Search

The accurate prediction of scour depth in culverts is crucial to ensure public safety and due to the uncertainties in classical empirical/deterministic equations, it remains challenging. This study presents new ensemble multi-models to predict scour depth and quantify model bias and uncertainty. In this study, culvert scour was predicted by gene expression programming (GEP) and least square support vector machine (LSSVM), firstly based on dimensionless parameters. In the multi-model ensemble strategy, four ensemble approaches, namely SM, WM, LSSVM, and BMA were applied. The LSSVM showed overall best skill for the multi-model approach, that reduced the error by 61.3% and increased the coefficient of determination by 52.1%. The results showed the proposed approach can satisfy the requirements of simplicity, applicability, and accuracy at the same time. For application purposes, we provided a simple code to obtain scour depth with great accuracy using LSSVM. BMA and LSSVM were used to quantify the scour culvert uncertainty. The results of the uncertainty analysis showed that the LSSVM predicts scour depth quite well and generates high-quality prediction intervals. The results revealed that a multi-model ensemble strategy improved accuracy, certainty and reliability for culvert scour compared to empirical equations. [ABSTRACT FROM AUTHOR]

Copyright of Atmosphere -- Ocean (Taylor & Francis Ltd) is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Extreme climate change events are major causes of devastating impacts on socioeconomic well-being and ecosystem damage. Therefore, understanding the performance of appropriate climate models representing local climate characteristics is critical for future projections. Thus, this study analyses the performance of 24 GCMs from the Coupled Model Intercomparison Project Phases 5 and 6 (CMIP5 and 6) and their multi-model ensembles in simulating climate variables including average rainfall, maximum (Tmax), and minimum (Tmin) temperatures at annual and seasonal scales over the Chungcheong region of South Korea from 1975 to 2015. A trend analysis was conducted to estimate the future trends in climate variables in the 2060s (2021–2060) and 2080s (2061–2100). Inverse distance weighting and quantile delta mapping were applied to bias-correct the GCM data. Further, six major evaluating indices comprising temporal and spatial performance assessments were used, after which a comprehensive GCM ranking was applied. The results showed that CMIP6 models performed better in simulating rainfall, Tmax, and Tmin at both temporal and spatial scales. For CMIP5, the top three performing models were GISS, ACCESS1-3, and MRI-CGCM3 for rain; CanESM2, GISS, and MPI-ESM-L-R for Tmax; and GFDL, MRI-CGCM3, and CanESM2 for Tmin. However, the top three performing models in the CMIP6 were MRI-ESM2-0, BCC_CSM, and GFDL for rain; MIROC6, BCC_CSM, and MRI-ESM2-0 for Tmax, and GFDL, MPI_ESM_HR, and MRI-ESM2-0 for Tmin. The multi-model ensembles (an average of the top three GCMs) performed better in simulating rain and Tmin for both CMIP5 and CMIP6 compared with multi-model ensembles (an average of all the GCMs), which only performed slightly better in simulating Tmax. The trend analysis of future projection indicates an increase in rain, Tmax, and Tmin; however, with distinct changes under similar radiative forcing levels in both CMIP5 and CMIP6 models. The projections under RCP4.5 and RCP8.5 increase more than the SSP2-4.5 and SSP5-8.5 scenarios for most climate conditions but are more pronounced, especially for rain, under RCP8.5 than SSP5-8.5 in the far future (2080s). This study provides insightful findings on selecting appropriate GCMs to generate reliable climate projections for local climate conditions in the Chungcheong region of South Korea. [ABSTRACT FROM AUTHOR]

Extreme climate change events are major causes of devastating impacts on socioeconomic well-being and ecosystem damage. Therefore, understanding the performance of appropriate climate models representing local climate characteristics is critical for future projections. Thus, this study analyses the performance of 24 GCMs from the Coupled Model Intercomparison Project Phases 5 and 6 (CMIP5 and 6) and their multi-model ensembles in simulating climate variables including average rainfall, maximum (Tmax), and minimum (Tmin) temperatures at annual and seasonal scales over the Chungcheong region of South Korea from 1975 to 2015. A trend analysis was conducted to estimate the future trends in climate variables in the 2060s (2021–2060) and 2080s (2061–2100). Inverse distance weighting and quantile delta mapping were applied to bias-correct the GCM data. Further, six major evaluating indices comprising temporal and spatial performance assessments were used, after which a comprehensive GCM ranking was applied. The results showed that CMIP6 models performed better in simulating rainfall, Tmax, and Tmin at both temporal and spatial scales. For CMIP5, the top three performing models were GISS, ACCESS1-3, and MRI-CGCM3 for rain; CanESM2, GISS, and MPI-ESM-L-R for Tmax; and GFDL, MRI-CGCM3, and CanESM2 for Tmin. However, the top three performing models in the CMIP6 were MRI-ESM2-0, BCC_CSM, and GFDL for rain; MIROC6, BCC_CSM, and MRI-ESM2-0 for Tmax, and GFDL, MPI_ESM_HR, and MRI-ESM2-0 for Tmin. The multi-model ensembles (an average of the top three GCMs) performed better in simulating rain and Tmin for both CMIP5 and CMIP6 compared with multi-model ensembles (an average of all the GCMs), which only performed slightly better in simulating Tmax. The trend analysis of future projection indicates an increase in rain, Tmax, and Tmin; however, with distinct changes under similar radiative forcing levels in both CMIP5 and CMIP6 models. The projections under RCP4.5 and RCP8.5 increase more than the SSP2-4.5 and SSP5-8.5 scenarios for most climate conditions but are more pronounced, especially for rain, under RCP8.5 than SSP5-8.5 in the far future (2080s). This study provides insightful findings on selecting appropriate GCMs to generate reliable climate projections for local climate conditions in the Chungcheong region of South Korea.

