Trends in Annual, Seasonal, and Daily Temperature and Its Relation to Climate Change in Puerto Rico

Abstract

Puerto Rico has experienced recent increases in annual, seasonal and daily temperatures that have been associated with climate change. More recently, the island has been experiencing an increase in the frequency of extremely warm days that are causing significant environmental and socio-economic impacts. This study focuses on examining how annual, seasonal and daily temperatures have changed over recent decades in 12 historical sites spread across the island for the 1970–2024 period and how it relates to climate change. The Mann–Kendall tests for trends were employed for the annual and seasonal series to identify areas of the island where warming has been found to be statistically significant. The 90th, 95th, and 99th percentiles of daily temperature series were also analyzed. This study found that Puerto Rico has experienced significant warming from 1970 to 2024, with the most consistent increases in minimum temperatures, especially during the summer and nighttime hours. The frequency of extreme heat events has increased across nearly all stations in different areas of the island. Stepwise regression models identified surface air temperature (SAT), sea surface temperature (SST), and total precipitable water (TPW) as the most influential regional climate predictors driving mean temperature trends and the occurrence of extreme heat events.

1. Introduction

Climate change has led to an increase in air temperatures and a higher frequency of extreme heat events worldwide, including in tropical regions like the Caribbean [1]. Puerto Rico, a small tropical island located in the eastern Caribbean, has not been exempt from these warming trends [1,2]. Rising temperatures have been observed on the island over the past decades [3], with 2024 breaking records for the highest number of consecutive extremely hot days in the metropolitan area of San Juan. Understanding Puerto Rico’s temperature trends at the annual, seasonal, and daily scales are crucial for the island’s adaptation and resilience planning, as warming can impact water resources, ecosystems, agriculture, and public health in ways that vary between the island’s urban and rural areas.

Understanding temperature trends is critical for addressing the far-reaching consequences of climate change, particularly in islands like Puerto Rico that are highly vulnerable to extreme events that can cause significant environmental, social and economic disruptions. Increasing temperatures have significant ecological implications, like prolonged heat waves that can worsen drought conditions, disrupt ecosystems, and threaten biodiversity [2]. From a socio-economic standpoint, rising temperatures can negatively impact agriculture, increase energy demands for cooling, and pose severe health risks, especially for vulnerable populations such as the elderly and low-income communities. The urgency of studying temperature trends in Puerto Rico is underscored by its geographical location in the tropics, where the impacts of a warmer planet are expected to be significantly pronounced.

Scientific research on temperature trends in Puerto Rico has steadily grown over the past few decades, providing valuable insights into the island’s changing climate. Some studies have documented increasing trends in maximum and minimum temperatures [4], especially in the metropolitan area of San Juan [5]. The work has pointed to urban heat island effects in urban areas, demonstrating how land-use changes can exacerbate warming trends. Such research has not only identified historical changes but has also spurred further inquiries into the drivers of these trends and their local manifestations. However, most of the work regarding temperature trends on the island has focused on the sites located in the metropolitan area of San Juan, with little attention given to other sites located in rural areas.

The relationship between temperature, land cover, and climate change in Puerto Rico provides a critical lens for understanding local warming patterns. Changes in land cover, such as deforestation, urbanization, and agricultural expansion, can amplify temperature increases by altering the land surface’s energy balance and reducing natural cooling processes. Urban areas, characterized by high concentrations of impervious surfaces, exhibit pronounced heat island effects, where localized temperatures are significantly higher than surrounding rural areas [4,6]. Additionally, deforestation in Puerto Rico’s mountainous regions not only contributes to biodiversity loss but also disrupts carbon sequestration and evapotranspiration processes, further exacerbating temperature rises [7,8]. These land cover changes, combined with broader climate change phenomena, underscore the need for integrated strategies to manage land use and mitigate warming.

Recent studies have extended this foundational research, focusing on the implications of future warming scenarios. For instance, a recent study utilized high-resolution climate models to project temperature increases under various greenhouse gas emission scenarios, predicting a significant rise in both mean and extreme temperatures by mid-century [9]. These findings emphasize the potential for increased heat stress and its detrimental impacts on public health and ecosystems. Moreover, advances in regional climate modeling have enabled more spatially detailed projections, helping researchers understand how warming may differentially affect urban areas, coastal zones, and mountainous regions across Puerto Rico [10]. Such information is critical for adaptation planning, infrastructure design, and public health preparedness.

