Precipitation and Temperature in Costa Rica at the End of the Century Based on NEX-GDDP Projected Scenarios

: The evaluation of intraseasonal, seasonal, and annual variability of rainfall and temperature extremes, while using climate change scenarios data, is extremely important for socio-economic activities, such as water resources management. Costa Rica, a climate change hotspot, is largely dependent on rainfall for socioeconomic activities; hence, the relevance of this study. Based on the NEX-GDDP, rainfall and temperature range were analyzed for Costa Rica at the end of the century (2070–2099), while using 1970–1999 as a baseline for six available meteorological stations. Differences between the multimodel ensembles of two prospective scenarios (RCP 4.5 and 8.5) and the historical information were computed. This study highlights Costa Rica as an inﬂexion region for climate change impacts in Central America, for which projected scenarios suggest an early onset of the rainy season, and a decline in the mid-summer drought (MSD) minimum. The assessment of model data in some regions of Costa Rica, for which historical data were available, suggests that the latter does not capture a well-known regional climate feature, the MSD, in both precipitation and temperature range well. The availability of observed past data sources is a major limitation of this research; however, with the station data used, it is still possible to draw some conclusions regarding future climate in some regions of Costa Rica, especially in the northwest side of the country, where past data are consistent with model information, providing a more reliable picture of changes in climate there that has potential implications for socioeconomic sectors.


Introduction
Costa Rica is located in the Central American Isthmus, which is the only region in the world whose position is both intercontinental (uniting two great continental masses, North and South America) and interoceanic (putting two oceans into communication, Pacific, and Atlantic Oceans [1]. Its geographical location in the tropics contributes to a unique climate that nurtures rich biodiversity, allowing for the creation of a receptive and positive environment for international conservation investment [2]. This has turned Costa Rica into a pioneer country in the environmental service payment system by establishing a formal country-wide subsidy program. Costa Rica has made substantial progress in charging water consumers, although there has been more limited progress in doing that for biodiversity and carbon sequestration users [3]. This preservation scheme, in which farmers receive payments for protecting existing forests and integrating trees into their agricultural practices, has served as a regional model [4][5][6]. Similar schemes are now being developed in Honduras, Guatemala, Mexico, and Nicaragua [6,7]. Costa Rican environmental protection policies have alleviated poverty in communities close to protected areas [8] and contribute to the national commitment to becoming calculated runoff climate change projections for the twenty-first century from a suite of 30 General Circulation Model (GCM) simulations for the A1B emission scenario in a 0.5 • by 0.5 • grid over Central America. Their main findings, of interest to this work, are that projected climate in the 2050-2099 period showed median significant reductions in precipitation (as much as 5-10%) and runoff (as much as 10-30%) in northern Central America (Guatemala, El Salvador and Belize), with no definite changes in other regions. In a more recent paper, Hidalgo et al. [64] found, while using CMIP5 data, that Panama is projected to be wetter in the future than the current climate by the mid of the twenty-first century, with Costa Rica being a region with no clear signal (see Figure 4 of the above paper). Maurer et al. [65] used historical gridded precipitation data and future projections to analyze the MSD. Their results, which are valid for most of the Pacific side of Central America, showed a decrease in the MSD minimum precipitation by the end of the twenty-first century. Recently, Gutierrez et al. [55] and Iwamura et al. [56] have employed NEX-GDDP data to analyze projected climate scenarios for public health applications. The latter work estimated the invasion potential of disease vector Aedes aegypti under climate change for several regions of the world, including Central America. Their findings for tropical areas presented a particularly high level of strong increases in the number of generations of the vector, a result that suggest the importance of implementing adaptation and mitigation strategies in this part of the world. Depsky and Pons [66] downscaled CMIP5 GCMs for different time scales to the analyzed projected dry patterns for the Central America Dry Corridor (CADC, [67]). Their results suggest a pronounced scenario in the length of the seasonal-scale droughts, such as the MSD.
One of the objectives of this research is to assess the potential applications of projected model data (NEX-GDDP) for future climate impacts in Costa Rica, especially in regions that are known to suffer dry spells on a seasonal and annual basis [63][64][65]67]. Another objective is to include in the analysis other regions of Costa Rica, such as the Caribbean and southwestern part of the country, which have been less studied in that respect. Last but not least, the authors also aim to continue the preparation of historical station data for comparison with model data in a region that is known to be deficient in data availability for climate studies [63][64][65]67].

