Theoretical and Applied Climatology https://doi.org/10.1007/s00704-020-03400-3 ORIGINAL PAPER Comparison of GCM and RCM simulated precipitation and temperature over Central America and the Caribbean Alejandro Vichot-Llano 1 Abel Centella-Artola 1 & Daniel Martinez-Castro 1,2 & Filippo Giorgi 3 & Arnoldo Bezanilla-Morlot 1 & Received: 24 May 2019 / Accepted: 17 September 2020 # Springer-Verlag GmbH Austria, part of Springer Nature 2020 Abstract The climate features over Central America and the Caribbean are simulated with the regional climate model RegCM4 to analyze the performance of the model in reproducing precipitation and temperature patterns over the region. Results from RegCM4 and the driving global climate model (GCM) HadGEM2-ES (HadG) are compared against gridded observations for a period from January 1980 to December 2004 using a perturbed physics ensemble formed by four configurations of the RegCM4 using different combinations of four convection schemes run over land and ocean areas: Kain Fritsch (Kf), Emanuel (Em), Grell (Ge), and Tiedtke (Tk). The added value of RegCM4 relative to the GCM simulation is also estimated, with focus over four subregions, using several metrics calculated on the RegCM4 as well as the GCM grids. The RegCM4 downscaling, expressed by the ensemble mean ENS, adds significant details to the HadG simulation of the temperature field patterns, showing smaller biases of up to ± 1.6 °C at the RCM resolution. Regarding the spatial patterns of precipitation, the HadG has overall higher correlation values with observations than ENS. However, the regional model provides more detailed spatial distributions, decreasing the bias by more than 1 mm/day over some of the islands. It also captures extreme precipitation events that are underestimated in the HadG simulation, even after upscaling to the GCM resolution grid. Concerning the performance of the different RegCM4 convection scheme combinations, KfEm and EmEm show the highest skill values for precipitation, while for temperature TkEm and GeEm are the best performing. This highlights that no individual scheme outperforms the others in all respects, while the application of the averaged ensemble technique provides the best results. 1 Introduction The use of regional climate models (RCMs) driven by global climate models (GCMs) to produce climate simulations or projections at a higher scale than that of the GCM has become a common practice in the last decades (Giorgi 2019). The advantages and disadvantages of RCMs have been discussed in many papers (Anthes et al. 1989; Centella-Artola et al. 2015; Feser 2006; Giorgi and Gutowski 2015; Giorgi and Mearns 1991; Castro et al. 2005; Martínez-Castro et al. 2006, 2016, 2018; Prömmel et al. 2010; Rockel et al. 2008; * Alejandro Vichot-Llano alejandrovichot.llano@gmail.com; alejandro.vichot@insmet.cu 1 Instituto de Meteorología, Loma de Casablanca S/N, Havana, Cuba 2 Instituto Geofísico del Perú, Lima, Perú 3 Earth Physics Section, International Centre for Theoretical Physics, Trieste, Italy Vichot-Llano et al. 2014; Winterfeldt and Weisse 2009). One of the most debated issues concerning regional climate modeling is the quantitative representation of the added value (AV) of going from the lower GCM resolution to the higher RCM one (Giorgi and Gutowski 2015; Torma et al. 2015; Rummukainen 2016; Di Luca et al. 2012, 2013a, b, 2015, 2016). This problem arises because the identification of AV depends on the analyzed variable and the applied metrics and is also related to the morphological structure of the domain and study areas, particularly if complex topography and coastlines are involved (Mearns et al. 2003). Several studies on AV have adopted diverse metrics to quantify it. For example, Sun et al. (2005) used a scale separation technique for precipitation in Northeast Brazil, finding that the regional model was able to provide accurate small-scale details that were missed by the global model. As another example, the RCM performance has been assessed by comparing with observations to evaluate the spatial and temporal patterns of the model results by Castro et al. (2005). On the other hand, Prömmel et al. (2010) used temporal correlation and bias metrics, to find the A. Vichot-Llano et al. RCM AV in temperature surrounding the Alps during the winter and in topographically complex regions in the summer. Another research aimed at finding improvement in the precipitation skills by RCMs was made by De Sales and Xue (2013), who found AV over the southern part of South America using a verification of the intensity scale. Di Luca et al. (2012) developed a potential AV metrics and determined that the highest potential for precipitation AV is on shorter time scales and near complex topography. Similar conclusions were found by Lucas-Picher et al. (2017) for North America, who also showed AV in the simulation of lake-effect snow, sea breeze–driven precipitation and wind intensity and direction over coastal areas. As a final example, Torma et al. (2015) used different metrics to show substantial AV by high-resolution RCMs in simulating precipitation patterns, daily precipitation rate distributions, and precipitation extremes over the Alpine region. This work was extended by Giorgi et al. (2016) to the case of summer precipitation change signals over the region. They found that the RCMs produced a change signal of the opposite sign compared with that of the driving GCMs because of increased high elevation convection not simulated by the GCMs. The RCM signal was also more consistent with observed trends over the region. All these examples clearly show that the issue of the identification of AV is complex and depends on the variables and regional climate types being considered, which may require targeted analyses, specific to different regional settings (Giorgi and Gutowski 2015; Rummukainen 2016; Di Luca et al. 2012, 2013a, b, 2015, 2016; Torma et al. 2015). According to these considerations, the main objective