Report No . 339 High-resolution Climate Projections for Ireland – A Multi-model Ensemble Approach

The method of regional climate modelling was employed to assess the impacts of a warming climate on the 21st-century climate of Ireland. The regional climate model (RCM) simulations were run at high spatial resolution (3.8 and 4 km), the first systematic study of its kind at this scale, thus allowing a better evaluation of the local effects of climate change. To address the issue of uncertainty, a multi-model ensemble approach was employed. Through the ensemble approach, the uncertainty in the projections can be partly quantified, thus providing a measure of confidence in the projections. Simulations were run for the reference period 1981–2000 and the future period 2041–2060. Differences between the two periods provide a measure of climate change. The Consortium for Small-scale Modeling–Climate Limited-area Modelling (COSMO-CLM) and Weather Research and Forecasting (WRF) RCMs were used to downscale the following Coupled Model Intercomparison Project – Phase 5 (CMIP5) global climate model (GCM) datasets: CNRM-CM5, EC-EARTH (four ensemble members), HadGEM2-ES, MIROC5 and MPI-ESM-LR. To account for the uncertainty in future greenhouse gas emissions, the future climate was simulated under both the Representative Concentration Pathway 4.5 (RCP4.5) and RCP8.5 scenarios.


Our Responsibilities Licensing
We regulate the following activities so that they do not endanger human health or harm the environment: • waste facilities (e.g. landfills, incinerators, waste transfer stations); • large scale industrial activities (e.g. pharmaceutical, cement manufacturing, power plants); • intensive agriculture (e.g. pigs, poultry); • the contained use and controlled release of Genetically Modified Organisms (GMOs); • sources of ionising radiation (e.g. x-ray and radiotherapy equipment, industrial sources); • large petrol storage facilities; • waste water discharges; • dumping at sea activities.

National Environmental Enforcement
• Conducting an annual programme of audits and inspections of EPA licensed facilities. • Overseeing local authorities' environmental protection responsibilities. • Supervising the supply of drinking water by public water suppliers. • Working with local authorities and other agencies to tackle environmental crime by co-ordinating a national enforcement network, targeting offenders and overseeing remediation. • Enforcing Regulations such as Waste Electrical and Electronic Equipment (WEEE), Restriction of Hazardous Substances (RoHS) and substances that deplete the ozone layer. • Prosecuting those who flout environmental law and damage the environment.

Water Management
• Monitoring and reporting on the quality of rivers, lakes, transitional and coastal waters of Ireland and groundwaters; measuring water levels and river flows. • National coordination and oversight of the Water Framework Directive. • Monitoring and reporting on Bathing Water Quality.

Monitoring, Analysing and Reporting on the Environment
• Monitoring air quality and implementing the EU Clean Air for Europe (CAFÉ) Directive. • Independent reporting to inform decision making by national and local government (e.g. periodic reporting on the State of Ireland's Environment and Indicator Reports).

Regulating Ireland's Greenhouse Gas Emissions
• Preparing Ireland's greenhouse gas inventories and projections.
• Implementing the Emissions Trading Directive, for over 100 of the largest producers of carbon dioxide in Ireland.

Environmental Research and Development
• Funding environmental research to identify pressures, inform policy and provide solutions in the areas of climate, water and sustainability.

Strategic Environmental Assessment
• Assessing the impact of proposed plans and programmes on the Irish environment (e.g. major development plans).

Radiological Protection
• Monitoring radiation levels, assessing exposure of people in Ireland to ionising radiation. • Assisting in developing national plans for emergencies arising from nuclear accidents. • Monitoring developments abroad relating to nuclear installations and radiological safety. • Providing, or overseeing the provision of, specialist radiation protection services.

Guidance, Accessible Information and Education
• Providing advice and guidance to industry and the public on environmental and radiological protection topics. • Providing timely and easily accessible environmental information to encourage public participation in environmental decision-making (e.g. My Local Environment, Radon Maps). • Advising Government on matters relating to radiological safety and emergency response. • Developing a National Hazardous Waste Management Plan to prevent and manage hazardous waste.

Awareness Raising and Behavioural Change
• Generating greater environmental awareness and influencing positive behavioural change by supporting businesses, communities and householders to become more resource efficient. • Promoting radon testing in homes and workplaces and encouraging remediation where necessary. The EPA is assisted by an Advisory Committee of twelve members who meet regularly to discuss issues of concern and provide advice to the Board.

Executive Summary
The method of regional climate modelling was employed to assess the impacts of a warming climate on the 21st-century climate of Ireland. The regional climate model (RCM) simulations were run at high spatial resolution (3.8 and 4 km), the first systematic study of its kind at this scale, thus allowing a better evaluation of the local effects of climate change.
To address the issue of uncertainty, a multi-model ensemble approach was employed. Through the ensemble approach, the uncertainty in the projections can be partly quantified, thus providing a measure of confidence in the projections. The climate projections of the current report are in broad agreement with previous research, which adds a measure of confidence to the projections. Moreover, the current report presents projections of additional climate fields and derived variables that are of vital importance to sectors such as agriculture, health, energy, biodiversity and transport. It is envisaged that the research will inform policy and further the understanding of the potential environmental impacts of climate change in Ireland at a local scale.

Temperature Projections
Mid-century mean annual temperatures are projected to increase by 1-1.2°C and 1.3-1.6°C for the RCP4.5 and RCP8.5 scenarios, respectively. Temperature projections show a clear west-to-east gradient, with the largest increases in the east. Warming is enhanced for the extremes (i.e. hot days and cold nights), with the warmest 5% of daily maximum temperatures projected to increase by 1.0-2.2°C compared with the baseline period. The coldest 5% of daily minimum temperatures are projected to rise by 1-2.4°C. Heatwave events are expected to increase by the middle of the century; over the 20-year period (2041-2060), increases in heatwave events range from 1 to 8 for the RCP4.5 scenario and from 3 to 15 for the RCP8.5 scenario, with the largest increases in the south-east. Averaged over the whole country, the number of frost days (days when the minimum temperature is lower than 0°C) is projected to decrease by 45% and 58% for the RCP4.5 and RCP8.5 scenarios, respectively. Similarly, the number of ice days (days when the maximum temperature is lower than 0°C) is projected to decrease by 68% and 78% for the RCP4.5 and RCP8.5 scenarios, respectively. It is worth noting that periods of frost and ice are important environmental drivers that trigger phenological phases in many plant and animal species. Changes in the occurrence of these weather types may disrupt the life cycles of these species. The projected increase in heatwaves will have a direct impact on public health and mortality but this may be offset by the projected decrease in frost and ice days. xiv

Precipitation, Snow and Surface Humidity Projections
Substantial decreases in precipitation are projected for the summer months, with reductions ranging from ≈0% to 11% for the RCP4.5 scenario and from 2% to 17% for the RCP8.5 scenario. Other seasons, and over the full year, show small projected changes in precipitation. However, the mid-century precipitation climate is expected to become more variable with substantial projected increases in both dry periods and heavy precipitation events.
The frequencies of heavy precipitation events show notable increases over the year as a whole and in the winter and autumn months, with "likely" projected increases of 5-19%. 1 The projected increase in evapotranspiration, noted for all seasons, may offset flooding events caused by the expected increases in heavy rainfall. However, it is recommended that additional hydrological modelling be undertaken to improve understanding of the potential impact on flooding. The number of extended dry periods (defined as at least 5 consecutive days for which the daily precipitation is less than 1 mm) is also projected to increase substantially by the middle of the century over the full year and for all seasons except spring. The projected increases in dry periods are largest for summer, with "likely" values of +11% and +48% for the RCP4.5 and RCP8.5 scenarios, respectively. The precipitation projections, summarised previously, were found to be generally robust with over 66% of the ensemble members in agreement.
Snowfall is projected to decrease substantially by the middle of the century with "likely" reductions of 51% and 60% for the RCP4.5 and RCP8.5 scenarios, respectively.
Specific humidity is projected to increase substantially (≈10%) for all seasons by the middle of the century. Relative humidity is projected to increase slightly (or show ≈0% change) for all seasons except summer. The largest increases are noted for spring (both RCP scenarios) and winter (RCP8.5). For summer, relative humidity is expected to decrease in the south-east and increase in the north-west (both RCP scenarios).

Wind Speed, Storm Tracks and Mean Sea Level Pressure Projections
Mid-century mean 10-m wind speeds are projected to decrease for all seasons. The decreases are largest for summer months under the RCP8.5 scenario. The summer reductions in 10-m wind speed range from 0.3% to 3.4% for the RCP4.5 scenario and from 2% to 5.4% for the RCP8.5 scenario. The frequency of "driving rain" events is projected to decrease for all seasons with the exception of the winter months (RCP8.5), when small increases are projected.
The projections indicate that the mean sea level pressure (MSLP) is projected to increase by ≈1 hPa by the middle of the century, with similar increases noted for all seasons. To assess the potential impact of climate change on extreme cyclonic activity in the North Atlantic, an algorithm was developed to identify and track cyclones as simulated by an ensemble of EURO-CORDEX 12-km downscaled CMIP5 RCMs.
The results show an overall reduction of ≈10% in the numbers of storms affecting Ireland and suggest an eastward extension of the more severe wind storms over Ireland and the UK from the middle of the century. It should be noted that because extreme storms are rare events, the storm projections should be considered with a level of caution. Future work will focus on analysing a larger ensemble of downscaled CMIP6 data, thus allowing a more robust statistical analysis of extreme storm track projections.

Agricultural Impacts
The projections, outlined previously, of increases in temperature, heatwaves, heavy precipitation and dry periods/droughts along with decreases in frost and ice days will have direct and substantial effects on agriculture in Ireland by the middle of the century. In addition, the projections indicate an average increase in the length of the growing season by the middle of the century of 12% and 16% for the RCP4.5 and RCP8.5 scenarios, respectively. Similarly, the grazing season, crop heat units (CHUs) and growing degree days (GDDs) for a range of crops are projected to increase substantially by the middle of the century. The results suggest a warming climate may present some positive opportunities for farming. However, the results should be viewed in the context that a warming climate will also result in an increase in pests as a result of an increase in pest-GDDs and a decrease in frost and ice days, as cold conditions are a key control mechanism for the survival of pests. Furthermore, the projected increase in the frequency of both droughts and heavy rainfall events could be detrimental to the potential gains of a warming climate to the agricultural sector.