Climate Scenarios
The NEX-GDDP dataset is composed of downscaled climate scenarios for the globe that is derived from the GCM runs conducted under CMIP5. The NEX-GDDP dataset includes statistical downscaled projections for Representative Concentration Pathways (RCPs; [68,69]) 4.5 and 8.5 scenarios from 21 models (see Table 1), for which daily scenarios were produced and distributed under CMIP5.
The RCPs scenarios provide information on the possible development trajectories of the main forcing agents of climate change, such as greenhouse gases [69]. The RCP 4.5 refers to the stabilization of the radiative forcing at 4.5 W · m −2 in the year 2100 without ever exceeding that value (approximately 650 ppm of CO 2 ). This scenario includes long-term, global emissions of greenhouse gases, short-lived species, and land-use-land-cover in a global economic framework [70][71][72][73][74]. The RCP 8.5 corresponds to concentrations of 1370 ppm of CO 2 , which makes it the route with the highest greenhouse gas emissions. This scenario combines assumptions regarding high population and relatively slow income growth with modest rates of technological change and energy intensity improvements, leading, in the long-term, to a high energy demand and greenhouse gas emissions in the absence of climate change policies [74][75][76]. The NEX-GDDP dataset is provided in order to assist the scientific community in conducting studies of climate change impacts at the local to regional scales, and enhance public understanding of possible future global climate patterns at the spatial scale of individual towns, cities, and watersheds. This dataset has a bias-corrected global grid with a high spatial (25 km × 25 km) and daily temporal resolution. It covers a period from 1950 to 2100, divided into two sub-periods, one from 1950 to 2005 (retrospective baseline) and the other one from 2006 to 2100 (prospective scenario) for the variables of precipitation, and maximum (t max ) and minimum (t min ) near-surface air temperature [77]. In this study, data for the RCP 4.5 and 8.5 was used in order to evaluate how rainfall and temperature over Costa Rica is projected to change at the end of the century (2070-2100). Additionally, with these variables, the diurnal temperature range (DTR) was derived as (Equation (1)).  Table 2 summarizes the specifications of this dataset used to carry out the aims of this research. Thrasher and Nemani [77] The Bias-Correction Spatial Disaggregation (BCSD) method that is used to generate the NEX-GDDP dataset is a statistical downscaling algorithm specifically developed for addressing the current limitations of global output from GCMs [78][79][80][81]. These limitations are related to (a) the coarse resolution grids used in most GCM runs (e.g., a few degrees or 10 2 km), which lessens their ability to capture spatial details in climate patterns that are often required or desired in regional or local analysis and (b) the local statistical biased characteristics (i.e., mean, variance, etc.) when compared with observations [77].
Because of the reliability and high spatial resolution provided by the NEX-GDDP, it is very convenient to perform climate projections analysis while using this dataset to evaluate the impacts of climate change on rainfall distribution in Costa Rica. The intraseasonal, seasonal, and annual cycle long-term changes projected in precipitation and DTR in Costa Rica at the end of the twenty-first century were computed while using the differences between the multimodel ensembles of the two prospective scenarios corresponding to the RCP 4.5 and 8.5 for the period of 2070-2099 and the retrospective scenario for the baseline period of 1970-1999 [82]. The multimodel ensemble mean of precipitation and DTR studied in this research were computed while using the following equation: where is the ensemble, x is the ensemble member variable, and n is given by the number of total ensemble members (n = 1, . . . , 21). In order to detect the ensemble atypical values we used an outliers detection algorithm that is based on the probabilistic approach described by Wilcox [83], where a value is declared as atypical if the absolute value of the difference between this one and the ensemble exceeds two standard deviations (2σ), as shown in Equation (3).
Central America and the Caribbean domain (5-25 • N, 95-60 • W) was selected in order to analyze the annual and boreal seasonal (summer (JJA) and winter (DJF)) multimodel ensemble mean long-term changes. The climate change impact assessment on annual distribution of precipitation and DTR was performed over five regions of analysis. The first region covers the entire Costa Rican territory (Costa Rica Continental), the second comprises the region from the Caribbean coast to the central mountain range (Caribbean Watershed), the third one covers the Pacific coast to the central mountain range (Pacific Watershed), the fourth only considers the North Pacific region (a.k.a Chorotega Region; the most arid area of Costa Rica, [67]), and the fifth takes the rest of the Pacific (South and Central Pacific Region) into account.

Surface Meteorological Stations
Rainfall from six and temperature from five surface meteorological stations for the retrospective baseline scenario  were selected in order to perform multimodel data assessments. The surface stations were selected as representative of the Pacific Watershed, namely: La Guinea (also in the Chorotega Region), Juan Santamaría, Fabio Baudrit, Rancho Redondo, and Piñera (located in southwestern Costa Rica, it was added to the analysis to separate this region from the northwest), and one as representative of the Caribbean Watershed, namely: Limón. Figure 1 shows the meteorological stations distribution across the country and Table 3 summarizes the metadata and their climatological  basics data (average, maximum, and minimum precipitation, and DTR) of the stations when comparing to the NEX-GDDP multimodel ensemble.
In Costa Rica, as in most Central America countries, past meteorological data have not been treated extensively to generate observed long term reliable data bases. During the last 2-3 decades, national meteorological services and some international institutions (i.e., CHIRPS, https://www.chc. ucsb.edu/data/chirps) have made efforts to gather and generate past weather and climate information and, although the situation has improved, not all data have undergone quality and homogeneity tests. Uncertainties in gridded meteorological datasets, such as CHIRPS, are also sources of error for these types of works [63]. For this work, the initial space and time coverage of observed data were deficient, so, authors performed quality control and testing on available information to complete the station data used here. As pointed out by Hidalgo et al. [63], there is a need for data to do research on climate variability and change in the region, especially with current limitations in the availability and quality of data. Recently, Amador et al. [84] also stressed the difficulties to gather both original historical documents and instrumental data for past climate studies.