Multiple empirical studies have supported these concerns. One of the studies found that the island’s average temperature rose by about 1.24 °C (2.24 °F) between 1950 and 2014 [11], a substantial warming trend. NOAA also reported that Puerto Rico’s average annual temperature has increased by more than 0.8 °C (1.5 °F) since 1950 [12]. These estimates vary slightly depending on methodology and location, but the trend is consistent. Another study observed a 1.17 °C (2.1 °F) increase in San Juan from 1956 to 1983 [13], suggesting that warming began early in the observational record and may be magnified in highly urbanized areas. Other studies show that the increasing trend is often greater in daily minimum temperatures [14,15], indicating that nighttime warming is outpacing daytime heating, a pattern consistent with other tropical regions.

Temperature extremes have also become more frequent, with studies documenting record-breaking heat events in San Juan during the summer of 2012 [16], while others noted a steady increase in the number of hot days and a decrease in cold nights [11]. These changes can have serious implications for public health and critical infrastructure. Concurrently, another study [17] found that urban areas have warmed significantly faster than nearby rural zones, with heat island effects contributing up to 2 °F of additional warming in some cases. These studies indicate that temperature increases in Puerto Rico are neither uniform nor isolated to a single factor; rather, they reflect an interplay of global climate dynamics, local land use, and socio-economic development.

The current state of knowledge reveals a clear and concerning pattern of temperature increase in Puerto Rico. While previous studies have demonstrated statistically significant warming trends, particularly in the metropolitan area of San Juan, there has been limited analysis of how these trends manifest across different regions of the island and at multiple temporal scales. In particular, the frequency of extreme heat events has not been thoroughly explored outside urban centers, and most existing research has relied on modeled data or focused on annual averages alone. This study addresses these gaps by using over five decades of quality-controlled, station-based temperature data from 12 long-term meteorological sites distributed across diverse urban, rural, coastal, and highland environments. By examining temperature trends at the annual, seasonal, and daily levels, including extremes defined by the 90th, 95th, and 99th percentiles, this research offers a uniquely detailed and spatially explicit analysis of Puerto Rico’s warming patterns over the last 5 decades.

2. Data and Methods

2.1. Data Sources and Processing

This study analyses historical air temperature data from twelve meteorological stations distributed across Puerto Rico for the 1970–2024 period (

Table 1. Stations with daily temperature data for the 1970–2024 period.

Official Station Name	Short Name	Lat	Lon	Elev (M)
Adjuntas Substation, US	Adjuntas	18.17	−66.80	557.8
Aguirre, US	Aguirre	17.96	−66.22	7.6
Arecibo Observatory, US	Arecibo Ob	18.35	−66.75	323.1
Coloso, US	Coloso	18.38	−67.16	12.2
Dos Bocas, US	Dos Bocas	18.34	−66.67	61
Isabela Substation, US	Isabela	18.47	−67.05	128
Juncos 1 SE, US	Juncos	18.23	−65.91	64.9
Lajas Substation, US	Lajas	18.03	−67.07	27.4
Manati 2 E, US	Manati	18.43	−66.47	76.2
Ponce 4 E, US	Ponce	18.03	−66.53	21.3
Roosevelt Roads, US	Roosevelt Rds	18.26	−65.64	10.1
San Juan LM Marin International Airport, US	San Juan	18.43	−66.01	3

), including both urban and rural settings and a range of elevations and climatic zones (Figure 1). The stations selected were those with at least 40 years of records and 85% or more of data availability of daily minimum (TMIN), mean (TAVG), and maximum (TMAX) temperature data. All data were obtained from the National Oceanic and Atmospheric Administration (NOAA) Global Historical Climatology Network (GHCN) and the National Centers for Environmental Information (NCEI) archives. Rigorous quality control procedures were implemented to ensure consistency and remove erroneous or incomplete records prior to analysis.

Monthly averages were calculated from the daily data, and annual mean temperatures were derived for each site and each temperature metric (TMIN, TAVG, TMAX). In addition, seasonal mean temperatures were computed for four climatological seasons: the colder and drier months of December–February (DJF) and March–May (MAM), and the warmer and wetter months of June–August (JJA) and September–November (SON). This approach allows for the assessment of intra-annual variability in temperature trends and the identification of seasonally specific warming signals.

2.2. Trend Analysis Using the Mann–Kendall Test

To evaluate the presence and significance of temporal trends in annual and seasonal temperatures, the Mann–Kendall (MK) test was applied to each site and each temperature metric. The Mann–Kendall test is a non-parametric, rank-based statistical method commonly used in climatological and hydrological studies to assess monotonic trends in time series data without requiring the data to follow a specific distribution [18,19]. It is particularly robust for climate datasets, as it is insensitive to outliers, missing values, and non-normal distributions, which are common in observational environmental records.