Energy Impacts
The energy content of the 120-m wind is projected to decrease for all seasons by the middle of the century. The decreases are largest for summer, with reductions ranging from 2.8% to 8.7% for the RCP4.5 scenario and from 6.5% to 14.1% for the RCP8.5 scenario.
To assess the impacts of climate change on solar power in Ireland, projections of solar photovoltaic (PV) power were analysed. Results show a small expected decrease in PV by the middle of the century ranging from ≈0 to 4%. The largest decreases are noted in the north of the country and for the RCP8.5 scenario. The projected change in heating degree days (HDDs) shows that by the middle of the century there will be a greatly reduced requirement for heating in Ireland, with HDDs projected to decrease by 12-17% and 15-21% for the RCP4.5 and RCP8.5 scenarios, respectively. A clear north-to-south gradient is evident for both RCP scenarios, with the largest decreases in the south. The projections show that cooling degree days (CDDs) are expected to slightly increase, suggesting a small increase in air conditioning requirements by the middle of the century. However, the amounts are small compared with HDDs and therefore have a negligible effect on the projected changes in the total energy demand, calculated using the first-order approximation: energy degree days (EDD) = HDD + CDD.

Introduction
The objective of this study is to evaluate the effects of climate change on the future climate of Ireland using the method of high-resolution regional climate modelling. There is a lack of research in dynamically downscaled high-resolution (finer than 7-km grid spacing) climate modelling of Ireland, for projections in the medium term. Existing studies have either focused on analysing relatively small ensembles of regional climate model (RCM) simulations at a relatively low spatial resolution (7-12 km) (e.g. McGrath et al., 2005;McGrath and Lynch, 2008;Nolan et al., 2012Nolan et al., , 2014Nolan et al., , 2017Gleeson et al., 2013;Nolan, 2015;O'Sullivan et al., 2015) or analysed a large ensemble of low-resolution RCM simulations (van der Linden and Mitchell, 2009;Jacob et al., 2014). The analysis presented in this study was undertaken to address this lack of research by analysing the output of three high-resolution (≈4 km) RCMs of Ireland, driven by an ensemble of eight global climate model (GCM) datasets, under the Representative Concentration Pathway 4.5 (RCP4.5) and RCP8.5 emission scenarios. Simulations were run for a reference period, 1981-2000, and a future period, 2041-2060. Differences between the two periods are used to provide a measure of the projected climate change.
The current research consolidates and expands on previous national RCM research (e.g. McGrath et al., 2005;McGrath and Lynch, 2008;Nolan et al., 2012Nolan et al., , 2014Nolan et al., , 2015Gleeson et al., 2013;O'Sullivan et al., 2015) by running a large ensemble of downscaled simulations, using the most up-to-date RCMs and the RCP4.5 and RCP8.5 scenarios to simulate the future climate of Ireland. Additionally, the accuracy and usefulness of the model predictions are enhanced by increasing the model grid spacings to ≈4 km. Although uncertainty can never be eliminated in climate projections (see sections 1.1 and 1.2.5 for further comments), the large ensemble size (using different RCMs, GCMs and RCPs) allows for a robust quantification of climate projection uncertainty and a measure of confidence to be assigned to the projections. Nevertheless, RCM-downscaling studies will always be limited by the available resources to process large ensembles and will likely underrepresent the full range of possible climate futures.
The climate projections of the current report are in broad agreement with previous research, which adds another measure of confidence to the projections. Moreover, the current report presents projections of additional climate fields and derived variables that are of vital importance to sectors such as agriculture, health, energy, biodiversity and transport. It is envisaged that the research will inform policy and further the understanding of the potential environmental impacts of climate change in Ireland at a local scale.

Regional Climate Models
The impact of increasing greenhouse gases and changing land use on climate change can be simulated using GCMs. However, on account of computational constraints, long climate simulations using GCMs are currently feasible only with horizontal resolutions of ≈50 km or coarser. Because climate fields such as precipitation, wind speed and temperature are closely correlated to the local topography, this is inadequate to simulate the detail and pattern of climate change and its effects on the future climate of Ireland. Furthermore, and of particular relevance to Ireland, numerous studies have shown that, even at 50-km grid spacing, GCMs severely underresolve both the number and intensity of cyclones (e.g. Zhao et al., 2009;Camargo, 2013;Zappa et al., 2013).
To overcome these limitations, the RCM method dynamically downscales the coarse information provided by the global models and provides highresolution information on a subdomain covering Ireland. The computational cost of running the RCM, for a given resolution, is considerably less than that of a global model. The approach has its flaws: all models have errors, which are cascaded in this technique, and new errors are introduced via the flow of data through the boundaries of the regional model. Nevertheless, numerous studies have demonstrated that high-resolution RCMs improve the simulation of fields, such as precipitation (Lucas-Picher et al., 2012;Kendon et al., 2012Bieniek et al., 2015;Nolan, 2015 and topography-influenced phenomena and extremes with relatively small spatial or short temporal character (Feser et al., 2011;Feser and Barcikowska, 2012;Shkol'nik et al., 2012;IPCC, 2013a). An additional advantage is that the physically based RCMs explicitly resolve more small-scale atmospheric features and provide a better representation of convective precipitation (Rauscher et al., 2010) and extreme precipitation (Kanada et al., 2008;Nolan et al., 2017). Other examples of the added value of RCMs are improved simulations of near-surface temperature (Feser, 2006;Di Luca et al., 2016), European storm damage (Donat et al., 2010), strong mesoscale cyclones (Cavicchia and Storch, 2011), North Atlantic tropical cyclone tracks (Daloz et al., 2015) and near-surface wind speeds (e.g. Kanamaru and Kanamitsu, 2007;Nolan et al., 2014;Nolan, 2015), particularly in coastal areas with complex topography (Feser et al., 2011;. The added value of RCMs in the simulation of cyclones is particularly important for the current study, as low pressure systems are the main delivery mechanism for precipitation and wind in Ireland. Furthermore, numerous studies have demonstrated that increased RCM spatial resolution results in a more accurate representation of the climate system. Low-resolution RCMs use parameterised convection schemes, meaning that the heaviest precipitation events (e.g. convective systems on hot summer days) may not be adequately represented in the simulations (Prein et al., 2013;. Zängl et al. (2015) investigated heavy rainfall events over the North-Alpine region and found that increasing the mesh size (9, 3 and 1 km) resulted in a stepwise improvement in skill. Similarly, Nolan et al. (2017) found that RCM accuracy increased with higher spatial resolution; however, reducing the horizontal grid spacing below 4 km provided relatively little added value.
The Intergovernmental Panel on Climate Change (IPCC) has concluded that there is "high confidence that downscaling adds value to the simulation of spatial climate detail in regions with highly variable topography (e.g., distinct orography, coastlines) and for mesoscale phenomena and extremes" (IPCC, 2013a).

Climate models and emission scenarios
The future climate of Ireland was simulated at high spatial resolution (3.8 and 4 km) using the Consortium for Small-scale Modeling-Climate Limited-area Modelling (COSMO-CLM; v4.0 and 5.0) and Weather Research and Forecasting (WRF; v3.8) RCMs. The COSMO-CLM RCM is the COSMO weather forecasting model in climate mode (Rockel et al., 2008). 2 The COSMO model 3 is the non-hydrostatic operational weather prediction model used by the German weather service (Deutscher Wetterdienst; DWD). A detailed description of the COSMO model is given by Doms and Schättler (2002) and Steppeler et al. (2003). The WRF model 4 is a numerical weather prediction system designed to serve atmospheric research, climate and operational forecasting needs. The WRF simulations of the present study adopted the Advanced Research WRF (ARW, v3.8.1) dynamical core, with development led by the US National Center for Atmospheric Research (NCAR) (Skamarock et al., 2008;Powers et al., 2017).  (Giorgetta et al., 2013).
To account for the uncertainty arising from the estimation of future global emission of greenhouse gases, downscaled GCM simulations based on two RCPs (RCP4.5 and RCP8.5) (Moss et al., 2010;van Vuuren et al., 2011) were used to simulate the future climate of Ireland.

Model domains and experiment setup
The RCMs were driven by GCM boundary conditions with the following one-way nesting strategies: GCM to 50 km to 18 km to 4 km (COSMO v4); GCM to 18 km to 4 km (COSMO v5); and GCM to 19 km to 3.8 km (WRF). The COSMO v4 50-, 18-and 4-km model domains are shown in Figure 1.1. The COSMO v5 (18 and 4 km) and WRF (19 and 3.8 km) domains are similar to the d02 and d03 domains of Figure 1.1. The advantage of high-resolution RCM simulations is highlighted in Figure 1.2, which shows how the  surface topography is better resolved by the higher resolution data. For the current study, only 3.8-km and 4-km grid spacing RCM data are considered. The higher resolution data allow improved estimates of the regional variations of climate projections. The climate fields of the RCM simulations were archived at 3-h intervals. An overview of the COSMO-CLM archived fields are provided in The RCM simulations were run on the Irish Centre for High-End Computing (ICHEC) supercomputers. Running such a large ensemble of high-resolution RCMs was a substantial computational task and required extensive use of the ICHEC supercomputer systems over a period of 3-4 years. This archive of data will be made available to the wider research community and general public through the EPA and the Climate Ireland platform. 5 The choice of domain size, nesting downscaling ratio and grid spacing was decided upon following the recommendations of previous studies. For example, several studies found that the parent-grid ratio should be no larger than approximately 1:12 (e.g. Denis et al., 2003;Antic et al., 2006). Brisson et al. (2015) investigated the sensitivity of simulating precipitation over Belgium by downscaling ERA-Interim data. They concluded that an intermediate nesting ratio of approximately 3 was essential for the correct representation of precipitation. Rummukainen (2010) recommended that an RCM domain should be "large enough to allow for desired phenomena related to topographic influence and small-scale atmospheric processes to develop, but still sufficiently small so that the flow solution does not deviate too much from the driving model". The 50-and 18-km domains of the current study are large enough to allow changes to synoptic scales. Ideally, the domain for the finest grid size would be larger in order to allow the RCM to fully develop small-scale dynamical structures in the interior of the domain, superposed on the coarse-scale information that enters through the lateral boundaries. However, the size of the inner domains was constrained by available computational resources. Finally, the choice of grid spacing was determined by both computational constraints and a careful preliminary validation experiment (e.g. Nolan et al., 2017;Flanagan et al., 2019). A number of 1-month validation simulations were run using different physics schemes in order to determine the most accurate physics options to use for the current study. It was found that although the RCM accuracy increased with a higher spatial resolution, reducing the horizontal grid spacing below 4 km provided relatively little added value . The results of the preliminary experiments determined the model configurations of the current study Flanagan et al., 2019).
An overview of the simulations is presented in Table 1.2. The GCM realisations result from running the same GCM with slightly different initial conditions, i.e. the starting date of historical simulations. Data from two time slices, 1981-2000 (the reference period) and 2041-2060 (the future period), were used for analysis of projected changes in the middle of the 21st-century Irish climate. These periods were chosen because they are the longest decadal time periods common to all RCM simulations. The historical period was compared with the corresponding future period for all simulations within the same RCM-GCM group. This results in future anomalies for each model run, i.e. the difference between future and past. In this study the ensemble members of the downscaled GCM simulations are treated as independent estimates of the climate system and are given equal weight. Only the differences between the simulations of the past and future climate for each model will be used in the analysis. While model biases may not be invariant under future scenarios of greenhouse gas emissions, this approach may reduce the impact of model bias.

