1 Introduction

Small Caribbean Islands are highly vulnerable to the impacts of climate change, including extreme weather events (Bissolli et al. 2023), food security including fishery and agriculture (Lincoln Lenderking et al. 2021) and other socio-economic indicators (Stennett-Brown et al. 2019) currently combined to a low adaptative climatic capacity and a low climate-adaptation financial attraction (Fernández et al. 2023). Due to their geographic location, physical characteristics such as small size, and development profiles, Small Island Developing States (SIDS) are widely recognized as being highly vulnerable to the impacts of climate change. However, very much less studied are the Small Caribbean Islands which are not part of the 29 Caribbean SIDS (Lincoln Lenderking et al. 2021). San Andres, Providencia, and Santa Catalina (SAI) is a Colombian archipelago in the Caribbean Sea at about 775 km northwest of mainland Colombia. Due to the geographic condition of insularity characterized by small oceanic islands (Gobernación del Departamento de San Andrés, Providencia y Santa Catalina 2020), SAI is highly vulnerable to tropical storms, storm surges, coastal flooding, and the effects of sea level rise, which may increase significantly in a context of climate change (Cooley et al. 2022; IPCC 2022).

The main factors affecting climate and ocean change, especially small islands, include air and ocean temperature variations, ocean chemistry, rainfall, evaporation, wind strength and direction, sea levels, and weather extremes such as cyclones or tropical storms (Cooley et al. 2022; IPCC 2022). Although each of these has different effects, depending on their magnitude and temporal and spatial frequency, they impact the natural, social, economic, and political environment (Cooley et al. 2022; IPCC 2022).

This vulnerability has been increasing due to the constant transformations of the territory, especially by anthropogenic processes (Gobernación del Departamento de San Andrés, Providencia y Santa Catalina 2020), despite the declaration of the archipelago by UNESCO as a Seaflower Biosphere Reserve in 2010, which promoted zoning of the islands into three special management areas related to the functions of conservation, development, and logistic support to promote the sustainable development of the island territories (Gobernación del Departamento de San Andrés, Providencia y Santa Catalina 2020).

In terms of territorial transformations, many of the archipelago's ecosystems have significantly deteriorated, mainly due to the uncontrolled development of productive activities and the effects of municipal waste pollution, making the islands more vulnerable to environmental and anthropogenic stresses (Lopez-Rodriguez et al. 2009). This vulnerability was particularly exposed in November 2020, when hurricane Iota, which reached category V with winds of up to 240 km/h, passed through the archipelago, leaving 6′300 people affected in Providencia and Santa Catalina and at least 700 families on the island of San Andres, with infrastructure damage of 98% on the island of Providencia (Sociedad Nacional de la Cruz Roja Colombiana 2020).

According to the Third National Communication of Colombia (IDEAM, PNUD, MADS, DNP, CANCILLERIA 2017), the archipelago is the department in Colombia with the highest vulnerability to climate change, which means that food security and infrastructure will be negatively affected by its effects. This is coupled with a very low adaptive capacity and a very high exposure. Moreover, IDEAM, PNUD, MADS, DNP, CANCILLERIA (2015) projected, amongst the 33 departments of Colombia, the largest departmental average precipitation changes within the next decades (towards drying: -30% by 2040 to -33% by 2100) for the archipelago of San Andrés, Providencia and Santa Catalina. This makes SAI a particular region of interest for accurate hydroclimatic assessment.

Overall, the Caribbean climatology in terms of temperature has a relatively small positive anomaly in 2022 compared to previous years, with the annual anomaly for the region being 0.45ºC above average, making 2022 the eighth warmest year since 1950. In terms of precipitation, most Caribbean islands had slightly above-average precipitation in 2022 with an average rainfall anomaly of -0.22 mm per day (Bissolli et al. 2023). Seasonally, during the months of December through February, the region's temperature was above average, while precipitation decreased and the region experienced moderate to severe drought, and during the months of March through May while temperature and precipitation were above average (Bissolli et al. 2023).

Valencia and Osorio (1999) analyzed the interdecadal variability of rainfall and temperature for the island of San Andres, taking as reference the data provided by the Institute of Hydrology, Meteorology and Environmental Studies of Colombia (IDEAM), for the period from 1958 to 1997. In this study, precipitation did not show significant trends. They found that La Niña episodes generated increased rainfall and decreased temperatures and that the greater incidence of climatic extremes during these episodes denotes notable alterations of the normal climate regime (Valencia and Osorio 1999). In line with the above, a study carried out by Guerrero Jimenez (2019) determined that since 2006, there has been a gradual decrease in annual mean rainfall. In addition, based on data obtained from the year 1965, it was concluded that rain did not show large changes related to climate variability events. Only 31% of the years with a La Niña report recorded rainfall above the upper limit of normal precipitation values, while for El Niño, rainfall was below the lower limit of normal values in only 39% of the years (Guerrero Jimenez 2019).

However, according to the literature, there is no comprehensive assessment of extreme weather over historical and future periods exists for SAI over the last recent decades. Most climate extremes studies focus on Colombia's land (e.g., Skansi et al. 2013; Ávila et al. 2019; Cerón et al. 2022), large areas of Central America Land (e.g., Aguilar et al. 2005; Imbach et al. 2018; Avila-Diaz et al. 2023) or specific islands over the Caribbean (e.g., Stephenson et al. 2014; Beharry et al. 2015).

The objective of this study is to perform a detailed evaluation of mean and extreme precipitation and temperature from the 1960s to 2020 using seven observational stations, and for future periods, an ensemble of CMIP6 models assessed under high-emission scenario by the end of the century. In addition, this paper aims to contribute to solving the following scientific questions: I) How have extreme temperatures and precipitation in the SAI Caribbean islands varied over the last decades? II) What are the relations between the region's extreme precipitation, ENSO, and hurricanes? And III) What are the projected changes in mean and extreme climate by CMIP6 models?