The MK test computes the Tau (τ) coefficient, which indicates the direction and strength of the trend. Positive τ values indicate upward trends (warming), while negative values denote cooling. The statistical significance of each trend is evaluated using a p-value threshold of 0.05 calculated from a two-sided statistical test. For each of the three-temperature metrics (TMIN, TAVG, and TMAX), MK tests were performed for the annual and four seasonal means across all 12 stations. The resulting Tau coefficients were spatially mapped using Geographic Information Systems (GIS) to illustrate the geographic distribution of warming or cooling trends across the island for each season and the full year.

2.3. Analysis of Daily Extreme Temperatures

To further investigate changes in extreme temperature events, we analyzed the distribution of daily TMIN, TAVG, and TMAX values for each site. For each temperature metric, the 90th, 95th, and 99th percentiles were calculated from the historical daily series at each station. These percentile thresholds represent the warmest 10%, 5%, and 1% of daily temperature values, respectively, and are commonly used to assess the frequency and intensity of extreme warm days in climate studies [20,21].

Using these thresholds, we calculated the annual count of days exceeding the 90th, 95th, and 99th percentile values for each station and temperature metric. This method allows for the detection of relative changes in extremes over time, regardless of differences in baseline climate conditions across stations. One of the main advantages of this percentile-based approach is that it normalizes extreme thresholds to the local climate, enabling meaningful comparisons between stations situated in different thermal regimes (e.g., high-elevation vs. coastal urban areas). Moreover, percentiles are less sensitive to arbitrary thresholds and better reflect changes in the tails of the temperature distribution, which are often more sensitive to climate change.

2.4. Trend Detection in Extreme Temperature Frequencies

To assess whether the frequency of extremely warm days has changed over time, the Mann–Kendall test was again employed for each of the annual time series representing the number of days exceeding the 90th, 95th, and 99th percentile thresholds for each temperature metric. This step enables the identification of statistically significant trends in the occurrence of extreme heat events across Puerto Rico. Tau coefficients were computed for each station and percentile threshold and were mapped spatially to visualize how the frequency of extreme temperature events has evolved over time and how these patterns vary geographically.

Following the Mann–Kendall trend analysis, Sen’s slope estimates were applied to determine the magnitude of temperature changes on both annual and seasonal scales. Sen’s method provides a robust, non-parametric estimate of the median rate of change, reducing the influence of outliers or non-normal data distributions [22]. The slopes represent the rate of temperature increase or decrease per year, offering a direct measure of the intensity of the observed trends. This approach allows for a clearer interpretation of long-term temperature changes across different time periods and regions.

Together, these methods provide a comprehensive analysis of long-term warming trends across Puerto Rico, examining not only the mean temperature at annual and seasonal scales but also changes in the frequency and intensity of extreme warm events, which are critical for understanding potential climate impacts on human health, infrastructure, ecosystems, and water resources.

2.5. Forward Stepwise Regression Procedures

In this analysis, we employed a forward stepwise regression procedure to identify the best climate predictors for TAVG, TMAX, and TMIN for each of the 12 meteorological stations in Puerto Rico. Forward stepwise regression is a multiple-step variable selection method that begins with no predictors in the model [23]. At each step, the predictor that produces the largest improvement in model performance, measured by the Adjusted R-squared, is added [24]. This procedure continues one independent variable at a time, with the model evaluating all remaining predictors at each iteration until no additional variable significantly improves the model’s explanatory power [25]. This technique is especially useful for identifying a subset of variables that best explain the variance in a dependent variable from a larger pool of potential predictors [23,24,25]. This method has been used to analyze trends in rainfall and tropical cyclone frequency in Puerto Rico and its relation to climatic factors [26,27].

To implement this analysis, annual data for six climate predictors covering the period from 1970 to 2024 was compiled for climatic factors such as surface air temperature (SAT), sea level pressure (SLP), sea surface temperature (SST), total precipitable water (TPW), carbon dioxide concentration (CO₂), and cloud cover (CCV). All predictors, except for CO₂, were obtained from the ECMWF ERA5 Reanalysis [28] as annual and seasonal averages for the grid cells spanning 16–20° N latitude and 64 to −70° W longitude, a region that captures the oceanic-atmospheric conditions influencing Puerto Rico and the northeastern Caribbean [29]. The CO₂ data was obtained from the Mauna Loa Observatory and represents global average annual concentrations of atmospheric carbon dioxide [30]. These predictors were merged with the annual temperature records from each site to generate a consistent and comparable dataset for regression modeling.