Model domains and experiment setup
Simulations were run for a reference period, 1981-2000, and a future period, 2041-2060. Differences between the two periods provide a measure of climate change. To provide a more comprehensive examination of climate change, projected changes in the standard deviation are considered in context with changes in the mean. Analyses of changes in the standard deviation provide information on projected changes in the shape (or variability) of the distribution of a climate field. In particular, analyses of changes in the mean and standard deviation provide a more comprehensive understanding of projections of extreme events.
To illustrate this concept, Figure 1.3 presents a schematic of past and future probability distributions of precipitation. 6 Figure 1.3a presents a future with increases in mean precipitation and no change in the standard deviation. In this future world, the total amount of precipitation increases, the amount of dry events decreases and the amount of wet events increases. Figure 1.3b presents a future with no change in mean precipitation and an increase in the standard deviation. In this future world, the total amount of precipitation remains constant, with an increase in both dry and wet events (i.e. increased variability). Conversely, Figure 1.3c shows that a decrease in variability, coupled with no change in mean precipitation, results in a decrease in both dry and wet events. Finally, Figure 1.3d illustrates how an increase in the mean and variability results in large increases in wet events.
To create a large ensemble, all RCM outputs were regridded to a common 4-km grid over Ireland using the method of bilinear interpolation. This results in 16 RCP4.5 and 16 RCP8.5 ensemble comparisons.
The relatively large number of comparisons allows for the uncertainty of the projections to be partly quantified, providing a measure of confidence in the predictions.

Overview of climate projection uncertainty
Climate change projections are subject to uncertainty, which limits their utility. Fronzek et al. (2012) suggest that there are four main sources of uncertainty: (1) the natural variability of the climate system; (2) uncertainties on account of the formulation of the models themselves; (3) uncertainties in future regional climate because of the coarse resolution of GCMs; and (4) uncertainties in the future atmospheric composition, which affects the radiative balance of the Earth. The uncertainties arising from (1) and (2) can be addressed, in part, by employing a multi-model ensemble approach (Déqué et al., 2007;van der Linden and Mitchell, 2009;Jacob et al., 2014). The ensemble approach of the current project analyses the output of three RCMs, driven by several GCMs, to simulate climate change (see Table 1.2). Through the ensemble approach, the uncertainty in the projections can be partly quantified, providing a measure of confidence in the predictions. The uncertainty arising from (3) is addressed in the current work by running the RCM simulations at the high spatial resolution of ≈4-km grid spacings. To account for the uncertainty arising from (4), the future climate is simulated under both the RCP4.5 and RCP8.5 emission scenarios.
A disagreement between RCM ensemble projections can result in large individual outliers, 7 skewing the mean ensemble projection. For this reason, it can be informative to also consider percentiles when analysing an ensemble of future projections. The relatively large ensemble size of the current study allows the construction of a probability density function (pdf) of climate projections. Likelihood values can then be assigned to the projected changes. For example, if the mean (and median) ensemble projection is positive for a particular climate field, the 33rd percentile of the ensemble of projected changes is considered and is defined as the "likely" projected increase. This is a projection such that over 66% of the RCM ensemble 7 However, there is information in the outliers that may be of relevance in specific circumstances and so they cannot be entirely discounted. For example, analysis of the outliers allows policymakers to plan for "low-probability, high-impact" climate projections.
members project greater increases. Similarly, if the mean (and median) ensemble projection is negative, the 66th percentile of the ensemble of projections is considered and is defined as the "likely" projected decrease. In this case, over 66% of the RCM ensemble members project greater decreases. In a similar manner, a "very likely" projection is defined as a projection for which at least 90% of the RCM ensemble members are in agreement; the "as likely as not" projection is defined as the 50th percentile (median) projection.
This method of analysing percentiles allows for a better understating of climate change uncertainty and allows for a quantification of conservative and robust ("likely") projections. Conversely, the likelihood method allows for policymakers to consider more "unlikely" (and possibly high-impact) climate projections. These definitions, based on an ensemble of 16 members for each RCP, provide a statistically based descriptive measure of the climate change projection uncertainty.
Note that the accuracy of these statistical descriptions is based on the assumption that the ensemble members represent an unbiased sampling of the (unknown) future climate. It is also important to stress that the likelihood values presented in the current study (and similarly in studies such as Murphy et al., 2009;IPCC, 2013b;and Lowe et al., 2018) are derived from the most up-to-date evidence currently available. Therefore, the "likelihood" values only apply to the specific sets of high-resolution models and experimental design of the current study. Future improvements in modelling may alter the projections, as uncertainty is expected to be further reduced. Future work will focus on reducing this uncertainty by increasing the ensemble size and employing more up-to-date RCMs (including fully coupled atmosphereocean-wave models) to downscale recently completed CMIP6 GCMs under the full range of the Scenario Model Intercomparison Project (ScenarioMIP) "tier 1" Shared Socioeconomic Pathways (SSPs), namely SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5 (Riahi et al., 2017). 9

Regional Climate Model Validations
The RCMs were validated by running 20-year simulations of the past Irish climate for the time period 1981-2000, driven by both ERA-Interim reanalysis (Dee et al., 2011) and the GCM datasets, and comparing the output against observational data. Uncertainty estimates (bias, absolute error and root mean square error -RSME) have been calculated for precipitation and 2-m temperature, utilising gridded datasets of observations made available by Met Éireann and the UK Met Office. The results of these analyses are presented in sections 2.1 and 2.2. The equivalent uncertainty estimates for 10-m winds and 2-m relative humidity have been calculated utilising station observations and are presented in sections 2.3 and 2.4.

RCM Precipitation Validations
Gridded datasets of (observed) accumulated daily precipitation, at 1-km resolution, covering Ireland (Walsh, 2012) for the period 1981-2000 were obtained from Met Éireann. Additionally, equivalent UK Met Office datasets covering Northern Ireland were acquired from the Centre for Ecology and Hydrology (Tanguy et al., 2016). The gridded datasets are available in monthly comma-separated values (CSV) files and require several processing steps before they can be used in any later analyses. These steps involve solutions that vary in complexity: the precipitation files contain spurious negatives that must be masked; easting and northing coordinates must be transformed to longitude and latitude pairs; and the gridded datasets are at 1-km resolution, whereas the modelled datasets are at 4 km (COSMO4 and 5) and 3.8 km (WRF  projections analysed for the current study use this configuration (see Table 1.2). It is noted that the RCM accurately captures the magnitude and spatial characteristics of the historical precipitation climate, e.g. higher rainfall amounts in the west and over mountains. The COSMO4-CLM RCM was found to give similar results (not shown). The WRF 3.8-km RCM was found to generally overestimate precipitation, whereas both COSMO4-CLM and COSMO5-CLM underestimate.
Figure 2.1c shows that the percentage errors range from approximately −30% to approximately +15% for COSMO5-CLM downscaled ERA-Interim data. The percentage error at each grid point (i, j) is given by: ( 2.1) where (2.2) and the RCM i,j ( ) and OBS i,j ( ) terms represent the RCM and observed values, respectively, at grid point (i, j), averaged over the period 1981-2000. Figure 2.1c highlights a clear underestimation of precipitation over the mountainous regions. This is probably because the RCMs underestimate heavy precipitation; previous validations studies (e.g. Nolan et al., 2017) have demonstrated a decrease in RCM skill with increasing magnitude of heavy precipitation events. To quantify the overall bias evident in Figure 2.1c, the mean was calculated over all grid points covering Ireland, resulting in an overall bias of −4.7%. The bias metric allows for the evaluation of the systematic errors of the RCMs but this can hide large errors, as positive and negative values can cancel each other out. For this reason, the percentage mean absolute error (MAE) metric was also used to evaluate the RCM precipitation errors: Again, the mean was calculated over all grid points covering Ireland, resulting in an overall MAE value of 8.3%. Additionally, the percentage RMSE metric was calculated: where N is the number of grid points covering Ireland. The COSMO5-CLM-ERA-Interim precipitation data, presented in Figure 2.1, has a per_RSME value of 14%.
The validations described previously were repeated for each RCM ensemble member (with 3.8-and 4-km horizontal grid spacings) outlined in Table 1.2. The mean bias (daily), absolute error (MAE) and RMSE (in both mm and as a percentage of observations) for each ensemble member has been calculated over the period 1981-2000. The results found for each ensemble member are presented in Table 2 It should be noted that the observed precipitation dataset has a margin of error of approximately ± 10%, so the RCM validations should be considered within this context.
To assess the added value of high-resolution RCM models, and to quantify the improved skill of RCMs over the GCMs, precipitation data were compared with both RCM and GCM data for the period 1976-2005. The analysis was limited to Ireland and the COSMO5-CLM RCM simulations as outlined in Table 1.2. Results, presented in Table 2.2, demonstrate improved skill of the RCMs over the GCMs. Moreover, an increase in grid resolution of the RCMs (from 18-to 4-km grid spacings) results in a general increase in skill. Nolan et al. (2017) analysed a larger ensemble of RCMs (both COSMO-CLM and WRF) with different grid spacings (18, 7, 6, 4, 2 and 1.5 km) and found that the RCMs demonstrated a general stepwise increase in skill with increased model resolution. Furthermore, it was shown that heavy precipitation events are more accurately resolved by the higher spatial resolution RCM data. However, it was found that although the RCM accuracy increased with higher spatial resolution, reducing the horizontal grid spacing below 4 km provided relatively little added value .