2 Methodology

2.1 Study area

The Archipelago of San Andres, Providencia, and Santa Catalina is a department of Colombia consisting of small oceanic islands distributed in San Andres (27 km2), Providencia (18 km2), and Santa Catalina (1 km2), in addition to different islets (Instituto de Investigaciones Marinas y Costeras José Benito Vives de Andréis 2012). The islands are located between parallels 10° and 18° North Latitude and meridians 78° and 82° West Longitude, with a sea area of 350,000 km2, of which 65,000 km2 are marine protected areas (Gobernación del Departamento de San Andrés, Providencia y Santa Catalina 2020), declared by the Colombian Ministry of Environment and Sustainable Development in 2005 for their special ecological, economic, social and cultural importance.

The archipelago was declared a Seaflower Biosphere Reserve by UNESCO in 2000 (see Fig. 1), where the zoning of the area was carried out based on several scenarios that led to its sustainable development and in which different types of dynamics, problems, and environmental, social, cultural, economic, and political contexts are included (Gobernación del Departamento de San Andrés, Providencia y Santa Catalina 2020).

Fig. 1
figure 1

Archipelago of San Andres, Providencia, and Santa Catalina, Colombia. Name and location of the metereological stations studied in colored dots. Historical precipitation expressed in mm/month from WorldClim (https://www.worldclim.org), for the 1970–2000 period with spatial resolutions of 30 s (~ 1 km2)

Full size image

In terms of weather conditions, the CIOH (Centro de Investigaciones Oceanográficas e Hidrográficas 2010) has described the climatology of the archipelago based on data collected from 1961 to 1991 from an IDEAM synoptic station, which shows that the area has a warm humid climate, with an average annual temperature of 27.4 °C and an annual relative humidity of 82%, where annual rainfall exceeds 1900 mm with 217 days of precipitation. However, there is no systematic analysis of trends related to climate change or climate extremes.

Rainfall for the archipelago is determined by highest rainfall amounts occurring between June and November and lowest rainfall between January and April, with the period from May to December being a transition period (Gobernación del Departamento de San Andrés, Providencia y Santa Catalina 2020).

The archipelago is in the lower part of the hurricane belt of the Caribbean (Centro de Investigaciones Oceanográficas e Hidrográficas 2010), where in the 200 km around San Andres 23 hurricanes were recorded from 1876 to 2020, including Eta and Iota in 2020 (NOAA Office for Coastal Management 2021).

During December to March, there are favorable positions for a constant flow of wind, causing the highest wind speeds of the year, which, according to IDEAM's multiannual records, range between 6.3 and 7.0 m/s. Likewise, from April to November, there are average speeds between 4.5 and 6.2 m/s, and from June to July, there are average speeds between 5.9 and 7.8 m/s, respectively, the latter being the average for the entire year (Centro de Investigaciones Oceanográficas e Hidrográficas 2010).

2.2 Estimation of climatology and extreme climate

Based on the data obtained for precipitation and temperature from IDEAM (2021) weather stations in SAI, time series on a daily basis were estimated to quantify in terms of frequency, intensity, and duration, the extreme weather events in comparison with precipitation thresholds and contrasted with the occurrence of hurricanes and/or tropical storms. In the context of climate change, this makes it possible to identify when these events could be repeated and whether, for example, they will increase in intensity, frequency, and duration.

Regarding the total number of data, it is important to point out that some stations do not have precipitation information for some years, which vary depending on the metereological station (see Supplementary Table 1), as for the period of time with available information, this also varies according to each meteorological station but the oldest record is from 1973 at the El Embrujo Airport station in Providencia and Santa Catalina.. As for the maximum temperature, this variable is only available for two stations, i.e., Sesquicentenario Airport for San Andres Island and El Embrujo Airport for Providencia and Santa Catalina, for the period 1978–2017 (see Supplementary Table 1).

We selected 7 stations (see Fig. 1) in the period where at least 90% of the daily data was available. For a maximum of two years where more than 75% of the daily data was available, we corrected the calculated extreme indices by multiplying them by the ratio of the number of days per year and the number of days with available data, this in order to reduce the bias. Compared to removing these years, this correction, has minor effects on trends and significance. We separate historical period (HIST) and current period (CURRENT) by the center of the available data period or at least containing 20 years for each period (cut-off year: 1995 ± 3 years for all stations except for Hoyo Soplador and Agua dulce stations having less than 35 years of data; see Supplementary Table 1). Our approach aims to maximize in priority the number of years included in the calculated climatologies (≥ 20 years, for HIST and for CURRENT) while minimizing overlap, thereby enhancing robustness.

For each station, the climatology of average precipitation (PR) and average daily, monthly, and annual temperature (T) was calculated over the entire available period, over a recent period (approximately the last two decades, "CURRENT"), and a past period (the first two available decades, "HIST"). In order to increase the robustness of our results, we decided to add ERA5, a widely used gridded climate dataset for climatology assessments and extreme climate events (Avila-Diaz et al. 2021; Lavers et al. 2022; Randriatsara et al. 2022; Vega-Durán et al. 2021). ERA5 is the fifth generation ECMWF atmospheric global climate reanalysis covering a much larger period from January 1940 to the present (Hersbach et al. 2020). It uses an advanced weather model that combines real-time measurements from satellites and ground stations to create a detailed picture of area climate patterns. Furthermore, it is very important to note that the ERA5 reanalysis data has a resolution of 0.25° which is equivalent to a pixel of 25 km × 25 km, i.e., 20 times larger than the San Andrés Island territory in terms of area. Hence, using stations on small islands becomes crucial for accurate historical assessments at the local scale. Our study will thus also evaluate the performance of ERA5 compared to island stations as performed in other studies (e.g. Lavers et al. 2022).

