PROMES Regional Climate Model: 30 Years of Ensemble Modelling Research Contributions

1. Introduction

Climate modelling is an essential tool for understanding the climate system, all its features, processes and mechanisms. The first step, of course, is their capability to accurately describe what the observations describe for current and past climate conditions. The huge increase of computational resources in the past decades has allowed the development of numerical and mathematical procedures to convert the physical conservation laws into a solvable and viable procedure to obtain results. Therefore, climate research groups and scientific institutions have developed numerical models able to reproduce atmospheric mechanisms, both on meteorological and climatological scales. Global climate models (GCMs) have allowed for a physically consistent description of the atmosphere, oceans and the rest of the components of the climate system. Therefore, they are the central part of the successive IPCC (Intergovernmental Panel on Climate Change) reports, the base of the climate change scientific studies [1,2].

Before CMIP7 [3] is fully ready, CMIP6 [4] is currently the reference ensemble database of GCMs for global climate studies. It has proven to improve previous intercomparison exercises when representing the main global climate features [5], including the spread and uncertainties when using a large ensemble of global climate models. The most recent IPCC Assessment Report (AR6) [2] results, analysis and findings are largely based on their results when dealing with climate projections for the XXIst century. At the same time, regional climate modelling [6] has become a perfect complementary dynamical downscaling methodology. It started in the late 1980s [7]. The aim was to try to overcome GCM limitations to resolve fine-scale processes that govern regional climate variability [8]. By using dynamic downscaling, that is, solving the same atmospheric physics evolution equations on climatic scales, nested with GCMs, and so using their large-scale fields as boundary conditions, RCMs are able to increase the spatial resolution by one order of magnitude. This physical consistency allows for a better and even explicit representation of processes related to complex orography, coastal contrasts or heterogeneous land cover conditions, that can be relevant to impact [9,10] or adaptation studies. RCM-based studies have achieved a large list of milestones, as shown in IPCC reports [11,12,13,14] or the Atlas [15], and also on many more specific regional and national studies during these years. The degree to which this additional information constitutes genuine scientific added value beyond that of the driving GCM is an active area of debate [16,17,18,19,20,21], analyzing their strengths and limitations when describing climate on regional scales [16,22,23,24,25,26]. One of the strongest assets of RCM studies has been the coordinated multi-model experiments, starting from Regionalization [27,28], PRUDENCE [29,30] and ENSEMBLES [31] EU projects in Europe, to other regions, such as North America [32], Africa [33], or South America [34,35]. These efforts become a global framework of the COordinated Regional Climate Downscaling EXperiment (CORDEX) since 2009 [36]. One of the more climatically complex regions is the Iberian Peninsula, but also with a broader focus, the Mediterranean basin [6,37]. And over these quite challenging areas, specifically EuroCORDEX [38,39,40] and MedCORDEX [41] initiatives have been developed in the past years. It is within this broader scientific context where PROMES regional climate model [42] emerged. It has contributed during more than three decades to climate research, in particular on the Mediterranean region, a well-known hot-spot for climate change studies [43]. This region represents a challenging and paradigmatic area for regional climate studies [37], due to its location in the transition zone between Atlantic and the Mediterranean oceans (from west to east), and in the transition between the mild climate above 50 degrees north and the subtropical to arid climate of northern Africa (from north to south). But it is also for its high degree of orographic complexities, that lead to a strong variety of climates [44], where the higher spatial resolution of RCMs is likely to play a relevant role.

It belongs to the pioneering ensemble of regional climate models (RegCM[45]; WRF [46]; COSMO-CLM[47]; PRECIS[48]; ETA [49]; REMO[50]; HIRHAM[51]; ALADIN[52]; RCA[53] or RACMO[54])) which have been the core of those international projects [6].

Despite its sustained scientific participation on those international projects, and also ESCENA project [55,56] (the first major spanish effort of ensemble regional climate modelling), no comprehensive synthesis of PROMES’s features, performance, and contributions has been published to date. This review aims to address it structured in the following sections: (i) document the main features of PROMES on its mature regional climate modelling configuration; (ii) validate its performance for present climate conditions (using observations), identifying both its strengths and its systematic biases on modelling ensembles; (iii) synthesize the scientific contributions of PROMES across several domains, with particular focus on western Europe; (iv) assess its role within multi-model ensemble frameworks and its contribution to the characterization ensemble regional climate projections and their uncertainties; and (v) reflect on the legacy and limitations of PROMES in the context of current and future directions in regional climate modelling.