RCM 2-m Temperature Validations
Daily 1-km gridded observations of 2-m temperature for Ireland for the period 1981-2000 were obtained from Met Éireann and processed for comparison with each ensemble member. As with the gridded precipitation observations, the (observed) temperature data require coordinate transformation (easting and northing to longitude and latitude) and regridding (1-km model grids to 3.8-km/4-km model grids). Unlike precipitation, however, no spurious values needed to be masked and the calculation of temperature at specific grid points is relatively straightforward; a bilinear interpolant was used.  Figure 2.2b presents the downscaled ERA-Interim data as simulated by the COSMO5-CLM model at 4-km resolution. It is noted that the COSMO5-CLM data accurately capture the magnitude and spatial characteristics of the observed temperature climate. This is confirmed by Figure 2.2c, which shows a small negative bias of a mean value of −0.32°C over Ireland. The corresponding MAE statistic has a value of 0.34°C.
In  For each metric, the best-and worst-performing scores are highlighted in green and red, respectively. For each metric, the best-and worst-performing scores are highlighted in green and red, respectively.
(COSMO5-CLM-CNRM-CM5) and RMSE values range The observed gridded 2-m temperature dataset has an estimated MAE of 0.19°C and an RMSE of 0.41°C, so the RCM validations should be considered within this context.

RCM 10-m Wind Speed Validations
Daily (  For each metric, the best-and worst-performing scores are highlighted in green and red, respectively. Before this latter step could be completed, time series of (daily) 10-m wind speeds were generated from each ensemble member at each station location. This was achieved by first estimating the (3-hourly) 10-m (zonal velocity -U, meridional velocity -V) wind components at each station location through bilinear interpolation and then calculating (3-hourly) 10-m wind speeds through the simple formula W = U 2 +V 2 . Initial test simulations showed that the WRF data exhibited a consistent overestimation in the wind speed. This overestimation was corrected by adapting the topo_wind parameterising scheme -a topographic correction for surface winds to represent extra drag from subgrid topography and enhanced flow at hill tops (Jimenez and Dudhia, 2012). However, adapting this parameterising scheme resulted in an underestimation of the WRF-MIROC5 wind speed (-0.94 m s −1 bias; see Table 2.4).

RCM 2-m Relative Humidity Validations
Hourly 2-m relative humidity data from the nine Met Éireann weather stations listed in section 2.3 were used for model validation. The data obtained have an earliest starting date of 1 January 1987, 01:00, and cover periods that extend beyond 2000. As with the daily data described in section 2.3, other Met Éireann station data exist but do not extend back to 1987 and were therefore not used. The observed time series were therefore systematically trimmed to cover the common period 1987-2000.
Bilinear interpolation was used to generate time series of 3-hourly 2-m relative humidity from each ensemble member at each station location. For each metric, the best-and worst-performing scores are highlighted in green and red, respectively.
these calculations are given in Table 2

RCM Validation Summary
The RCMs were validated by downscaling ERA-Interim reanalyses and the GCM datasets for the period 1981-2000, and comparing the output with observational data. Extensive validations were carried out to test the ability of the RCMs to accurately model the temperature, precipitation, wind and humidity climate of Ireland. Results confirm that the output of the RCMs exhibit reasonable and realistic features as documented in the historical data record. The skill of the individual RCM datasets was dependent on the field under analysis (e.g. WRF performed well for precipitation but less well for wind speed). This variation in RCM skill stresses the importance of using an ensemble of RCMs to simulate climate change.
For an in-depth validation of additional climate fields, please refer to Nolan et al. (2014Nolan et al. ( , 2015Nolan et al. ( , 2017 and Flanagan et al. (2019). Additional experiments were carried out to assess the added value of high-resolution RCM models, the results of which demonstrated improved skill of RCMs over the GCMs.
Moreover, an increase in the spatial resolution of the RCMs was found to result in a general increase in skill (e.g. Nolan et al., 2017). However, it was found that although the RCM accuracy increased with higher spatial resolution, reducing the horizontal grid spacing below 4-km provided relatively little added value . Werner et al. (2019) completed a validation of agri-climate fields derived from downscaled ERA-Interim COSMO5-CLM5 and WRF datasets. The authors compared derived fields, such as evapotranspiration and soil moisture deficits, with observations and found that both RCMs exhibit high skill, with WRF slightly outperforming COSMO5-CLM.
The analysis presented in this chapter confirms that the RCM configurations and domain size of the current study are capable of accurately simulating the climate of Ireland.
Future validation work will focus on downscaling and analysing the more up-to-date and accurate ERA5 global reanalysis dataset from the European Centre for Medium-Range Weather Forecast (ECMWF), in place of ERA-Interim. ERA5 is the fifth generation of the ECMWF global climate reanalysis dataset (C3S, 2017). The ERA5 dataset was not available at the time that the current research was carried out. Furthermore, additional WRF historical simulations will be completed, which will allow for a robust quantification of the relative skill of the COSMO5-CLM and WRF RCMs. For each metric, the best-and worst-performing scores are highlighted in green and red, respectively.

Mid-century Climate Projections
8 Since all ensemble members project increases in temperature, the 33rd percentile is denoted the "likely" projection in this case (conversely, if projections are negative, the 66th percentile is denoted the "likely" projection). Figure 3.1 presents the spatial distribution of annual temperature changes for 2041-2060 relative to 1981-2000. The mean annual temperature is projected to increase by 1-1.2°C and by 1.3-1.6°C for the RCP4.5 and RCP8.5 scenarios, respectively. Temperature projections show a clear west-to-east gradient, with the largest increases in the east.

Temperature Projections
The seasonal temperature projections are presented in Figure 3.2; winter temperatures show increases ranging from 0.9°C in the south-west to 1.2°C in the north-east for the RCP4.5 scenario (1.2°C in the south-west and 1.6°C in the north-east for RCP8.5).
The patterns for spring are similar to winter, with a projected increase in temperature of 0.9°C to 1.0°C for RCP4.5 (1.0°C to 1.3°C for RCP8.5) with a southwest to north-east gradient. Summer temperatures show increases ranging from 1.0°C in the north-west to 1.3°C in the south-east for RCP4.5 (1.3°C in the north-west and 1.8°C in the south-east for RCP8.5).
Autumn shows a west-to-east pattern with expected increases of 1.3°C to 1.5°C and 1.6°C to 1.9°C for the RCP4.5 and RCP8.5 scenarios, respectively. In summary, the temperature change gradient is from south-west to north-east in winter and spring, northwest to south-east in summer and from west to east in autumn and over the full year. These trends are consistent with previous studies (e.g. Gleeson et al., 2013;Nolan, 2015;O'Sullivan et al., 2015) and all RCM-GCM simulations, RCPs and future time periods assessed to date.
As outlined in Chapter 1, a disagreement between RCM ensemble members can result in large individual outliers skewing the mean ensemble projection. For this reason, it can be more informative to consider percentiles when analysing an ensemble of future projections. The relatively large ensemble size of the current study allows the construction of a pdf of climate projections. Likelihood values can then be assigned to the projected changes. This method of analysing percentiles allows for a better understating of climate change uncertainty and it allows for a quantification of conservative and robust ("likely") projections. Conversely, the likelihood method allows for policymakers to consider more "unlikely" (and possibly high-impact) climate projections.
To this end, the 33rd, 50th and 66th percentiles of annual and seasonal mean 2-m temperature projections are presented in Figure 3.3. For example, the annual figures (top panels) show that over 67% (P33) of the ensemble members project an annual increase in temperatures of 0.8-1.1°C and 1.1-1.5°C for the RCP4.5 and RCP8.5 scenarios, respectively. That is to say, it is "likely" that increases in temperature will be greater than or equal to these values. 8 Similarly, the 50th percentile figures (P50) provides information on the "as likely as not" projection. The 66th percentile (P66) provides information on the "unlikely" projection and can be useful for the analysis of high-impact, lowprobability projections. The warming gradient of the annual (Figure 3.1) and seasonal (Figure 3.2) mean projections are also evident in the percentile projections of Figure 3.3. Furthermore, there exists a small variation between the 33rd, 50th and 66th projection percentiles, which demonstrates good agreement (small spread) between ensemble members. Finally, the annual and seasonal warming gradients are similar for both the RCP4.5 and RCP8.5 scenarios. This agreement increases the confidence in the regional projections of temperature.
The annual change in the standard deviation (Figure 3.4) shows small changes of between ≈−0.1°C and ≈0.2°C for both the RCP4.5 and RCP8.5 scenarios. 9 Similarly, the seasonal projected changes in the standard deviation of temperature are small (Figure 3.5); small increases (decreases) are noted for summer (winter) for both RCPs and a mixed signal is noted for spring and autumn. It should be noted that large increases in the mean summer temperature (Figure 3.2) coupled with increases in the standard 9 Please refer to section 1.2.4 for an overview of the effects of changes in the standard deviation on the distribution of a climate field. deviation will lead to enhanced increases in extreme high temperatures (refer to Figure 1.3d for a schematic example of such an outcome). Similarly, increases in mean winter temperature (Figure 3.2) coupled with a decrease in standard deviation will lead to enhanced decreases in extreme low temperatures. However, it should be noted that the projected changes in standard deviation are small for all seasons. The results suggest that although future temperatures will increase substantially for all seasons, the shape of the temperature distribution will remain broadly similar.

Extreme Temperature Projections
Changes in the daily maximum and daily minimum temperatures are arguably of more immediate importance, since extreme events have an abrupt and much larger impact on lives and livelihoods than a gradual change in mean values (Easterling et al., 2000;O'Sullivan et al., 2015). A sustained increase in the daily maximum temperature is associated with heatwaves, whereas an increase in the daily minimum temperature will typically imply warmer nights. Figure 3.6a shows how the warmest 5% of daily maximum temperatures are projected to change (TMAX-95%). A strong warming is evident, which is greater than the projected mean summer increase (Figure 3.2), ranging from 1.0°C to 1.6°C for the RCP4.5 scenario and from 1.4°C to 2.2°C for the RCP8.5 scenario. Warming is greater in the south than in the north. Figure 3.6b shows how the coldest 5% of daily minimum temperatures are projected to change (TMIN-5%). Again, the projected increase of TMIN-5% is greater than the mean winter increase (Figure 3.2), ranging from 0.9°C to 1.8°C for the RCP4.5 scenario and from 1.2°C to 2.4°C for the RCP8.5 scenario.
Warming is greater in the north than in the south.