Similarly, the following ETCCDI (Expert Team on Climate Change Detection and Indices) precipitation extreme indices were estimated for each year (ARC Centre of Excellence for Climate Extremes University of New South Wales 2021): R1mm ("wet days", annual number of days with precipitation > 1 mm), R10mm (idem with precipitation > 10 mm), R20mm (idem with precipitation > 20 mm), Rx1day (annual record daily precipitation), Rx5day (highest annual five day precipitation amount), R95p (annual precipitation amount during wet days belonging to the 95th percentile), R99p (annual precipitation amount during wet days belonging to the 99th percentile), PRCPTOT (annual precipitation amount during wet days), CWD (maximum annual number of consecutive wet days, > 1 mm) and CDD (maximum annual number of consecutive non-wet days, < 1 mm).

In addition to the ETCCDI (World Climate Research Programme 2021) extreme temperature indices, we incorporated key indices related to extreme heatwaves. These include the count of short and long heatwaves (annual count of waves with TX > 90th percentile, spanning at least 3 or 5 consecutive days, respectively) and the maximum duration of a warm spell (the maximum annual number of consecutive days with TX > 90th percentile).

Finally, to evaluate the sensitivity of our results to the type of extreme indices, we built the RXXp and TXXp indices, defined as the Extreme wet days (i.e., Annual total precipitation when daily rainfall is greater than the XXth percentile) and Extreme hot days (i.e., % of days when daily maximum temperature is greater than the XXth percentile), respectively, where XX vary from 90 (i.e., R90p and T90p) to 99.5 (R99.5p and T99.5p) percentile (considering a 0.5 percentile step).

To quantify the changes in precipitation and temperature extremes that occurred in the archipelago, we calculated: i) the trend of change over the maximum period of available data and ii) the percentage difference between the CURRENT and HIST periods ("CURRENT- HIST"). Finally, the significance of the changes (difference) was evaluated with a Student t-test at the 90% ("p < 0.1") and 95% ("p < 0.05") confidence level.

To evaluate the change in hurricanes and tropical storms and their characteristics (pressure, distance, maximum winds, precipitation), data from the NOAA National Hurricane Center HURDAT2 & NOAA National Centers for Environmental Information IBTrACS (2021) were collected. We selected: i) Category I, II, III, IV, and V Hurricanes and Tropical storms according to the Saffir-Simpson scale, ii) being relatively close to San Andres Island (less than 600 km radius from the point of latitude 12.541077°N and longitude 81.715031°W), and iii) from 1964 to 2020. The category associated to each hurricane or tropical storm is determined here by the category with maximum duration within the 600 km-radius from San Andrés Island center. Our decision to choose a radius of 600 km is grounded in the average findings of the literature (e.g. Englehart and Douglas 2001; Kimball and Mulekar 2004; Pérez-Alarcón et al. 2021), which suggest that the impacts of most North Atlantic tropical storms and hurricanes, including winds and rainfall, are typically within this distance.

El Niño-Southern Oscillation (ENSO) data from 1962 to 2020 were extracted from NOAA's Climate Prediction Center corresponding to quarterly values of the Oceanic Niño Index (ONI) [3-month running average of sea surface temperature anomalies in the Niño 3.4 region (5°N-5°S, 120°- 170°W)], based on 30-year centered base periods updated every 5 years (NOAA National Weather Service & National Centers for Environmental Prediction 2021). Global cold La Niña and global warm El Niño episodes are characterized by ONI values < -0.5 and ONI > 0.5, respectively, while neutral episodes correspond to ONI values between -0.5 and 0.5. The associated month corresponds to the central month of each quarter.

2.3 Daily data for CMIP6 climate models

For this study, nine models (see Table 1) were downloaded from the WCRP Coupled Model Intercomparison Project (https://esgf-node.llnl.gov/search/cmip6/). Criteria for selecting the CMIP6 models are the following: (a) simulations (1980–2014) and projections (2015–2100) under Shared Socioeconomic Pathways: SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 shall be available, (b) the three variables used to calculate extremes climate indices are available (i.e., daily precipitation, minimum and maximum temperature), (c) models with a horizontal resolution equal or less than 100 km. We considered the first ensemble member (r1i1p1f1). These criteria were established to avoid the same values information from San Andrés and Providencia, the data were extracted at geographical coordinates centered around the islands: 12.52°N (latitude), 81.72°W (longitude) for San Andrés and 13.35°N, 81.37°W for Providencia. To ensure a reasonable sample size of pixels per model, the rectangular box chosen was [84°W;80°W]x[11°N;15°N] (see Supplementary Table 2 for the number of pixels per model). Results were very similar when gridpoint interpolation was chosen.

Table 1 CMIP6 models were used in this analysis. The plus sign (asterisk) indicates information availability (unavailability) in a particular scenario
Full size table

3 Results and discussion

3.1 Historical mean and extreme precipitation changes

The monthly climatology of the study area is based on the data obtained from the 7 IDEAM weather stations based on the current time series (CURRENT) that accumulates ~ 20 years, expressed in mm/month (Fig. 2), considering that time period varies according to station and amount of data. In this sense each station has a different time period which in general ends in 2020 but started in 1992 (Sesquicenten Airport), 1995 (El Embrujo Airport), 1997 (Empoislas, Pueblo Viejo y San Felipe), or 2001 (Hoyo Soplador y Agua Dulce). The climate of SAI is tropical semi-humid and is characterized by a dry season centered around February-April (~ 1.1 mm/day on average) and a wet season with a quasi-bimodal behavior: first wet ("modest") season from May to August (with rainfall of ~ 5.4 mm/day on average) and a second ("high") wet season from September to November (~ 9 mm/day on average), reaching its peak in mid-October (Fig. 2).