of the present paper is to analyze the AV for variables such as precipitation and temperature as simulated by a driving GCM, the HadGEM2-Es (HadG), and the regional climate model RegCM4 (Giorgi et al. 2012) over Central America and the Caribbean. This region is especially suitable for such an analysis because of the presence of islands, complex topography, and coastlines. Our study is based on data from four simulations using different cumulus schemes, which allows us to determine how the findings are specific to a particular configuration of the model. The analysis is focused in the comparison with the GCM output patterns, assessing the dependence on the quality of the boundary forcing fields. Both the individual simulations and the ensemble mean (ENS) are assessed using several metrics at both the GCM and RCM resolution grids, so that we can assess the AV also when the RCM data are upscaled onto the GCM grid. 2 Methods and data 2.1 Model description, data, and numerical experiments The latest version of the regional climate model RegCM4 developed at the International Centre for Theoretical Physics (ICTP) (Giorgi et al. 2012) is used in this work (Giorgi et al. 1993, 1999; Pal et al. 2007). The model uses a finitedifference discretization and a hydrostatic dynamical core (Grell et al. 1994) on an Arakawa B horizontal grid (Arakawa and Lamb 1977). The RegCM4 has capability of multiple physics schemes, and here we use the NCAR radiation scheme (Kiehl et al. 1996), the biosphere–atmosphere transfer scheme (BATS) (Dickinson et al. 1993) to represent soil–vegetation–atmosphere interaction processes, the nonlocal planetary boundary layer scheme of Holtslag et al. (1990), the ocean surface flux scheme of Zeng and Dickinson (1998), and the explicit moisture representation of resolvable scale cloud and precipitation of Pal et al. (2007). 2.1.1 Convective schemes The RegCM4 system has multiple options for representing cumulus convection, with the capability of using different schemes over land and ocean areas (Giorgi et al. 2012). In this paper, four different cumulus parameterization scheme configurations are used over land: the Grell (1993) scheme (GE), the Emanuel (Emanuel 1991) scheme (EM), the Tiedtke (TK) scheme (Tiedtke 1989), and the parameterization of Kain and Fritsch (1989) (KF). The EM scheme is applied over ocean in all cases. The influence of the GE, EM, and TI schemes in the performance of RegCM4 was tested by the authors in previous studies (Martínez-Castro et al. 2016, 2018; Vichot-Llano et al. 2014), while the KF was implemented in the last version of the model and has been recently tested with good results in the tropical regions of India and West Africa by Huang and Gao (2017), Adeniyi (2019), and Maity (2020). 2.2 Initial conditions and reference observation datasets HadGEM2-ES (hereafter as HadG) (Martin et al. 2011) is used to produce initial and boundary conditions for RegCM4. Its atmospheric component has 38 levels, extending to nearly 40-km height. The global grid used by the model has 192 × 145 grid cells, with a horizontal resolution of 1.25 degrees of latitude by 1.875 degrees of longitude. The variables used to drive the RegCM4 are as follows: air temperature, relative humidity, horizontal wind components, and mean sea level pressure. Sea surface temperature (SST) is obtained by interpolation of the monthly average values of the Reynolds and Smith (1994) dataset. With a horizontal resolution of 10 min, the topography and land-use data are obtained from the United States Geological Survey (USGS) and Global Land Cover Characterization (GLCC, Loveland et al. 2000), respectively. To evaluate the model, different datasets are used: the reanalysis ERA-Interim (ERAI) from the European Center of Medium-Range Weather Forecasts (ECMWF) (Dee et al. 2011; Simmons et al. 2007) at a Comparison of GCM and RCM simulated precipitation and temperature over Central America and the Caribbean Table 1 Definitions of the 5 experiments Experiment acronym Definition GeEm EmEm TkEm KfEm ENS RegCM4: GE over land, EM over ocean driven by HadGM2-ES RegCM4: EM over land and ocean RegCM4:TK over land, EM over ocean RegCM4: KF over land, EM over ocean Average of GeEm, EMEm, TkEM and KfEm HadG HadGM2-ES output horizontal resolution of 1.5° × 1.5° (ERAI); the Climate Research Unit (CRU) of the University of East Anglia dataset for temperature and precipitation with a horizontal resolution of 0.5° × 0.5°, version 3.10 (Mitchell and Jones 2005; Harris et al. 2014); and the Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) (Funk et al. 2015), 5 × 5 km horizontal resolution also for precipitation. Other datasets with coarser temporal and horizontal resolution are mainly used to evaluate the larger-scale precipitation patterns and the annual cycle. These include the climate prediction center merged analysis of precipitation (CMAP) dataset (Xie and Arkin 1997) and the GPCP (Global Precipitation Climate Project, Adler et al. 2003). The model data and the observations are interpolated onto two representative regular grids, at grid spacing of 0.25° and 1.875° corresponding with the RCM and GCM model grids, respectively. This allows a direct assessment and intercomparison of the results across resolutions. Following Torma et al. (2015), for the interpolation we use the distance-weighted method in CDO (Schulzweida 2019). The domain (Fig. 1) includes the Greatest and Lesser Antilles, and the Interamerican Seas and its surrounding territories, including part of the Atlantic and pacific Oceans. However, most of the analysis is focused in a study area including the Caribbean Region and part of central America (central rectangle in Fig. 1), and particularly in four subregions, defined following previous work (Martinez-Castro et al. 2018). 