Heatwaves
The large projected increase in high summer temperatures (TMAX-95%; Figure 3.6a) suggests an increase in the number of heatwave events by the middle of the century. This is confirmed by Figure 3.7, which presents the projected change in the number of heatwave events over the 20-year period 2041-2060. The increases range from 1 to 8 for the RCP4.5 scenario and from 3 to 15 for the RCP8.5 scenario.
Both scenarios exhibit a north-west to south-east gradient. For comparison, the observed number of heatwave events over the period 1981-2000 is presented in Figure 3.7b (derived from daily maximum temperature data provided by Walsh, 2012). The projected increase in heatwaves will have a direct impact on public health and mortality, but this may be offset by the projected decrease in frost and ice days (see section 3.4).
For the analysis of the change in number of heatwaves, the following definition as described in Jacob et al. (2014) was used: heatwaves are considered as periods of more than 3 consecutive days exceeding the 99th percentile of the daily maximum temperature of the May-to-September season of the control period (1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000). Jacob et al. (2014) analysed a large ensemble of relatively lowresolution (12.5 km to 25 km) RCMs and showed small projected increases in heatwave events over Ireland , with a slight north-west to south-east gradient evident over the country. These results are consistent with the findings of the current report.

Frost and Ice Days
The large projected decrease in cold nights ( Figure  3.6b; TMIN-5%) implies a decrease in the number of frost and ice days by the middle of the century. This is confirmed by Figure 3.8a and b, which present the projected annual change in the number of frost and ice days, respectively. Averaged over the whole country, the number of frost days (days when the minimum temperature is < 0°C) is projected to decrease by 45% and 58% for the RCP4.5 and RCP8.5 scenario, respectively. Similarly, the number of ice days (days when the maximum temperature is < 0°C) is projected to decrease by 68% and 78% for the RCP4.5 and RCP8.5 scenarios, respectively. For comparison, the observed annual mean numbers of frost and ice days for 1981-2000 are presented in Figure 3.9a and b, respectively (data from Walsh, 2012). Note that the observed number of ice days is small, particularly in coastal regions. It is worth noting that periods of frost and ice are important environmental drivers that trigger phenological phases in many plant and animal species. Changes in the occurrence of these weather types may disrupt the life cycles of these species (e.g. Williams et al., 2015;Bigler and Bugmann, 2018).

The Growing Season
Within a period of 12 months, the thermal growing season length is officially defined as the number of days between the first occurrence of at least 6 consecutive days with a daily mean temperature > 5°C and the first occurrence of at least 6 consecutive days with a daily mean temperature < 5°C. Figure 3.10a shows a large projected increase in the average length of the growing season over Ireland by the middle of the century. There exists a clear south-to-north gradient, with values ranging from 5% to 18% and from 6% to 23% for the RCP4.5 and RCP8.5 scenarios, respectively. Averaged over the whole country, the length of the growing season is projected to increase by 12% and 16% for the RCP4.5 and RCP8.5 scenarios, respectively. Figure 3.10b, the projected change in the start of the growing season, shows that by the middle of the century the growing season is expected to start 10-31 days earlier for the RCP4.5 scenario and 10-36 days early for the RCP8.5 scenario. Averaged over the whole country, the growing season is projected to start 15 and 24 days early for the RCP4.5 and RCP8.5 scenarios, respectively.
For comparison, the observed length and start of the growing season over the period 1981-2000 (derived from daily mean temperature data provided by Walsh, 2012) are presented in Figure 3.11a and b, respectively.

The Grazing Season
The growing season calculation is based solely on temperature and does not take into account the delay before sufficient plant cover is available to support grazing animals or the ability of animals and machinery to pass over land. The approximate length of the grazing season, in days per year, can be approximated from the following equation (Smith, 1976;Collins and Cummins, 1996): where T is the mean annual 2-m temperature (°C) and R is the mean annual rainfall (mm year −1 ). Figure 3.12a shows that, by the middle of the century, the grazing season is projected to increase by 12-46 and 17-55 days per year for the RCP4.5 and RCP8.5 scenarios, respectively. Averaged over the whole country, the grazing season is projected to increase by It should be noted that not all areas presented in Figures 3.10 and 3.12a are suitable for agriculture and/or forestry. The projections should therefore be considered in the context of an observed soil/grass map, as presented in Figure 3.12b.

Growing Degree Days (Crops and Pests)
A degree day, an estimate of accumulated heat, is defined as the deviation (°C) from a reference temperature value (Fraisse et al., 2010;Project Team ECA&D, 2013;Kendon et al., 2015). Degree days represent the number of degrees by which the temperature has gone above or below a threshold.

(b) (a)
Growing degree days (GDDs) are used to predict the growth and development of plants, insects and diseases of which the developments are very dependent on temperature and the daily accumulation of heat. The amount of heat required to advance a plant or pest to the next development stage remains constant from year to year; however, the actual amount of time (days) can vary considerably because of weather conditions. Each crop, insect and disease species has a minimum base temperature (T b ) or threshold below which development does not occur (Fraisse et al., 2010;OMAFRA, 2017). For example, in Europe, 5.5°C applies to wheat, barley, rye, oats and lettuce, 8°C to sunflowers and potatoes and 10°C to American maize, rice, corn, and tomato (McMaster and Wilhelm, 1997;Miller et al., 2001;Spinoni et al., 2015). See Table 3.1 for a list of base temperatures for crops and pests (McMaster and Wilhelm, 1997;Johnson et al., 1998;Miller et al., 2001;Spinoni et al., 2015).
The GDD was computed using the daily mean temperature (T M ) for different base temperatures (T b ), as described in Spinoni et al. (2015) and Project Team ECA&D (2013): (3.2) (3.3) Figure 3.13 shows that the GDDs for "crop base temperatures" 5.5°C, 8°C and 10°C are projected to increase substantially by the middle of the century. The projected increases are largest for the higher baseline temperatures. Averaged over the whole country, GDDs are projected to increase by 23% for RCP4.5 and 30% for RCP8.5 (T b = 5.5°C); 32% for RCP4.5 and 42% for RCP8.5 (T b = 8°C); and 45% for RCP4.5 and 59% for RCP8.5 (T b = 10°C) (see Table 3.1). The results suggest a warming climate may present some positive opportunities for farming. However, the results should be viewed in the context that a warming climate will also result in an increase in pests as a result of an increase in heating and a decrease in frost and ice days (see section 3.4). Figure 3.14 and Table 3.1 show that the GDDs for "pest base temperatures" 6°C, 7°C, 9°C and 10°C are similarly projected to increase substantially by the middle of the century. Furthermore, projected increases in extreme temperatures (section 3.2), heatwaves (section 3.3), heavy precipitation (section 3.10) and dry periods/ droughts (section 3.11) will have substantial adverse effects on agriculture in Ireland by the middle of the century.

Ontario Crop Heat Units
The Ontario Crop Heat Unit (OCHU) is a variant of a degree day accumulation and is widely used to rate the suitability of regions for production of corn/ maize (Collins and Cummins, 1996;Bootsma et al., 1999Bootsma et al., , 2007OMAFRA, 2017). The OCHU model uses separate calculations for maximum and minimum temperatures. The maximum or daytime relationship (Y max ) uses 10°C as the base temperature and has a curvilinear response to temperature with a maximum GDD = GDD daily ∑ and T min and T max are the daily minimum and maximum temperatures, respectively. Figure 3.15 shows that, by the middle of the century, OCHUs (May to September) are projected to increase by 18-38% and 23-49% for the RCP4.5 and RCP8.5 scenarios, respectively. Averaged over the whole country, OCHUs are projected to increase by 23% and 29% for the RCP4.5 and RCP8.5 scenarios, respectively.
Daily OCHU= Y max + Y min 2  Figure 3.16 presents the mean annual percentage change in precipitation for the RCP4.5 and RCP8.5 scenarios. There is an indication of a slight reduction in the annual precipitation of ≈0-6% for the RCP4.5 scenario. However, projected reductions are small (≈0%) over most of the country for both scenarios. Figure 3.17a presents the seasonal change (%) in precipitation for the RCP4.5 scenario; the corresponding plots for RCP8.5 are presented in Figure 3.17b. The strongest signals are a projected decrease for summer, with the largest impacts for the RCP8.5 scenario. The summer reductions range from ≈0% to 11% for the RCP4.5 scenario and from 2% to 17% for the RCP8.5 scenario. Other seasons show a small projected change in precipitation, with the exception of winter under the RCP8.5 scenario, where precipitation is expected to increase by ≈0 to 11%.

Mid-century Precipitation Projections
The projected precipitation changes of the current study vary greatly between ensemble members, much more so than for the temperature projections (see section 3.1). The regional details of Figures 3.16 and 3.17 are therefore not reliable. Furthermore, the disagreement between RCM projections can result in large individual outliers skewing the mean ensemble  projection. It is therefore more informative to consider the 33rd, 50th and 66th percentile projections as presented in Figure 3.18. The likelihood values are calculated at each grid point using the full RCM ensemble of projections, as outlined in Table 1 are summer (both RCP4.5 and RCP8.5 scenarios) and annual and autumn (RCP4.5 projections), when the signals are more robust. The strongest signal is for a substantial drying during summer, when there is a clear agreement between the percentiles (for both the RCP4.5 and RCP8.5 scenarios).

Changes in the variability of the precipitation climate
The uncertainty of the mean precipitation projections may be partly attributed to a projected increase in the variability of the future Irish precipitation climate, resulting in an increase in both dry periods and heavy rainfall events. This is clearly demonstrated in Figures 3.19 and 3.20, which present the annual and seasonal projected change in the standard deviation of 3-hour precipitation, respectively. It is noted that the standard deviation of precipitation is expected to increase substantially for all seasons by the middle of the century, with the exception of spring, when the signal is less pronounced for the RCP8.5 scenario. The large projected change in the standard deviation of precipitation, coupled with small changes in the mean, imply an increase in both dry periods and heavy rainfall events (i.e. the tails of rainfall distribution will become more pronounced; see Figure 1.3b for a schematic example of such an outcome). This is confirmed by Figures 3.21 and 3.22, which show large projected changes in intense rainfall events and dry periods, respectively. The projections of increased variability of the precipitation climate will have adverse implications for society (e.g. droughts, flooding, water management and housing) and sectors of the economy, such as agriculture. Furthermore, the increase in frequency of both droughts and heavy rainfall events could be detrimental to potential gains of a warming climate to the agricultural sector, as discussed in sections 3.5-3.8.