Fig. 2
figure 2

Monthly precipitation climatology expressed in mm/month, based on the current period (CURRENT) for the Archipelago of San Andres, Providencia and Santa Catalina

Full size image

We found a bimodal regime in most stations which is more pronounced on the historical period compared to the current one (1964–1992 for San Andres Island and 1973–1995 for Providencia and Santa Catalina, blue light bars; Fig. 3). Note that this regime classification found in SAI agrees with the one found by Guzmán et al. (2014) that evaluated the precipitation patterns over Colombia during 1971–200. However, the regime in SAI does not coincide with some descriptions of the island’s climatology, which is defined as an unimodal cycle by other studies (Gobernación del Departamento de San Andrés, Providencia y Santa Catalina 2020; Urrea et al. 2019). These different regimes can be attributed to the distinct selection criteria, climatogical period chosen and methodologies employed.

Fig. 3
figure 3

Monthly precipitation climatology for the Sesquicentenario Airport station in San Andres Island and for the El Embrujo Airport station between the current period (CURRENT, red bars) and the historical period (HIST, blue light bars)

Full size image

There is a marked spatial variability in annual precipitation between nearby stations (over the common 1986–2019 period) on the island of San Andres: Sesquicentennial Airport (1939 mm/yr), EmpoIslas (2015 mm/yr), Hoyo Soplador (1740 mm/yr) but less pronounced over Providencia and Santa Catalina islands: El Embrujo Airport (1739 mm/yr), Agua Dulce (1756 mm/yr), Pueblo Viejo (1665 mm/yr), San Felipe (1794 mm/yr). San Andres Island has a significant south-north precipitation gradient with 10–15% less rainfall in the south (Hoyo Soplador) vs. middle (EmpoIslas) / north (Sesquicentenario Airport) locations.

To further detail the results obtained, a specific analysis for the Sesquicentennial Airport and the El Embrujo airport stations is performed (Fig. 3). Between two 20 + years climatological periods, a significant increase in precipitation (+ 7%, + 78 mm/year; p < 0.05) was found in the wettest month: November (+ 40%, + 96 mm/month; Fig. 3, bars “November” of the upper panel) in San Andrés. In recent decades, November is now the wettest month in San Andres (336 mm/month on average during 1992–2020), while previously October was (302 mm/month from 1964 to 1992). The opposite is true for Providencia, where October was and still is the rainiest month (276 mm/month on average during 1973–1995 – blue light bar – and 402 mm/month on average during 1995–2017, red bar in lower panel).

The amount of precipitation during most extreme rainfall events has increased significantly between the two periods: the annual record daily precipitation increased by 18%, while the amount of precipitation during very wet days (95th percentile) and extreme rainy days (99th percentile) is rising by 30% and 24% respectively. Consecutive rainy days are also increasing by 18% (Table 2).

Table 2 Summary of significant extreme precipitation changes (p < 0.1: * y p < 0.05: **) between CURRENT and HIST periods in each SAI meteorological station. The corresponding change for each significant extreme precipitation change is related between brackets. ↗ y ↘ indicate increase (blue colors) and decrease (red colors), respectively
Full size table

Moreover, we contrasted the current climatology (CURRENT: 1992–2020 (Sesquicenten Airport), 1995 -2017 (El Embrujo Airport), 1997–2020 (Empoislas, Pueblo Viejo y San Felipe) and 2001–2020 (Hoyo Soplador y Agua Dulce)) with the previous climatology (HIST: 1964–1992 (Sesquicenten Airport), 1973 -1995 (El Embrujo Airport), 1974–1997 (Empoislas, Pueblo Viejo y San Felipe) and 1986–2005 (Hoyo Soplador y Agua Dulce)), to determine whether and where precipitation has significantly varied in the recent decades. The significance of this variability was estimated by applying a Student t-test, determining with two asterisks (**) those months where the change is very significant (95% confidence level), and with an asterisk (*) those months where it is significant (90% confidence level) (Fig. 4). We found that the late-rainy season (since October) is significantly more rainy, particularly the two wettest months (October–November). By contrast, the early dry season (February) and early wet season (June–September) tend to be slightly drier (Fig. 4).The dry season in SAI is associated with the San Andres jet and North-East (NE) trade winds. When the San Andres jet is strongest, the dry season occurs along the entire Caribbean coast and coincides with very strong NE trade winds. On the other hand, the wet season coincides with the highest intensity of the Chocó jet and the lowest intensity of the San Andres jet (Poveda and Mesa 1999; Bernal et al. 2006). Moreover, as found by Carmona Ramírez et al. (2010): “the variation in surface temperatures of the North Tropical Atlantic has a significant influence on the annual rainfall cycle of SAI, with increases of over 60% in the months of October and November”. Thus, there is an anticorrelation between overall monthly rainfall in SAI and monthly trade wind speed and a positive correlation in the tropical North Atlantic between surface temperatures in October–November and October–November rainfall in SAI. In consequence, the potential reasons behind the significant rainfall increase in October and November in most SAI stations are: 1) the historical increase in the tropical North Atlantic under recent global warming and, 2) a weakening of NE trade wind speed around SAI islands, particularly in the NE sector (Bustos Usta and Torres Parra 2023).

Fig. 4
figure 4

Difference in monthly precipitation climatology (mm/month) between the current period (CURRENT) and the previous period (HIST) for each month and each station in the Archipelago of San Andres, Providencia, and Santa Catalina. Significance, obtained by applying the Student t-test, is indicated as follows: p < 0.1 with * and p < 0.05 with **

Full size image

Once the mean climatology was determined, we proceeded to analyze the percentage difference in the amount of precipitation during precipitation extremes (ETCCDI index: RXXp) between the current period (CURRENT 1992- 2020) and the historical period (HIST 1964- 1992) as a function of the degree of extreme (XX from 90 to 99.5).This aims to estimate, in percentage terms, whether the precipitation extremes have increased over time. In other words, it allows us to identify whether extreme weather events have increased in the archipelago as a function of the increase in precipitation, comparing two climatic periods, which allows us to estimate in the future whether the number of days with more rainfall will increase and exceed these extremes. This makes it possible to estimate whether the number of days with more rainfall will increase and exceed these indexes.