2.3 Numerical experiments 2.4 Metrics The study areas, resolution, and domain were chosen from previous studies (Vichot-Llano et al. 2014; Martinez-Castro et al. 2018;). In total, four RCM simulations were completed at 25-km grid spacing for the 25-year-long present-day period January 1979–December 2004 using the GCM output for initial and lateral boundary conditions. The first year (1979) is removed from the analysis as model spin-up. Table 1 shows the acronyms for the four experiments, along with the convective parameterizations used over land and sea. Fig. 1 Domain with five subregions defined as: study area (SA), Greater Antilles (GA), and Central America: north-east (CAMNE), east (CAME), and west (CAMW). The analysis of the AV is focused on the spatial distribution of mean precipitation and temperature and on the daily precipitation probability distribution functions (PDFs), whose right tail is evaluated using as metrics the fraction of precipitation accounted for by events above the 95th percentile (R95) (Giorgi et al. 2014; Sillmann et al. 2013). In addition, we also use Taylor diagrams, which provide information on the spatial correlation and standard deviation, the bias, the root mean square error (RMSE), and the t A. Vichot-Llano et al. Fig. 2 Spatial variability of the mean precipitation (a, c, e, and g) and bias (b, d, f, and h; statistically significant biases are shown by hatched areas) during the dry season against CHIRPS for the period 1980–2004, interpolated on a 1.875 grid (GCM-res) and 0.25 (RCM-res) degrees. From top to bottom: HadG (a, b; e, f) and the ENS simulations (c, d; g, h) (mm/day) test value to calculate statistical significance. The dry (Nov–Apr) and rainy (May–Oct) seasons are analyzed separately, as defined by Lecha and Paz (1994). Following Di Luca et al. (2012), a specific AV metrics, the AV index, or AVI, is used to estimate whether the RCM adds value over the lower resolution GCM. This is defined by: 3 Results AVI ¼ ðGCM−ObsÞ2 −ðRCM−ObsÞ2 ð1Þ In this way, the RCM generates some AV if its RCM square error is smaller than the GCM one, i.e., if AVI is positive. Fig. 3 Same as Fig. 2, but during the rainy season 3.1 Precipitation 3.1.1 AV in the representation of the mean dry and wet season climate The increase in detail for the precipitation field with higher resolution is visually evident during both the dry and rainy seasons in ENS, as is the deterioration of this signal at the coarsest resolution for the GCM simulations relative to the CHIRPS observations (Figs. 2 and 3). ENS reproduces quite Comparison of GCM and RCM simulated precipitation and temperature over Central America and the Caribbean well the observed spatial pattern of precipitation in the dry and rainy seasons, capturing the relative maxima over both Central American coasts. However, the Pacific coast maximum is overestimated and the Atlantic one is underestimated, with biases statistically significant at the 90% confidence level (Figs. 2 and 3, d and h). The detailed spatial information represented by the observations is only captured by ENS, even at the coarse resolution. For the Caribbean Islands, the improvement is more marked, as can be seen from the simulation of precipitation maxima, coincident with the observed ones in the mountainous regions of Cuba and La Española. ENS outperforms the HadG’s results with lower biases over the smallest islands, and even regions such as the Bahamas without significant biases during the dry season. In general, the ENS performance is better throughout the domain, with smaller values of root mean squared error (RMSE) (between 1 and 3 mm/day) than the HadG during the rainy season, except over Guatemala and El Salvador, where the values are between 3 and 5.8 mm/day, with permanent biases over this region. The range of the bias values during the dry season (around 1.8 and 3 mm/day) are similar to the ones in HadG. Figure 4 presents Taylor diagrams for the four individual RegCM4 configurations, HadG and ENS for precipitation in the two seasons over the study area at the two resolution grids. At the finer scale, the standard deviation and spatial correlation values are similar across simulations and resolutions. For the wet season, the highest correlation value is shown by the global model, even if the spatial details of the observed spatial patterns are better reproduced by the regional model. This can be explained by the relative displacement of the precipitation regions, which is greater for the high-resolution case, and occurs mainly in the rainy period, when precipitation is mostly Fig. 4 Taylor diagram of mean precipitation over land points (1980–2004). From top to bottom: dry and rainy seasons. From left to right: GCM resolution and RCM resolution. Reference data: CHIRPS A. Vichot-Llano et al. convective, and related to local processes. In the lower resolution distributions, the precipitation areas are more expanded and the agreement with observations is better. Only TkEm appears out of range because of its poor ability to produce precipitation during the wet season, a result consistent with Vichot-Llano et al. (2014) and Martínez-Castro et al. (2016, 2018). At the GCM resolution, all the simulations show similar values for the standard deviation and correlation. 3.1.2 AV in the representation of the annual cycle The dominant annual cycle of the Central American region, except for the central part of its Atlantic coast, is monsoonal, with highest temperatures in April, just before the summer rains, and minimum temperatures in January, related with strong trade winds. Precipitation over most of Central America, including the Yucatan peninsula, is characterized by two maxima in June and September, an extended dry season from November to May, and a shorter dry season in July– August, known as the midsummer drought (MSD). The main dry season of winter and early spring is more intense on the Pacific slopes of the isthmus, due to the seasonal reversal of the winds on the Pacific side which blow offshore during winter and to the migration of the intertropical convergence zone (ITCZ) which shifts to its southernmost position from February to March (Taylor and