Hydrodynamic Modelling of the Guajira Upwelling System (Colombia)

Abstract

Coastal upwelling off La Guajira, Colombia, is an atypical system where persistent easterly winds drive upwelling along a zonally oriented coastline. To characterize its seasonal cycle and variability, the ROMS AGRIF hydrodynamic model was implemented under climatological forcing. Three indicators were analyzed: the 25 °C isotherm, the 36.5 isohaline, and sea-level anomalies. The simulations showed that upwelling initiates in December, reaches maximum intensity during February–April, and weakens from September to November. At maturity, vertical velocities up to 8.5 m·day⁻¹ and the shoaling of Subtropical Underwater (T = 22–25 °C; S = 36.5–37.0) dominate the coastal domain, producing widespread surface cooling (23–24 °C) and salinity enhancement. During relaxation, weaker winds and the influence of the Caribbean Coastal Undercurrent displace the upwelled waters to below 80–100 m in depth, with surface temperatures above 27 °C. Model performance against MODIS Aqua SST was high (d > 0.99; RMSE < 1.7 °C), confirming its reliability to reproduce the observed thermal cycle. The multiparametric approach reveals that upwelling persistence depends on both seasonal trade wind forcing and regional circulation. This framework provides a more integrated description of the Guajira upwelling system than previous studies and supports applications in fisheries management, ecosystem monitoring, and maritime operations.

1. Introduction

Coastal upwelling systems are among the most productive oceanic regions on Earth, accounting for approximately 20% of global marine fish catches despite occupying less than 1% of the ocean surface area [1]. These systems sustain exceptionally high levels of primary productivity, fisheries yield, and biodiversity through the persistent injection of cold, nutrient-rich subsurface waters into the euphotic zone, where light availability permits photosynthesis. The resulting enhancement of phytoplankton biomass supports complex food webs and substantial carbon sequestration, making upwelling regions critical components of both marine ecosystems and global biogeochemical cycles [2,3].

The classical Eastern Boundary Upwelling Systems (EBUS)—including the California, Canary, Humboldt (Peru-Chile), and Benguela systems—have been extensively studied due to their economic importance and relatively predictable dynamics. In these systems, equatorward winds generate offshore Ekman transport, drawing deep waters to the surface along western continental margins [4]. The intensity and spatial structure of upwelling in EBUS are modulated not only by local wind stress but also by the planetary beta effect, a consequence of the meridional variation in the Coriolis parameter (f) with latitude [5,6]. This beta effect influences the propagation of coastal trapped waves, generates poleward undercurrents, and contributes to the complex three-dimensional circulation patterns observed in these regions [7,8]. Furthermore, EBUS exhibit characteristic mesoscale and sub mesoscale variability, including filaments, eddies, and frontal structures that play crucial roles in cross-shelf transport and ecosystem dynamics [9].

Coastal upwelling also occurs in regions not associated with the classical eastern boundary upwelling systems. These secondary upwelling zones—such as those in the East and South China seas, southern Australia, western India, Java-Sumatra, and the Gulf of Mexico—are typically driven by local wind patterns, topographic effects, or seasonal monsoon systems rather than persistent boundary currents [10]. As we can see, upwelling systems along east–west oriented coastlines have received comparatively less research attention, including those in the Campeche Bank [11], and the southern Caribbean [12,13,14,15,16].

The southern Caribbean upwelling system has been studied in two distinct sectors: the western (La Guajira, Colombia) and the eastern (Cariaco, Venezuela) basins. Several oceanographic cruises, satellite observations, ocean general circulation models, and high-resolution oceanic and atmospheric reanalysis products [17].

La Guajira upwelling system in the southern Caribbean Sea represents a distinctive case within the global spectrum of upwelling dynamics. Located along the northern Colombian coast between approximately 11° N–13° N and 71° W–74° W, this system is primarily driven by the direct action of strong, persistent easterly trade winds, with minimal influence from planetary-scale effects due to its low-latitude position. The absence of significant beta-plane dynamics and the semi-enclosed nature of the Caribbean basin create a unique physical environment that differs fundamentally from the classical EBUS [15,18]. Evidence indicates that during the peak rainy season, weak wind impulses allow onshore geostrophic flow to modulate Ekman dynamics. Short-lived atmospheric disturbances can interrupt upwelling off La Guajira for several days. At times, cold fronts (Nortes) and warm fronts from the eastern tropical Pacific induce upper-layer downwelling. During these brief relaxation periods, the mixed layer warms and subsurface isotherms deepen due to onshore surface transport [17]. Understanding these alternative upwelling configurations is essential for developing a comprehensive theory of coastal ocean dynamics and for predicting ecosystem responses to climate variability across diverse geographical settings.

1.1. Regional Setting and Atmospheric Forcing

The La Guajira upwelling system exhibits pronounced seasonality, with maximum intensity typically occurring during the boreal winter and summer months (December–March and June–August), coinciding with periods of enhanced trade wind activity. The primary atmospheric driver is the Caribbean Low-Level Jet (CLLJ), a prominent feature of the regional circulation characterized by a westward wind maximum at approximately 925 hPa [19,20]. The CLLJ intensifies semi-annually in response to meridional sea surface temperature gradients and the seasonal migration of the Intertropical Convergence Zone (ITCZ). During peak periods, wind speeds frequently exceed 10 m s⁻¹, with sustained velocities sometimes reaching 12–15 m s⁻¹ over several days [18,21,22].