From data shown in Fig. 5, it was found that in general the more extreme the precipitation considered, the more important the increase in the current period. For example, for the Sesquicentenario Airport station for the rainfall events of the 90th percentile, an increase of less than 20%; for the 99.5th percentile, an increase of 140% is found, similar to the Agua Dulce and El Embrujo airport stations, which show a significant percentage increase. In other words, trends are generally more pronounced in SAI while more extreme rainfall is considered (90th → 99.5th, see Fig. 5 and Supplementary Fig. 1). Note that dry trends are not significant (not shown).

Fig. 5
figure 5

Percentage difference in the amount of precipitation during precipitation extremes (Adapted ETCCDI index: RXXp, see Methods) between the current period (CURRENT, 1992–2020) and the historical period (HIST, 1964–1992) as a function of the degree of precipitation extremes (XX from 90 to 99.5). Dotted lines indicate non-significant changes in those specific adapted ETCCDI indices while solid lines indicate significance (p < 0.05). Example: For days where precipitation is above the 99.5.th percentile, there is a significant increase of ~ 140–150% of the precipitation amount in the current period compared to the historical one in Sesquicentenerario Airport, Agua Dulce and El Embrujo Airport stations. See also Supplementary Fig. 1. (Same as Fig. 5 but for ERA5 precipitation series in two pixels)

Full size image

To summarize, extreme rainfall events significantly increased in frequency and intensity in most stations, particularly Sesquicentenario Airport, El Embrujo Airport, and Agua Dulce, but opposite and heterogeneous extreme rainfall trends are also found across a small territory (< 30 km2): significant extreme dry trends in Hoyo Soplador at the South of the San Andrés Island are particularly different (Table 2). As shown in other analyzes of climatic trends in small islands (McGree et al. 2019; Leon and Oculi 2023; Yeo et al. 2023), this large spatial variability in precipitation trends can be attributed to local changes in topography, land use, land cover, microclimatic conditions, and/or certain inhomogeneities could have acted together to make precipitation trends opposite between nearby parts of the SAI. First, the centers of the Islands have some elevated areas (up to 90 m in San Andrés islands and up to 345 m in Providencia island). While the dominant direction for winds and currents in SAI is East-North-East (ENE) according to the wind and wave rose, topography could have shadow rainfall in the stations of the West (San Felipe and Pueblo Viejo) and South (Hoyo Soplador) in the historical period, where there are non-significant or negative trends in precipitation extremes. Moreover, between 1985 and 2022, San Andrés island witnessed a decrease in forest cover, especially along its Western, Eastern, and Southern coasts (see the customized map of the “MapBiomas Amazonia Colección 5” consulted on February 7, 2023: https://bit.ly/LandCoverMapSAI). In particular, a significant reduction in forests is noticeable in the surroundings of the Hoyo Soplador station which is associated in the Tropics with a local reduction in evapotranspiration and a subsequent local reduction in precipitation (Quesada et al. 2017). However, considering that it is a small island with a limited number of trees and total evapotranspiration compared to the nearby ocean, the precipitation impact of forest loss is likely to be minimal. Finally, we contacted officers in the IDEAM institute (IDEAM, pers. comm.), who guarantee a “large homogeneity” (the data systematically undergoes quality control and an approbation score for homogeneity assessment) of the precipitation data across decades in SAI except in many days in Hoyo Soplador station during the year 1986. As a robustness test, we recalculated the trends removing the year 1986 in this specific Hoyo Soplador meteorological station (not shown). We find that it does not change the overall historical trend towards drying for most significant indices (Mean precipitation, R1mm, PRCPTOT, CWD, CDD, R99.2–99.4p), but it reduces the significance and magnitude of the changes. Historical trends of mean precipitation go from a significant -21% to a non-significant -11%, for R10mm and PRCPTOT, it becomes non-significant, for R1mm it goes from -30% to -25% while CDD and CWD trends are slightly reduced as well (not shown). This suggests a significant impact of some heterogeneities (mainly present during 1986; IDEAM, pers.comm.) although it is not the primary factor explaining the drying in this specific zone. All these reasons underscore an urgent need in the near future to analyze the relative contributions of each factor to precipitation variability in such small territories.

The analysis of the correlation between stations and reanalysis show an overall great agreement for precipitation: 0.44–0.83 (see Supplementary Fig. 3). Moreover, reanalysis underestimate the historical increasing trend of the most rainfall extremes, defined as 95th/99.5th percentile, in some parts of San Andrés, for instance in Sesquicentenario Airport: + 35%/ + 151% at the station and + 9%/ + 25% for ERA5-San Andrés (see Supplementary Fig. 1).

3.2 Historical mean and extreme temperature changes

The maximum temperature climatology displays a quasi-bimodal regime with a high peak in September (31.27ºC in El Embrujo Airport and 30.93 °C in Sesquicentenario Airport in the CURRENT period) and a low peak in January (29.22 °C and 28.86 °C, respectively; see Fig. 6). Only two stations were analyzed due to the limited temperature data available for the archipelago, being the stations located at both airports the ones with the most data and the most representative as evidenced by the maximum correlation with a coarser resolution dataset (ERA-5) for these stations (see Supplementary Fig. 3 and Supplementary Table 1). We also found an increase in maximum temperature of 0.4 °C and 0.22 °C in San Andrés and Providencia islands, respectively, from 1978 to 2017. Although Providencia and Santa Catalina tend to have slower recent warming than San Andrés (Fig. 6), those islands are warmer than San Andrés Island (Fig. 7). The warmest month (September) is the month with the most significant warming (+ 0.5–0.7 °C, Fig. 7), while November, the rainiest month, is the one with less recent warming in SAI (≤ 0.10 °C).