Alfaro 2005). The Caribbean islands have two main seasons. The rainy season, with higher values of temperature and accumulated precipitation, occurs during the boreal summer, including part of spring and autumn, although its limits depend on the subregion of the Caribbean. The dry season occurs during the boreal winter, with minimum temperatures above 18 °C. In the small islands and coastal areas of the Greater Antilles, the sea breeze attenuates the diurnal temperature cycle, limiting the temperature extremes. The direct influence of the ITCZ on precipitation in the Caribbean islands is not significant. The MSD is also present in most of the Caribbean, particularly in the Greater Antilles (Taylor and Alfaro 2005). An additional way to detect the AV from the RCMs is through their representation of the annual cycle of precipitation, which is shown in Fig. 5. In general, HadG, ENS, and all its members are capable to reproduce the two maximum peaks that are characteristic of the region, with a bimodal precipitation regime during the rainy season. However, some discrepancies arise about the timing and month of occurrence of the maxima and the MSD. There is a general agreement between the model simulations and observations in the time of the first maximum in May-July, with the exception of CAME, for which the models reproduce the first maximum in June, while the observations show it in July. Regarding the intensity of the maxima, there is a general agreement among observations, except for CRU, with a precipitation value well over those of the other datasets for three of the subregions. There is also a good agreement among model simulations except TkEm, which in CAMW overestimates substantially the value of the maximum. CAME is a particular region, where there is not a marked climatological bimodal precipitation regime (Magaña et al. 1999; Reuscher et al. 2008); however, the observational datasets show a second weak peak in October (Curtis and Gamble 2007). This is captured by ENS and its members, while HadG simulates it incorrectly in time (two months in advance) and intensity. The models have more discrepancies in simulating the second maximum, especially its time of occurrence. In the case of Fig. 5 Precipitation annual cycle at both horizontal resolution over the four subregions defined in Fig. 1 (mm/day) Comparison of GCM and RCM simulated precipitation and temperature over Central America and the Caribbean GA, the observations show it in September but HadG and all the RegCM4 configurations produce it in October. In the case of CAME, where the bimodal pattern is not so marked, ENS correctly situates the second maximum in October, while HadG produces a very weak maximum in September–October. For CAMW, there are discrepancies among observational datasets, so that ENS follows CHIRPS and CRU, while HadG is closer to GPCP and CMAP, and for CAMNE, RegCM4 and particularly ENS locate the peak correctly while HadG lags 1 month. In general, the models underestimate the second peak, but ENS gives a much better estimation than HadG, which is a clear indication of AV. Regarding the relative minimum of precipitation, defining the MSD, it is underestimated by all the model simulations, which leads to an overestimate of the MSD. The effect of the increase in resolution does not influence much the annual cycle, as can be seen by comparing each of the four panels at the right half of the figure with the corresponding ones of the left half. The difference in the simulations by ENS and HadG is significant in CAME and CAMNE, but it does not depend much on the output grid resolution. 3.1.3 AV in representation of extreme rainfall The R95 metrics provides information about the extremes tail of the PDF distribution, accounting for the fraction of precipitation above the 95th percentile (Giorgi et al. 2014; Sillmann et al. 2013; Torma et al. 2015). Figure 6 shows the spatial distribution of R95 during the dry and rainy seasons comparing the HadG and the ENS performance against CHIRPS. The observed and simulated R95 during the dry season have common areas with high values over the western coast of Central America and the Lesser Antilles. A second maximum over the Bahamas is not captured, while a third maximum over the Colombia coast can be seen in both simulations, although it is underestimated by HadG. The performance of the models is quite good at both resolutions, but in general the GCM tends to underestimate this metrics, missing the high values and high-resolution details, specifically over the Caribbean sea and over the Greater and Lesser Antilles, where ENS captures the patterns even after upscaling to the GCM resolution grid. The performances of ENS and HadG are quite similar during the rainy season, reproducing well the maximum over the Bahamas and around the Lesser Antilles. Even if HadG reproduces these two maxima, the signal is weaker than ENS and CHIRPS, and the maximum observed over the north of SouthAmerica is only captured by ENS. Comparing against CHIRPS with information only over land and at the highest original grid resolution, the ENS performance is particularly good and better than that of HadG, showing AV in this Fig. 6 Spatial distribution of simulated and observed R95 during the dry and rainy season. From top to bottom: CHIRPS (a–b; g–h), HadG (c–d; i–j), and ENS (e–f; k–l) A. Vichot-Llano et al. metrics. The moderate precipitation maximum shown in CHIRS over the western part of Cuba is greatly overestimated by HadG, with an improvement by the regional model which shows a much reduced overestimation limited to the southwest coastal and sea region including the small island Isla de la Juventud. Figure 7 shows the Taylor diagram for R95, clearly revealing the AV of the RegCM4 at both resolutions against the GCM. During the dry season, ENS and all its members show higher correlation values than HadG, with the AV being even clearer at the coarse resolution, where the HadG shows the poorest correlation and