In the context of global warming, the frequency and intensity of extreme weather and climate events are increasing. However, the impact of these changes that is directly felt by people is the day-to-day temperature change. Extreme temperature changes between neighboring days (ETCNs) carry substantial disease risks and socioeconomic impacts. Evaluation studies of ETCN events with global climate models (GCMs) remain unknown in China. This study quantitatively evaluates the performances of 35 GCMs and the multi-model ensemble (MME) of the Coupled Model Intercomparison Project 6 (CMIP6) in simulating the extreme cooling (EC) and extreme warming (EW) events of two consecutive days as defined by relative thresholds. The results showed that from 1981 to 2013, the annual average EW frequencies showed an increasing trend over China, but a decreasing trend for EC events, and the frequency of EW events was higher than that of EC events. EW events mostly occurred in spring, while EC events occurred in autumn. Additionally, the performances of the CMIP6 models were quite different between EC and EW events. The models could capture the annual cycle of EC and EW events well, and the simulations of EW events were generally more reliable than those of EC events. Furthermore, most CMIP6 models overestimated the frequency of EW events but underestimated the frequency of EC events in China. The CMIP6 models could capture the trends in EC events in China but fail to simulate them in EW events. The interannual variability of EW events exhibited relatively better performance than that of EC events. The CMIP6 MME effectively improved the capabilities of the models to simulate the climatology of ETCN events. Individual CMIP6 models exhibited better performances than the CMIP6 MME in terms of the trend and interannual variability. Finally, according to the overall ranking of the CMIP6 models, MPI-ESM-1–2-HAM and FGOALS-f3-L achieved the best performance in simulating EW and EC events, respectively. This study selected the optimal models in different regions at the seasonal and annual scales, providing theoretical support for the frequency projection and modeling improvement of ETCN events. [ABSTRACT FROM AUTHOR]

For regional‐scale studies on climate change and relevant impact assessment, the projections of regional climate models (RCMs) are used due to their advantage of high resolution and better representation of the local climate relative to the global climate models. However, direct use of RCM outputs is prone to uncertainties and biases that may significantly diminish the accuracy of results. EURO‐COordinated Regional Downscaling EXperiment (CORDEX) initiative that is a part of the global Coordinated Regional Downscaling Experiment Project provides high‐resolution RCM projections for the European domain and bias‐adjusted regional projections under the "CORDEX‐Adjust" Project for climate change impact assessment studies. This study aims to perform a multi‐model analysis of precipitation data using bias‐adjusted and raw/non‐bias adjusted CORDEX RCMs to obtain an evaluation of their representativeness for local climate conditions and adequacy to be used for climate change impact assessment. For this purpose, the analysis focuses on four CORDEX RCMs and their 12 bias‐adjusted versions generated with cumulative distribution function transformation, quantile mapping, and distribution‐based scaling methodologies. For the analysis in total, 16 hindcast results of raw and bias‐adjusted RCMs, and three superensembles (SEs) generated through multiple linear regression are compared for their performance regarding their goodness of fit to the ground‐based precipitation monitoring data from eight meteorological stations in the Mediterranean region in Turkey. The analysis verified that the skill of individual simulations including the bias‐adjusted outputs is significantly variable in spatial and temporal means. On the other hand, SE formed by using all 16 hindcast outputs has the highest skill for the representation of variability in precipitation in time as well as for the reproduction of annual climatology at all stations, although potential drawbacks concerning seasonality and the range of anomaly may still exist which might be significant depending on the specific aim of impact assessment. [ABSTRACT FROM AUTHOR]

The shifts in socio-economic development and climate conditions currently become the challenge for water resources system security in the Himalayan region. The aforesaid concern was found pertinent to the main objective of this study, which is to evaluate the combined impacts of climate and socioeconomic changes on likely future water security in the Himalayan basin, India. The future climate was projected by Multi-model Ensembles under the Representative Concentration Pathway (RCP) 4.5 scenario. Land use projection under the Shared Socioeconomic Pathway (SSP) 1 scenario was performed using Markov Chain, whose transition probabilities were derived using multi-layer perceptron neural networks. The results showed that future annual precipitation and temperature at the downstream part will increase from baseline by 5% - 10% and 1.0oC - 1.55oC, respectively. The land use projections showed that irrigated areas will decrease for Punjab by 10% - 30% and Haryana by 5% - 10% due to the increase in urbanisation, whereas it will be increased in Rajasthan by 12% - 18%. Consequently, the annual irrigation water demand was found to be decreased by 10% for Punjab and 5% for Haryana, while it will be increased by 13% for Rajasthan. Eventually, the obtained findings will be beneficial for planning strategies to ensure water security in the Himalayan region, in particular the Beas-Sutlej basin. [ABSTRACT FROM AUTHOR]

High-frequency eddy-covariance measurements of net ecosystem CO2 exchange (NEE) with the atmosphere are valuable resources for model parameterization, calibration, and validation. However, uncertainties in measured data, i.e., data gaps and inherent random errors, create problems for researchers attempting to quantify uncertainties in model projections of terrestrial ecosystem carbon cycling. Here, we demonstrate that a model-data fusion method (residual bootstrap) produces defensible annual NEE sums, through mimicking the behavior of random errors, filling missing values, and simulating gap-filling biases. This study estimated annual NEE sums for 53 site-years based on nine eddy-covariance tower sites in the USA, and found that our annual estimates were, in most cases, comparable in magnitude with those obtained from AmeriFlux gap-filled data. Additionally, compared to the AmeriFlux standardized gap-filling, our approach provides better NEE estimates for moderate to longer, and more frequent, data gaps. Annual accumulated uncertainties in NEE at the 95% confidence level were ±30gCm−2year−1 for evergreen needleleaf forests; ±60gCm−2year−1 for deciduous broadleaf forests; and ±80gCm−2year−1 for croplands. The residual bootstrap performed worst when gap length was greater than one month or data exclusion greater than 90% during the growing season, common to other gap-filling techniques. However, this study produced robust results for most site years when monthly data coverage during the growing season is not extremely low. We therefore suggest that the inclusion of NEE uncertainty estimates and better estimation for moderate to longer, and more frequent, data gaps as provided by the residual bootstrap approach can be beneficial for ecosystem model evaluation.

Abstract Reconstructing past climates remains a difficult task because pre‐instrumental observational networks are composed of geographically sparse and noisy paleoclimate proxy records that require statistical techniques to inform complete climate fields. Traditionally, instrumental or climate model statistical relationships are used to spread information from proxy measurements to other locations and to other climate variables. Here ensembles drawn from single climate models and from combinations of multiple climate models are used to reconstruct temperature variability over the last millennium in idealized experiments. We find that reconstructions derived from multi‐model ensembles produce lower error than reconstructions from single‐model ensembles when reconstructing independent model and instrumental data. Specifically, we find the largest decreases in error over regions far from proxy locations that are often associated with large uncertainties in model physics, such as mid‐ and high‐latitude ocean and sea‐ice regions. Furthermore, we find that multi‐model ensemble reconstructions outperform single‐model reconstructions that use covariance localization. We propose that multi‐model ensembles could be used to improve paleoclimate reconstructions in time periods beyond the last millennium and for climate variables other than air temperature, such as drought metrics or sea ice variables.