2. PROMES Description

PROgnostic at MESoscale (PROMES) model development started around 1990 [42,57,58,59,60] at Complutense University (UCM), and since 2000s at Castilla-La Mancha University (UCLM), in the frame of MOMAC research group (MOdelling the environMent And Climate: www.uclm.es/grupos/momac). PROMES model was, in parallel, used as a meteorological prognostic model [61,62], although its main development and focus was directed towards a regional climate model. PROMES was, among other achievements, the first regional modelling exercise addressed over the Iberian Peninsula [28].

2.1. Main Model Characteristics

One major aspect of climate modelling group works is their continuous development and improvements [6,63], related to numerical methods, adaptation to different computer architectures, but specially the analysis and selection of parameterizations and schemes that are relevant for a better performance to describe regional climate features.

The more detailed and complete description of PROMES characteristics can be found in [64], with detailed numerical and parameterization aspects. It was a hydrostatic model with vertical $\sigma$ levels and horizontal Lambert projection, using an Arakawa-C grid distribution for variables. Lateral boundary conditions were updated every 6 h from GCM fields, with a vertical interpolation to model levels following [59]. Contour zone with 8 points was used to relax external information [65]. PROMES model was also parallelized [66,67], for supercomputer usage.

The physical core of the model [68] included the following parameterizations: radiation scheme based on [69], including fractional cloud cover parameterization from [70,71]; turbulent vertical exchange for the prognostic variables at the planetary boundary layer (PBL) with a 1.5 turbulent kinetic energy closure scheme [72] and precipitation [73] and convective processes based on [74,75]. Convection processes at that 25-50 km resolution were a major issue, as it was mainly subgrid-scale, and was identified [76] as a primary source of spread for precipitation on RCMs ensembles, particularly over the Iberian Peninsula. Land-surface scheme was a major part of PROMES model and its development and modelling studies, based on ORCHIDEE scheme [77], with numerous complexities included, such as dynamical vegetation, hydrological features and vegetation characterization. Latest efforts on model development were focused on atmosphere-ocean coupling [78,79].

2.2. Principal Application Features

PROMES has been applied for multi-decadal simulations from 50 to 12 km horizontal resolution. The more frequent domain for studies was the whole European continent, although the Iberian Peninsula specifically was the subject of several analyses. Also Africa [80] and South America [81] were regions where the model has been used. One major contribution has been the participation with other RCMs in intercomparison exercises, with both the characterization of present-day climate variability and the simulation of future climate change scenarios under multiple greenhouse gas emission scenarios. Specific analysis were particularly focused on land surface–atmosphere interactions [82], PBL characterization [83] and the analysis of numerous extreme climatic events such as droughts [84] or intense precipitation episodes [68].

3. Results

An overall view of PROMES achievements, limitations and biases is presented, for a wider as possible picture of its performance during these thirty years.

3.1. Validation—Model Performance

Modelling regional climate features needs first a validation analysis, that is, the capability to describe current climate conditions. Observational available databases, typically on gridded format (therefore, with a mathematical and physical procedure to convert direct observations to cell-averaged values), such as CRU [85], E-OBS (v6) [86], Spain02 [87] or MW [88], allow a straightforward comparison with model outputs. Sometimes also point/location observations can be used against the corresponding model cells, sometimes reanalysis boundary forcing fields are also used on the validation (as it contains observational information, and also allows for the downscaling capabilities of RCMs from that boundary conditions), and finally also the ensemble comparison with the other RCMs that run with the same boundary conditions is frequently used, as it gives the opportunity to inspect common and consistent features, agreements, disagreements and spread among them when describing regional climate features. Obtained systematic biases are a major topic that has been strongly studied and analyzed [89,90] with the aim to understand the limitations of RCMs to accurately describe the regional climate, and in particular to be used for impact studies.