standard deviation values. For the rainy season, the ENS performance is even better in relation to HadG, who has metrics values out of range. However, the correlation decreases with increasing resolution of the grid, from a value of 0.6 during the dry season at the GCM grid and to less than 0.4 at the RCM grid. During the rainy season, the values are around 0.3 and for the high-resolution grid. This indicates that, even if in general the increase of resolution leads to more realistic fine scale detail, the exact placement of the orographic maxima deviates from observations. This may be due to the relatively smooth model topography even at 25 km resolution, resulting from a topography smoother employed in the model to increase numerical stability (Giorgi et al. 2012). The PDFs of all individual simulations along with the corresponding observations at both resolution grids during the rainy season are shown in Fig. 8. Here and also during the dry season (not shown), the AV of the RCM simulations relative to the GCM ones is noticeable. The simulations by the individual configurations of RegCM4 overestimate the high-intensity tail of the observed distribution, producing Fig. 7 Taylor diagram of R95 over land points (1980–2004). From top to bottom: dry and rainy seasons. From left to right: GCM grid and RCM grid Comparison of GCM and RCM simulated precipitation and temperature over Central America and the Caribbean Fig. 8 Probability distribution functions during the rainy season at the coarsest and finest resolution grids events of the highest magnitudes exceeding 600 mm/day, where the observations only reach between 200 and 300 mm/day. This may be due to the occurrence of occasional numerical grid point storms which are characteristic of climate models especially in tropical regimes (Giorgi 1991; Giorgi and Marinucci 1991). ENS improves this problem by filtering out the contributions of individual point storm events, and indeed, over the entire study region, the RegCM4 PDFs match well the CHIRPS ones. Conversely, HadG underestimate the PDFs substantially when looking at the entire analysis region. Clearly, ENS matches best the CHIRPS observations compared with HadG, while the HadG matches best ERAI, as both have a relatively coarse resolution. For the coarse-resolution grid, ENS substantially improves the PDFs compared with the HadG revealing significant AV. Here, even when the GCM matches the ERAI performance, it still underestimates medium to high intensity events compared with CHIRPS. Generally, the tails in the GCM run are shorter than the RCM ones and underestimate the intensity of the events because of the poor representation of topography or low spatial resolution. Figure 9 shows the spatial distribution of the added value index, AVI, defined by Di Luca et al. (2015), and previously applied by Torma et al. (2015), according to Eq. (1). Positive AVI regions represent ENS AV. In this case, the highresolution model tends to improve the performance of HadG to a greater extent over the Caribbean Islands, specifically where HadG fails in the precipitation performance, because of the resolution and the topography information at the original resolution. Also over Central America and Florida, the performance of ENS adds value. The regions with positive AVI are more extended during the rainy season for both the coarse and high-resolution grids. 3.2 Temperature 3.2.1 AV in the representation of the temperature patterns Figure 10 shows the temperature biases for the wet and dry seasons in HadG and ENS. HadG heavily overestimates the temperature field at both grids, the bias being statistically significant at the 95% confidence level over La Española, the Bahamas, and Central America, and being also high over Cuba and Florida. This is clearly a result of the coarse resolution of the GCM which does not capture properly the coastlines and topography of islands and peninsulas. ENS reproduces quite well the spatial pattern of temperature in both seasons, with biases around ± 1.6 at the RCM resolution. This is one of the clearest examples of the AV introduced by the regional model, due to the better representation of the fine scale physiography of the region. The Taylor diagrams in Fig. 11 produced for the entire study area at the two resolution grids show that for ENS both the standard deviation (close to the reference) and correlation values (around 0.9) are close to observations and better than HadG at both resolutions and seasons. In particular, also at the coarse resolution, ENS and its members show higher performance comparing with HadG. A. Vichot-Llano et al. Fig. 9 Added value index AVI of the RegCM4 precipitation relative to the driving GCM (only over land), using CHIRPS as reference, at the coarse resolution (left) and at the high resolution (right). From top to bottom: dry (a–b) and rainy (c–d) seasons The AVI index for temperature is shown in Fig. 12. The application of the RCM improves the performance of HadG to a great extent over the whole analysis area in both seasons, but more clearly during the dry season. It is clear how the regional Fig. 10 Statistical significance biases over the hatched areas (b, d, f, and h) during the dry (top) and rainy (bottom) season against CRU for the period 1980–2004 interpolated on 1.875 (GCM-res) and 0.25 (RCM-res) degrees. To the left: HadG and to the right: ENS simulations (°C) Comparison of GCM and RCM simulated precipitation and temperature over Central America and the Caribbean Fig. 11 Taylor diagram of mean temperature (1980–2004). From top to bottom: dry and rainy seasons. From left to right: GCM grid and RCM grid, against CHIRPS model with a better representation of the topography and distribution of small islands in the region makes the temperature field closer to observations relative to the GCM, particularly in mountainous regions of Central America, Cuba and La Española, the narrow western extreme of Cuba, The Bahamas and the Lesser Antilles. 