These strong easterly winds, oriented nearly parallel to the northern South American coastline, generate significant offshore Ekman transport. The resulting divergence in the surface layer induces vertical velocities of 1–3 m day⁻¹ in the coastal zone, drawing subsurface waters from depths of 50–100 m toward the surface [15,23]. Satellite-derived Sea Surface Temperature (SST) imagery reveals characteristic upwelling signatures, with coastal SST anomalies of −2 °C to −4 °C relative to offshore waters during active events. The upwelled waters are characterized by temperatures of 22–24 °C (compared to typical surface values of 27–29 °C) and elevated nutrient concentrations, with nitrate levels reaching 5–8 μmol L⁻¹ in the upwelling core versus <1 μmol L⁻¹ in surrounding oligotrophic waters [18,23].

The spatial extent of the upwelling influence is variable but substantial. While the strongest signal is consistently observed along the La Guajira Peninsula (particularly between Cabo de la Vela and Punta Gallinas), the coastal cooling and biological response can extend westward beyond 74° W, affecting waters off the Magdalena Department and occasionally reaching as far as the Tayrona region [15,21]. The alongshore extent typically spans 200–400 km, with a cross-shore scale of 20–50 km, though these dimensions vary considerably depending on wind forcing intensity and duration. This spatial variability has important implications for regional fisheries, as several commercially important species, including small pelagics and coastal demersal fish, exhibit population dynamics closely linked to upwelling-driven productivity [21].

1.2. Previous Research and Knowledge Gaps

Despite its ecological and economic significance, the La Guajira upwelling system remains substantially under-studied compared to the major EBUS. Earlier investigations have primarily relied on satellite remote sensing, limited in situ observations from research cruises, and reanalysis products to characterize the system’s gross features. Corredor (1979) [23] provided the first comprehensive description of the upwelling’s nutrient dynamics and phytoplankton response. Andrade and Barton (2005) [15] analyzed satellite SST and altimetry data to document the system’s seasonal cycle and relationship to CLLJ variability. More recently, Rueda-Roa and Müller-Karger (2013) [18] employed a multi-sensor approach combining SST, ocean colour, and wind fields to examine interannual variability and its connection to basin-scale climate modes, such as the El Niño-Southern Oscillation (ENSO).

While these studies have established the basic phenomenology of the La Guajira upwelling, several critical gaps remain. First, the three-dimensional thermohaline structure of the system and its temporal evolution are poorly resolved. Existing studies have largely focused on surface manifestations, leaving fundamental questions about the vertical distribution of upwelled water masses, the depth of the source layer, and the role of subsurface stratification unanswered. Second, the interaction between upwelling dynamics and regional circulation features—particularly the Caribbean Coastal Undercurrent (CCU) and the Panama-Colombia Gyre—has not been systematically investigated. These circulation elements potentially modulate the cross-shelf transport of nutrients and planktonic organisms, yet their role in the upwelling system remains speculative [24,25].

Third, existing numerical modelling studies of the Colombian Caribbean have either employed relatively coarse spatial resolution (>5 km) or focused on other aspects of regional oceanography (e.g., sediment dynamics, basin-scale circulation) rather than upwelling processes specifically [25,26]. No previous study has implemented a high-resolution (≤3 km), eddy-resolving numerical model specifically configured to examine the La Guajira upwelling system’s detailed dynamics. The resolution is essential for resolving mesoscale and sub mesoscale processes that can significantly influence cross-shelf exchange and biological productivity in upwelling zones [9,27]. There is a negative relationship between eddy kinetic energy (EKE) and net primary production (NPP) across the four major EBUS. Specifically, high levels of eddy activity correlate with low biological production, indicating a suppressive effect of eddies on productivity [9].

Furthermore, the relative contributions of different physical forcings (wind stress magnitude, wind stress curl, buoyancy fluxes, and remote forcing via coastal trapped waves) to upwelling intensity and variability have not been quantitatively assessed. While wind forcing is recognized as the primary driver, the influence of wind stress curl patterns, freshwater inputs from riverine discharge, and interactions with basin-scale pressure gradients remain poorly understood. Additionally, the potential role of sea level anomalies as tracers of upwelling dynamics—through their relationship to thermocline depth variations—has not been explored in this region.

1.3. Numerical Modelling of Upwelling Systems

Over the past two decades, regional ocean modelling has emerged as an indispensable tool for investigating coastal upwelling dynamics. Models based on primitive equations, such as the Regional Ocean Modelling System (ROMS) [28,29], have proven particularly successful in simulating the complex interactions between wind forcing, topography, stratification, and mesoscale variability that characterize upwelling regions. Applications in EBUS have demonstrated that eddy-resolving configurations (horizontal resolution ≤ 5 km) can accurately reproduce observed SST patterns, upwelling intensity, cross-shelf transport, and even ecosystem responses when coupled with biogeochemical models [8,9,30]. Others studies like Pérez and Calil [31] use a regional model with different spatial resolutions (6, 3, and 1 km), focusing on the Guajira Peninsula and the Lesser Antilles in the Caribbean Sea, to evaluate the impact of sub mesoscale processes on the regional KE energy cascade.