The forward stepwise regression procedure was applied not only to annual temperature values but also to seasonal averages corresponding to the DJF, MAM, and SON periods. This allowed for the identification of seasonal drivers of temperature variability in addition to long-term annual trends. For each site, season, and temperature metric, the process produced a best-fit model identifying the most important predictors and reporting the final Adjusted R-squared, a metric that accounts for model complexity and indicates the proportion of variance explained by the independent variables. It is important to note that the adjusted R-squared may overestimate model skill due to selection bias in choosing the best predictors among six candidates [31]. These results provide insight into the relative influence of regional oceanic, atmospheric, and radiative factors on seasonal temperature trends across Puerto Rico.

3. Results & Discussion

3.1. Annual Temperature Trends

The MK results for annual temperature trends across Puerto Rico show consistent warming trends, particularly in average and minimum temperatures. Stations such as Aguirre, Coloso, Juncos, and Isabela (Figure 2a) exhibit some of the strongest positive trends in annual average temperature (TAVG), with Kendall’s Tau coefficients exceeding 0.5. These trends suggest a notable intensification of heat across the northwestern and southeastern regions of the island (Figure 2b,c). It is important to note that the northwest region has also been found to be experiencing a drying trend when compared to other areas of the island [26]. In contrast, coastal sites like Roosevelt Roads and San Juan, located in the northeast, show more moderate trends, while locations farther inland, like Arecibo Ob, display relatively weak or variable trends across temperature metrics (Figure 2d,e). Notably, TMIN shows stronger and more widespread positive trends (Figure 2d) than TMAX (Figure 2e), pointing to warmer nights being a key driver of overall warming.

The analysis of annual mean temperature change between the 1970–1997 and 1998–2024 periods across Puerto Rico reveals a consistent warming trend, with the most pronounced mean temperature changes and Sen’s slope (Figure 3) observed in minimum temperatures (Figure 3). Stations like Coloso experienced a 2.06 °C (3.71 °F), Isabela a 1.89 °C (3.40 °F), Aguirre a 1.83 °C (3.29 °F) and Juncos a 1.89 °C (2.90 °F) increase in TMIN, indicating a significant rise in nighttime temperatures (Figure 3). The top five sites in terms of mean annual temperature change between the two periods exhibited Sen’s slope estimates (Figure 3) ranging from 0.02–0.06 °C (0.04–0.11 °F). TMAX was found to have the lowest increase among most sites, except for the station of Ponce in the south, which had a mean increase of 0.089 °C (0.16 °F), a Sen’s slope close to 0.04 °C/year. (0.07 °F) (Figure 3). These results suggest that significant warming trends are stronger outside of the metropolitan region of San Juan, even with a very pronounced and well-documented urban heat island [1,16,17].

3.2. Seasonal Temperature Trends

The seasonal temperature trends for DJF exhibit consistent warming trends across Puerto Rico, especially in TAVG and TMIN temperatures (Figure 4a,b). For the DJF season, TMAX shows weaker trends, with only five sites showing significant increases (Figure 4c). Stations such as Juncos, Isabela, and Aguirre showed strong positive trends in both DJF and MAM seasons, with Kendall’s Tau values indicating a steady increase over time (Figure 4a,d,e). The eastern and northwestern regions show the most robust warming signals, especially in TMIN (Figure 4e), suggesting that overnight minimums have increased significantly during the cooler months. While TMAX increased more modestly, some sites like Ponce and Juncos still showed upward trends (Figure 2c and Figure 4c).

During the DJF and MAM seasons, nearly all stations across Puerto Rico exhibited significant warming in both TAVG and TMIN (Figure 5), with many locations showing annual mean changes of 1.5 °C (2.7 °F) or higher and yearly Sen’s slope increases above 0.05 °C (0.09 °F) for the 1970–1997 and 1998–2024 periods. Particularly notable were northwestern stations like Coloso and Arecibo, where TMIN rose by 2 °C (3.6 °F) in DJF and 2.56 °C (4.6 °F) in DJF and MAM (Figure 5a,b). In the eastern region sites like Aguirre and Juncos, TMIN changes went over 1.5 °C (2.7 °F) in both seasons (Figure 5a,b). This sharp rise in minimum temperatures reflects a substantial reduction in cooler nighttime conditions during the colder and drier months of the year in more rural areas of the island in both the west and east.