4 Summary and conclusions In this study we compared an ensemble of RegCM4 simulations of present-day climate with the driving GCM (HadG) over Central America and the Caribbean region for precipitation and temperature. The results clearly show the need to use high-resolution regional models for climate studies in the region because of the complex orography and land-sea contrasts associated with the presence of islands and peninsulas. High resolution is also needed to simulate climate extremes, and our results in fact show that HadG fails to represent the tail of the PDFs and the R95 patterns, whereas the RegCM4 ensemble considerably improved this aspect of the simulation. The added value of using the regional model nested in the global model was found for most of the tested variables and climate features. Substantial AV of RCM downscaling and upscaling was found when using the AVI metric for both variables, with the ENS showing improved performance also compared with the individual ensemble members. Spatial patterns, standard deviation, and correlations also mostly improved relative to HadG, even when upscaled at the GCM grid resolution, except for the case of fine scale precipitation correlations due to some displacement of small-scale orographic maxima. The improvement compared with the GCM was particularly evident for temperature (all metrics) and for precipitation PDFs and extremes. A. Vichot-Llano et al. Fig. 12 Added value of the RegCM4 temperature relative to the driving GCM, using CRU as reference, at the coarse resolution (left) and at the highest resolution (right) for temperature. From top to bottom: dry (a–b) and rainy (c–d) seasons Concerning the performance of the different RegCM4 convection schemes, for precipitation, KfEm and EmEm showed the highest skill values, while for temperature TkEm and GeEm are the best performing, consistently with previous results (Martínez-Castro et al. 2016, 2018). Therefore, there is no individual scheme that outperforms the others in all respects, and in fact, in general terms, the application of the averaged ensemble technique provided the best results. The global model and the configurations of the regional model reproduced the bimodal annual precipitation cycle for the Greater Antilles and three subregions in the Central America isthmus, with the presence of the midsummer drought, with different time intervals and intensity, but the use of the regional model improved the estimations of the positions and magnitudes of the maxima and minimum of the cycle. Our results confirm the potential of RCMs to substantially improve the representation of fine scale climate processes compared with coarse-resolution GCMs in areas characterized by complex morphological features (such as Central America and the Caribbean). They also point to the usefulness of using ensemble approaches compared with individual models. We are using the same ensemble approach to the production of future climate change scenarios over the region, which will be reported in future papers. Acknowledgments Special thanks to the ICTP “Earth System Physics” (ESP) group for providing the regional climate model RegCM4, access to the “Argo” supercomputer cluster and technical assistance and to the Associateship Program, funded by the Simons Foundation. We also thank the “Caribbean Community Climate Change Centre” (CCCCC) for its support. Funding This study was supported by funding from Technology and Environment (CITMA) of Cuba through the project SUPERCLIMA of the Institute of Meteorology of Cuba (INSMET) and the “Sandwich Training Educational Program” (STEP) of the International Centre for Atmospheric Physics (ICTP). References Adeniyi MO (2019) Sensitivities of the Tiedtke and Kain-Fritsch convection schemes for RegCM4.5 over West Africa. Meteorology Hydrology and Water Management 7(2):27–37. https://doi.org/10. 26491/mhwm/103797 Adler RF, Huffman GJ, Chang A, Ferraro R, Xie P-P, Janowiak J, Rudolf B, Schneider U, Curtis S, Bolvin D, Gruber A, Susskind J, Arkin P, Nelkin E (2003) The Version-2 Global Precipitation Climatology Project (GPCP) monthly precipitation analysis (1979–Present). J Comparison of GCM and RCM simulated precipitation and temperature over Central America and the Caribbean Hydrometeorol 4(6):1147–1167. https://doi.org/10.1175/15257541(2003)004<1147:TVGPCP>2.0.CO;2 Anthes RA, Kuo Y-H, Hsie E-Y, Low-Nam S, Bettge TW (1989) Estimation of skill and uncertainty in regional numerical models. Q J R Meteorol Soc 115:763–806. https://doi.org/10.1002/qj. 49711548803 Arakawa A, Lamb VR (1977) Computational design of the basic dynamical process of the UCLA General Circulation Model. Methods Comput Phys 17:173–265. https://doi.org/10.1016/B978-0-12460817-7.50009-4 Castro CL, Pielke R Sr, Leoncini G (2005) Dynamical downscaling: assessment of value retained and added using the Regional Atmospheric Modeling System (RAMS). J Geophys Res 110. https://doi.org/10.1029/2004JD004721 Centella-Artola A, Taylor MA, Bezanilla-Morlot A, Martinez-Castro D, Campbell JD, Stephenson TS, Vichot A (2015) Assessing the effect of domain size over the Caribbean region using the PRECIS regional climate model. Clim Dyn 44(7–8):1901–1918. https://doi.org/10. 1007/s00382-014-2272-8 Curtis S, Gamble D (2007). Regional variations of the Caribbean midsummer drought. Theor Appl Climatol 94(25) https://doi.org/10. 