Empirical Statistical Downscaling Methods Research Articles

Regional climate model emulator based on deep learning: concept and first evaluation of a novel hybrid downscaling approach

Providing reliable information on climate change at local scale remains a challenge of first importance for impact studies and policymakers. Here, we propose a novel hybrid downscaling method combining the strengths of both empirical statistical downscaling methods and Regional Climate Models (RCMs). In the longer term, the final aim of this tool is to enlarge the high-resolution RCM simulation ensembles at low cost to explore better the various sources of projection uncertainty at local scale. Using a neural network, we build a statistical RCM-emulator by estimating the downscaling function included in the RCM. This framework allows us to learn the relationship between large-scale predictors and a local surface variable of interest over the RCM domain in present and future climate. The RCM-emulator developed in this study is trained to produce daily maps of the near-surface temperature at the RCM resolution (12 km). The emulator demonstrates an excellent ability to reproduce the complex spatial structure and daily variability simulated by the RCM, particularly how the RCM refines the low-resolution climate patterns. Training in future climate appears to be a key feature of our emulator. Moreover, there is a substantial computational benefit of running the emulator rather than the RCM, since training the emulator takes about 2 h on GPU, and the prediction takes less than a minute. However, further work is needed to improve the reproduction of some temperature extremes, the climate change intensity and extend the proposed methodology to different regions, GCMs, RCMs, and variables of interest.

Evaluation of statistical downscaling methods for climate change projections over Spain: Future conditions with pseudo reality (transferability experiment)

AbstractThe Spanish Meteorological Agency (AEMET) is responsible for the elaboration of downscaled climate projections over Spain to feed the Second National Plan of Adaptation to Climate Change (PNACC‐2) and this is the last of three papers aimed to evaluate and intercompare five empirical/statistical downscaling (ESD) methods developed at AEMET: (a) Analog, (b) Regression, (c) Artificial Neural Networks, (d) Support Vector Machines and (e) Kernel Ridge Regression, in order to decide which methods and under what configurations are more suitable for that purpose. Following the framework established by the EU COST Action VALUE, in this experiment we test the transferability of these methods to future climate conditions with the use of regional climate models (RCMs) as pseudo observations. We evaluate the marginal aspects of the distributions of daily maximum/minimum temperatures and daily accumulated precipitation, over mainland Spain and the Balearic Islands, analysed by season. For maximum/minimum temperatures all methods display certain transferability issues, being remarkable for Support Vector Machines and Kernel Ridge Regression. For precipitation all methods appear to suffer from transferability difficulties as well, although conclusions are not as clear as for temperature, probably due to the fact that precipitation does not present such a marked signal of change. This study has revealed how an analysis over a historical period is not enough to fully evaluate ESD methods, so we propose that some type of analysis of transferability should be added in a standard procedure of a complete evaluation.

Subsampling Impact on the Climate Change Signal over Poland Based on Simulations from Statistical and Dynamical Downscaling

ABSTRACTMost impact studies using downscaled climate data as input assume that the selection of few global climate models (GCMs) representing the largest spread covers the likely range of future changes. This study shows that including more GCMs can result in a very different behavior. We tested the influence of selecting various subsets of GCMs on the climate change signal over Poland from simulations based on dynamical and empirical–statistical downscaling methods. When the climate variable is well simulated by the GCM, such as temperature, results showed that both downscaling methods agree on a warming over Poland by up to 2° or 5°C assuming intermediate or high emission scenarios, respectively, by 2071–2100. As a less robust simulated signal through GCMs, precipitation is expected to increase by up to 10% by 2071–2100 assuming the intermediate emission scenario. However, these changes are uncertain when the high emission scenario and the end of the twenty-first century are of interest. Further, an additional bootstrap test revealed an underestimation in the warming rate varying from 0.5° to more than 4°C over Poland that was found to be largely influenced by the selection of few driving GCMs instead of considering the full range of possible climate model outlooks. Furthermore, we found that differences between various combinations of small subsets from the GCM ensemble of opportunities can be as large as the climate change signal.