During JJA (June–August) and SON (September–November), the warming trends continue, though the regional distribution and intensity of change differ slightly (Figure 6). All sites exhibited significantly increasing trends for TAVG and TMIN for the JJA period, with sites like Aguirre, Coloso, and Isabela showing stronger warming trends in both (Figure 6a,b). Most sites also exhibited increasing trends in TMAX for JJA, yet the Tau coefficients were overall weaker than TAVG and TMIN (Figure 6c). SON trends show all sites with significant warming, with sites like Aguirre, Juncos, Isabela, and Coloso standing out for their strong increase (Figure 6d,e). However, TMAX in the SON period only had four sites with significant warming trends (Figure 6f). Overall, these findings suggest that summer and fall seasons are also warming, but the signal is most consistent in nighttime temperatures, reinforcing the broader pattern of asymmetric warming across Puerto Rico.

In the JJA and SON seasons, the mean temperature increase between the 1970–1997 and 1998–2024 periods shows various sites across different regions of the island with increases of 1.5 °C (2.7 °F) or more in TMIN (Figure 7a,b). Stations such as Coloso and Arecibo Ob, located in the west, experienced increases of over 2 °C (3.6 °F) in TMIN during the JJA period, with Sen’s slopes of 0.08 °C (0.14 °F) and 0.05 °C (0.09 °F) correspondingly (Figure 7a). Notably, SON featured the highest increase in TMIN, with Coloso showing a 2.67 °C (4.8 °F) rise, highlighting a strong warming trend in late wet-season months. These results indicate that nighttime warming dominates Puerto Rico’s seasonal temperature trends (TMIN and TAVG), particularly during the warmer half of the year, with potential implications for energy use, heat stress, and public health during summer and early fall.

3.3. Daily Temperature Events Trends

At the 90th percentile, all sites show statistically significant positive trends for TAVG and TMIN, suggesting a clear increase in the frequency of the top 10% of extremely warm days across the island (Figure 8a,b). For TMAX at the 90th percentile, seven sites located in the western and southern regions exhibited significant warming trends, yet with overall lower Tau coefficients (Figure 8c). Stations located in different regions of the island, such as Juncos (TAVG), Lajas (TAVG), Aguirre (TAVG), Ponce (TMAX), and Arecibo Ob (TMIN), show notable increments in the mean of extreme heat days ranging from 42–47 (Figure 8d). Those four sites also exhibited Sen’s slopes ranging from 0.9 to 1.63, showing that significant warming is occurring across different regions of the island and outside the metropolitan area of San Juan.

In the 95th percentile group, similar trends persist, with most sites showing significant warming trends in TAVG, TMIN, and TMAX (Figure 9). While TMIN and TAVG continue to show significant warming across different regions of the island (Figure 9a,b), TMAX shows significant trends in the western and southern sites (Figure 9c). At the 95th percentile level, the mean of extremely warm temperature events shows notable increases between the 1970–1997 and 1998–2024 (Figure 9d) period in sites like Aguirre (TMAX and TMIN) with 36, Juncos with 33 (TAVG), Lajas (TAVG) with 22 and San Juan (TMIN) with 21. It is important to note that even though San Juan experienced significant increases in TAVG and TMIN, sites like Aguirre and Juncos showed higher differences in mean events and larger Sen’s slopes ranging from 0.7–1 per year. Other studies have documented significant warming in the metropolitan area of San Juan [14,16,17], yet the findings of this study suggest that sites outside of the main urban area are experiencing a more significant increase in the number of extreme heat events at or above the top 5%.

At the 99th percentile, which reflects the top 1% most intense temperature extremes, significant warming trends continue to emerge across different regions of the island (Figure 10). Some stations, such as Aguirre, Isabela, and Arecibo Ob, maintain strong positive trends (Tau > 0.35) in TMIN 99th and TAVG 99th (Figure 10a,b), reinforcing the observation that even the most extreme warm nights are becoming more frequent in coastal and lowland areas. The TMAX trends at the 99th percentile are strong (Tau > 0.4) in southern sites like Ponce, Aguirre, and Lajas (Figure 10c). The sites with the most notable increase in the mean of extreme heat events between the 1970–1997 and 1998–2024 periods are Aguirre (TMAX) with 11 and Ponce (TMAX) with 9 (Figure 10d). These findings highlight that the increase in TMAX 99th days is occurring in the southern coastal plains and the eastern region of the island. San Juan was the site with the highest increase in the mean of extreme heat events, with an increment of 9 between the periods and a Sen’s slope of 0.132 for TMIN 99th. Overall, the results of the daily temperature extremes at the three different thresholds show that significant warming is more pronounced in TMIN and TAVG across different regions of the island, while TMAX increases are concentrated in the southern region of the island.