Reconstructing past climates remains a difficult task because pre‐instrumental observational networks are composed of geographically sparse and noisy paleoclimate proxy records that require statistical techniques to inform complete climate fields. Traditionally, instrumental or climate model statistical relationships are used to spread information from proxy measurements to other locations and to other climate variables. Here ensembles drawn from single climate models and from combinations of multiple climate models are used to reconstruct temperature variability over the last millennium in idealized experiments. We find that reconstructions derived from multi‐model ensembles produce lower error than reconstructions from single‐model ensembles when reconstructing independent model and instrumental data. Specifically, we find the largest decreases in error over regions far from proxy locations that are often associated with large uncertainties in model physics, such as mid‐ and high‐latitude ocean and sea‐ice regions. Furthermore, we find that multi‐model ensemble reconstructions outperform single‐model reconstructions that use covariance localization. We propose that multi‐model ensembles could be used to improve paleoclimate reconstructions in time periods beyond the last millennium and for climate variables other than air temperature, such as drought metrics or sea ice variables. Plain Language Summary: Understanding past climate variability is important for contextualizing climate change as well as for testing the ability of climate models to simulate the climate system before global warming. However, reconstructing past climate variability remains a complex task because pre‐instrumental paleoclimate proxy records, such as tree rings, corals, ice cores, and sediment cores, are geographically sparse and are not perfect recorders of climate information. Exactly how to extrapolate information from paleoclimate proxies to other locations and climate variables remains an outstanding issue. Traditionally, information about how one location varies with another location or variable (covariance) is derived from one climate model or from one instrumental data source. Here we find that reconstructions using covariance estimated from combinations of multiple climate models produce less error than reconstructions that use just one climate model. Key Points: Ensembles drawn from single climate models and combinations of multiple models are used to reconstruct surface air temperature variabilityReconstructions from multi‐model ensembles produce lower error than reconstructions from single‐model ensemblesReconstructions from multi‐model ensembles show the largest decreases in error in regions with few observations such as high‐latitude oceans [ABSTRACT FROM AUTHOR]

This paper reviews the sources of uncertainty in physical climate impact assessments. It draws on examples from related fields such as climate modelling and numerical weather prediction in discussing how to interpret the results of multi-model ensembles and the role of model evaluation. Using large-scale, multi-model simulations of hydrological extremes as an example, we demonstrate how large uncertainty at the local scale does not preclude more robust conclusions at the global scale. Finally, some recommendations are made: climate impact studies should be clear about the questions they want to address, transparent about the uncertainties involved, and honest about the assumptions being made. [ABSTRACT FROM AUTHOR]

Multi-model ensembles are one of the most common ways to deal with epistemic uncertainty in hydrology. This is a problem because there is no known way to sample models such that the resulting ensemble admits a measure that has any systematic (i.e., asymptotic, bounded, or consistent) relationship with uncertainty. Multi-model ensembles are effectively sensitivity analyses and cannot - even partially - quantify uncertainty. One consequence of this is that multi-model approaches cannot support a consistent scientific method - in particular, multi-model approaches yield unbounded errors in inference. In contrast, information theory supports a coherent hypothesis test that is robust to (i.e., bounded under) arbitrary epistemic uncertainty. This paper may be understood as advocating a procedure for hypothesis testing that does not require quantifying uncertainty, but is coherent and reliable (i.e., bounded) in the presence of arbitrary (unknown and unknowable) uncertainty. We conclude by offering some suggestions about how this proposed philosophy of science suggests new ways to conceptualize and construct simulation models of complex, dynamical systems. [ABSTRACT FROM AUTHOR]

Large ensembles of climate models are increasingly available either as ensembles of opportunity or perturbed physics ensembles, providing a wealth of additional data that is potentially useful for improving adaptation strategies to climate change. In this work, we propose a framework to evaluate the predictive capacity of 11 multi-model ensemble methods (MMEs), including random forest (RF), to estimate reference evapotranspiration (ET0) using 10 AR5 models for the scenarios RCP4.5 and RCP8.5. The study was carried out in the Segura Hydrographic Demarcation (SE of Spain), a typical Mediterranean semiarid area. ET0 was estimated in the historical scenario (1970–2000) using a spatially calibrated Hargreaves model. MMEs obtained better results than any individual model for reproducing daily ET0. In validation, RF resulted more accurate than other MMEs (Kling–Gupta efficiency (KGE) M=0.903, SD=0.034 for KGE and M=3.17, SD=2.97 for absolute percent bias). A statistically significant positive trend was observed along the 21st century for RCP8.5, but this trend stabilizes in the middle of the century for RCP4.5. The observed spatial pattern shows a larger ET0 increase in headwaters and a smaller increase in the coast.

• Forecasting Brazilian wheat yield two months before harvest with < 8% RMSE. • Using monthly temperature and rainfall features from four locations from Aug-Oct. • Employing forecasted features from seasonal climate models. • Comparing the approach with features from climatology and multi-model-ensembles. • Providing a transparent, reproducible, and data-inexpensive approach. National wheat yield depends on climate conditions and usually remains unknown until harvest. In-season knowledge can be provided by wheat yield forecast systems, supporting the decision-making of farmers, food traders, or policymakers. In this study, we improved a previously developed statistical wheat yield model to forecast trend-corrected wheat yield in Brazil with monthly temperature and precipitation data from seasonal climate models (SCM) from the last three months before harvest. We chose SCM from the European Center for Medium-Range Weather Forecasts (ECMWF), the National Centers for Environmental Prediction (NCEP), and the UK-based Met-Office (UKMO). A multi-model ensembles (MME) approach from the three individual models as well as a climatology (CLIMATE) approach were also tested. Wheat yield forecasts were issued at the beginning of each month from planting in April to harvest in November. Each month, features from future months are forecasted by SCM, and past features are supplemented with observations from weather stations. Our approach shows a 12% RMSE in forecasting yield early in the season, from April to June. Forecasts start to improve from July onwards, with shorter lead times and including observed features from September onwards. At the beginning of October, about two months before harvest is completed, wheat yield can be forecasted with 7.6%, 7.9%, 7.9%, 9.1%, and 9.3% RMSE using climate data from UKMO, ECMWF, MME, NCEP, and CLIMATE respectively. Seasonal climate models can be useful tools to forecast national wheat yield, even shortly before harvest to prepare for possible food shortages. Our approach could be applied to other staple crops and regions. [ABSTRACT FROM AUTHOR]