Fig. 6
figure 6

Monthly maximum temperature climatology for the Sesquicentenario Airport station in San Andres Island and the El Embrujo Airport station between the current period (CURRENT, red bars) and the historical period (HIST, blue light bars)

Full size image
Fig. 7
figure 7

Difference in monthly maximum temperature climatology (ºC) between the current period (CURRENT, 1997–2017) and the previous period (HIST, 1978–1998) for each month, for Sesquicentenario Airport (red) and El Embrujo (green) station in San Andres and in Providencia islands, respectively. Significance, obtained by applying the Student t-test, is indicated as follows: p < 0.1 with * and p < 0.05 with **

Full size image

For the extreme events, between CURRENT and HIST periods, we found that hot days (≥ 90th percentile) are multiplied by 2 and 3.1 for El Embrujo and Sesquicentenario stations, while very extreme hot days (> 99th percentile) are multiplied by 2.2 and 4.4, respectively (Fig. 8).

Fig. 8
figure 8

Multiplication factor of extreme hot days between current (1997–2017) and historical (1978–1998) period in function of extremeness of the event (XXth percentile, X from 90 to 99.5) for the two available meteorological stations for temperature (Sesquicentenario Airport in red and El Embrujo airport in green). Example: For an extreme hot day defined as a daily maximum temperature above the 99th percentile, there are 2.2 times more extreme hot days in the current period than the historical one in Sesquicentenerario Airport and 4.4 times more in the El Embrujo Airport. See Supplementary Fig. 2 same as Fig. 8 but for ERA5 temperature series in two pixels

Full size image

We also found that short heatwaves (≥ 90th percentile during at least 3 days) and long heatwaves (≥ 90th percentile during at least 5 days) increased by a factor 3.8 and 5.2 in Sesquicentenario Airport and by a factor 3.4 and 3.5 in El Embrujo Airport between the two periods. The duration of consecutive hot days has been multiplied by 2.2 (+ 3.8 days on average) and 1.7 (+ 2.9 days). The multiplication factor between 90 to 99 percentile is lower than between 99 to 99.5 on average, indicating a slightly increased frequency of the more extreme maximum temperature and of the more extreme heatwaves (Fig. 8). Across a larger period and number of stations of meteorological measurements compared to (Valencia and Osorio 1999), our study finds significant hotter-and-drier trends in mean climate over the recent decades but also on extreme weather indices (not studied before).

Applying the same methodology with ERA5, described in 3.1 for precipitation, but in this case for temperature, we find that, the analysis of correlation between stations and reanalysis show a overall great agreement particularly for temperature: 0.85–0.88 (see Supplementary Fig. 3). However, we show that the ERA5 reanalysis overestimates the historical increasing trend in very extreme hot days compared to the stations. In the ERA5 reanalysis, for an extreme hot day defined as a daily maximum temperature above the 99th percentile, there are 5.1 times more extreme hot days in the current period than the historical one in San Andrés and 6.2 times more in Providencia (see Supplementary Fig. 2). By contrast, stations show a multiplication of those very extreme hot days by 2.2 and 4.4, respectively.

3.3 Relations between extreme precipitation, ENSO, and hurricanes

During El Niño, La Niña, and neutral periods, the recent climatologies of annual precipitation at Sesquicentenario are 1568 mm/year, 1723 mm/year, and 1851 mm/year, respectively. Figure 9 shows that, on average, during each month, except July and October, the Neutral episodes correspond to a higher monthly precipitation in the archipelago than La Niña or El Niño episodes. During transition months, precipitation is still substantially lower in El Niño vs. Neutral episodes: February (-66%), December (-48%), August (-29%), and January (-26%).

Fig. 9
figure 9

Monthly precipitation climatology (mm/month) during El Niño (red), La Niña (blue light) and Neutral months (grey) in the Sesquicentenario Airport (San Andrés Island) over the 1962–2020 period. (p < 0.1: * y p < 0.05: **). Note that El Niño Southern Oscillations months are defined according to the ONI index of the centered month (see Methodology)

Full size image

Over the period 1962–2020, there is no significant correlation (r = 0.06, p > 0.1) between the amount of precipitation in San Andres and the intensity of the El Niño-Southern Oscillation. Similarly, with a lag of 1 month between the two-time series (r = 0.09, p > 0.1; not shown). This makes ENSO not a substantial predictor of precipitation in SAI, particularly in the rainiest months, and therefore cannot be used as a proxy for precipitation in the archipelago on a multi-annual or multi-decadal scales. Although we find rainfall deficit during El Niño and rainfall excess during La Niña events as (Carmona Ramírez et al. 2010) found in a shorter time period (until 2008), our results indicate a lower rainfall sensitivity to ENSO phenomenon. In our study, rainfall magnitude is modulated by around 10% on average between El Niño and La Niña events (compared to the “high influence of ENSO” in the annual cycle of precipitations with rainfall “variations between 25 and 30%” found by Carmona Ramírez et al. 2010).

Complementarily, Supplementary Figures 4-6 show that: (i) There is overall more rainfall (+ 27%, + 35 mm/month on average) in SAI during La Niña compared to El Niño events (all stations display this significant behavior). This is particularly clear during July, October, and December (+ 93, + 130 and + 61 mm/month respectively, on average) (Supplementary Fig. 4), (ii) There is overall less rainfall (-13%, -21 mm/month on average) in SAI during El Niño compared to neutral events (all stations display this significant behavior except Hoyo Soplador). This is particularly clear during July, October and December (-44, -53 and -56 mm/month on average) (Supplementary Fig. 4), (iii) Very low negative or non-significant correlations between ONI indices and rainfall are found (-0.10 ± 0.06, ranging from -0.28 to + 0.08) (Supplementary Fig. 5). With a lag of 1 month, correlations are on average similar (-0.09 ± 0.06). Those correlations are slightly lower (-0.08) in the recent period (CURRENT) compared to the historical period (-0.12) and (iv) 4) Extreme rainfall are not particularly associated with strong La Niña o El Niño events (Supplementary Fig. 6).