PROMES validation performance, including biases analysis, have been extensively studied on specific studies [27,28,91], but the main evaluation has been done within the framework of RCMs intercomparison projects: PRUDENCE over Europe [29,92], ESCENA over Spain [55], MedCORDEX over the Mediterranean [56,93], EuroCORDEX [38], CLARIS-LPB over South America [81,94] and the AMMA intercomparison over West Africa [95]. Many indices and metrics have been used for the analysis, such as seasonal fields, annual cycles, interannual variability or spatial and temporal correlations, mainly for temperature and precipitation, although some other variables, such as wind [96,97] were also inspected. Just a brief summary of the main validation results indicated that it is able to describe the main features of regional climate over all these regions, with comparable performance with the other regional climate models. A couple of examples on both behaviours can be: it performs best over the Iberian Peninsula in summer temperature, where it was formerly designed; a tendency to overestimate precipitation in convection-dominated regimes across all domains (Mediterranean coast in autumn, NE-Brazil during the rainy season, and West Africa throughout the year), probably related to model’s convection scheme, as it is not resolved by increasing horizontal resolution exercises.

3.2. PROMES and the Coordinated Ensemble Framework: From PRUDENCE to EuroCORDEX

The participation of PROMES in the main coordinated multi-model ensemble projects is one of its most significant scientific contributions, in terms of the continuous long-term contribution to merged results inside the best international regional climate groups [6]. PRUDENCE (2001–2004) [92] was the first systematic European RCM intercomparison exercise, producing end-of-century climate change projections under the SRES-A2 greenhouse gases scenario using HadAM3H as a common boundary condition for all participating models [30]. One key analysis of PRUDENCE projections [76] identified four sources of uncertainty: sampling, model formulation, boundary conditions, and natural variability. Thus, for example, in summer precipitation over southern Europe, model formulation uncertainty was dominating over boundary conditions, in which PROMES, with its specific physical parameterizations (convection, land-surface and PBL treatment) was a relevant contributor and a genuinely informative ensemble member. Then, ENSEMBLES (2004–2009) extended the PRUDENCE design significantly, combining thirteen RCMs and six GCMs under the SRES-A1B scenario. The uncertainty analysis studies [98] identified three sources of spread: simulation lengths, RCMs parameterization choices, and driving GCMs. In particular, RCM choice dominated the uncertainty in summer precipitation over southwestern Europe, rather than the GCM. Again, the specific physical formulation of PROMES significantly contributed then to the range of plausible future precipitation projections of the RCM ensemble. AMMA (2005-2009) [33] was parallel to ENSEMBLES project, but over West African region, producing the first coordinated multi-decadal RCM intercomparison over Africa, using ERA-Interim boundary conditions for 1990–2007 at approximately 50 km resolution. PROMES developed specific analysis of land-surface sensitivity to precipitation as part of the project [64]. ESCENA (2008–2012) was the Spanish national counterpart of ENSEMBLES initiative, producing an ensemble of four RCMs (PROMES, WRF, MM5 [99], and REMO), nested within ERA-Interim reanalysis at 25 km resolution and within three GCMs under multiple SRES scenarios [55,56,97,100,101]). ESCENA was specifically designed to obtain higher spatial resolution, denser observational validation, and a domain optimized for Spain, Portugal, and northern Morocco compared with ENSEMBLES. Numerous specific studies over these climatically complex and challenging regions were also made based on this project [102,103,104,105]. CLARIS (2004-2007) [106], but specially CLARIS-LPB (2008-2012) [35], an EU FP7 project, produced the first coordinated ensemble of RCM simulations over South America. PROMES showed a comparable performance with the other six RCMs. The main ensemble deviations were a warm bias relative to observations in South Eastern South America (especially for minimum winter temperatures), and overestimating convective precipitation in the tropics and orographic precipitation along the Andes and the Brazilian Highlands [81,94,107,108]. Finally, within the framework of CORDEX initiative, MedCORDEX [41,93] and EuroCORDEX [39] are the more recent ensemble studies, over Europe, where PROMES has been involved. During all these ensemble framework initiatives, PROMES has proven to contribute as a very valuable member, with consistent overall results compared with the average values of the models and the available databases. The essential objective of these projects, once models have been validated, is how they project future climate conditions. Future climate projections, based on different emissions scenarios and a matrix of the evolving CMIP GCMs that allow to construct GCM/RCM matrices, are the key procedure to contribute to climate change studies [14,29,109,110,111,112]. One major aspect of multi-model ensembles utility is the members differences. The larger the spread among RCMs with different parameterizations and numerical methods, the more relevant would be the estimated uncertainty. This is the principle that RCM diversity is essential for uncertainty characterization [98,113,114].