Data Assessment Evaluation
From Table 3, the ability of the BCSD algorithm to adjust the precipitation data close to that of La Guinea station is noteworthy. Additionally, in the cases of Juan Santamaría, Fabio Baudrit, and Rancho Redondo stations as compared to the NEX-GDDP data, it can be seen precipitation overestimation. In the case to DTR, an overestimation also exists in all stations; however, the smallest overestimation is close to that of Fabio Baudrit station. Figure 2 shows the monthly multimodel ensemble distributions for historical NEX-GDDP precipitation and DTR for the period 1970-1999 as compared to six surface meteorological stations data (column A and C, respectively, see Figure 1 for stations data location). All stations, except Limón in the Caribbean, present similar observed annual rainfall distributions; however, what makes them different is the intensity of the rainfall totals in June and September-October and the MSD minimum, as defined by Maldonado et al. [21]. This is a crucial factor, for agriculture, tourism, and other economic sectors that the NEX-GDDP data are not capturing. Moreover, model data are not able to reproduce the MSD in any sense, except in northwestern and southwestern Pacific, presenting a feature in August that is not observed in the surface data in Fabio Baudrit and Juan Santamaría stations, confirming the shortcomings of the NEX-GDDP data. Differences between NEX-GDDP information and observed data may be due to the lack of ability of the model parameterization schemes to deal with mountainous regions at the 0.25 • × 0.25 • horizontal resolution. For the DTR, it can be observed that the NEX-GDDP data adequately captured the pattern of the annual cycle in all five stations, however, an overestimation of that cycle is also exhibited by NEX-GDDP being more remarkable in the Limón station. A consequence that results from this analysis is that the results that are associated with the Chorotega Region and the Pacific Watershed are more reliable than those related to the Caribbean Watershed. This limitation was identified by Hidalgo and Alfaro [86], and it is possibly associated with the lack of ability of current climate models to simulate the poor current knowledge of processes driving the MSD and its evolution. Future changes in the tropical circulations (i.e., fluctuations in the trade winds intensity and strength of the CLLJ and SSTs) under climate change are likely to be another cause for this discrepancy [87][88][89]. Figure 3 shows the multimodel ensemble mean projected changes in precipitation for RCP 4.5 and 8.5 scenarios over Central America and the Caribbean. The results show a reduction of rainfall over most of the domain, mainly in Central America (except for Panama), and the Greater and Lesser Antilles. The rainfall decrease seems to be more important during boreal summer for the RCP 8.5 scenario. This could be related to a future differential warming in the Pacific and Atlantic Oceans [90], where the projected rainfall deficit over Central America and the surplus in the Pacific coastal regions of northern South America could be associated with the potential intensification of ENSO and SST increase linked to future warming [91].

Rainfall
These results are in agreement with Hidalgo et al. [63], for projected climate in the 2050-2099 period that showed a decrease in precipitation (as much as 5-10%) in northern Central America. Similarly, Maurer et al. [65] reported a projected enlargement in the MSD length by an average of approximately a week and a decrease in its precipitation minimum by nearly 26% and a decline of 9.6% in annual total precipitation within the RCP 8.5 scenario for most of Nicaragua, Honduras, El Salvador, and Guatemala. The projected precipitation drop is a potential threat in terms of severe water stress affecting the water resources supply (for both consumption and agriculture) and hydroelectric power generation over this region. Table 4 shows the precipitation trends (mm/decade) over Costa Rica different regions from 1950-2100 while using the Mann-Kendall test at the 95% confidence level [83]. It can be seen that all regions have negative significant statistical trends, but Caribbean Watershed and Chorotega Region have the higher significant statistical trends. Figure 4 presents monthly multimodel ensemble distributions for historical NEX-GDDP precipitation for the period 1970-1999 (column A), and their corresponding future changes for the period 2070-2099 for RCP 4.5 and 8.5 scenarios (column B), for Costa Rica different regions (column E). It should be noted that precipitation changes indicate an early onset of the rainy season for Costa Rica (signal found in all country's different regions), projecting rainfall increasing in the transition season from a dry to a rainier season (March to April). Additionally, a small rainfall increase (decrease) is projected in the RCP 8.5 (RCP 4.5) scenario in the transition month from rainy to dry conditions (November) over all different regions (just in Caribbean Watershed and Chorotega Region).
Precipitation during the rainy season (May to October) and the minimum of the MSD are projected to decline.    Figure 2 in a like manner shows the monthly multimodel ensemble distributions for historical NEX-GDDP precipitation (column A) and their corresponding future changes (column B). It can be observed that the distributions follow the same patterns that are shown in Figure 4; however, in this figure more details are exhibited.

