Inter-comparison of past and projected climatic trends in Puerto Rico: 1950–2100

Urbanization and climate change affect the Antilles Islands of the sub-tropical Caribbean via coastal erosion, air/land/water pollution, resource limitations and declining aesthetic value (Taylor et al. 2012a). Warming seas give off more humidity, leading to shorter winters and mosquito-borne diseases (Colón-González et al. 2021). Many of these trends could continue into the 21st century (Taylor et al. 2012b, 2018), requiring adaptation to a changing climate and development of local mitigating strategies. Most Antilles Islands have mountainous interiors, narrow coastal plains and high population density, so agricultural land-use competes with tourism and residential demand (Reyer et al. 2017).

Future projections of global warming make use of coupled general circulation models (GCMs) to quantify interactions between atmospheric composition, land–ocean feedback and thermodynamic–kinematic responses. Coupled Model Inter-comparison Project (CMIP) simulations tend to have a cool dry windy bias in the Caribbean (IPCC 2014; Ryu & Hayhoe 2014) and under-represent multi-year variability. Local circulations and their impacts may go undetected for small islands (IPCC 2021) due to coarse resolution (>100 km). Although many CMIP version 6 models have similar issues (Eyring et al. 2016), some offer improved coupling, finer resolution and realistic moist convection. Simulated mean patterns, annual cycles and trends can be compared with fine resolution (<50 km) reanalysis products that interpolate station and satellite data in the period 1950–2020 to represent small island climates.

The circulation of the sub-tropics around Puerto Rico (17.7°–18.7 °N, 67.5°–65.2 °W) is characterized by persistent trade winds and a shallow marine layer beneath subsidence from the Hadley Cell. Historical trends and projections indicate a widening of this meridional overturning atmospheric circulation (Lu et al. 2007; IPCC 2014; Jury 2018).

Here, key trends in rainfall, winds, sea temperature, sea level and related climatic elements over Puerto Rico are analyzed in the context of earlier studies (cf. Karmalkar et al. 2013). Revealing the uptake of global warming and its impacts on the local water budget are the goals of this work. Second-order trends in past and future eras are linked to regional kinematic and local thermodynamic forcing and recommendations for adaptation and mitigation are given. This island is densely populated in the central Caribbean, with a modern observing network and a size that is mid-way between the larger and smaller Antilles islands to the west and east. It is therefore amenable to simulation by high-resolution coupled GCM and comparison with data assimilation products.

Air temperature, winds, specific humidity, evaporation (surface moisture flux) and boundary layer (BL) height are analyzed from quality-controlled data assimilated into European Community Reanalysis v5 (ERA5) (Hersbach et al. 2020) at 25-km resolution. The surface water balance is estimated from Climate Research Unit of UEA v4 (CRU4) station-based interpolation of the Palmer Drought Severity Index (PDSI, Harris et al. 2014). Historical rainfall trends are evaluated via Climate Hazards InfraRed Precipitation with Station (CHIRPS) (Funk et al. 2014), Global Precipitation Climatology Centre v8 (GPCC8) (Schneider et al. 2018) and ERA5 reanalysis. The rainfall products are based on gauge network, satellite infrared and microwave radiance, and model physics and parameterization; interpolated to 5- to 25-km resolution. Observed and projected sea temperature, sea level and polar ice cover are studied using European Community Ocean Reanalysis v5 (ORA5) reanalysis (Zuo et al. 2019) and Hadley reanalysis & earth system MM model v3 (HAD3es) model simulations (Andrews et al. 2020). Dataset acronyms and attributes are listed in Table 1; products are selected that minimize discontinuities and offer the most realistic outcomes.

Regional circulation trends are analyzed via National Centre for Environmental Prediction (NCEP2) reanalysis (Kanamitsu et al. 2002) in N–S height section. Transpiration from island vegetation is estimated from satellite colour fraction (normalized difference vegetation index (NDVI), Pinzon & Tucker 2014) in the period 1980–2020. Evaluating all CMIP6 outputs determined that the CNRM6 high-resolution model (Voldoire et al. 2019) adequately simulates local rainfall over Puerto Rico in comparison with the Global Precipitation Measurement (GPM) satellite product (Hou et al. 2014). The CNRM6 temporal record over the 21st century using the ssp3-7.0 scenario (Gidden et al. 2019) is compared with extrapolated historical rainfall measurements (GPCC8).