This study examines the change in daily characteristics of seasonal precipitation under 1.5 °C, 2.0 °C, and 3.0 °C global warming levels (GWLs) above the pre-industrial period over four southern African basins. We compare the projected changes among large ensembles of climate models during sub-seasons of the rainy season. In particular, we used future projections under the highest emission scenario from the Coupled Model Intercomparison Project (CMIP) versions 5 and 6, the Coordinated Regional Climate Downscaling Experiment (CORDEX), and the Coordinated Output for Regional Evaluations (CORE). The results show that all ensemble projects increase in the length of dry spells (CDD) and daily rainfall intensity (SDII), while wet spell duration (CWD) and the number of rainy days (RR1) will decrease under all GWLs. However, discrepancies are found in the signs of change for indices of heavy precipitation (R10mm and R20mm), especially among regional climate model (RCM) ensembles (CORDEX and CORE). The discrepancies among RCM ensembles may be associated with the difference in spatial resolution, number of downscaled global climate models (GCMs) and downscaling RCMs. Spatial inconsistencies are also found in terms of the magnitude and robustness of future changes under all GWLs. Overall, the projected changes are uncertain over most parts of the basins in CMIP6, whereas CORE projects a stronger signal and larger areas of robust changes over the basins under all GWLs compared to other ensembles. Stronger signals and larger areas of robust change in RCMs may be associated with an improved ability to capture topographical characteristics in RCMs compared to GCMs. Our results show that even if the Paris Agreement goals are met, some areas of the basin will be exposed to robust changes in extreme precipitation. However, our findings highlight that limiting global warming at 1.5 °C will reduce the exposure of southern Africa's major basins to global warming-induced risks. These findings are useful for Southern African policymakers to refine adaptation and mitigation strategies to account for warming-induced extreme precipitation. • Future changes in extreme precipitation over southern Africa major basins • Changes in extreme precipitation under 1.5, 2.0 and 3.0 global warming levels • Southern Africa extreme precipitation under global warming • Comparison of future changes in extreme precipitation from CMIP and CORDEX multi-model ensembles [ABSTRACT FROM AUTHOR]

We have quantified and examined the South Pacific convergence zone (SPCZ) characteristics for the purpose of its seasonal prediction, by defining two orientation indices, strength and area. The multi-model ensemble (MME) tends to simulate the ENSO-associated shift of SPCZ orientation, especially for the 1-month forecast lead. The migration of the SPCZ orientation indices associated with ENSO phases is clear in the observation and the MME. The variation of the SPCZ strength and area associated with ENSO phases is not as clear as in the SPCZ orientation. In spite of marginal changes in the SPCZ strength and area related to ENSO phases, the SPCZ strength becomes a bit stronger during El Niño and weaker during La Niña, which is represented in individual models and MME. The performance of the MME in simulating the variability of the SPCZ orientation, strength and area is also examined. We found that the MME reasonably predicts the observed interannual variability of the western portion of the SPCZ, with systematic and marginal shift southward. Compared to the western part of the SPCZ, the MME seems to have a limitation in predicting the variability of the eastern part. In comparison to the SPCZ orientation, the MME is not capable of predicting the strength and area of the SPCZ. The interannual variability of the SPCZ strength in the MME is systematically weaker compared to that in the analysis. By comparison with SPCZ orientation and strength, the SPCZ area is not resolved in the MME. The SPCZ is a main source of precipitation in the South Pacific, and the SPCZ predictability also influences high impact weather prediction such as tropical cyclones. Therefore, skillful predictions of seasonal variability of the SPCZ could benefit users who utilize the seasonal forecasting information for their decision making in many applicable sectors. [ABSTRACT FROM AUTHOR]

ABSTRACT: Regional climate models (RCMs) from the North American Regional Climate Change Assessment Program (NARCCAP) are compared with the two gridded precipitation data sets [Climate Prediction Center (CPC) and the University of Washington (UW)] and the North American regional reanalysis (NARR) to examine if RCMs are able to reproduce very heavy precipitation under similar physical conditions seen in observations. The analysis focuses on contemporary climate (1982–1999) in an upper Mississippi region during the summer (June–July–August) months and utilizes output from NARCCAP RCMs forced with a reanalysis and atmosphere–ocean global climate models (AOGCMs). The NARCCAP models generally reproduce the precipitation frequency versus intensity spectrum seen in observations up to around 25 mm day−1, before producing overly strong precipitation at high intensities. CRCM simulations produce lower precipitation amounts than the rest of the models and observations past the 25 mm day−1 threshold. Further analysis focuses on precipitation events exceeding the 99.5th percentile that occur simultaneously at several points in the region, yielding ‘widespread events’. Apart from the CRCM and EPC2 simulations, models and observations produce peaks in widespread events during 0300–0900 UTC, although the models typically produce slightly weaker intensities compared to observations. Widespread precipitation falls too frequently throughout the day, especially between 1500 and 2100 UTC, compared to observations. Composite precipitation shows inter‐model differences in magnitude and location of widespread events. Examination of additional fields shows that NARCCAP models produce credible representations of very heavy precipitation and their supporting environments when compared to the NARR. [ABSTRACT FROM AUTHOR]

Multi-model ensembles (MMEs) are powerful tools in dynamical climate prediction as they account for the overconfidence and the uncertainties related to single-model ensembles. Previous works suggested that the potential benefit that can be expected by using a MME amplifies with the increase of the independence of the contributing Seasonal Prediction Systems. In this work we combine the two MME Seasonal Prediction Systems (SPSs) independently developed by the European (ENSEMBLES) and by the Asian-Pacific (APCC/CliPAS) communities. To this aim, all the possible multi-model combinations obtained by putting together the 5 models from ENSEMBLES and the 11 models from APCC/CliPAS have been evaluated. The grand ENSEMBLES-APCC/CliPAS MME enhances significantly the skill in predicting 2m temperature and precipitation compared to previous estimates from the contributing MMEs. Our results show that, in general, the better combinations of SPSs are obtained by mixing ENSEMBLES and APCC/CliPAS models and that only a limited number of SPSs is required to obtain the maximum performance. The number and selection of models that perform better is usually different depending on the region/phenomenon under consideration so that all models are useful in some cases. It is shown that the incremental performance contribution tends to be higher when adding one model from ENSEMBLES to APCC/CliPAS MMEs and vice versa, confirming that the benefit of using MMEs amplifies with the increase of the independence the contributing models. To verify the above results for a real world application, the Grand ENSEMBLES-APCC/CliPAS MME is used to predict retrospective energy demand over Italy as provided by TERNA (Italian Transmission System Operator) for the period 1990-2007. The results demonstrate the useful application of MME seasonal predictions for energy demand forecasting over Italy. It is shown a significant enhancement of the potential economic value of forecasting energy demand when using the better combinations from the Grand MME by comparison to the maximum value obtained from the better combinations of each of the two contributing MMEs. The above results demonstrate for the first time the potential of the Grand MME to significantly contribute in obtaining useful predictions at the seasonal time-scale. [ABSTRACT FROM AUTHOR]