Wind Power Potential in Near Future Climate Scenarios: The Case for Burundi (East Africa)

This paper assessed projected changes of wind power potential in near future climate scenarios over four sites from two contrasting regions of Burundi. Observed and MERRA-2 data sets were considered for the historical period 1981-2010, and a computed Multi-model ensemble for future projections data of eight Regional Climate Models under RCP 4.5 and 8.5 over the period 2011-2040 was used. Regional Climate Models were downscaled at local climate using Empirical Statistical Downscaling method. Mann-Kendall’s test was used for trend analysis over the historical period, while future changes in wind power density (WPD) quartiles were computed for each climate scenario by 2040. The findings revealed an increase in wind power potential all over the area studied with higher values during summer time. Indeed, over the period 2011-2040, the lowest WPD change is projected at Northern highlands (NHL) under RCP 4.5 with 27.03 W.m-2, while the highest WPD change of 46.34 W.m-2 is forecasted under RCP 8.5 at Southern Imbo plain (SIP). The month of August and September are expected to have higher WPD change in RCP 4.5 and RCP 8.5, respectively while January is projected to have the lowest WPD. Places near by the Lake Tanganyika are the most favorable areas for wind power generation.

Evaluation of the empirical\u2013statistical downscaling method EPISODES

A comprehensive performance assessment of the empirical–statistical downscaling (ESD) technique named EPISODES is presented. Pertaining evaluation analyses consist of multifarious validation experiments as well as various comparisons of EPISODES’ projections with those of three RCMs and two ESD methods based on the same GCM scenarios driven by two distinct representative concentration pathways (RCPs). EPISODES combines the downscaling of GCM simulations with a follow-up production of synthetic local time series. EPISODES is a comparably simple, computationally rather inexpensive technique, providing multi-variable and multi-site data that are suitable for being merged in an ensemble of RCM projections. This allows (e.g. for different RCPs) the compilation of large multi member ensembles derived from various GCM simulations via both main downscaling strategies (ESD and RCMs). Evaluation experiments reveal satisfying degrees of compliance between various results generated by EPISODES and observations. The grid cell bias for yearly values, for instance, is mostly less than 0.1,^circC for temperature and 10% for precipitation totals. Recorded temperature values and precipitation totals corresponding to their 1st and the 99th percentiles are well represented by EPISODES too. Comparisons of various climate change signals derived by EPISODES and other downscaling approaches, present high levels of agreement as well. Many more findings referring to evaluation experiments and climate change projections are to be found in the paper as well as throughout the “Appendix”.

Evaluating the stationarity assumption in statistically downscaled climate projections: is past performance an indicator of future results?

Empirical statistical downscaling (ESD) methods seek to refine global climate model (GCM) outputs via processes that glean information from a combination of observations and GCM simulations. They aim to create value-added climate projections by reducing biases and adding finer spatial detail. Analysis techniques, such as cross-validation, allow assessments of how well ESD methods meet these goals during observational periods. However, the extent to which an ESD method’s skill might differ when applied to future climate projections cannot be assessed readily in the same manner. Here we present a “perfect model” experimental design that quantifies aspects of ESD method performance for both historical and late 21st century time periods. The experimental design tests a key stationarity assumption inherent to ESD methods – namely, that ESD performance when applied to future projections is similar to that during the observational training period. Case study results employing a single ESD method (an Asynchronous Regional Regression Model variant) and climate variable (daily maximum temperature) demonstrate that violations of the stationarity assumption can vary geographically, seasonally, and with the amount of projected climate change. For the ESD method tested, the greatest challenges in downscaling daily maximum temperature projections are revealed to occur along coasts, in summer, and under conditions of greater projected warming. We conclude with a discussion of the potential use and expansion of the perfect model experimental design, both to inform the development of improved ESD methods and to provide guidance on the use of ESD products in climate impacts analyses and decision-support applications.