Heavy Precipitation Events
Changes in the occurrence of heavy rainfall events are of particular importance because of the link with flooding. In this section, mid-century projections of "wet days" and "very wet days" are presented. A "wet day" is defined as one on which the daily precipitation amount is greater than 20 mm. A "very wet day" is defined as one on which the daily precipitation is greater than 30 mm. 10 Figure 3.21a indicates an increase in the annual number of wet days for the RCP4.5 (mean value 10%) and RCP8.5 (mean value 14%) scenarios. The largest increases in wet days are expected for the autumn (9% for RCP4.5; 16% for RCP8.5) and winter (17% for RCP4.5; 26% for RCP8.5) months; a small increase is projected for summer (≈10% for both scenarios) and a mixed signal was found for spring. The projected increase in the annual number of very wet days, presented in Figure 3.21b, is substantial, with mean values of 21% and 31% for the RCP4.5 and RCP8.5 scenarios, respectively. Again, the largest increases were noted for the autumn (25% for RCP4.5; 42% for RCP8.5) and winter (33% for RCP4.5; 61% for RCP8.5) months.
A "likely" increase was also noted in the number of annual wet days (5% for RCP4.5; 8% for RCP8.5), annual very wet days (8% for RCP4.5; 19% for RCP8.5), autumn wet days (1% for RPC4.5; 6% for RCP8.5), autumn very wet days (3% for RCP4.5; 18% for RCP8.5), winter wet days (4% for RCP4.5; 12% for RCP8.5) and winter very wet days (2% for RCP4.5; 18% for RCP8.5). 11 It follows that it is "likely" that increases in heavy rainfall events will be greater than or equal to these values. The increased frequency of heavy precipitation is well marked in winter and autumn and over the full year, particularly for the RCP8.5 scenario, but regional details are not reliable because of a large spread in the ensembles. The projected increase in heavy rainfall events during autumn, winter and over the full year is consistent with the large projected change in standard deviation of the 3-hour precipitation presented in Figures 3.19 and 3.20.
The projected increase in the number of wet and very wet days should be considered in the context of historical values. Figure 3.23 presents the annual number of observed wet and very wet days, averaged The projected change in heavy rainfall events are in line with previous RCM studies for Ireland, which showed large projected increases in intense rainfall by the middle of the century, particularly during the winter and autumn months (e.g. Gleeson et al., 2013;Nolan, 2015;Nolan et al., 2017).

Dry Periods
To quantify the potential impact of climate change on future drought events, the change in the number of dry periods was analysed. A dry period is defined as at least 5 consecutive days on which the daily precipitation is less than 1 mm. Figure 3.22a indicates an increase in the annual number of dry periods for the RCP4.5 and RCP8.5 scenarios (mean value ≈16% for both RCPs). The largest increases in dry periods are expected for the summer months (20% for RCP4.5; 27% for RCP8.5 ; see Figure 3.22b). Substantial increases in dry periods are also expected for winter and autumn; projections are similar for both seasons, with an expected increase of ≈20% for both scenarios. A mixed signal was found for spring.
A "likely" increase was also noted in the number of annual (≈9% for both RCPs), winter (≈9% for both RCPs), summer (11% for RCP4.5; 18% for RCP8.5) and autumn (≈10% for both RCPs) dry periods. It follows that it is "likely" that increases in dry periods will be greater than or equal to these values. The projected increase in the frequency of dry periods is well marked in summer, winter, autumn and over the full year, but regional details are not reliable because of a large spread in the ensembles. The projected increase in dry periods during winter, summer, autumn and over the full year is consistent with the large projected change in standard deviation of precipitation (presented in Figures 3.19 and 3.20).
The projected percentage increase in the number of dry periods should be considered in the context of the historical number of dry periods. The projected changes in dry periods are in line with previous RCM studies for Ireland, which showed large projected increases in dry events by the middle of the century, particularly during the summer and autumn months (e.g. Gleeson et al., 2013;Nolan, 2015;Nolan et al., 2017). The current study has additionally demonstrated a "likely" increase in dry periods during mid-century winter months.  Figure 3.25 shows that annual snowfall is projected to decrease substantially by the middle of the century for the RCP4.5 (mean value 52%) and RCP8.5 scenarios (mean value 63%). The largest decreases are noted over low-lying regions. Figure 3.26 shows a small variation between the 33rd, 50th and 66th projection percentiles, which demonstrates good agreement (small spread) between ensemble members. Averaged over the whole country, the "likely" decreases in midcentury snowfall are 51% and 60% for the RCP4.5 and RCP8.5 scenarios, respectively.  of the surrounding sea. The projections show a slight reduction in the 10-m wind speed of 1-2.7% (mean value 1.8%) for the RCP4.5 scenario and 1.6-3.3% (mean value 2.6%) for the RCP8.5 scenario. Figure 3.28a presents the seasonal change (%) in 10-m wind speed for the RCP4.5 scenario; the corresponding plots for RCP8.5 are presented in Figure 3.28b. All seasons show a projected decrease in mean 10-m wind speed. The decreases are largest for summer under the RCP8.5 scenario. The summer reductions range from 0.3% to 3.4% for the RCP4.5 scenario and from 2% to 5.4% for the RCP8.5 scenario.

10-m Wind Speed Projections
With the exception of spring and autumn under the RCP4.5 scenario, Figure 3.29 shows a small variation between the 33rd, 50th and 66th 10-m wind speed projection percentiles. This agreement adds a level of confidence to the projected reductions during summer (both RCPs), winter (both RCPs), spring (RCP8.5), autumn (RCP8.5) and over the full year (both RCPs).   in mean 10-m wind speed, coupled with a decrease in standard deviation, implies a shift to the left of the wind speed distribution and a decrease in the frequency of higher wind speeds. Figure 3.32a shows that annual specific humidity (the amount of water vapour in the atmosphere calculated as the ratio of the mass of water vapour to the total mass of the air parcel) is projected to increase substantially by the middle of the century for both the RCP4.5 (mean value 8%) and RCP8.5 (mean value 11%) scenarios. There exists a clear south-west to north-east gradient in the projections, with the largest increases in the north. For reference, the "observed" mean annual specific humidity (g kg −1 ), as resolved by a high-resolution (1.5-km) downscaled ERA-Interim climate simulation, is presented in Figure 3.32b. Please refer to Flanagan et al. (2019) for an overview of the climate simulation configuration and validation results. good agreement (small spread) between ensemble members and adds a high level of confidence to the projections. Averaged over the whole country, the "likely" projected increases in mid-century specific humidity are 7% (annual RCP4.5), 10% (annual RCP8.5), 7% (winter RCP4.5), 9% (winter RCP8.5), 6% (spring RCP4.5), 8% (spring RCP8.5), 7% (summer RCP4.5), 10% (summer RCP8.5), 9% (autumn RCP4.5) and 12% (autumn RCP8.5).

Specific Humidity Projections
Specific humidity has direct impacts on animal and human health. Epidemiological studies indicate that low levels of specific humidity are associated with greater influenza mortality (e.g. Shaman et al., 2010Shaman et al., , 2011Tamerius et al., 2013). Barreca and Shimshack (2012) showed that the humidity-influenza relation is nonlinear, with lower specific humidity levels resulting "in greater influenza mortality at mean daily specific humidity levels below 6 g kg −1 ", and that "incremental changes in humidity do not significantly affect influenza mortality when mean daily specific humidity exceeds a 6 g kg −1 threshold". An increase in specific humidity will amplify the adverse effects of increases in extreme temperatures (section 3.2) and heatwaves (section 3.3) and lead to higher mortality by limiting heat loss through evaporative cooling (Coffel et al., 2017).