Climate change is a major challenge for the energy sector, particularly for wind energy onshore and offshore. Climate models are the only tool which is able to produce physical-based projections of future changes in response to increasing greenhouse gas emissions. In the present study, the Western Iberian offshore wind resource is analysed for present and future climates, using a set of regional climate models (RCMs) simulations produced in the framework of the CORDEX experiment at 0.11° resolution (∼12 km), and a regional climate simulation produced with the WRF model at higher resolution (9 km). All these simulations are firstly, evaluated against wind buoy measurements and Cross-Calibrated Multi-Platform (CCMP) wind data, and used to generate two high quality multi-model ensembles based on the individual model’s performance. The results of the WRF simulation and of the two multi-model ensembles are then used to describe the wind resource both for the present and future climates, according to the RCP4.5 and RCP8.5 emission scenarios. This allows the assessment of the climate change signal on the offshore wind and to provide an uncertainty measure of these projections. The vast majority of climate models project reductions of wind speed and wind power for all seasons, with the exception of summer. For the RCP8.5 emission scenario the multi-model ensembles project reductions in power density of around 7% for winter, 4% for spring and 12% for autumn, and increases of 5% for summer. In the latter, and increase up to 20% in power density is forecasted for the Iberian northwest coast. This is sufficient to offset the yearly balance, in as much as no change is expected at a yearly scale for this area. For the remaining west Iberian coast, a yearly reduction of less than 5% is estimated. These results are shared by the two multi-model ensembles and by WRF higher resolution simulation (9 km). The projected changes have the consequence of reducing the annual cycle of power density availability and of its yearly mean values. Finally, for the less aggressive scenario, RCP4.5, the changes have the same signal but with smaller values. [ABSTRACT FROM AUTHOR]

Portugal, which is located in the west limit of the Mediterranean subtropics, is a small region with a complex orography with large precipitation gradients and interannual variability. In this study, the newer and higher resolution regional climate simulations, covering Portugal, are evaluated in present climate and used to investigate the rainfall projections for the end of the twenty-first century, following the RCP4.5 and RCP8.5 emission scenarios. The EURO-CORDEX historical simulations, at 0.11° and at 0.44° resolution, are evaluated against gridded observations of precipitation, which allows the assembly of four multi-model ensembles. An extra simulation, at even higher resolution (9 km) with WRF is also analysed. In present climate, the models are able to describe the precipitation temporal and spatial patterns as well its distributions, although there is a large spread and an overestimation of larger rainfall quantiles. The multi-model ensembles show that selecting the best performing models adds quality to the overall representation of rainfall. The high-resolution simulations augment the spatial details of precipitation, but objectively do not seem to add value with respect to the coarse resolution. Regarding the RCP8.5 scenario, WRF and the multi-model ensembles consistently predict important losses of precipitation in Portugal in spring, summer and autumn, ranging from −10% and −50%. For all seasons, the changes are more severe in the southern basins. The precipitation distributions show, for all models, important reductions of the contribution from low to moderate/high precipitation bins and augments of days with strong rainfall. Furthermore, a prominent growth of high-ranking percentiles is predicted reaching values over 70% in some regions. Generally, the changes associated with the RCP4.5 scenario have the same signal and features, but with smaller magnitudes. [ABSTRACT FROM AUTHOR]

Crop simulation models play an increasingly important role in agricultural management. Therefore, improving crop models' accuracy and reliability has become a key issue. Previous studies have shown that multi-model ensembles (MMEs) perform better than individual models and are capable of assessing model uncertainty. Here, we study the use of the Bayesian model averaging (BMA) method to improve the simulation performance of rice phenology models. The main objective of this paper is to apply BMA to five individual rice phenology models (CERES-Rice, the beta model, ORYZA2000, the rice clock model, and the simulation model for the rice-weather relationship) and to compare its simulation performance with an MME based on the mean value (Mean-MME). Furthermore, we quantify and evaluate the uncertainty of the individual and ensemble models. Finally, we analyze the prediction results of two ensemble models and five individual models under future climate scenarios. The result indicates that, compared with the accuracies of individual models, the accuracies of BMA and Mean-MME are 12.75% and 11.48% higher, respectively. BMA performs the best when the individual models had the best parameters, effectively optimizing the phenology simulation results of the individual models. For individual and ensemble models, the squared bias accounts for 48–88%, which is the largest contributor to overall prediction uncertainty. Under the future climate scenarios, the five models predict a range of 0.40–9.78 days in the phenological period, whereas the two ensemble models predict a smaller range of 0.54–1.63 days, which shows that the MMEs reduce the uncertainty of the individual models. We conclude that BMA is suitable for improving the accuracy and reliability of crop modeling. • Five widely used rice phenology models are calibrated and validated. • Two ensemble models are investigated to improve phenological simulation performance. • The prediction results of individual and ensemble models under future climate scenarios are analyzed. • We find that Bayesian model averaging effectively optimizes the phenological simulation results. • Uncertainties of individual models and ensembles are quantified and assessed. [ABSTRACT FROM AUTHOR]

Decision-makers aiming to improve food security, livelihoods and resilience are faced with an uncertain future. To develop robust policies they need tools to explore the potential effects of uncertain climatic, socioeconomic, and environmental changes. Methods have been developed to use scenarios to present alternative futures to inform policy. Nevertheless, many of these can limit the possibility space with which decision-makers engage. This paper will present a participatory scenario process that maintains a large possibility space through the use of multiple factors and factor-states and a multi-model ensemble to create and quantify four regional scenarios for Southeast Asia. To do this we will explain 1) the process of multi-factor, multi-state building was done in a stakeholder workshop in Vietnam, 2) the scenario quantification and model results from GLOBIOM and IMPACT, two economic models, and 3) how the scenarios have already been applied to diverse policy processes in Cambodia, Laos, and Vietnam. [ABSTRACT FROM AUTHOR]