3.3. Scientific Applications Examples

PROMES early studies [27,28,82,115] demonstrated its capability to describe present climate over Western Europe and compare with other RCMs, becoming the earliest dynamical downscaling reference in the Spanish scientific literature. Then the participation on the major ensemble climate change regional projections described on previous section demonstrated that its signal for scenario simulations with a robust warming magnitude was broadly consistent with the ensemble of RCMs that have been the modelling reference during these years [6,14]. Now, we present three examples of different aspects of regional climate change, among the huge amount of studies where PROMES has been involved. The aim is to give an idea of the wide range of different capabilities that represents its contribution to regional climate modelling community studies during these years.

3.3.1. Characterization of Length, Start and End of Seasons

Definition of seasons based meteorological (mainly temperature) thresholds is a challenging feature [116,117,118,119], in contrast with the fixed astronomical (solstices and equinoxes) or climatological (calendar months), and is likely to be favoured by using regional information, for a more accurate relation of how temperature evolves at each location. Therefore, regional climate models, such as PROMES, allow a study on that direction, not only to propose how seasons can be defined [117,120], but also how it can evolve for future conditions, using the regional details, in contrast with GCMs [120]. An example of the results that can be obtained, based on the work [120] is shown with a polar diagram in Figure 1.

By the end of XXI^st century over the whole Iberian Peninsula summers would be enlarged by a couple of months or more, both early from spring, and later towards autumn. Meanwhile, winters, defined as all the other seasons with present climate thresholds, are likely to disappear. Spring and autumn are also slightly enlarged and displaced, occupying winter time. This type of results are consistent with several other more recent studies, using similar methodologies [116,118,119].

3.3.2. Dry Spells Statistics

Dry spells (defined as the consecutive days without rain) analysis, due to the local nature of precipitation is likely to be favoured by using regional models. Thus, a first analysis over the Iberian Peninsula [84] was made with PROMES, using PRUDENCE data, which highlighted the advantage of RCMs at 25 km when compared with low resolution GCMs to describe observed dry spells from observational databases. Then, future projections of these dryness patterns were obtained and described. In Figure 2 the analysis was extended to ENSEMBLE RCM models, together with E-OBS (v6) gridded observations [86], allowing to both see how accurate RCMs were in relation to observations and among each other, in a consistent way. The north-south pattern over land can be clearly seen, with around 150 days/year over northern Africa, and less than 5 days/year for the northern part of the analyzed domain, on Central Europe. Clear coastal and orographic features were obtained. Over the Mediterranean Sea, where observations from E-OBS are not available, RCMs were able to add information about this variable.

Then, other studies took advantage of this first analysis [93,121,122,123], allowing to extend the lines of research related to the absence of precipitation features over Europe, in particular over the Mediterranean, where it is clearly a major issue from a climatological perspective, but also over other regions, such as South America.

3.3.3. South America Regional Climate Precipitation Extremes

PROMES showed its capability to describe regional climate in other regions around the world apart from Europe. It was the case of Africa, in the frame of AMMA project [64,95,124], but the larger effort, first inside CLARIS project [34], but mainly in CLARIS-LPB project, where an ensemble of RCMs modelled present and future climate over the whole South American continent [81,125], with particular attention to La Plata basin area. This project has been the base of more recent studies with a similar structure, as part of CORDEX initiative, with the aim to further study the climate of the region [126,127,128]. An exhaustive analysis of different climatic features has been made throughout a large list of studies [121,129,130]. But, in particular, a whole Special Issue [35] was conducted, where analysis of regional processes and feedbacks [131] or extremes [107], among many other works, was made. A simple example of these studies is presented on Figure 3, precipitation extremes of the RCMs ensemble in terms of monthly values is compared against the small amount of available observations.