Relative Humidity Projections
Relative humidity is the ratio of the amount of water vapour present in the air to the greatest amount possible at the same temperature. Warm air can hold substantially more moisture than cold air, meaning that the relative humidity of cold air is far higher than 12 For relative humidity, rather than a relative percentage change, climate projections (%) are calculated as "future (%) minus past (%)". that of warm air for equal absolute humidity levels. Relative humidity is expressed as a percentage, with 0% corresponding to totally dry air and 100% to totally saturated air (leading to increased probability of precipitation). Figure 3.35a presents the mean annual change (%) 12 in surface relative humidity for the RCP4.5 and RCP8.5 scenarios. The projections show a slight increase in relative humidity of ≈0-0.6% (mean value 0.1%) for the RCP4.5 scenario and 0-1% (mean value 0.35%) for the RCP8.5 scenario. There exists a south-east to north-west gradient in the projections, with the largest increases in the north. For reference, "observed" mean annual relative humidity (%), as resolved by a high-resolution (1.5-km) downscaled ERA-Interim climate simulation, is presented in Figure  3.35b. Please refer to Flanagan et al. (2019) for an overview of model configuration and validation. Figure 3.36 presents the seasonal change (%) in relative humidity for the RCP4.5 and RCP8.5 scenarios. All seasons, with the exception of summer, show an increase (or ≈0% change) in relative humidity. The largest increases are noted for spring (both RCP scenarios) and winter (RCP8.5). For summer, relative humidity is projected to decrease in the south-east and increase in the north-west (both RCP scenarios). The percentile projection figures, presented in Figure 3.37, show a small variation between the 33rd, 50th and 66th for winter (RCP8.5), spring (both scenarios), over the full year (RCP8.5) and, to a lesser extent, during autumn (RCP8.5). The signal of decreases in the south-east and increases in the north-west during summer is evident for all percentiles and both RCPs. This agreement adds a level of confidence to the projected changes for summer (both RCPs) and the projected increases for spring (both RCPs), winter (RCP8.5), autumn (RCP8.5) and annual (RCP8.5).
The projections are somewhat contrary to the general consensus in the scientific literature that relative humidity will decrease over land in a warming climate. For example, Byrne and O'Gorman (2018) showed that in recent decades  the land surface has warmed substantially more than the ocean surface and, consequently, relative humidity has fallen over land. Declining relative humidity over land is also the dominant feature of future climate projections, with models predicting that future changes in surface temperature will be strongly amplified over land (Sutton et al., 2007;Byrne and O'Gorman, 2013) and that relative humidity will decline over land and either increase (M. Collins et al., 2013) or remain approximately constant (O'Gorman and Muller, 2010;Byrne and O'Gorman, 2016) over the oceans. However, the projected trend of decreasing relative humidity over land is not universal. For example, the CMIP5 ensemble of global projections show a projected increase in relative humidity in some coastal land regions, such as West Africa, the Middle East and India (e.g. Figure 1a of Byrne et al., 2016;Figure 12.21 of M. Collins et al., 2013). The projections over Ireland are small and a large projected increase is noted over the North Atlantic region directly to the north and west of Ireland (which is particularly noticeable in Figure  1a of Byrne et al., 2016). Sloth et al. (2012) analysed the ENSEMBLES dataset, a large ensemble of RCM climate projections for Europe (van der Linden et al., 2009), and found a mixed signal for mid-century projections of relative humidity, with a number of ensemble members showing an increase in relative humidity, particularly over Ireland and Northern Europe (see Figure 9 of Sloth et al., 2012). The authors found that the regions where large changes in relative humidity are projected generally correspond to those with large projected increases in mean temperature (relative humidity decreases) or large projected increases in precipitation (relative humidity increases). The projections of the current study, of small projected increases in relative humidity for all seasons except summer, may be partly attributed to the large influence of the North Atlantic Ocean on the Irish climate. The relative humidity projections for summer (decreases in the south-east and increases in the north-west) may be partly attributed to the temperature projections for summer; note that Figure 3.2 shows a similar southeast to north-west gradient in projected temperature increases, with enhanced warming in the south-east. However, further investigation of these factors is necessary to attribute causation to the relative humidity projections of the current study.
Relative humidity is an important climate field that has a direct impact on many sectors, including public health, agriculture and the built environment. For example, the link between low relative (and specific) humidity and greater influenza mortality is well established (e.g. Noti et al., 2013). The incidences of Lyme borreliosis (Lyme disease), a vector-borne illness caused by the bacterium Borrelia and spread by ticks, increase with high relative humidity; ticks require a minimum 80% humidity to avoid drying out during the early stages of life (Cullen, 2010) and air temperatures greater than 6°C during host questing (Süss et al., 2008). Potato crop failures in Ireland can result when high relative humidity and temperature combine to provide the warm, wet conditions in which the Phytophthora infestans fungi (potato blight) thrive (Cucak et al., 2019). Changes in relative humidity will have an impact on the built and archaeological heritage of Ireland, affecting deterioration mechanisms such as salt weathering, mould growth and corrosion (Daly, 2019). Relative humidity is also an important field for derived variables, such as fire risk indexes; the risk of wildfire decreases with increasing relative humidity (e.g. Dowdy et al., 2010).   Figure 3.40 shows a small variation between the 33rd, 50th and 66th projection percentiles for all seasons and both RCP scenarios. This result demonstrates good agreement (small spread) between ensemble members and adds a high level of confidence to the MSLP projections. Averaged over the whole country, the "likely" projected increases in the mid-century average MSLP are 1.2 hPa (annual RCP4.5), 0.9 hPa (annual RCP8.5), 0.6 hPa (winter RCP4.5), 0.2 hPa (winter RCP8.5), 0.8 hPa (spring RCP4.5), 0.6 hPa (spring RCP8.5), 0.6 hPa (summer RCP4.5), 1.3 hPa (summer RCP8.5), 1.6 hPa (autumn RCP4.5) and 1 hPa (autumn RCP8.5).

Mean Sea Level Pressure Projections
The projected increases in MSLP are some of many possible factors that could contribute to the projections of decreases in wind speed (section 3.13) and wind power (section 3.18), and increases in dry periods (section 3.11) and heatwave (section 3.3) events.
The projected increase in MSLP may be attributed to the projected decrease in the number of overall cyclones (section 3.17). A discussion on possible mechanisms for a reduction in the number of mid-latitude storms in a warming world is provided in section 3.17. However, further investigation is necessary to attribute causation to the MSLP projections of the current study. Future work will attempt to address this issue by analysing a substantially larger RCM ensemble of downscaled CMIP6 data. Furthermore, the impact of changes in both the frequency and intensity of low pressure systems on MSLP will be quantified.

Storm Track Projections
Given the large societal impacts of extreme storms, there is considerable interest in the potential impact of climate change on extreme cyclonic activity in the North Atlantic. Windstorms and associated high wind speeds are a major source of natural hazard risk for Ireland and many countries across Europe. For example, Ireland and the UK were severely affected by an exceptional run of storms during the winter of 2013/2014, culminating in serious coastal damage and widespread, persistent flooding. Reports issued by the meteorological agencies of Ireland and the UK have confirmed that records for precipitation totals and extreme wind speeds were set during this period (Met Éireann, 2014;Kendon et al., 2015). Matthews et al. (2014) found that the UK/Ireland winter of 2013/2014 was the stormiest for at least 143 years when storm frequency and intensity are considered together. In addition to the potential widespread flooding and structural damage associated with intense storms, the wind energy supply can be negatively affected, as wind turbines are shut down during periods of high wind speeds to prevent damage. Feser et al. (2014) conducted a review of studies of storms over the North Atlantic and north-western Europe to identify potential long-term trends. Storm trends derived from reanalyses data and climate model data for the past were mostly limited to the last four to six decades. They found that "the majority of these studies find increasing storm activity north of about 55-60°N over the North Atlantic with a negative tendency southward" (Feser et al., 2014). Furthermore, "future scenarios until about the year 2100 indicate mostly an increase in winter storm intensity over the North Atlantic and western Europe. However, future trends in total storm numbers are quite heterogeneous and depend on the model generation used" (Feser et al., 2014). Matthews et al. (2016) analysed cyclone trends in the British and Irish Isles (BI) by assessing a 142-year (1871-2012) record of cyclone frequency, intensity and "storminess" derived from the 20th-century Reanalysis V2 (20CR) dataset. They found an "upward trend in cyclone intensity for the BI region, which is strongest in winter and consistent with model projections" (Matthews et al., 2016). Zappa et al. (2013) analysed a CMIP5 ensemble of 19 GCMs and found a small, but significant, increase in the number and intensity of winter cyclones associated with strong wind speeds over the UK by the end of the century. A 2013 study with a very high-resolution version of the EC-Earth model (Haarsma et al., 2013) suggests an increase in the frequency of extreme wind storms affecting Western Europe in future autumn seasons as a result of climate change.
As part of the current study, an algorithm was developed to identify and track cyclones. The algorithm was applied to an ensemble subset of EURO-CORDEX 12-km downscaled CMIP5 data.
Results show a reduction of ≈10% in the numbers of less intense storms affecting Ireland and suggest an eastward extension of the more severe wind storms over Ireland and the UK from the middle of the century. 13 Figure 3.41 presents intense storm tracks as simulated by the European Coordinated Regional climate Downscaling Experiment (EURO-CORDEX) ensemble. Previous studies that analysed RCM projections of future extreme storm events over Ireland are in broad agreement with these results (Semmler et al., 2008a,b;Nolan, 2015;McGrath and Nolan, 2017). It should be noted that extreme storms, as presented in Figure 3.41, are rare events. Therefore, the storm projections should be considered with a high level of caution. Future work will focus on analysing a larger ensemble of downscaled CMIP6 data, thus allowing a robust statistical analysis of extreme storm track projections.
13 Note that because extreme storms are very rare events, a slight increase has no noticeable effect on the mean wind speed statistics presented in section 3.13. The projected decrease in less intense (very common) storms has a substantially greater effect on decreasing the frequency of higher wind speeds than the increase in very rare intense storms.
14 A more accurate measure of wind power is given by wind power = 0.5 × air density × (wind speed) 3 . Because air density was not archived for all RCM ensemble members, we focus on the cube of the wind speed as a measure of wind power.
The warming of the climate system on account of greenhouse gas forcing is expected to change the thermal structure of the lower atmosphere; the enhanced warming of the poles, particularly in the Arctic, will reduce the equator-to-pole temperature gradient and this effect is often appealed to as a mechanism for a reduction in the number of midlatitude storms in a warming world. For example, Geng and Sugi (2003) and Semmler et al. (2008b) proposed that a decreased meridional temperature gradient and the associated reduced baroclinicity in the future climate could be responsible for the decrease of the total number of cyclones. Furthermore, the higher moisture supply as a result of a generally higher sea surface temperature and the related increase in latent heat fluxes could trigger strong-intensity cyclones (Hall et al., 1994;Semmler et al., 2008b). Further work, analysing a large ensemble of downscaled CMIP6 datasets, is required to fully investigate these issues.