Changes in climate are likely to induce changes in precipitation characteristics including intensity, frequency, duration and patterns of events. In this paper, we evaluate the performance of multiple regional climate models (RCMs) in the North American Regional Climate Change Assessment Program (NARCCAP) to simulate storm properties including storm duration, inter-storm period, storm intensity, and within-storm patterns at eight locations in the continental US. We also investigate the future projections of them based on precipitation from NARCCAP historic runs and future runs. Results illustrate that NARCCAP RCMs are consistent with observed precipitation in the seasonal variation of storm duration and inter-storm period, but fail to simulate the magnitude. The ability to simulate the seasonal trend of average storm intensity varies among locations. Within-storm patterns from RCMs exhibit greater variability than from observed records. Comparisons between RCM-historic simulations and RCM projections indicate that there is a large variation in the future changes in storm properties. However, multi-model ensembles of the storm properties suggest that most regions of the United States will experience future changes in storm properties that includes shorter storm duration, longer inter-storm period, and larger average storm intensity. [ABSTRACT FROM AUTHOR]

The Fifth Assessment Report of the Intergovernmental Panel on Climate Change highlights that climate change and ocean acidification are challenging the sustainable management of living marine resources (LMRs). Formal and systematic treatment of uncertainty in existing LMR projections, however, is lacking. We synthesize knowledge of how to address different sources of uncertainty by drawing from climate model intercomparison efforts. We suggest an ensemble of available models and projections, informed by observations, as a starting point to quantify uncertainties. Such an ensemble must be paired with analysis of the dominant uncertainties over different spatial scales, time horizons, and metrics. We use two examples: (i) global and regional projections of Sea Surface Temperature and (ii) projection of changes in potential catch of sablefish (Anoplopoma fimbria) in the 21st century, to illustrate this ensemble model approach to explore different types of uncertainties. Further effort should prioritize understanding dominant, undersampled dimensions of uncertainty, as well as the strategic collection of observations to quantify, and ultimately reduce, uncertainties. Our proposed framework will improve our understanding of future changes in LMR and the resulting risk of impacts to ecosystems and the societies under changing ocean conditions. [ABSTRACT FROM AUTHOR]

This paper deals with the issue of reconciling gaps between stochastic estimates (probability distributions) provided by alternative statistically inaccurate observation/estimation techniques. We employ a posterior reconciliation (integration) method based on selection of mutually compatible test outcomes. Unlike other methods used in this context, the posterior integration method employed does not include assessment of the credibility of the original (prior) estimation sources, which is usually based on analysis of their past performance. The quality of the resulting posterior integrated distribution is evaluated in terms of change in the variance. The method is illustrated by integration of stochastic estimates of the annual net primary production (NPP) of forest ecosystems in seven bioclimatic zones of Russia. The estimates result from the use of two alternative NPP estimation techniques — the landscape–ecosystem approach based on empirical knowledge, and an ensemble of dynamic global vegetation models. The estimates differ by up to 23%. Elimination of these gaps could help better quantify the terrestrial ecosystems' input to the global carbon cycle. The paper suggests a set of candidates for credible integrated NPP estimates for Russia, which harmonize those provided by two alternative sources. [ABSTRACT FROM AUTHOR]

This work focuses on elucidating issues related to an increasingly common technique of multi-model ensemble (MME) forecasting. The MME approach is aimed at improving the statistical accuracy of imperfect time-dependent predictions by combining information from a collection of reduced-order dynamical models. Despite some operational evidence in support of the MME strategy for mitigating the prediction error, the mathematical framework justifying this approach has been lacking. Here, this problem is considered within a probabilistic/stochastic framework which exploits tools from information theory to derive a set of criteria for improving probabilistic MME predictions relative to single-model predictions. The emphasis is on a systematic understanding of the benefits and limitations associated with the MME approach, on uncertainty quantification, and on the development of practical design principles for constructing an MME with improved predictive performance. The conditions for prediction improvement via the MME approach stem from the convexity of the relative entropy which is used here as a measure of the lack of information in the imperfect models relative to the resolved characteristics of the truth dynamics. It is also shown how practical guidelines for MME prediction improvement can be implemented in the context of forced response predictions from equilibrium with the help of the linear response theory utilizing the fluctuation-dissipation formulas at the unperturbed equilibrium. The general theoretical results are illustrated using exactly solvable stochastic non-Gaussian test models. [ABSTRACT FROM AUTHOR]

Attribution studies investigate the causes of recent global warming. For a few decades the scientific community generally adopted dynamical models – the so-called Global Climate Models (GCMs) – for such an investigation. These models show the essential role of anthropogenic forcings in driving the temperature behaviour of the last half century. In the last period even other (data-driven) methodological approaches were adopted for attribution studies. This allows the scientific community to compare the results coming from these different approaches and to possibly increase their robustness. For such a purpose, the paper explores the possibility of applying a robustness framework, so far used only in the case of multi-model GCM ensembles, to a strategy including models from different methodological orientations, assessing such an application especially in the light of the independence issue. [ABSTRACT FROM AUTHOR]

ABSTRACT Present and future climate change information is required to develop adaptation and mitigation strategies at national and international levels. This study assessed the simulated surface air temperature ( SAT) and precipitation ( PR) over China from 24 models involved in the Coupled Model Intercomparison Project Phase 5 ( CMIP5). The reliability ensemble average ( REA) is applied to project the SAT and PR change under representative concentration pathway ( RCP) scenarios over China in the 21st century. The results show that most CMIP5 models tend to underestimate SAT and overestimate PR in China. Models generally agree better with the observed SAT than PR. For SAT, the ensemble prediction shows that warming is expected all over China for all RCPs. The warming trend from 2006 to 2099 in China is 0.87 ± 0.14 °C 100 year−1, 2.47 ± 0.48 °C 100 year−1, 5.85 ± 0.73 °C 100 year−1 for RCP 2.6, RCP 4.5 and RCP 8.5, respectively. Northern regions experience more warming than southern regions. The Songhua River basin warms the most, considering the ten studied basins for RCP 4.5 and RCP 8.5. Under RCP 2.6, the largest warming trend occurs in the Huaihe River basin. For PR, the spatial pattern of PR change has zonal characteristics. The girds with the maximum linear trend, i.e. >7.5 mm decade−1, are concentrated in the upper Yangtze River basin. For temporal scale, PR in China is also projected to increase during the 21st century by 4.89 ± 2.30% 100 year−1, 8.67 ± 6.27% 100 year−1 and 13.39 ± 12.58% 100 year−1 for RCP 2.6, RCP 4.5 and RCP 8.5, respectively. PR tends to decrease in the Yangtze River basin, Southeast River Drainage and Pearl River basin during the early period (2011-2030) for all RCPs, largely increase thereafter. However, uncertainties are unavoidable for SAT and PR projections. The PR uncertainty exceeds the temperature uncertainty. More studies regarding the analysis of narrowing uncertainties are essential for a better understanding of climate change. [ABSTRACT FROM AUTHOR]

