated through application to the catchment of Malaprabha reservoir in India. For downscaling, the probable predictor variables are extracted from the (1) National Centers for Environmental Prediction reanalysis data set for the period 1978–2000 and (2) simulations of the third-generation Canadian Coupled Global Climate Model for the period 1978–2100. The performance of the downscaling and disaggregation models is evaluated by split sample validation. Results show that among the SVM models, the model developed using predictors pertaining to only land location performed better. The R _H is projected to increase in the future for A1B and A2 scenarios, while no trend is discerned for B1 and COMMIT.     

A comparison of statistical downscaling and climate change factor methods: impacts on low flows in the River Thames, United Kingdom

Strategic-scale assessments of climate change impacts are often undertaken using the change factor (CF) methodology whereby future changes in climate projected by General Circulation Models (GCMs) are applied to a baseline climatology. Alternatively, statistical downscaling (SD) methods apply climate variables from GCMs to statistical transfer functions to estimate point-scale meteorological series. This paper explores the relative merits of the CF and SD methods using a case study of low flows in the River Thames under baseline (1961–1990) and climate change conditions (centred on the 2020s, 2050s and 2080s). Archived model outputs for the UK Climate Impacts Programme (UKCIP02) scenarios are used to generate daily precipitation and potential evaporation (PE) for two climate change scenarios via the CF and SD methods. Both signal substantial reductions in summer precipitation accompanied by increased PE throughout the year, leading to reduced flows in the Thames in late summer and autumn. However, changes in flow associated with the SD scenarios are generally more conservative and complex than that arising from CFs. These departures are explained in terms of the different treatment of multidecadal natural variability, temporal structuring of daily climate variables and large-scale forcing of local precipitation and PE by the two downscaling methods.     

Long-Term Trend of Temperature Derived by Statistical Downscaling Based on EOF Analysis

This study analyzes the ability of statistical downscaling models in simulating the long-term trend of temperature and associated causes at 48 stations in northern China in January and July 1961–2006. The statistical downscaling models are established through multiple stepwise regressions of predictor principal components (PCs). The predictors in this study include temperature at 850 hPa (T850), and the combination of geopotential height and temperature at 850 hPa (H850+T850). For the combined predictors, Empirical Orthogonal Function (EOF) analysis of the two combined fields is conducted. The modeling results from HadCM3 and ECHAM5 under 20C3M and SERS A1B scenarios are applied to the statistical downscaling models to construct local present and future climate change scenarios for each station, during which the projected EOF analysis and the common EOF analysis are utilized to derive EOFs and PCs from the two general circulation models (GCMs). The results show that (1) the trend of temperature in July is associated with the first EOF pattern of the two combined fields, not with the EOF pattern of the regional warming; (2) although HadCM3 and ECHAM5 have simulated a false long-term trend of temperature, the statistical downscaling method is able to well reproduce a correct long-term trend of temperature in northern China due to the successful simulation of the trend of main PCs of the GCM predictors; (3) when the two-field combination and the projected EOF analysis are used, temperature change scenarios have a similar seasonal variation to the observed one; and (4) compared with the results of the common EOF analysis, those of the projected EOF analysis have been much more strongly determined by the observed large-scale atmospheric circulation patterns.     

Statistical downscaling of daily mean temperature, pan evaporation and precipitation for climate change scenarios in Haihe River, China

A statistical downscaling method (SDSM) was evaluated by simultaneously downscaling air temperature, evaporation, and precipitation in Haihe River basin, China. The data used for evaluation were large-scale atmospheric data encompassing daily NCEP/NCAR reanalysis data and the daily mean climate model results for scenarios A2 and B2 of the HadCM3 model. Selected as climate variables for downscaling were measured daily mean air temperature, pan evaporation, and precipitation data (1961–2000) from 11 weather stations in the Haihe River basin. The results obtained from SDSM showed that: (1) the pattern of change in and numerical values of the climate variables can be reasonably simulated, with the coefficients of determination between observed and downscaled mean temperature, pan evaporation, and precipitation being 99%, 93%, and 73%, respectively; (2) systematic errors existed in simulating extreme events, but the results were acceptable for practical applications; and (3) the mean air temperature would increase by about 0.7°C during 2011~2040; the total annual precipitation would decrease by about 7% in A2 scenario but increase by about 4% in B2 scenario; and there were no apparent changes in pan evaporation. It was concluded that in the next 30 years, climate would be warmer and drier, extreme events could be more intense, and autumn might be the most distinct season among all the changes.     