Furthermore, within years 1964–2020, all hurricanes (categories I, II, III, IV, and V on the Saffir-Simpson hurricane scale) and tropical storms (hereafter referred to as "H&TS") events near San Andres Island (< 600 km) have been studied. We estimated that there were 61 H&TS recorded in this period, representing ~ 1 H&TS per year (including 0.8 hurricanes/year), with an average maximum wind speed of 92 km/h, an average duration of 9 days, and an average distance from San Andres Island of 340 km. In turn, 14 tropical storms were found, 12 category I hurricanes, 10 category II hurricanes, 8 category III hurricanes, 9 category IV hurricanes and 8 category V hurricanes. Year 2020 showed the highest hurricane activity of the period studied, with 5 hurricanes: 1 category V hurricane (Iota), 2 category IV hurricanes (Eta and Delta), 1 category II hurricane and 1 category I hurricane.

Compared to the HIST period, this 2020 hurricane season means a 3.8-fold increase in hurricane frequency, a 27% increase in maximum wind speed and one additional day of average duration. The number of such tropical storms and hurricanes doubled between the "historical" period 1964–1992 (n = 21, TS and category IV) and the "current" period 1992–2020 (n = 40), which means 0.72 (21/28) H & TS/year to 1.38 (40/28) H & TS/year in the current period.

Between these two periods, the maximum wind speed during hurricane and tropical storm episodes increased on average by ~ 10% (from 86 km/h to 95 km/h). The estimated distance of hurricanes and tropical storms did not vary on average between the two periods (341 km from San Andres Island, for both periods, i.e., < 1% difference). Likewise, the average duration of hurricanes and tropical storms did not recently change (< 1% difference). From 1964 to 2020, category IV and V hurricanes near the archipelago had an average maximum wind speed of 135 km/h, an average duration of 12 days and an average distance from San Andres Island of 369 km. Category IV and V hurricanes near the archipelago more than doubled (× 2.4) between the current and historical periods.

In recent decades, hurricanes and tropical storms occur later in the year (November–December), while in the historical period, their number peaked in October. Surprisingly, precipitation during hurricanes and tropical storms does not vary significantly between categories (Fig. 10): the average precipitation during H&TS days is ~ 6–10 mm/day on average and the maximum daily precipitation during H&TS reaches ~ 18–50 mm/day on average. In other words, the mean precipitation during H&TS is below the 90th daily percentile (i.e., more around 80%), while the maximum daily precipitation during H&TS is below the 99th daily percentile (i.e., plus about 95%). Moreover, not daily mean nor daily maximum precipitation increases with an increasing category of hurricanes (Fig. 10). The increase in average daily precipitation attributed to H&TS shows a very limited relative contribution (only + 8% monthly precipitation during H&TS compared to monthly precipitation climatologies; refer to Supplementary Table 3).

Fig. 10
figure 10

Box-plot of average and maximum daily precipitation values (mm/day, Sesquicentenario Airport) during tropical storm (TS) periods and during hurricanes of each category on Saffir-Simpson scale (1, 2, 3, 4 and 5) near the Archipelago of San Andres, Providencia and Santa Catalina (< 600 km from San Andres Island center). Grey to black lines show the 90th, 95th, and 99th percentile daily precipitation values

Full size image

More surprisingly, of the approx. 200 record daily precipitation events (99th percentile) on San Andres Island (Sesquicentennial Airport), only 13 days (6%) coincide with the dates of hurricanes and tropical storms passing nearby (< 600 km from San Andres Island) or 24 days (12%) if we add daily precipitation records ± 5 days before/after the H&TS dates (not shown). Moreover, the heaviest rainfall events do not occur with growing categories of hurricanes. This indicates that overall, the extreme rainfall or flash floods coincide marginally with tropical storms or hurricanes.. It is logical to inquire whether heightened hurricane activity results in higher levels of rainfall, potentially indicating a faster water cycle and more associated floods. However, if record rainfall events can be synchronous with hurricanes and tropical storms in several parts of the Northwestern Atlantic basin (Shepherd et al. 2007; Vosper et al. 2020), this does not seem to be necessarily the case for the SAI.

3.4 Future projected extreme precipitation and temperature

Table 3 shows the future average change from the 1981–2010 reference period to 2071–2100 over the domain centered on SAI under four scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5) for the Multi-Model Ensemble (MME; see last column of Table 3). The number of pixels per model used for the statistics can be found in the Supplementary Table 2. The results reveal a consistent dry and warmer pattern over SAI. This is consistent with the likely projected Caribbean drying (Cooley et al. 2022; IPCC 2022), potentially explained by an enhancement of the Caribbean Low-Levet jet under global warming (Taylor et al. 2013) and an anomalous subsidence over the Caribbean Sea and northeastern South America (Herrera et al. 2020).