1007/s00704-007-0342-0 De Sales F, Xue Y (2013) Dynamic downscaling of 22-year CFS winter seasonal hindcasts with the UCLA-ETA regional climate model over the United. Clim Dyn 41:255–275. https://doi.org/10.1007/ s00382-012-1567-x Dee DP, Uppala SM, Simmons AJ, Berrisford P, Poli P, Kobayashi S, Andrae U, Balmaseda MA, Balsamo G, Bauer P, Bechtold P, Beljaars ACM, van de Berg L, Bidlot J, Bormann N, Delsol C, Dragani R, Fuentes M, Geer AJ, Haimberger L, Healy SB, Hersbach H, Hólm EV, Isaksen L, Kållberg P, Köhler M, Matricardi M, McNally AP, Monge-Sanz BM, Morcrette JJ, Park BK, Peubey C, de Rosnay P, Tavolato C, Thépaut JN, Vitart F (2011) The ERA-Interim reanalysis: configuration and performance of the data assimilation system. Q J R Meteorol Soc 137(656):553– 597. https://doi.org/10.1002/qj.828 Di Luca A, de Elía R, Laprise R (2012) Potential for added value in precipitation simulated by high-resolution nested Regional Climate Models and observations. Clim Dyn 38:1229–1247. https://doi.org/ 10.1007/s00382-011-1068-3 Di Luca A, de Elía R, Laprise R (2013a) Potential for small scale added value of RCM’s downscaled climate change signal. Clim Dyn 40: 601–618. https://doi.org/10.1007/s00382-012-1415-z Di Luca A, de Elía R, Laprise R (2013b) Potential for added value in temperature simulated by high-resolution nested RCMs in present climate and in the climate change signal. Clim Dyn 40:443–464. https://doi.org/10.1007/s00382-012-1384-2 Di Luca A, de Elía R, Laprise R (2015) Challenges in the quest for added value of regional climate dynamical downscaling. Curr Clim Change Rep 1:10–21. https://doi.org/10.1007/s40641-015-0003-9 Di Luca A, Argüeso D, Evans JP, Elía R, Laprise R (2016) Quantifying the overall added value of dynamical downscaling and the contribution from different spatial scales. J Geophys Res Atmos 121:1575– 1590. https://doi.org/10.1002/2015JD024009 Dickinson RE, Henderson-Sellers A, Kennedy PJ (1993) Biosphereatmosphere Transfer Scheme (BATS) Version 1e as coupled to the NCAR Community Climate Model (No. NCAR/TN-387+STR). University Corporation for Atmospheric Research. https://doi.org/ 10.5065/D67W6959 Emanuel KA (1991) A scheme for representing cumulus convection in large-scale models. J Atmos Sci 48(21):2313–2329. https://doi.org/ 10.1175/1520-0469(1991)048<2313:ASFRCC>2.0.CO;2 Feser F (2006) Enhanced detectability of added value in limited-area model results separated into different spatial scales. Mon Weather Rev 134(8):2180–2190. https://doi.org/10.1175/MWR3183.1 Funk C, Peterson P, Landsfeld M, Pedreros D, Verdin J, Shukla S et al (2015) The climate hazards infrared precipitation with stations—a new environmental record for monitoring extremes. Sci Data 2: 150066. Retrieved from. https://doi.org/10.1038/sdata.2015.66 Giorgi F (1991) Sensitivity of Simulated Summertime Precipitation over the Western United States to Different Physics Parameterizations. Mon Weather Rev 119:2870–2888. https://doi.org/10.1175/15200493(1991)119<2870:SOSSPO>2.0.CO;2 Giorgi F (2019) Thirty years of regional climate modeling: where are we and where are we going next? J Geophys Res Atmos 124:5696– 5723. https://doi.org/10.1029/2018JD030094 Giorgi F, Gutowski WJ (2015) Regional dynamical downscaling and the CORDEX initiative. Annu Rev Environ Resour 40(1):467–490. https://doi.org/10.1146/annurev-environ-102014-021217 Giorgi F, Marinucci MR (1991) Validation of a regional atmospheric model over Europe: Sensitivity of wintertime and summertime simulations to selected physics parametrizations and lower boundary conditions. Q J R Meteorol Soc 117:1171–1206. https://doi.org/ 10.1002/qj.49711750204 Giorgi F, Mearns LO (1991) Approaches to the simulation of regional climate change: a review. Rev Geophys 29(2):191–216. https://doi. org/10.1029/90RG02636 Giorgi F, Marinucci MR, Bates GT, De Canio G (1993) Development of a second-generation regional climate model (RegCM2). Part II: Convective processes and assimilation of lateral boundary conditions. Mon Weather Rev 121(10):2814–2832. https://doi.org/10. 1175/1520-0493(1993)121<2814:DOASGR>2.0.CO;2 Giorgi F, Huang Y, Nishizawa K, Fu C (1999) A seasonal cycle simulation over eastern Asia and its sensitivity to radiative transfer and surface processes. J Geophys Res Atmos 104(D6):6403–6423. https://doi.org/10.1029/1998JD200052 Giorgi F, Coppola E, Solmon F, Mariotti L et al (2012) RegCM4: model description and preliminary tests over multiple CORDEX domains. Clim Res 52:7–29. https://doi.org/10.3354/cr01018 Giorgi F, Coppola E, Raffaele F (2014) A consistent picture of the hydroclimatic response to global warming from multiple indices: models and observations. J Geophys Res Atmos 119(20):11,611– 695,708. https://doi.org/10.1002/2014JD022238 Giorgi F, Torma C, Coppola E, Ban N, Schär C, Somot S (2016) Enhanced summer convective rainfall at Alpine high elevations in response to climate warming. Nat Geosci https://doi.org/10.1038/ ngeo2761 Grell GA (1993) Prognostic evaluation of assumptions used by cumulus parameterizations. Mon Weather Rev 121(3):764–787. https://doi. org/10.1175/1520-0493(1993)121<0764:PEOAUB>2.0.CO;2 Grell GA, Dudhia J, Stauffer D (1994) A description of the fifthgeneration Penn State/NCAR Mesoscale Model (MM5) (No. NCAR/TN-398+STR). University Corporation for Atmospheric Research. https://doi.org/10.5065/D60Z716B Harris I, Jones PD, Osborn TJ, Lister DH (2014) Updated high-resolution grids of monthly climatic observations – the CRU TS3.10 Dataset. Int J Climatol 34(3):623–642. https://doi.org/10.1002/joc.3711 Holtslag AAM, De Bruijn EIF, Pan H-L (1990) A high resolution air mass transformation model for short-range weather forecasting. Mon Weather Rev 118(8):1561–1575. https://doi.org/10.1175/ 1520-0493(1990)118<1561:AHRAMT>2.0.CO;2 Huang D, Gao S (2017) Impact of different cumulus convective parameterization schemes on the simulation of precipitation over China. Tellus A 69:1406264. https://doi.org/10.1080/16000870.2017. 