As an example of measuring extreme precipitation conditions, Figure 3 shows the different precipitation regimes over the continent, by using fraction of monthly values above or below 50 mm, during the year, which represents the quite heterogeneous pattern that the continent presents in terms of precipitation regimes. Thus, the amazonian basin or the southern part of the continent over the Pacific ocean, monthly precipitation is always above that threshold, or the opposite behaviour, such as northeast Brazil, or larger parts of Argentina or the Atacama desert. The South Atlantic convergence zone presents an intermediate result, which is challenging for the RCMs simulations capabilities. The spread and consistency of results among the regional models depending on the region is a clear picture of the importance of using several models to analyze the regional climate features.

3.3.4. Other Relevant Regional Analysis Lines of Research

A comprehensive catalogue of regional climate studies where PROMES has been part would be too large and it is out of the scope of this review, but some brief list can be of relevance to be mentioned: extreme events [68,102,132,133]; relevant processes where regional climate could be interesting, such as thermal lows at the Iberian Peninsula [134], hurricane-like mediterranean pressure systems (medicanes) [135,136], boundary layer characterization [83], vegetation and land impact on climate [82,115,133,137,138], climate classifications [139,140,141], wind analysis [96,142,143], and numerous subregional specific analysis [144,145,146,147,148,149,150,151,152].

It is worth noting that a large number of impact studies have also been favoured by using PROMES and other RCM results, related to hydrology [144,153,154,155,156,157,158,159], health [160], climate risks [161], vegetation [162] or agriculture [163,164].

This brief list of applied studies demonstrates that PROMES has been deployed across a wide and coherent range of scientific problems. This scientific record, produced over more than three decades, is one of the basis for the legacy assessment presented here. The most scientifically distinctive PROMES contributions (land surface sensitivity studies, climate characterization over several complex regions around the globe or the focus on extreme events), reflect the specific scientific priorities of the PROMES group besides what is needed to contribute to the standard ensemble needs, underscoring the value of model-specific research programs alongside coordinated ensemble efforts. Besides, the consistency of PROMES behaviour across domains: Mediterranean, West Africa, and South America, demonstrates its capabilities in terms of describing regional climate characteristics, including the limitations and biases, that also contribute to obtain a robust RCM ensemble.

4. Discussion: Legacy, Limitations and Relevance

Some aspects of the scientific legacy of PROMES can be better understood with the following distinctive contributions made to the regional climate modelling community:

• A sustained and coherent scientific focus on the Iberian Peninsula along the ensemble RCM exercises, in particular related to the choice of parameterizations specially developed for the characteristics of the regions. The result is a consistent group of present-climate simulations, climate change projections, and sensitivity studies that can be compared across periods and scenarios.
• A pioneering research line on Mediterranean tropical-like cyclones: [135]) was one of the first RCM-based studies to address this phenomenon, followed by [132,136,165] studies. It constitutes the most internationally visible scientific contribution of PROMES to the Mediterranean high-impact weather systems in a changing climate, that has grown substantially in the recent years [166].
• A systematic investigation of land surface–atmosphere coupling at the regional scale, from sensitivity studies [82] on land degradation, deforestation [115], or vegetation description uncertainties [137] constitutes a coherent program of land–atmosphere interaction research that used PROMES for understanding how surface conditions modulate regional climate. These studies were among the first to use an RCM to quantify the non-local effects of land surface change on precipitation over the Iberian Peninsula, and anticipated land use change experiments on EuroCORDEX community [167].