Temperature
Despite that a majority of climate projections tend to focus on the changes in mean temperature or temperature extremes, here we focus on the DTR instead, as it has been shown that ecosystems are more sensitive to the thermal amplitude of the environment [92,93]. Figure 5 shows the projected multimodel ensemble mean changes in DTR for RCP 4.5 and 8.5 scenarios over Central America and the Caribbean. It can be noticed that DTR and precipitation changes exhibit an opposite behavior, DTR decreases, while precipitation increases and vice versa. Even though causality is not directly implied, it should be considered that the variations in the availability of moisture due to rainfall changes is expected to have a local impact over surface temperature. A local reduction in rainfall affects humidity, as plants may increase transpiration due to water stress, with the latter being a highly relevant constraint for most ecosystems. Under such conditions in densely forested areas, via transpiration, water vapor is transferred to the atmosphere increasing the capacity of the air above the canopy to warm up. Furthermore, DTR variations imply that temperature extremes are changing, which is relevant according to the Clausius-Clapeyron relationship, saturation vapor pressure increases by approximately 7% per • C of warming [94]. As a result, warming would affect saturation and as a result, the potential for evapotranspiration is also affected, which at the same time impinges an impact on the surface-atmosphere feedback within the hydrological cycle. In addition, there are species that need specific temperature extremes for their functioning and, as DTR decreases, such species become endangered at a faster rate. The monthly multimodel ensemble distributions for DTR are presented in Figure 4 (column C for baseline and column D for future changes) over Costa Rica different regions (column E) and locally in Figure 2 (column C for baseline and column D for future changes). As we mentioned before, DTR change projections are opposite to future precipitation change. Finally, it was found that, in most of Costa Rica regions, there are no significant statistical trends (m • C/decade) for DTR based on the Mann-Kendall test at the 95% confidence level. Only Chorotega Region and South and Central Pacific Region for RCP 4.5 and Caribbean Watershed for RCP 8.5 showing significant statistical trends (see Table 4). It can be observed that projected DTR changes and their trends have very small magnitudes (∼10 −1 and ∼10 −3 /decade, respectively). This is because the maximum and minimum temperatures are projected to increase in a very similar future rate in Costa Rica.

Conclusions
This work suggests that Costa Rica is an inflexion region for climate change impacts in Central America, because the projected precipitation decreases in Costa Rica until Guatemala and it increases in Panama. This result agrees with Hidalgo et al. [63,64] and Maurer et al. [65], whose studies projected significant reductions in precipitation in northern Central America.
We find that DTR changes are opposite to future precipitation change. The very small magnitudes of projected DTR changes and their trends suggest that the maximum and minimum temperatures are projected to increase in a very similar future rate in Costa Rica.
Additionally, the precipitation changes indicate an early onset of the rainy season for Costa Rica, since the projected rainfall increases during the transition period from the dry to rainy season (March to April). Additionally, the precipitation during the rainy season (May to October) and the minimum of the MSD are projected to decline. The results that are associated with the Chorotega Region and the Pacific Watershed are more reliable than those associated with the Caribbean Watershed. This is because the NEX-GDDP data could not reproduce the rainfall features particularities in all Costa Rica different regions. This may be a limitation of this dataset and a more realistic downscaled model might be needed.
The authors recognized the shortcomings in the availability of station data to assess past and future model results in this research; however, as has been noted by other works, this has also been a major regional limitation in clarifying the uncertainty of future climate projections [63][64][65]67]. Despite this limitation with the station data used, it is still possible to draw some conclusions regarding future climate in some regions of Costa Rica, especially in the northwest side of the country, where past data are consistent with model information, providing a more reliable picture of changes in climate there that has potential implications for socioeconomic sectors.
The results that are presented herein are to be implemented, as part of a set of tools devoted to supporting decision-making, for both territory planning as well as in relation to economic activities, such as those that are related to agriculture, tourism, food security, and hydroelectric power generation. These contributions become highly relevant for areas that are identified as those becoming dryer than the current climate and whose economies and livelihood are largely dependent on rainfall.
Author Contributions: Conceptualisation was developed by R.C. and J.A.A.; data preparation and results were carried out by R.C.; writing was performed by R.C.; review and editing by R.C. and J.A.A. All authors have read and agreed to the published version of the manuscript.