1406264 Kain JS, Fritsch JM (1989) A one-dimensional entraining/detraining plume model and its application in convective parameterization. J Atmos Sci 47(23):2784–2802. https://doi.org/10.1175/15200469(1990)047<2784:AODEPM>2.0.CO;2 Kiehl JT, Hack JJ, Bonan GB, Boville BA, Briegleb BP, Williamson DL, Rasch PJ (1996) Description of the NCAR Community Climate A. Vichot-Llano et al. Model (CCM3) (No. NCAR/TN-420+STR). University Corporation for Atmospheric Research. https://doi.org/10.5065/ D6FF3Q99 Lecha LB, Paz LB (1994) El clima de Cuba. Editorial Academia, La Habana, 186 pp Loveland TR, Reed BC, Brown JF, Ohlen DO, Zhu Z, Yang L, Merchant JW (2000) Development of a global land cover characteristics database and IGBP DISCover from 1km AVHRR data. Int J Remote Sens 21(6–7):1303–1330. https://doi.org/10.1080/ 014311600210191 Lucas-Picher P, Laprise R, Winger K (2017) Evidence of added value in North American regional climate model hindcast simulations using ever-increasing horizontal resolutions. Clim Dyn 48:2611–2633. https://doi.org/10.1007/s00382-016-3227-z Maity S (2020) Comparative assessment of two RegCM versions in simulating Indian Summer Monsoon. J Earth Syst Sci 129(1):75. https:// doi.org/10.1007/s12040-020-1340-1 Magaña V, Amador JA and Medina S (1999) The mid-summer drought over Mexico and Central America. J Climate 12:1577–1588 Martin GM, Bellouin N, Collins WJ, Culverwell ID, Halloran PR, Hardiman SC, Hinton TJ et al (2011) The HadGEM2 family of Met Office Unified Model climate configurations. Geosci Model Dev 4(3):723–757 Martínez-Castro D, da Rocha RP, Bezanilla-Morlot A, AlvarezEscudero L, Reyes-Fernández JP, Silva-Vidal Y, Arritt RW (2006) Sensitivity studies of the RegCM-3 simulation of summer precipitation, temperature and local wind field in the Caribbean Region. Theor Appl Climatol 86:5–22. https://doi.org/10.1007/s00704-005-0201-9 Martínez-Castro D, Vichot-Llano A, Benzanilla-Morlot A, CentellaArtola A, Campbell J, Viloria-Holguin V (2016) Performance of RegCM-4.3 over the Caribbean region using different configurations of the Tiedtke convective parametrization scheme. Revista de Climatología 16:77–98 http://www.climatol.eu/reclim/reclim16f. pdf. Accessed 26 Oct 2016 Martínez-Castro D, Vichot-Llano A, Bezanilla-Morlot A, Centella-Artola A, Campbell J, Giorgi F, Viloria-Holguin, CC (2017) The performance of RegCM4 over the Central America and Caribbean region using different cumulus parameterizations. Climate Dynamics, 50(11–12):4103–4126. https://doi.org/10.1007/s00382-017-3863-y Martínez-Castro D, Vichot-Llano A, Bezanilla-Morlot A, Centella-Artola A, Campbell J, Giorgi F, Viloria-Holguin CC (2018) The performance of RegCM4 over the Central America and Caribbean region using different cumulus parameterizations. Clim Dyn 50(11–12): 4103–4126. https://doi.org/10.1007/s00382-017-3863-y Mearns LO, Giorgi F, Whetton P, Pabon D, Hulme M, Lal M (2003) Guidelines for use of climate scenarios developed from regional climate model experiments. Intergovernmental Panel on Climate Change (IPCC): Data Distribution Centre (DDC), Norwich, Hamburg and New York, 38 pp Mitchell DT, Jones P (2005) An improved method of constructing a database of monthly climate observations and associated highresolution grids. Int J Climatol 25:693–712. https://doi.org/10. 1002/joc.1181 Pal JS, Giorgi F, Bi X, Elguindi N, Solomon F, Gao X, Francisco R, Zakey A, Winter J, Ashfaq M, Syed F, Bell JL, Diffanbaugh NS, Kamacharya J, Konare A, Martinez D, da Rocha RP, Sloan LC, Steiner A (2007) The ICTP RegCM3 and RegCNET: regional climate modeling for the developing world. Bull Am Meteorol Soc 88: 1395–1409 Prömmel K, Geyer B, Jones JM, Widmann M (2010) Evaluation of the skill and added value of a reanalysis-driven regional simulation for Alpine temperature. Int J Climatol 30:760–773. https://doi.org/10. 1002/joc.1916 Rauscher SA, Giorgi F, Diffenbaugh NS, Seth A (2008) Extension and Intensification of the Meso-American mid-summer drought in the twentyfirst century. Clim Dyn 31:551–571 Reynolds WR, Smith TM (1994) Improved global sea surface temperature analyses using optimum interpolation. J Clim 7:929–948. https://doi.org/10.1175/1520-0442(1994)007<0929:IGSSTA>2.0. CO;2 Rockel B, Will A, Hense A (2008) The regional climate model COSMOCLM (CCLM). Meteorol Z 17(4):347–348. https://doi.org/10.1127/ 0941-2948/2008/0309 Rummukainen M (2016) Added value in regional climate modeling. Clim Chan 7:145–159. https://doi.org/10.1002/wcc.378 Schulzweida Uwe (2019) CDO User Guide (Version 1.9.8). https://doi. org/10.5281/zenodo.3539275 Sillmann J, Kharin VV, Zhang X, Zwiers FW, Bronaugh DR (2013) Climate extremes indices in the CMIP5 multimodel ensemble: Part 1. Model evaluation in the present climate. J Geophys Res Atmos 118:1716–1733. https://doi.org/10.1002/jgrd.50203 Simmons JA, Uppala S, Dee D, Kobayashi S (2007) ERAInterim: New ECMWF reanalysis products from 1989 onwards. In ECMWF Newsletter (Vol. 110) Sun L, Moncunill DF, Li H, Moura AD, Filho FAS (2005) Climate downscaling over Nordeste, Brazil, using the NCEP RSM97. J Clim 18(4):551–567. https://doi.org/10.1175/JCLI-3266.1 Taylor MA, Alfaro E (2005) Climate of Central America and the Caribbean. In: Oliver JE (ed) Encyclopedia of world climatology. Springer, Berlin, pp 183–189 Tiedtke M (1989) A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon Weather Rev 117(8): 1779–1800. https://doi.org/10.1175/1520-0493(1989)117<1779: ACMFSF>2.0.CO;2 Torma C, Giorgi F, Coppola E (2015) Added value of regional cli-mate modeling over areas character-ized by complex terrain— precipitation over the Alps. J Geophys Res Atmos 120:3957– 3972. https://doi.org/10.1002/2014JD022781 Vichot-Llano A, Martínez-Castro D, Centella-Artola A, Bezanilla-Morlot A (2014) Sensibilidad al cambio de dominio y resolución de tres configuraciones del modelo climático regional RegCM 4.3 para la región de América Central y el Caribe. Revista de Climatología 14: 45–62 http://www.climatol.eu/reclim/reclim14e.pdf. Accessed 29 Sept 2014 Winterfeldt J, Weisse R (2009) Assessment of value added for surface marine wind speed obtained from two regional climate models. Mon Weather Rev 137(9):2955–2965. https://doi.org/10.1175/ 2009MWR2704.1 Xie P, Arkin PA (1997) Global precipitation: a 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model outputs. Bull Am Meteorol Soc 78(11):2539–2558. https:// doi.org/10.1175/1520-0477(1997)078<2539:GPAYMA>2.0.CO;2 Zeng X, Dickinson RE (1998) Effect of surface sublayer on surface skin temperature and fluxes. J Clim 11(4):537–550. https://doi.org/10. 1175/1520-0442(1998)011<0537:EOSSOS>2.0.CO;2 Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.