Monthly time series are constructed from an all-island average (17.9–18.55 °N, 67.3–65.6 °W) and smoothed to remove periods <18 months (de-seasonalized). A first- or second-order regression is fitted, and the slope and r² ‘explained’ variance is determined. For past observations 1950–2020, 95% confidence level is reached with r² > 0.27 and ∼47 degrees of freedom. Historical trends are extrapolated into the 21st century and compared with CMIP6 model projections, to determine whether long-term change emerges from multi-year fluctuations.

To understand thermal orographic forcing of island rainfall, the case of 28 May 2015 makes use of daily National Weather Service (NWS) radar reflectivity, mid-day North American Mesoscale (NAM) surface winds and moisture flux (transpiration), and changes in ERA5 BL height. Given the prevailing trade winds, we analyze the area east of Puerto Rico (16°–20.5 °N, 68°–60 °W) for changes in SST from HAD reanalysis (Rayner et al. 2003) and marine BL height calculated from ERA5.

Information on modern data assimilation of in situ and satellite measurements via numerical models are given in Funk et al. (2014) and Hersbach et al. (2020). The choice of datasets and parameters is guided by the latest high-resolution technology, the need to describe both gains and losses in the local water balance and comprehensive exploratory analyses. Thus, our methods reduce to (i) gather optimal datasets, (ii) calculate an area-average, (iii) apply a filter to remove noise, (iv) apply a regression to obtain slope and fit and (v) compare historical trends with model projection.

The island of Puerto Rico in the central Antilles covers an area of 17.9°–18.6 °N, 67.5°–65.25 °W. The >$100 B economy declines 10–20% following drought or hurricane landfall, despite importing much of its food and financial resources to support 3.2 M people <data.worldbank.org/country/puerto-rico>.

ERA5 reanalysis rainfall trends 1950–2020 in Figure 1(a) are upward: 0.008 mm day⁻¹ year⁻¹ across the island and weaken to the southwest, consistent with Hall et al. (2012) and Karmalkar et al. (2013). A large diurnal cycle (Figure 1(b)) prevails over the island due to seabreeze convergence and transpirated moisture: ∼2/3 of rain falls between 14:00 and 17:00 h local time. 850 hPa vertical motion trends (Figure 1(c)) are upward/downward over northeastern/southwestern sectors of Puerto Rico, while 850 hPa humidity trends are upward uniformly (not shown). The smoothed time series of their product (q·W, Figure 1(d)) exhibits a linear upward trend. Past reanalysis and CNRM6 projected island-averaged rainfall trends (Figure 1(e)) are weak (r² = 0.09) and second-order: rising from 1950 to 2020 and falling thereafter, thus remaining ∼2.5 mm/day. Comparison of CNRM6 and GPM rainfall climatology (Figure 1(f) and 1(g)) shows orographic enhancement over Puerto Rico (>4 mm/day) and dry conditions across the ocean (<2 mm/day). The mean annual cycle of rainfall reveals a dry winter, May peak and lengthy wet season from July to November. The gauge product is wetter/drier than satellite in winter/summer, while the model does not capture the May peak. All products indicate a decrease/increase of rainfall in spring/autumn.

The pattern of evaporation (Figure 2(a)) is upwards of 0.008 mm day⁻¹ year⁻¹ over the western half of the island, a small value similar to rainfall that reflects local recycling. Upwind to the east of the island, the marine evaporation trend is weak and downward. The evaporation time series (Figure 2(b)) shows rising second-order trends (+1.3 mm/150 year, r²=0.97) amid weak multi-year variability consistent with the station-based PDSI. Model evaporation rates are ∼10% higher than estimates from reanalysis. The MODIS satellite vegetation climatology (Figure 2(c)) identifies low fraction around the coastal margins and cities, and high fraction within dense forest canopies that release diurnal transpiration at elevations from 300 to 900 m. The vegetation time series, illustrated in Figure 2(d), reflects raw monthly and smoothed inter-annual records. The slope is the same (+5% over 40 years) but the r² fit naturally improves from 0.06 to 0.31 due to filtering. The seasonal cycle follows 2–3 months behind rainfall with an April min/November max.