The new national climate change documents of Mexico: what do the regional climate change scenarios represent?

This paper presents a review of the methodology applied for generating the regional climate change scenarios utilized in important National Documents of Mexico, such as the Fourth National Communication to the United Nations Framework Convention on Climate Change, the Fourth National Report to the Convention on Biological Diversity and The Economics of Climate Change in Mexico. It is shown that these regional climate change scenarios, which are one of the main inputs to support the assessments presented in these documents, are an example of the erroneous use of statistical downscaling techniques. The arguments presented here imply that the work based on such scenarios should be revised and therefore, these documents are inadequate for supporting national decision- making.     

Assessment of a stochastic downscaling methodology in generating an ensemble of hourly future climate time series

This study extends a stochastic downscaling methodology to generation of an ensemble of hourly time series of meteorological variables that express possible future climate conditions at a point-scale. The stochastic downscaling uses general circulation model (GCM) realizations and an hourly weather generator, the Advanced WEather GENerator (AWE-GEN). Marginal distributions of factors of change are computed for several climate statistics using a Bayesian methodology that can weight GCM realizations based on the model relative performance with respect to a historical climate and a degree of disagreement in projecting future conditions. A Monte Carlo technique is used to sample the factors of change from their respective marginal distributions. As a comparison with traditional approaches, factors of change are also estimated by averaging GCM realizations. With either approach, the derived factors of change are applied to the climate statistics inferred from historical observations to re-evaluate parameters of the weather generator. The re-parameterized generator yields hourly time series of meteorological variables that can be considered to be representative of future climate conditions. In this study, the time series are generated in an ensemble mode to fully reflect the uncertainty of GCM projections, climate stochasticity, as well as uncertainties of the downscaling procedure. Applications of the methodology in reproducing future climate conditions for the periods of 2000–2009, 2046–2065 and 2081–2100, using the period of 1962–1992 as the historical baseline are discussed for the location of Firenze (Italy). The inferences of the methodology for the period of 2000–2009 are tested against observations to assess reliability of the stochastic downscaling procedure in reproducing statistics of meteorological variables at different time scales.     

A comparison of explicit and implicit spatial downscaling of GCM output for soil erosion and crop production assessments

Spatial downscaling of climate change scenarios can be a significant source of uncertainty in simulating climatic impacts on soil erosion, hydrology, and crop production. The objective of this study is to compare responses of simulated soil erosion, surface hydrology, and wheat and maize yields to two (implicit and explicit) spatial downscaling methods used to downscale the A2a, B2a, and GGa1 climate change scenarios projected by the Hadley Centre’s global climate model (HadCM3). The explicit method, in contrast to the implicit method, explicitly considers spatial differences of climate scenarios and variability during downscaling. Monthly projections of precipitation and temperature during 1950–2039 were used in the implicit and explicit spatial downscaling. A stochastic weather generator (CLIGEN) was then used to disaggregate monthly values to daily weather series following the spatial downscaling. The Water Erosion Prediction Project (WEPP) model was run for a wheat–wheat–maize rotation under conventional tillage at the 8.7 and 17.6% slopes in southern Loess Plateau of China. Both explicit and implicit methods projected general increases in annual precipitation and temperature during 2010–2039 at the Changwu station. However, relative climate changes downscaled by the explicit method, as compared to the implicit method, appeared more dynamic or variable. Consequently, the responses to climate change, simulated with the explicit method, seemed more dynamic and sensitive. For a 1% increase in precipitation, percent increases in average annual runoff (soil loss) were 3–6 (4–10) times greater with the explicit method than those with the implicit method. Differences in grain yield were also found between the two methods. These contrasting results between the two methods indicate that spatial downscaling of climate change scenarios can be a significant source of uncertainty, and further underscore the importance of proper spatial treatments of climate change scenarios, and especially climate variability, prior to impact simulation. The implicit method, which applies aggregated climate changes at the GCM grid scale directly to a target station, is more appropriate for simulating a first-order regional response of nature resources to climate change. But for the site-specific impact assessments, especially for entities that are heavily influenced by local conditions such as soil loss and crop yield, the explicit method must be used.     