Present and future climate change information is required to develop adaptation and mitigation strategies at national and international levels. This study assessed the simulated surface air temperature (SAT) and precipitation (PR) over China from 24 models involved in the Coupled Model Intercomparison Project Phase 5 (CMIP5). The reliability ensemble average (REA) is applied to project the SAT and PR change under representative concentration pathway (RCP) scenarios over China in the 21st century. The results show that most CMIP5 models tend to underestimate SAT and overestimate PR in China. Models generally agree better with the observed SAT than PR. For SAT, the ensemble prediction shows that warming is expected all over China for all RCPs. The warming trend from 2006 to 2099 in China is 0.87 ±0.14 °C 100 year-1, 2.47 ±0.48 °C100 year-1, 5.85 ±0.73 °C100 year-1 for RCP 2.6, RCP 4.5 and RCP 8.5, respectively. Northern regions experience more warming than southern regions. The Songhua River basin warms the most, considering the ten studied basins for RCP 4.5 and RCP 8.5. Under RCP 2.6, the largest warming trend occurs in the Huaihe River basin. For PR, the spatial pattern of PR change has zonal characteristics. The girds with the maximum linear trend, i.e. >7.5 mm decade-1, are concentrated in the upper Yangtze River basin. For temporal scale, PR in China is also projected to increase during the 21st century by 4.89±2.30% 100 year-1, 8.67±6.27% 100 year-1 and 13.39±12.58% 100 year-1 for RCP 2.6, RCP 4.5 and RCP 8.5, respectively. PR tends to decrease in the Yangtze River basin, Southeast River Drainage and Pearl River basin during the early period (2011-2030) for all RCPs, largely increase thereafter. However, uncertainties are unavoidable for SAT and PR projections. The PR uncertainty exceeds the temperature uncertainty. More studies regarding the analysis of narrowing uncertainties are essential for a better understanding of climate change. [ABSTRACT FROM AUTHOR]

Use of multi-model ensembles from global climate models to simulate the current and future climate change has flourished as a research topic during recent decades. This paper assesses the performance of multi-model ensembles in simulating global land temperature from 1960 to 1999, using Nash-Sutcliffe model efficiency and Taylor diagrams. The future trends of temperature for different scales and emission scenarios are projected based on the posterior model probabilities estimated by Bayesian methods. The results show that ensemble prediction can improve the accuracy of simulations of the spatiotemporal distribution of global temperature. The performance of Bayesian model averaging (BMA) at simulating the annual temperature dynamic is significantly better than single climate models and their simple model averaging (SMA). However, BMA simulation can demonstrate the temperature trend on the decadal scale, but its annual assessment of accuracy is relatively weak. The ensemble prediction presents dissimilarly accurate descriptions in different regions, and the best performance appears in Australia. The results also indicate that future temperatures in northern Asia rise with the greatest speed in some scenarios, and Australia is the most sensitive region for the effects of greenhouse gas emissions. In addition to the uncertainty of ensemble prediction, the impacts of climate change on agriculture production and water resources are discussed as an extension of this research. [ABSTRACT FROM PUBLISHER]

We investigate the performance of the newest generation multi-model ensemble (MME) from the Coupled Model Intercomparison Project (CMIP5). We compare the ensemble to the previous generation models (CMIP3) as well as several single model ensembles (SMEs), which are constructed by varying components of single models. These SMEs range from ensembles where parameter uncertainties are sampled (perturbed physics ensembles) through to an ensemble where a number of the physical schemes are switched (multi-physics ensemble). We focus on assessing reliability against present-day climatology with rank histograms, but also investigate the effective degrees of freedom (EDoF) of the fields of variables which makes the statistical test of reliability more rigorous, and consider the distances between the observation and ensemble members. We find that the features of the CMIP5 rank histograms, of general reliability on broad scales, are consistent with those of CMIP3, suggesting a similar level of performance for present-day climatology. The spread of MMEs tends towards being 'over-dispersed' rather than 'under-dispersed'. In general, the SMEs examined tend towards insufficient dispersion and the rank histogram analysis identifies them as being statistically distinguishable from many of the observations. The EDoFs of the MMEs are generally greater than those of SMEs, suggesting that structural changes lead to a characteristically richer range of model behaviours than is obtained with parametric/physical-scheme-switching ensembles. For distance measures, the observations and models ensemble members are similarly spaced from each other for MMEs, whereas for the SMEs, the observations are generally well outside the ensemble. We suggest that multi-model ensembles should represent an important component of uncertainty analysis. [ABSTRACT FROM AUTHOR]

We dynamically downscaled Japanese reanalysis data (JRA-25) for 60 regions of Japan using three regional climate models (RCMs): the Non-Hydrostatic Regional Climate Model (NHRCM), modified RAMS version 4.3 (NRAMS), and modified Weather Research and Forecasting model (TWRF). We validated their simulations of the precipitation climatology and interannual variations of summer and winter precipitation. We also validated precipitation for two multi-model ensemble means: the arithmetic ensemble mean (AEM) and an ensemble mean weighted according to model reliability. In the 60 regions NRAMS simulated both the winter and summer climatological precipitation better than JRA-25, and NHRCM simulated the wintertime precipitation better than JRA-25. TWRF, however, overestimated precipitation in the 60 regions in both the winter and summer, and NHRCM overestimated precipitation in the summer. The three RCMs simulated interannual variations, particularly summer precipitation, better than JRA-25. AEM simulated both climatological precipitation and interannual variations during the two seasons more realistically than JRA-25 and the three RCMs overall, but the best RCM was often superior to the AEM result. In contrast, the weighted ensemble mean skills were usually superior to those of the best RCM. Thus, both RCMs and multi-model ensemble means, especially multi-model ensemble means weighted according to model reliability, are powerful tools for simulating seasonal and interannual variability of precipitation in Japan under the current climate. [ABSTRACT FROM AUTHOR]