3.4. Forward Stepwise Regression Models

Before exploring the results of the forward stepwise regression models, it is important to examine the correlations among the predictors used to explain TAVG, TMIN and TMAX trends across various temporal scales and sites. Strong positive correlations (dark red) are evident between SAT, SST, and TPW, especially in annual (Figure 11a) and all seasons (Figure 11b–e), indicating these variables tend to vary together. Conversely, the SLP exhibits negative correlations with SAT and SST (Figure 11b–e), reflecting the inverse relationship often observed between pressure systems and surface temperatures. CO₂ concentrations also show significant correlations with SAT and SST in most seasons, reflecting the long-term influence of greenhouse gas increases on warming trends (Figure 11a–e). It is important to note that even when predictors have strong correlations, the forward stepwise approach reduces the likelihood of including redundant independent variables by evaluating each factor’s unique contribution to model improvement. Thus, even though some predictors are strongly correlated, only those that meaningfully enhance the model’s explanatory power were retained in the models in this section.

The results of the annual forward stepwise regression reveal consistent patterns in the predictors that best explain temperature variability across Puerto Rico’s weather stations (

Table 2. Top Forward Stepwise Regression Model Procedures for Annual TAVG, TMIN, and TMAX by site, and highest Adjusted R².

Sites	Target	Best Predictors	Adj. R2
Isabela	TMIN	CO2, SST, TPW	0.80
Coloso	TMIN	CO2, CCV, SST	0.79
Juncos	TAVG	SAT, CO2, SST, SLP, CCV	0.72
Isabela	TAVG	SAT, SLP, SST, CO2	0.68
Ponce	TMAX	CO2, SST, CCV	0.66
Aguirre	TMIN	CO2, SST, TPW, SLP, SAT	0.62
Juncos	TMIN	CO2, SLP	0.61
Lajas	TMAX	SAT, CCV, CO2	0.57
Arecibo Ob	TMAX	CO2, SAT, SST, CCV	0.57
Lajas	TAVG	SST, CO2	0.55
Dos Bocas	TMIN	SAT, CO2, SLP	0.54
Aguirre	TAVG	CO2, SAT, SLP, SST	0.53
Coloso	TAVG	CO2, CCV	0.51
San Juan	TAVG	SAT, CO2, CCV, TPW	0.48
San Juan	TMIN	SAT	0.40
Dos Bocas	TAVG	SAT	0.44
Roosevelt Rds	TMIN	CO2, SLP, SAT	0.38
Arecibo Ob	TMIN	SST, SAT, CO2	0.37
Manati	TMAX	CO2, SAT, SLP, SST	0.37

). For TAVG, the most frequently selected predictors were SST, SAT, and CO₂, which appeared in multiple models and indicated their strong influence on mean temperatures annually. For TMIN, SST emerged as a key predictor, often selected alone or in combination with TPW, highlighting the role that ocean warmth and atmospheric moisture can have on nighttime temperatures (Table 2). In the case of TMAX, SAT and CO₂ were more commonly selected, suggesting that land-based warming and greenhouse gas concentrations have a stronger association with daytime high temperatures (Table 2). Overall, SST, SAT, and CO₂ stand out as the most influential predictors across the island, though their importance varies slightly depending on whether annual average, minimum, or maximum temperatures are being modeled.

The seasonal forward stepwise regression results show that the influence of climate predictors on temperature patterns in Puerto Rico varies by season (

Table 3. Top Stepwise Forward Regression Model Procedures for the periods DJF and MAM by site, and highest Adjusted R2.

Site	Target	Best Predictors	Adjusted R²
Selected DJF Model Results
Isabela	TMIN	CO2, SST	0.78
Coloso	TMIN	CO2, SLP	0.75
Juncos	TAVG	CO2, SAT, SST, SLP	0.68
Juncos	TMIN	CO2, SAT, SST	0.66
Ponce	TMAX	CO2, SAT, SLP	0.65
Isabela	TAVG	CO2, SAT	0.63
Aguirre	TMIN	CO2, TPW	0.59
Dos Bocas	TMIN	CO2, SAT, SST, CCV	0.58
Aguirre	TAVG	CO2, TPW	0.55
Coloso	TAVG	CO2, TPW	0.49
Selected MAM Model Results
Coloso	TMIN	CO2, TPW, SST	0.78
Isabela	TMIN	CO2, SST, TPW, CCV	0.78
Juncos	TAVG	CO2, SAT, SLP, SST	0.69
Ponce	TMAX	CO2, SST	0.65
Isabela	TAVG	CO2, SAT, TPW	0.65
Juncos	TMIN	CO2, SLP, CCV	0.64
Aguirre	TMIN	CO2, SAT, SLP, SST, TPW	0.60
Aguirre	TAVG	CO2, SAT, SLP	0.55
Lajas	TMAX	CO2, SAT, SST, SLP, TPW	0.55
Coloso	TAVG	CO2, TPW, SAT, SST	0.54