Robust climate scenarios for sites with sparse observations: a two‐step bias correction approach

ABSTRACTObserved and projected climatic changes demand for robust assessments of climate impacts on various environmental and anthropogenic systems. Empirical‐statistical downscaling (ESD) methods coupled to output from climate model projections are promising tools to assess impacts at regional to local scale. ESD methods correct for common model deficiencies in accuracy (e.g. model biases) and scale (e.g. grid vs point scale). However, most ESD methods require long observational time series at the target sites, and this often restricts robust impact assessments to a small number of sites. This paper presents a method to generate robust climate model based scenarios for target sites with short and (or) sparse observational data coverage. The approach is based on the well‐established quantile mapping method and incorporates two major steps: (1) climate model bias correction to the most representative station with long‐term measurements and (2) spatial transfer of bias‐corrected model data to represent target site characteristics. Both steps are carried out using the quantile mapping technique. The resulting output can serve as end user–tailored input for climate impact models. The method allows for multivariate and multi‐model ensemble scenarios and additionally enables to approximately reconstruct data for non‐measured periods. The method's applicability is validated using (1) long‐term weather stations across the topographically and climatologically complex territory of Switzerland and (2) sparse data sets from Swiss permafrost research sites located in challenging conditions at high altitudes. It is shown that the two‐step approach performs well and offers attractive quality, even for extreme target locations. Uncertainties, however, remain and primarily depend on (1) data availability and (2) the considered variable. The two‐step approach itself involves large uncertainties when applied to short reference data sets or spatially heterogeneous variables (e.g. precipitation, wind speed). For temperature, results are promising even when using very short calibration periods.

Comparison of empirical statistical methods for downscaling daily climate projections from CMIP5 GCMs: a case study of the Huai River Basin, China

Empirical statistical downscaling methods are becoming increasingly popular in climate change impact assessments that require downscaling multi-global climate model (GCM) projections. Here, empirical statistical downscaling methods are classified based on calibration strategies [bias correction (BC) and change factor (CF)] and statistical transformations (mean based, variance based, quantile mapping, quantile correcting and transfer function methods). Ten combinations of calibration strategies and transformation methods are used to represent a range of empirical statistical downscaling methods. To test the performance of these methods in downscaling daily precipitation and temperature, an inter-model cross-validation is carried out using an ensemble of 16 GCMs from the Coupled Model Intercomparison Project Phase 5 (CMIP5) dataset over the Huai River Basin in China. These downscaling methods are further applied to downscale the climate for the future period to assess the associated uncertainties. The results show that the CF-based methods outperform the BC-based methods in projecting the probability distribution of downscaled daily temperature, while both calibration strategies give comparable results in the case of precipitation. With the CF calibration strategy, simply adding (for temperature) or multiplying (for precipitation) the mean CF is sufficient to represent most of the relative changes projected by GCMs. The use of quantile-based methods appears to be advantageous only at the tails of the distribution. More sophisticated BC methods are needed to remove the biases in the higher-order statistics of the GCM outputs. The two calibration strategies lead to fundamentally different temporal structures and spatial variability of the downscaled climatic variables. The BC-based methods produce larger uncertainty bands of inter-annual variability than the CF-based methods. For downscaled future precipitation, the uncertainty arising from the downscaling methods is comparable to the uncertainty arising from GCMs, while more uncertainty is introduced by calibration strategies than by statistical transformation methods.

Reliability of day-ahead solar irradiance forecasts on Reunion Island depending on synoptic wind and humidity conditions

Day ahead solar irradiance forecasting is essential to manage solar energy systems and assist decision-making. Available global and mesoscale Numerical Weather Prediction (NWP) models show limitations for regions with high complex terrain in solar irradiance forecasts, due to the usual coarse resolution of the models and the difficulties in reproducing the correct physics (in terms of cloud evolution) at the correct moment. This is the case of Reunion Island (21°S, 55.5°E) a 60-km wide tropical island located in the Southwest of the Indian Ocean basin.This work explores the use of NWP thermodynamic variables at synoptic scale as predictors of solar irradiance as they are commonly considered robust and well predicted by these models. Our analyses show that the diurnal cycle of solar irradiance on Reunion Island is sensitive to synoptic wind and relative humidity at 700hPa (RH700). Moreover, this sensitivity varies depending on the site location on the Island with respect to synoptic wind direction. Hence, stations that are most exposed to synoptic wind, show greatest sensitivity. From these results, we have developed an empirical-statistical downscaling (ESD) method for day-ahead local-scale solar irradiance forecast and forecast reliability for the Reunion Island case, which searches for past analogous conditions of synoptic wind and RH700. Our study uses 12years of ECMWF reanalysis data together with ground-based measurements at 21 stations on the island. Forecasting results for 4 of these stations plus an aggregate virtual station (derived as the mean of 11 coastal stations) show that the ESD method outperforms a climatology-based method in all cases. The contribution of this method for solar irradiance forecasting and its reliability are discussed.