Evaluating the marine climate: the mean map of SST (Figure 2(e)) reflects a gradient from 27 °C east to 28 °C west. The upstream islands slow the ocean currents thereby increasing the residence time for warming. The SST time series (Figure 2(f)) reveals a second-order upward trend (r²=0.96) in both past and future records that could see mean values of 29 °C by 2100 under the ssp3-7.0 pathway. This will make coral reefs less resilient to acidification, leading to slower growth and more erosion from storm-surge (Magnan et al. 2019). One positive spin-off of warming seas is the deepening of the marine BL (700–760 m 1950–2020) that envelopes shallow clouds. Intra-seasonal fluctuations of SST control ∼14% of local convective variance (net OLR) while wind speeds have little effect, according to statistical tests (Appendix-a).

Analysis of linear trends in the NCEP2 meridional circulation, zonal wind and relative humidity 1979–2019 in N–S height section over Puerto Rico (Figure 3(a)–3(c)) reveals poleward winds above 300 hPa, equatorward winds below 800 hPa and subsidence in the mid-troposphere across the entire Caribbean. The resulting compression warms and dries the air, so humidity trends are downward particularly above accelerating trade winds at 12 °N and 24 °N. Zonal wind trends are negative above 500 hPa north of 20 °N, indicating a retreat of the sub-tropical jet stream and reduced wind shear, more so in early winter. A delayed wet season, alongside dust and smoke plumes from Africa, are potential consequences of regional circulation trends evident in prior work (Perez & Jury 2013). Many features appear localized: the Hadley Cell is accelerating over the Caribbean although weakening elsewhere (Lu et al. 2007) and trade winds over the Antilles Islands have remained steady despite expectations for an enhanced low-level jet (Taylor et al. 2012b).

A case study is used to highlight sub-grid scale thermal orographic processes. On 28 May 2015, moderate precipitation was generated in the absence of synoptic wave activity. The radar reflectivity map (Figure 4(a)) illustrates an axis of values >40 dBz extending W–E over the centre of Puerto Rico, and values ∼20 dBz upstream. The cloud band is underpinned by a mid-day moisture flux >400 W/m² under easterly airflow (Figure 4(b)). Land–sea temperature gradients induce seabreezes along the coast, turning the trade winds ∼30° onto the island and concentrating the transpired air on 28 May 2015. The BL height over the island alternates from 100 m (night) to ∼1,000 m (day) as evident in Figure 4(c), while the upstream marine BL height remains ∼600 m. Cloud band rainfall occurs between 14:00 and 17:00 h (local), lagging the solar angle by 3 h. Such processes are reflected in the CNRM6 model, ERA5 reanalysis, GPM satellite and CHIRPS gauge products. However, most other CMIP6 models and low-resolution reanalysis (NCEP2) under-represent thermal orographic rainfall driven by surface processes (Figure 4), and over-play the drying trend from mid-tropospheric subsidence (Figure 3).

Sea level rise in Puerto Rico is evaluated using gauge reanalysis, satellite altimeter and HAD3esm projection with ssp3-7.0 scenario (Figure 5(a)). Trends are consistently second-order: rising from 0.2 to 0.4 m from 1958 to 2020, then rising another ∼0.6 m to the end of the 21st century. The model rise is slightly ahead of gauge and satellite extrapolated trends. Few other CMIP6 models simulate such a steep rise consistent with measurements. Attribution of sea level rise considers polar ice and thermal expansion in Figure 5(b) and 5(c). The northern hemisphere polar ice cover 60°–90° latitude declines rapidly whereas the southern hemisphere is gradual. Only ∼5% is over land and relevant to global sea level. Depth-averaged sea temperatures 60 °S–60 °N latitude exhibit second-order trends that are steeper from 1 to 400 m than in the Caribbean 1–100 m. Thermal expansion underpins linear sea level rise, while polar ice melt adds an exponential contribution (Griggs et al. 2017). This translates into beach recession of ∼2 m/year at many places around Puerto Rico (USGS 2007), aggravated by storm-surge from hurricanes such as Maria 2017. Without coastal reclamation, beach tourism will suffer. Erosion control at public access points is recommended to preserve the environmental resource.

This research extends prior work on Caribbean climate change, using modern high-resolution reanalysis and coupled models from the latest IPCC iteration which reflect significant second-order trends in most variables. New results here include the following: (i) ERA5 data assimilation of historical trends since 1950, (ii) CNRM6 simulations that quantify thermal orographic rainfall over small islands and (iii) measurements and HAD3 simulations that reflect accelerating sea level rise.