120-m Wind Power Projections
There is considerable interest among policymakers and the energy industry in renewable energy resources as a means of reducing carbon dioxide emissions to minimise climate change (Solomon et al., 2007). Within this context, the current section assesses the impacts of climate change on the future wind energy resource of Ireland. Because the energy content of the wind is proportional to the cube of the wind speed, we focus on projections of the mean cube of the wind speed. 14 Furthermore, wind energy projections at 120 m above the surface were analysed, so the results provide information at a typical turbine height. 15 Figure 3.42 presents the mean annual percentage change in 120-m wind power for the RCP4.5 and RCP8.5 scenarios. For the purpose of onshore and offshore wind energy applications, the analyses of wind power cover all land points and a small portion of the surrounding sea. The projections show a slight reduction in the 120-m wind power of 3.1-5.8% (mean value 4.5%) for the RCP4.5 scenario and 4.5-7.5% (mean value 6%) for the RCP8.5 scenario. Figure 3.43a presents the seasonal change (%) in 120-m wind power for the RCP4.5 scenario; the corresponding plots for RCP8.5 are presented in Figure 3.43b. All seasons show a projected decrease in mean 120-m wind power. The decreases are largest for summer under the RCP8.5 scenario. The summer reductions range from 2.8% to 8.7% for the RCP4.5 scenario and from 6.5% to 14.1% for the RCP8.5 scenario.
With the exception of spring and autumn under the RCP4.5 scenario, Figure 3.44 shows a small variation between the 33rd, 50th and 66th 120-m wind power projection percentiles. This agreement adds a level of confidence to the projected reductions in wind power during summer (both RCPs), winter (both RCPs), spring (RCP8.5), autumn (RCP8.5) and over the full year (both RCPs). The annual change in the standard deviation of 120-m wind power (Figure 3.45) shows small changes of between ≈−5.5% and −0.6% for both the RCP4.5 and RCP8.5 scenarios. All seasons -except winter and autumn (RCP4.5), when small (≈0%) changes are noted -show reductions, with the largest decreases noted during summer (Figure 3.46). A reduction in 120-m wind power, coupled with a decrease in The projected changes in wind power are in line with previous RCM studies for Ireland, which showed projected decreases in wind power by the middle of the century during summer, spring and over the full year (e.g. Nolan et al., 2014;Nolan, 2015). Figure 3.47a, the projected change in mean annual surface shortwave radiation, shows ≈0% or small decreases by the middle of the century. Small projected decreases are also noted for winter and spring, whereas summer and autumn show small increases, particularly in the south (not shown). However, the projected changes are small and so will have minimal effect on solar energy, agricultural and health impacts (e.g. skin cancer and vitamin D deficiency). The radiation projections are in line with similar European studies. Jerez et al. (2015) analysed an ensemble of EURO-CORDEX RCM projections and found the ensemble mean pattern of change for mean surface-downwelling shortwave radiation (2070( -2099( vs 1970( -1999 shows "positive signals (about 5 W m −2 ) in Southern Mediterranean regions" and "negative signals northwards (about −10 W m −2 , down to −20 W m −2 in the northernmost areas)". Bartók et al. (2017) also analysed a EURO-CORDEX RCP85 ensemble and found that "the multi-model mean of RCMs indicates a decrease in surface solar radiation of −0.60 W m −2 per decade over Europe" for the period 2006-2100. The authors proposed that the reduction of surface solar radiation in the RCMs "can be attributed to increasing atmospheric absorption in line with the increase of water vapor content" (Bartók et al., 2017).

Surface Shortwave Radiation and Solar Photovoltaic Power
A more comprehensive analysis of the impacts of climate change on solar photovoltaic (PV) power in Ireland was evaluated using the following method outlined in Jerez et al. (2015) and Mavromatakis et al. (2010): (3.7) where SW refers to surface-downwelling shortwave radiation (W m -2 ) and SW S refers to surfacedownwelling shortwave radiation at standard test conditions (SW S = 1000 W m -2 ), for which the nominal capacity of a PV device is determined as its measured power output. P R is the "performance ratio", formulated to account for changes of the PV cells, efficiency as a result of changes in their temperature as: ( 3.8) where T cell is the PV cell temperature, T s = 25°C (standard test conditions) and γ = −0.005°C -1 , considering the typical response of monocrystalline silicon solar panels (Tonui et al., 2008). T cell is modelled considering the effects of surface temperature, T (°C), SW (W m -2 ) and wind speed, W (m s −1 ), on it as: T cell (t) = c 1 +c 2 T + c 3 SW +c 4 W (3.9) with c 1 = 4.3°C, c 2 = 0.943, c 3 = 0.028°C m 2 W -1 and c 4 = −1.528°C s m -1 , as recommended by Jerez et al. (2015) and Tonui et al. (2008).
The projected change in PV, presented in Figure 3.47b, shows a small expected decrease in PV by the middle of the century, ranging from ≈0 to 4%. The largest decreases are noted in the north of the country and for the RCP8.5 scenario. The results are consistent with Jerez et al. (2015), who analysed the effects of climate change on PV in Europe using an ensemble of Euro-CORDEX datasets.

Heating Degree Days
A degree day, an estimate of accumulated heat, is defined as the deviation (°C) from a base temperature value (Fraisse et al., 2010;Kalogirou, 2013;Project Team ECA&D, 2013;Kendon et al., 2015). Heating degree days (HDDs) are used by power companies and consumers to estimate the amount of energy required for residential or commercial space heating during the cold season. Conversely, cooling degree days (CDDs) are used to estimate the amount of air conditioning usage during the warm season.
The HDD was computed using a base temperature of 15.5°C (i.e. a temperate below which heating is required) and the daily mean temperature (T M ), as described by Spinoni et al. (2015) and Project Team ECA&D (2013): HDD daily = max{(15.5°C -T M ),0} (3.10) HDD = ΣHDD daily (3.11) Conversely, CDD was computed using a base temperature of 22°C (i.e. a temperate above which air conditioning is required) and the daily mean temperature (T M ): CDD daily = max{(T M -22°C),0} (3.12) CDD = ΣCDD daily (3.13) Figure 3.48, the projected change in HDDs, shows that by the middle of the century there will be a greatly reduced requirement for heating in Ireland, with HDDs projected to decrease by 12-17% and 15-21% for the RCP4.5 and RCP8.5 scenarios, respectively. A clear north-to-south gradient is evident for both RCP scenarios. Averaged over the whole country, the expected decreases in HDDs are 14% and 18% for the RCP4.5 and RCP8.5 scenarios, respectively. The projections show that CDDs are projected to slightly increase (not shown), particularly over the east and midlands, suggesting a very small increase in air conditioning requirements by the middle of the century. However, the amounts are small compared with HDD and therefore have a negligible effect on the projected changes in the total energy demand (EDD = HDD + CDD).
The projected changes in heating and cooling energy demand are in line with previous RCM studies for Ireland. Semmler et al. (2010) found that the midcentury (2021-2060) heating demand is projected to decrease by ≈10% for the A1B and A2 emissions scenarios. The authors found a small projected increase in summer CDDs, which may intensify a weak demand for air conditioning towards the end of the century (Semmler et al., 2010). However, the "main influence of a warming climate will be reflected in a decrease in energy requirements for commercial and domestic heating in Ireland" (Semmler et al., 2010).

Driving Rain
The "driving rain" metric (m 2 s -1 year -1 ) can be approximated from the following equation (Collins and Cummins, 1996;Met Éireann, 2010): where W is the mean annual 10-m wind speed (m s −1 ) and R is the mean annual rainfall (m year −1 ).
The driving rain metric is a useful parameter for agriculture, construction and transport applications. Figure 3.49a shows that by the middle of the century, "driving rain" is projected to decrease by 1-7% and 0-8% for the RCP4.5 (mean decrease 4.5%) and RCP8.5 (mean decrease 4%) scenarios, respectively. Decreases are projected for all seasons (except winter under the RCP8.5 scenario). The largest decreases are noted for summer, with overall mean decreases of 7% (RCP4.5) and 12% (RCP8.5). Increases in driving rain are projected for winter under the RCP8.5 scenario, with the largest increases noted near the south and east coastlines (see Figure 3.49b).

Evapotranspiration
Projections of evapotranspiration are presented in Figure 3.50a. The projections show that by the middle of the century, evapotranspiration is projected to increase by 2.3-7% and 1.8-8% for the RCP4.5 and RCP8.5 scenarios, respectively. The largest increases are noted in the east. All seasons show increases in evapotranspiration by the middle of the century, with the largest increases noted for summer and autumn. The projected increase in evapotranspiration may offset flooding events arising from the expected increases in heavy rainfall events (see section 3.10). For reference, "observed" annual evapotranspiration (mm day −1 ), derived from a high-resolution (1.5-km) downscaled ERA-Interim climate simulation, is presented in Figure 3.50b. Please refer to Werner et al. (2019) for validations and additional maps and information. 16 Evapotranspiration was calculated using the output of RCMs (see Table 1.1). The Penman-Monteith FAO-56 method of Zotarelli et al. (2010) was followed. A mathematical description is provided below.
(3.15) 16 In summary, Werner et al. (2019) compared modelled evapotranspiration with observational data at 22 stations spanning Ireland and found that the COSMO-CLM RCM resolved evapotranspiration to "within 10% of values calculated from station measurements for all stations analysed".
• ET sz is the reference evaporation, mm day -1 ; • R n is the net surface radiation, MJ m -2 day -1 (see equation 3.16); • G is the surface sensible heat flux, MJ m -2 day -1 (see equation 3.17); • T mean is the mean daily 2-m temperature,°C; • u 2 is the mean daily 2-m wind speed, m s -1 (see equation 3.18); • e s is the saturation vapour pressure (daily average), kPa (see equation 3.19); • e a is the actual vapour pressure (daily average), kPa (see equation 3.20); • Δ is the slope of the vapour pressure curve, kPa°C -1 (see equation 3.21); • γ is the psychrometric constant, kPa°C -1 (see equation 3.22); • C n is the reference crop type constant numerator; • C d is the reference crop type constant denominator.
For the calculation above, C n = 900 and C d = 0.34 were used.

Recommendations
There is higher confidence in the temperature (and derived variables related to temperature) projections than in the rainfall projections. This is reflected in a rather large spread, particularly at a regional level, in the rainfall projections between the individual RCM ensemble members. Current RCM research aims to reduce climate change projection uncertainty and provide sharper estimates of expected climate change in the decades ahead. This is being achieved by running a large ensemble of high-resolution downscaled simulations using the most up-to-date RCMs (both standard and coupled atmosphere-ocean-wave), additional CMIP5 GCM datasets including the RCP2.6 and RCP6.0 scenarios and recently completed CMIP6 GCM simulations (e.g. Nolan and McKinstry, 2020) under the full range of ScenarioMIP "tier 1" SSPs: SSP1-2.6, SSP2-4.5, SSP3-7.0 and SSP5-8.5 (Riahi et al., 2017). Additionally, the accuracy and usefulness of the model predictions will be enhanced by increasing the model resolution (≈3 km) and using fully coupled atmosphere-ocean-wave RCMs. Preliminary RCM projection results are in line with previous work showing, for example, enhanced temperature rises by  the end of the century (Figure 4.1), wetter winters with a clear north-west to south-east gradient (Figure 4.2) and a general decrease in wind speeds during summer. The preliminary results shown in Figures 4.1 and 4.2 were obtained by extending the COSMO5-CLM and WRF simulations outlined in Table 1.2 to cover the period 1975-2100 and inclusion of the RCP2.6 and RCP6.0 scenarios in the simulation of the future climate.
It is also important to stress that the likelihood values presented in the current study are derived from the most up-to-date evidence available. Therefore, the "likelihood" values only apply to the specific sets of high-resolution models and experimental design of the current study. It is expected that future improvements in modelling will alter the projections, as uncertainty is expected to be gradually reduced.
As extreme storm events are rare, the storm-tracking research needs to be extended. Future work will focus on analysing a larger ensemble, thus allowing a robust statistical analysis of extreme storm track projections.