A central aspect of RCM ensemble modelling is their utility as a tool for uncertainty quantification. It rests on the assumption that its members are sufficiently independent (despite sharing partly numerical methods and parameterizations) to explore different regions of the uncertainty space in a probabilistic way [113]. Uncertainties in regional climate projections come from internal variability and mode and emissions scenarios uncertainties [114], that are potentially reducible through progress in climate science. The model uncertainty component of total uncertainty is better sampled if the models in the ensemble are structurally diverse, and so PROMES then represents a relevant contribution to model diversity. The PROMES independent model development [28,42,59,82,83] made it a genuinely different model, and so a contributor to such uncertainty quantification [168]. Its divergence in some regional climatic features, inside the overall climatic coherence and consistency on the validation studies, mentioned previously on the review, is then a signature of this value and genuine added value [19] for the model ensemble analysis. The potential and interest of GCM/RCM combinations in multi-MIP ensembles for Spain was confirmed in [56].

A complete account of the PROMES legacy requires also to point that in the recent years it has been unable to follow the development of RCM community towards: convection-permitting configuration when resolution approaches to 4-5 km, where deep convection can be explicitly resolved without parameterization [169,170]; interactive ocean coupling, where just some efforts were made [79], and studies have shown their importance for a better description of climate features in areas such as the Mediterranean Sea [171]. It has not run scenario simulations with CMIP5-6 boundary conditions, the latest forcing GCMs for RCM downscaling simulations [14,40], which prevents direct comparison with the latest/current RCM ensembles, such as at the EuroCORDEX initiative.

The PROMES trajectory, with its achievements and limitations, and finally the decline as an active modelling tool contains several lessons that are relevant to the broader regional climate modelling community, independent of the scientific results it produced. The coherence and depth of the PROMES scientific record, particularly over the Western Mediterranean area, is attributable to a focused small group specialized on a challenging domain with the same model over three decades. This reinforces the importance of producing a regional-level knowledge that is difficult to replicate in a large community models where no regional expertise sometimes exists. In the recent movements in the regional modelling toward increasingly standardized community frameworks [40,172], there is a risk to lose such local experience. The maintenance costs, without a community of developers, users and infrastructure, did not allow the model to evolve at the pace required to remain competitive with community models that receive continuous updates.

5. Conclusions

The PROMES regional climate model, developed in the spanish UCM and UCLM universities from 1990 onward, represents one of the few sustained and coherent regional climate modelling programs produced by a single research group (MOMAC) in the European scientific community. This review has documented its development, performance and evaluation across multiple domains to describe present and future climate conditions on regional scales. Pleny of scientific applications and studies have been favoured from the huge modelling effort produced by the group within the framework of multi-model international initiatives. The core legacy of the PROMES research programme can be summarized with the following conclusions:

• PROMES produced physically rigorous results across its three decades of application. Its known biases are consistent, physically explicable, related to specific parameterization choices (convection, radiation, boundary layer or land-surface schemes). The cross-domain ensemble modelling consistency of PROMES behaviour makes it a scientifically reliable instrument, even when it includes limitations to describe all the regional climate aspects. It has shown to be particularly successful when representing western Mediterranean regional climate and has led specific lines of research, such as the pioneering research line on Mediterranean tropical-like cyclones.
• PROMES, as a structurally independent model, has contributed significantly to RCM ensemble uncertainties analysis, an argument that generalizes beyond this model. To lead to robust conclusions, the assessment of added value needs to be based on multiple models, over multiple domain settings, using the same metrics and simulation protocols, is crucial for uncertainty quantification. Ensemble quality is strongly proportional to the genuine structural diversity of the contributing models, as it was clearly stated in several studies [6,56].
• The full scientific value of a regional climate modelling program is acquired along the whole active lifetime, by means of individual model testing and development against observations, with several periods of study, regions of analysis, but mainly when establishing and participating in multi-model ensemble spread over the different domains, closely collaborating with other modelling groups, with a constant and strong interaction and learning process. The development of community RCMs applicable to a wide variety of studies and regional contexts, and the inception of intercomparison projects, culminating in the CORDEX initiative, are probably the main achievements in RCM research over the past thirty years, and PROMES has been part of this process from its beginning [6].

Regional climate modelling is a growing very active area of research [6,14,40,173], and it is expected to increasingly contribute to our understanding of climate on those scales, and so on other studies on the climate analysis chain [174]. At the same time, as is continuously stated in perspective works on the known deficiencies and challenging issues related to regional climate modelling [175], summarized in proposals about the future of climate modelling [176].