), with distinct drivers emerging during the drier and colder months (DJF and MAM). For these seasons, SAT and SLP frequently appear as significant predictors, particularly for TAVG and TMAX, suggesting that broader regional temperature patterns and atmospheric pressure exert a strong influence on Puerto Rico’s climate during the cooler and drier months of the year (Table 3). CO₂ also emerges more prominently during these seasons, especially in TMAX stepwise models, indicating a possible link between increasing greenhouse gas concentrations and enhanced daytime heating during dry periods.

During the warmer and wetter months (JJA and SON), the models highlight a stronger role for SAT, SST, TPW and CCV, especially in TAVG and TMIN predictions (

Table 4. Top Stepwise Forward Regression Model Procedures for JJA and SON by site, and highest Adjusted R2.

Site	Target	Best Predictors	Adjusted R²
Selected JJA Model Results
Coloso	TMIN	CO2, TPW	0.77
Isabela	TMIN	CO2, SST	0.77
Ponce	TMAX	CO2, TPW	0.65
Juncos	TAVG	CO2, SAT	0.63
Isabela	TAVG	CO2, SAT, SLP	0.61
Juncos	TMIN	CO2, TPW	0.60
Arecibo Ob	TMAX	CO2, SAT, SST, SLP	0.57
Aguirre	TMIN	CO2, SAT, TPW, SST	0.55
Dos Bocas	TMIN	CO2, SST	0.51
Coloso	TAVG	CO2, CCV	0.50
Selected SON Model Results
Isabela	TMIN	CO2, SST, SAT, SLP	0.79
Coloso	TMIN	CO2, SST, SAT, TPW, CCV	0.79
Isabela	TAVG	CO2, SST, SLP, SAT	0.66
Ponce	TMAX	CO2, SST, SAT	0.62
Juncos	TAVG	CO2, SAT, CCV	0.62
Juncos	TMIN	CO2, TPW	0.61
Aguirre	TMIN	CO2, SST, SLP, SAT	0.57
Dos Bocas	TMIN	CO2, TPW, SLP, SST	0.56
Lajas	TAVG	SST, SAT	0.54
Lajas	TMAX	SAT, CO2, CCV	0.52

). These variables might capture the enhanced ocean-atmosphere interactions in the waters around the island, the moisture availability, and the convective activity that characterizes the wetter and warmer months in the Caribbean [26,29]. TPW and CCV, in particular, show increased relevance in TMIN forward stepwise regression models (Table 4), suggesting that warm, moist air and frequent cloud cover help retain heat during nighttime hours. Overall, while SAT and CO₂ are more dominant in the dry and cool seasons, increasing SST’s, TPW, and CCV show strong correlations with temperature trends during Puerto Rico’s wetter and hotter months.

The forward stepwise regression models for the total number of annual extreme temperature events at the 90th, 95th, and 99th percentiles of TMAX, TMIN, and TAVG show several consistent patterns across stations and percentile thresholds (

Table 5. Top Stepwise Forward Regression Model Procedures for the 90th, 95th and 99th percentiles by site.