基于时空统计降尺度的淮河流域夏季分月降水概率预测

针对淮河流域水资源短缺、洪涝、干旱并存的问题,基于国家气候中心第二代季节气候模式的集合回报数据集(1991—2014年),建立时空相结合的统计降尺度模型,提前1—3个月预测该流域夏季分月降水,应用ROC(relative operating characteristics)评分评估比较了不同集合预测方案的预测技巧。交叉检验结果表明,样本数取18、20、22、28时,集合预测方案对3、4、5月三个起报时次预测的夏季各月降水技巧预测均高于模式预测技巧。2015—2017年的独立样本检验进一步表明该统计降尺度模型能够明显降低3月、5月起报的6月和8月的降水预测偏差。认为可尝试将该降尺度方法应用于淮河流域夏季降水预测及进一步的流域水文预测。     

使用岭回归对哈萨克斯坦月平均气温的统计降尺度研究

哈萨克斯坦是世界最大的内陆国家,拥有典型的大陆性气候和多样的地理环境及生态系统,同时哈萨克斯坦的自然环境和人类社会对于气候变化这一全球性问题是敏感的、脆弱的,需要运用科学的研究方法应对气候变化的挑战。通常,区域或局地尺度的气候变化影响研究需要对气候模式输出或再分析资料进行降尺度以获得更细分辨率的气候资料。近年来,大量验证统计降尺度方法在各个地区能力的研究见诸文献,然而在哈萨克斯坦地区验证统计降尺度方法的研究非常少见。本文使用了岭回归的方法对哈萨克斯坦地区11个气象站点1960~2009年的月平均气温进行了统计降尺度研究。结果显示,使用前30年数据和岭回归模型建立大尺度预报因子和观测资料的统计关系可以较好地预测后20年的月平均气温,预测能力在各站各月均有不同程度的差异,地形复杂的站点预测效果较差,夏季预测结果好于冬季;此外,将哈萨克斯坦地区平均来看则与观测数据相吻合。     

Evaluation of an urban land surface scheme over a tropical suburban neighborhood

A two-step statistical downscaling method has been reviewed and adapted to simulate twenty-first-century climate projections for the Gulf of Fonseca (Central America, Pacific Coast) using Coupled Model Intercomparison Project (CMIP5) climate models. The downscaling methodology is adjusted after looking for good predictor fields for this area (where the geostrophic approximation fails and the real wind fields are the most applicable). The method’s performance for daily precipitation and maximum and minimum temperature is analysed and revealed suitable results for all variables. For instance, the method is able to simulate the characteristic cycle of the wet season for this area, which includes a mid-summer drought between two peaks. Future projections show a gradual temperature increase throughout the twenty-first century and a change in the features of the wet season (the first peak and mid-summer rainfall being reduced relative to the second peak, earlier onset of the wet season and a broader second peak).     

统计降尺度法对华北地区未来区域气温变化情景的预估

迄今为止,大部分海气耦合气候模式（AOGCM）的空间分辨率还较低,很难对区域尺度的气候变化情景做合理的预测。降尺度法已广泛用于弥补AOGCM在这方面的不足。作者采用统计降尺度方法对1月和7月华北地区49个气象观测站的未来月平均温度变化情景进行预估。采用的统计降尺度方法是主分量分析与逐步回归分析相结合的多元线性回归模型。首先,采用1961～2000年的 NCEP再分析资料和49个台站的观测资料建立月平均温度的统计降尺度模型,然后把建立的统计降尺度模型应用于HadCM3 SRES A2 和 B2 两种排放情景, 从而生成各个台站1950～2099年1月份和7月份温度变化情景。结果表明:在当前气候条件下,无论1月还是7月,统计降尺度方法模拟的温度与观测的温度有很好的一致性,而且在大多数台站,统计降尺度模拟气温与观测值相比略微偏低。对于未来气候情景的预估方面,无论1月还是7月,也无论是HadCM3 SRES A2 还是B2排放情景驱动统计模型,结果表明大多数的站点都存在温度的明显上升趋势,同时7月的上升趋势与1月相比偏低。