Table 3 Mean future changes in precipitation and temperature extreme indices relative to 1981–2010, averaged over San Andrés and Providencia for the 2071–2100 period under the SSP1-2.6, SSP2-4.5, SSP3-7.0, SSP5-8.5 scenario. Green, yellow, and red colors indicate wetting, drying, and warming changes, respectively. We highlight large inter-model agreements (> 80%) in bold
Full size table

For SSP5-8.5 warming scenario the consensus of signal that indicates a dry pattern is from 56 to 100% in precipitation indices (see Supplementary Table 4). And, for temperature indices, there is an agreement of 100% for all indices and models. The most robust and significant future extreme precipitation indices projections in both San Andrés and Providencia are a decrease in wet days (R1mm: -17% and -20% on average, respectively), in the amount of precipitation during wet days (PRCPTOT: -29%), and in consecutive wet days (CDD: -40% and -20%) by 2100 (Table 3). For extreme temperature indices, increases of + 3.4 °C for the maximum value of daily maximum temperature, + 89% in the number of hot days, and almost all days (> 99%) being hot days by 2100 (with a daily maximum temperature greater than the historical 90th percentile; the final value of WSDI is near 365 days). Other extreme precipitation indices as the maximum consecutive 1-day and 5-day precipitation or annual count of days above the 95th and 99th percentiles, are reduced (–9% to –19% across the Islands) but with less agreement across CMIP6 models (< 80%, see Table 3). Except for the high mitigation scenario SSP1-2.6, the other warming scenarios confirm a warmer-and-drier trend in extreme events in SAI, with a lower magnitude for lower emissions scenarios, in general. For instance, R10mm significantly decreases by 37% in the worst scenario, only by half in SSP3-7.0, and by less than one tenth of it in SSP2-4.5. In SSP1-2.6, trends in extreme rainfall are not robust across models. Following this latter scenario would reduce by 20% the number of extreme hot days in SAI compared to SSP3-7.0 or by almost 30% compared to SSP5-8.5.

As far as we know, the only study that provides future climatic projections is from the official IDEAM, a government agency of the Ministry of Environment and Sustainable Development of Colombia (IDEAM 2015, 2017). This is grey literature, only based on mean climate, using a downscaling methodology with a multi-CMIP5 model ensemble and with a scenario equivalent to the “averaged RCP scenarios” (i.e. the arithmetic mean of the four simulations from RCP2.6, RCP4.5, RCP6.0 and RCP8.5) which poses considerable problems (Arias et al. 2022) from the scientific, socio-economic and adaptation point-of-views, among others. Under the worst business-as-usual emission scenario (SSP5-8.5), our multi-CMIP6 model mean precipitation project decreases by -3%, -11% and -26% in the archipelago for the 2011–2040, 2041–2070 and 2071–2100 periods, respectively (compared to -30%, -33% and -33% for the same time periods under the “averaged” RCP scenario, see IDEAM 2015, 2017). Thus, our study improves the current mean and extreme hydroclimatic knowledge in the highly vulnerable department of SAI. Note that we tested the strong robustness of our projections on a gridpoint level instead of a large domain, and the differences are very low (not shown).

4 Conclusions

For the first time, we comprehensively assess mean and extreme climate over historical and future periods of the Archipelago of San Andrés, Providencia, and Santa Catalina, small Caribbean islands highly vulnerable to climate change.

First, analyzing historical trends in duration, frequency, and intensity of extreme precipitation and temperature, we found that heatwaves have more than tripled in frequency and doubled their maximum duration since the end of the 80’s. Extreme rainfall events significantly increased in frequency and intensity in most stations but opposite and heterogeneous extreme rainfall trends are also found across a small territory (< 30 km2). This proves again that small islands need ad-hoc downscaling from global and regional climate simulations assimilating local meteorological data. San Andrés, Providencia, and Santa Catalina benefit from data from a reasonable number (7) of meteorological stations. However, data gaps (e.g., since 2017, no maximum temperature in SAI has been published) and potential local unknown biases should be solved in the near future to provide complete homogenized meteorological time series, potentially implying remote sensing data.

Second, our multi-model analysis showed that extremely hotter-and-drier trends mark the future in SAI. Our results improve the incorrect official climatic projections from (IDEAM 2017; Arias et al. 2022): under a future business-as-usual emission scenario, our multi-model mean precipitation projection decreases by -3%, -11% and -26% in the archipelago for 2011–2040, 2041–2070 and 2071–2100, respectively (compared to -30%, -33%, -33% for the same periods under an “averaged” RCP scenario, see IDEAM 2015, 2017). Those official projections should be urgently reassessed with the most updated climate models and downscaling techniques to prevent incorrect scenario averaging and inadequate downscaling results (Arias et al. 2022). Colombia's national communication (CNC) reports have shown remarkable progress regarding the clarity and methodological detail employed, and the comprehensiveness of the different socio-environmental aspects addressed. We invite the institutional leaders of the Fourth CNC of Colombia to consider the participation of a greater number of scientists from the country, including SAI, who represent research groups with experience in the use and implementation of climate models and their simulations, as well as in risk analysis and decision-making processes, so that they can contribute both in the development of methodologies and in the obtaining and analysis of results (Arias et al. 2022).

Third, although hurricanes Categories IV and V near SAI (< 600 km) more than doubled and hurricanes and tropical storms doubled since the’60 s, in line with global warming, only a small proportion of rainfall extremes (average daily rainfall > 99th percentile) are associated with ENSO events or hurricanes, even the heaviest ones. However, it is important to implement preventive strategies to reduce the vulnerability of the inhabitants of SAI to the increase in the number and intensity of heavy hurricanes, such as climate-adaptation investment plans, infrastructure resilience, education strategies on climate change and implementation of Nature-Based Solutions based on ecosystem services, among others (McConney et al. 2015).

Hydroclimatological research and monitoring (Quesada and Avila-Diaz 2023), bias correction (Romero-Hernández et al. 2024) and high-resolution modeling (Manciu et al. 2023) are urgently required in Colombia to strengthen societal resilience to climate change. Finally, the existence of the Seaflower marine protected area, hosted by the SAI archipelago, reinforces even more the need for accurate hydroclimatological data to be able to reduce the loss in distribution, functioning, and interactions of terrestrial and marine ecosystems, which are increasingly under threat of climate change (Pörtner et al. 2021; Pereira et al. 2024).