Station	Metric	Best Predictors	Adjusted R²
90th Percentile Results
Juncos	TMIN	SAT, CO2, SLP, SST, CCV	0.72
Dos Bocas	TMIN	SST, CO2, SLP, SAT	0.69
Juncos	TAVG	SAT, CO2, SLP, SST	0.69
Aguirre	TAVG	SAT, SLP, SST, CCV	0.68
Dos Bocas	TAVG	SAT, SST	0.58
Aguirre	TMIN	SAT	0.56
Arecibo Ob	TAVG	SAT	0.56
Aguirre	TMAX	CO2, SAT, SST	0.55
Ponce	TMAX	SAT, CO2, SLP, SST	0.55
Arecibo Ob	TMIN	SST, SLP, CO2, TPW, CCV	0.49
95th Percentile Results
Juncos	TAVG	SAT, SLP, SST, CO2, CCV, TPW	0.67
Juncos	TMIN	SAT, SLP, SST, CO2, CCV	0.64
Aguirre	TMAX	CO2, SLP, SAT, TPW	0.63
Aguirre	TAVG	SAT, SLP, SST, CCV	0.61
Dos Bocas	TMIN	SST, SLP, CO2, SAT	0.60
Arecibo Ob	TAVG	SAT, TPW, CO2	0.56
Aguirre	TMIN	SAT	0.50
Arecibo Ob	TMIN	SST, SLP, CO2, TPW, CCV	0.49
Adjuntas	TMAX	SST, SLP, CCV, CO2	0.49
Ponce	TMAX	CO2, SAT, SST, CCV	0.47
99th Percentile Results
Juncos	TAVG	SAT, SLP, SST, CCV, TPW, CO2	0.56
Aguirre	TMAX	CO2, SLP, TPW	0.55
Ponce	TMAX	CO2, SLP, SAT, SST	0.49
Adjuntas	TMAX	SAT, SLP, SST, TPW	0.46
Juncos	TMIN	SAT, SST, SLP, TPW	0.45
Arecibo Ob	TMIN	SST, TPW, SLP, CO2	0.45
Adjuntas	TMIN	SAT, CO2, CCV, TPW	0.39
Dos Bocas	TMIN	SAT, SLP, CCV	0.38
Ponce	TMIN	SAT, CO2, TPW	0.37
Arecibo	TAVG	CO2, SLP, TPW, SST	0.37

). Among the six tested climate predictors, SAT and TPW emerged as the most frequently selected variables across models, indicating their strong association with the frequency of extreme temperature days (Table 5). SAT was particularly dominant in TMAX-related models, reflecting its role in driving daytime temperature extremes, while TPW was often selected in TMIN and TAVG models, suggesting that elevated atmospheric moisture may help sustain elevated minimum and mean temperatures (Table 5). SST and SLP also contributed to several models, though with slightly less consistency. Interestingly, CO₂, despite its importance in long-term planetary warming, was less frequently selected in the extreme daily event models, yet this might be due to its global nature and more gradual temporal variability. Overall, the results suggest that atmospheric and oceanic factors like SAT and TPW are likely key drivers of the increasing trends in extreme temperature events in Puerto Rico.

4. Conclusions

This study focused on examining how annual, seasonal, and daily temperatures have changed in Puerto Rico across the 1970 to 2024 period, using data from 12 long-term meteorological stations distributed across the island. The results show clear and statistically significant warming trends across most stations, with minimum temperatures (TMIN) exhibiting the most consistent and pronounced increases. Warming trends were evident across all analyzed periods, particularly during the warmer and wetter months (JJA), with most sites showing significant increases in TAVG or TMIN. Daily extreme events also increased significantly, especially in the number of warm nights exceeding the 90th and 95th percentile thresholds. While TMAX increases were generally weaker and more variable, the number of extremely warm days at the 99th percentile still exhibited an upward trend in several coastal and southern stations. These findings suggest that Puerto Rico is warming more significantly in sites in the western and southern regions, outside the metro area of San Juan.

The forward stepwise regression analyses highlighted the climatic factors strongly associated with these warming trends across the island. For annual and seasonal temperature averages, SST, SAT and CO₂ emerged as the most frequently selected predictors across sites and metrics. During the colder and drier seasons (DJF and MAM), SAT and CO₂ were dominant, particularly for TMAX, while SST played a more consistent role in TAVG across all periods. In the wetter and warmer months (JJA and SON), TPW and CCV became more important factors, particularly for TMIN, suggesting that increased atmospheric moisture and cloud cover might help retain heat during nighttime hours. For the daily extreme temperature events, SAT and TPW were the most consistent predictors across percentile levels, reinforcing their strong association with trends in both average and extreme temperatures across Puerto Rico.

While this study offers a comprehensive analysis of long-term warming patterns in Puerto Rico, it is important to acknowledge several key limitations. The analysis relied on just 12 meteorological stations with consistent long-term data, which may not fully capture the island’s microclimatic and land cover variability. Additionally, while the ERA5 reanalysis predictors provide a useful spatial context, they may not fully resolve localized processes such as urban heat islands or land-use-driven warming that vary within smaller spatial scales. Moreover, the CO₂ data used in the models represent global averages and may not accurately reflect local atmospheric conditions. Future research could benefit from integrating finer-resolution land use, vegetation, and socio-environmental data to further understand localized drivers of warming, particularly in urban, coastal, and mountainous regions of Puerto Rico. Nonetheless, the study’s findings highlight critical warming trends that are already underway and reinforce the urgency for climate adaptation strategies tailored to the island’s diverse climatic and geographic contexts.