Abstract: This paper reports on a project to compare predictions from a range of catchment models applied to a mesoscale river basin in central Germany and to assess various ensemble predictions of catchment streamflow. The models encompass a large range in inherent complexity and input requirements. In approximate order of decreasing complexity, they are DHSVM, MIKE-SHE, TOPLATS, WASIM-ETH, SWAT, PRMS, SLURP, HBV, LASCAM and IHACRES. The models are calibrated twice using different sets of input data. The two predictions from each model are then combined by simple averaging to produce a single-model ensemble. The 10 resulting single-model ensembles are combined in various ways to produce multi-model ensemble predictions. Both the single-model ensembles and the multi-model ensembles are shown to give predictions that are generally superior to those of their respective constituent models, both during a 7-year calibration period and a 9-year validation period. This occurs despite a considerable disparity in performance of the individual models. Even the weakest of models is shown to contribute useful information to the ensembles they are part of. The best model combination methods are a trimmed mean (constructed using the central four or six predictions each day) and a weighted mean ensemble (with weights calculated from calibration performance) that places relatively large weights on the better performing models. Conditional ensembles, in which separate model weights are used in different system states (e.g. summer and winter, high and low flows) generally yield little improvement over the weighted mean ensemble. However a conditional ensemble that discriminates between rising and receding flows shows moderate improvement. An analysis of ensemble predictions shows that the best ensembles are not necessarily those containing the best individual models. Conversely, it appears that some models that predict well individually do not necessarily combine well with other models in multi-model ensembles. The reasons behind these observations may relate to the effects of the weighting schemes, non-stationarity of the climate series and possible cross-correlations between models. [Copyright &y& Elsevier]

• VB-LSTM provides ensemble predictions with 20 hydrological models as inputs. • We apply a model-agnostic algorithm to interpret the deep network LSTM. • We show VB-LSTM produces better deterministic and probabilistic predictions than BMA. Multi-model ensembles enable assessment of model structural uncertainty across multiple disciplines. Bayesian Model Averaging (BMA) is one of the most popular ensemble averaging approaches in hydrology but its predictions are easily impacted by the type of ensemble members selected. Here, we propose a regression-based ensemble approach, namely a Variational Bayesian Long Short-Term Memory network (VB-LSTM) to address this issue. In this approach, a state-of-the-art variational inference (VI) algorithm that is faster and more scalable than Bayesian Markov chain Monte Carlo (MCMC) is employed to approximate the posterior distributions of thousands of parameters in the LSTM networks. To interpret the behavior of deep learning methods, the Permutation Feature Importance (PFI) algorithm is introduced to understand the degree to which VB-LSTM relies on each ensemble member. Twenty conceptual hydrological models are considered to evaluate BMA and VB-LSTM in four catchments from China. Four scenarios with different ensemble members are established to investigate the impacts of ensemble members on model results. Our results show that compared with BMA, VB-LSTM improves deterministic and probabilistic predictions by approximately 10%–30% in terms of Mean Absolute Error (MAE), Sharpness and Continous Ranked Probability Score (CRPS). In addition, the VB-LSTM predictions are more robust and less impacted by the selection of ensemble members. Furthermore, our study encourages the use of Bayesian deep learning in hydrology as an alternative to other approaches tackling model structural uncertainty. [ABSTRACT FROM AUTHOR]

Analysis of variance (ANOVA) is a powerful statistical technique for making inferences about experiments that are influenced by multiple factors. Whilst common in many other scientific fields, its use within the climate community has been limited to date. Here we review the basis for ANOVA and how, in particular, it can be applied to partition the variance in a multi-model ensemble of Atmospheric General Circulation Model simulations. We examine an ensemble of four AGCMs forced with observed twentieth century sea surface temperatures (SST). We show that the dominant contributions to the total variance of seasonal mean sea level pressure arise from between-model differences (the bias term) and internal noise (the noise term). However, which term is most important varies from region to region. Of particular interest is the interaction term, which describes differences between the models in their responses to common SST forcing. The interaction term is found to be largest over the Indian Ocean (in all seasons), and over the subtropical Northwest Pacific in boreal summer. The differences between the model responses in these regions suggest differences in their simulation of atmospheric teleconnections, with potentially important implications, e.g. for seasonal predictions of the South and East Asian Monsoons. Examination of these differences may lead to an understanding of the reasons why models respond differently to common forcing, and ultimately to improvements in the performance of climate models. [ABSTRACT FROM AUTHOR]

Assessing the performance of climate models in surface air temperature (SAT) simulation and projection have received increasing attention during the recent decades. This paper assesses the performance of the Coupled Model Intercomparison Project phase 5 (CMIP5) in simulating intra-annual, annual and decadal temperature over Northern Eurasia from 1901 to 2005. We evaluate the skill of different multi-model ensemble techniques and use the best technique to project the future SAT changes under different emission scenarios. The results show that most of the general circulation models (GCMs) overestimate the annual mean SAT in Northern Eurasia and the difference between the observation and the simulations primarily comes from the winter season. Most of the GCMs can approximately capture the decadal SAT trend; however, the accuracy of annual SAT simulation is relatively low. The correlation coefficient R between each GCM simulation and the annual observation is in the range of 0.20 to 0.56. The Taylor diagram shows that the ensemble results generated by the simple model averaging (SMA), reliability ensemble averaging (REA) and Bayesian model averaging (BMA) methods are superior to any single GCM output; and the decadal SAT change generated by SMA, REA and BMA are almost identical during 1901–2005. Heuristically, the uncertainty of BMA simulation is the smallest among the three multi-model ensemble simulations. The future SAT projection generated by the BMA shows that the SAT in Northern Eurasia will increase in the 21st century by around 1.03 °C/100 yr, 3.11 °C/100 yr and 7.14 °C/100 yr under the RCP 2.6, RCP 4.5 and RCP 8.5 scenarios, respectively; and the warming accelerates with the increasing latitude. In addition, the spring season contributes most to the decadal warming occurring under the RCP 2.6 and RCP 4.5 scenarios, while the winter season contributes most to the decadal warming occurring under the RCP 8.5 scenario. Generally, the uncertainty of the SAT projections increases with time in the 21st century.

Large ensembles of climate models are increasingly available either as ensembles of opportunity or perturbed physics ensembles, providing a wealth of additional data that is potentially useful for improving adaptation strategies to climate change. In this work, we propose a framework to evaluate the predictive capacity of 11 multi-model ensemble methods (MMEs), including random forest (RF), to estimate reference evapotranspiration (ET 0 ) using 10 AR5 models for the scenarios RCP4.5 and RCP8.5. The study was carried out in the Segura Hydrographic Demarcation (SE of Spain), a typical Mediterranean semiarid area. ET 0 was estimated in the historical scenario (1970–2000) using a spatially calibrated Hargreaves model. MMEs obtained better results than any individual model for reproducing daily ET 0 . In validation, RF resulted more accurate than other MMEs (Kling–Gupta efficiency (KGE) M = 0.903 , S D = 0.034 for KGE and M = 3.17 , S D = 2.97 for absolute percent bias). A statistically significant positive trend was observed along the 21st century for RCP8.5, but this trend stabilizes in the middle of the century for RCP4.5. The observed spatial pattern shows a larger ET 0 increase in headwaters and a smaller increase in the coast. [ABSTRACT FROM AUTHOR]