Rainfall and evaporation trends are small (slope +0.008 mm day⁻¹ year⁻¹) because of competing influences at regional and local scale (cf. Figures 3 and 4). The rainfall trend accounts for only 9% of inter-annual variance, whereas the evaporation trend reaches 97%. Alternative parameters, datasets and models could yield different outcomes, but a growing consensus – made possible by improving technology – has demonstrated the alignment of trends in extrapolated measurements and model projections. A few model limitations remain: (i) convection does not reach the west side of the island (Figure 1(f)), (ii) spring rainfall is too low (Figure 1(g)), (iii) evaporation is too high (Figure 2(b)) and (iv) eastern SST exhibit cold bias (Figure 2(f)).

Mid-tropospheric drying by the Hadley circulation (Held & Soden 2006) and locally intensified transpiration are competing processes that are inadequately handled by coarse resolution products. Although tropical troughs and cyclones will entrain dry air subsiding from the Hadley Cell, a deeper marine BL over warmer SST and rising diurnal transpiration underpins the long-term stability of Puerto Rico rainfall. Seabreeze convergence −10⁻⁴ s⁻¹ induces low-level uplift of ∼0.1 m/s and diurnal rainfall of >5 mm/day on a regular basis (cf. Figure 1(b)). This contrasts with the widely held notion that summer precipitation will decline ∼30% (IPCC 2021), based on ensemble CMIP6 projections averaged over the Caribbean Antilles Islands. Unfortunately, these coupled GCMs have an area-average elevation of 15 m (88% of grid points are at sea level, cf. Appendix-b) and cannot simulate sub-grid scale thermal orographic effects which concentrate and recycle local moisture, as illustrated in Figure 4. Diurnal transpiration benefits shallow convection over mountainous islands like Puerto Rico.

New historical and projected results suggest that a drier spring relates to jet stream retreat, while a wetter autumn coincides with the annual cycle of vegetation. CRU4 maximum temperatures are rising faster than minimum temperatures (0.021 °C/year vs 0.017 °C/year 1950–2020), enhancing the diurnal cycle mainly in summer. Rainfall over Puerto Rico is amplified by seabreezes and transpiration, but upward trends may crest soon (cf. Figure 1(e)). Greater humidity contributes to insect-borne diseases and air conditioner use strains health and energy resources for an aging population.

Sea levels are rising with second-order trends in extrapolated gauge/satellite records and HAD3 projections. A 21st century, attribution budget includes over-land northern and southern polar ice melt ΔZ_N ΔZ_S, and thermal expansion from ΔZ_T = β (Z_h) ΔT, with coefficient β = 1.5×10⁻⁴, global heating depth Z_h = 400 m, sea temperature change ΔT = +4 °C by 2100. Based on Figure 5(c) – the rise due to thermal expansion is ΔZ_T ∼ 0.24 m. Northern hemisphere ice volume divided by ocean area with 4% over-land ice melted yields ΔZ_N ∼ 0.3 m. For Z_S in the southern hemisphere, we assume only 0.5% ice is melted by 2100 and obtain ΔZ_S ∼ 0.37 m. The sum is ∼ 0.9 m: the end-point of second-order trends under the ssp3-7.0 scenario (cf. Figure 5(a)) consistent with Nauels et al. (2017). The IPCC (2021) assessment from the CMIP6 ensemble gives a range of sea level rise from 0.3 to 1.2 m by the end of the 21st century under the worst-case ssp5-8.5 scenario.

Mitigation strategies begin with state funding for coastal erosion control and set-back of new development using official guidance. Popular beaches need access roads, parking facilities, pedestrian boardwalks, run-off drainage and dune restoration. These efforts need local adaptation and cyclical renewal following storm surges in late summer.

Puerto Rico can mitigate climate change by incentivizing renewable energy and developing green alternatives to gas turbine power generation. Per capita greenhouse gas emissions (∼1.5 T/year ) can be cut by electrifying vehicles and avoiding local generator use, with centralized reliable power distribution. Urban pollution threatens water quality and calls for waste recycling to keep landfills from overflowing. At the personal level, there is no better way to mitigate climate change than to minimize consumption. Affluent citizens can make do with less, to ensure the health of our planet's finite resources.

The author acknowledges support from the South Africa Department of Education, and data from sources are listed in Table 1.

The author can supply a spreadsheet of data analyses on request. The author declares no conflict of interest and received no specific funding for this work.

All relevant data are included in the paper or its Supplementary Information.