Interrogating empirical-statistical downscaling

The delivery of downscaled climate information is increasingly seen as a vehicle of climate services, a driver for impacts studies and adaptation decisions, and for informing policy development. Empirical-statistical downscaling (ESD) is widely used; however, the accompanying responsibility is significant, and predicated on effective understanding of the limitations and capabilities of ESD methods. There remain substantial contradictions, uncertainties, and sensitivity to assumptions between the different methods commonly used. Yet providing decision-relevant downscaled climate projections to help support national and local adaptation is core to the growing global momentum seeking to operationalize what is, in effect, still foundational research. We argue that any downscaled climate information must address the criteria of being plausible, defensible and actionable. Climate scientists cannot absolve themselves of their ethical responsibility when informing adaptation and must, therefore, be diligent in ensuring any information provided adequately addresses these three criteria. Frameworks for supporting such assessment are not well developed. We interrogate the conceptual foundations of statistical downscaling methodologies and their assumptions, and articulate a framework for evaluating and integrating downscaling output into the wider landscape of climate information. For ESD there are key criteria that need to be satisfied to underpin the credibility of the derived product. Assessing these criteria requires the use of appropriate metrics to test the comprehensive treatment of local climate response to large-scale forcing, and to compare across methods. We illustrate the potential consequences of methodological choices on the interpretation of downscaling results and explore the purposes, benefits and limitations of using statistical downscaling.

The Weather Generator Used in the Empirical Statistical Downscaling Method, WETTREG

In this paper, the weather generator (WG) used by the empirical statistical downscaling method, weather situation-based regionalization method (in German: WETTerlagen-basierte REGionalisierungsmethode, WETTREG), is described. It belongs to the class of multi-site parametric models that aim at the representation of the spatial dependence among weather variables with conditioning on exogenous atmospheric predictors. The development of the WETTREG WG was motivated by (i) the requirement of climate impact modelers to obtain input data sets that are consistent and can be produced in a relatively economic way and (ii) the well-sustained hypothesis that large scale atmospheric features are well reproduced by climate models and can be used as a link to regional climate. The WG operates at daily temporal resolution. The conditioning factor is the temporal development of the frequency distribution of circulation patterns. Following a brief description of the strategy of classifying circulation patterns that have a strong link to regional climate, the bulk of this paper is devoted to a description of the WG itself. This includes aspects, such as the utilized building blocks, seasonality or the methodology with which a signature of climate change is imprinted onto the generated time series. Further attention is given to particularities of the WG’s conditioning processes, as well as to extremes, areal representativity and the interface of WGs and user requirements.

Climate Signals on the Regional Scale Derived with a Statistical Method: Relevance of the Driving Model’s Resolution

When assessing the magnitude of climate signals in a regional scale, a host of optional approaches is feasible. This encompasses the use of regional climate models (RCM), nested into global climate models (GCM) for an area of interest as well as employing empirical statistical downscaling methods (ESD). In this context the question is addressed: Is an empirical statistical downscaling method capable of yielding results that are comparable to those by dynamical RCMs? Based on the presented ESD method, the comparison of RCM and ESD results show a high amount of agreement. In addition the empirical statistical downscaling can be applied directly to a GCM or a GCM-RCM cascade. The paper aims at comparing the consequences of employing various CGM-RCM-ESD combinations on the derived future changes of temperature and precipitation. This adds to the insight on the scale dependency of downscaling strategies. Results for one GCM with several scenario runs driving several RCMs with and without subsequent empirical statistical downscaling are presented. It is shown that there are only small differences between using the GCM results directly or as a GCM-RCM-ESD cascade.