The choice of atmospheric forcing represents a critical design decision in upwelling simulations. While higher-frequency forcing (daily to 6-hourly) captures individual wind events and synoptic variability, climatological forcing (monthly means) offers advantages for process studies aimed at understanding mean circulation patterns, seasonal cycles, and the fundamental response to sustained wind stress [7]. Climatological simulations effectively filter out high-frequency atmospheric noise while retaining the essential physics of Ekman transport, geostrophic adjustment, and mesoscale instabilities. Such approaches have successfully characterized the mean state and dominant modes of variability in several upwelling systems [7,32].

Recent advances in ROMS development, particularly through the AGRIF (Adaptive Grid Refinement in Fortran) nesting capability, allow for efficient downscaling from basin-scale to coastal-scale processes while maintaining dynamical consistency across nested domains [33,34]. This approach is especially valuable in regions like the Caribbean, where large-scale features (e.g., the Caribbean Current) interact with local coastal processes. Previous applications of ROMS-AGRIF in similar contexts have demonstrated the model’s capacity to resolve both regional circulation and fine-scale coastal dynamics simultaneously [35,36]. A comparison of model results with observations demonstrates that the regional ocean circulation model has skill in simulating circulation and associated variability in the Caribbean Sea [37].

1.4. Study Objectives and Significance

The present study addresses the aforementioned knowledge gaps by implementing a high-resolution ROMS-AGRIF configuration specifically designed to simulate the La Guajira upwelling system under climatological forcing. Our specific objectives are to:

(1)

Characterize the three-dimensional thermohaline structure of the upwelling system, including the vertical distribution of temperature and salinity anomalies, the depth of upwelling source waters, and the cross-shelf extent of subsurface intrusions.

(2)

Quantify the seasonal cycle of upwelling intensity using complementary indicators based on temperature, salinity, and sea level anomalies, and assess their consistency and relative sensitivity.

(3)

Examine the spatial variability of upwelling response along the coast, particularly comparing the northern (La Guajira Peninsula) and southern (toward Magdalena Department) regions.

(4)

Investigate the interaction between wind-driven upwelling circulation and regional oceanographic features, including the Caribbean Coastal Undercurrent and mesoscale eddies.

(5)

Validate model performance against available observations, including satellite SST, altimetry-derived sea level anomalies, and historical in situ measurements.

(6)

Assess the utility of sea level anomalies as a diagnostic variable for monitoring upwelling intensity and identifying event onset and termination.

(7)

Take the first steps towards demonstrating the model’s operational relevance beyond the theoretical scope by applying it to Search and Rescue (SAR) contexts.

We hypothesize that the model will reveal significant subsurface complexity beyond what is observable from satellite data alone, including cross-shelf circulation cells, alongshore variability in upwelling source depth, and episodic intrusions of Caribbean Subtropical Underwater. Furthermore, we anticipate that the high-resolution simulation will capture mesoscale variability (eddies, filaments) that modulates the upwelling’s ecological impact through enhanced cross-shelf nutrient transport.

This study contributes to the broader understanding of non-canonical upwelling systems that deviate from classical EBUS paradigms. By providing the first detailed, three-dimensional numerical simulation of the La Guajira upwelling, we establish a framework for future investigations of ecosystem dynamics, biogeochemical cycling, and climate change impacts in this region. The validated model can serve as a foundation for coupled physical-biological simulations, support fisheries management through improved environmental characterization, and contribute to regional ocean forecasting capabilities. Moreover, insights gained from this low-latitude, wind-driven system may inform comparative studies across the global spectrum of coastal upwelling regimes.

2. Materials and Methods

2.1. Study Area

The southern Caribbean Sea is one of the most productive areas in the region due to the relatively large coastal upwelling that occurs in the offshore basins of Colombia and Venezuela [17]. The study area is located within the La Guajira upwelling system of the southern Caribbean Sea, near the coasts of Colombia’s La Guajira and Magdalena regions; see Figure 1. This area is recognized as one of the most productive in the southern Caribbean due to intense coastal upwelling [17,38]. The variability of the La Guajira upwelling is primarily influenced by the atmospheric Caribbean Low-Level Jet and significant ocean mesoscale activity, including filaments [15,38]. Strong easterly winds blowing nearly parallel to the coast, particularly those associated with the CLLJ, induce an offshore Ekman transport, leading to the ascent of nutrient-rich waters into the otherwise oligotrophic sea. This process results in high primary productivity, with surface chlorophyll-a concentrations reaching up to 3 mg·m⁻³ during the windy season [15,38].

The region’s climate is characterized by the predominance of trade winds, modulated by the Intertropical Convergence Zone, leading to two distinct seasons: a dry season and a rainy season. The upwelling system exhibits a marked weakness during the rainy season and intensifies during the dry season and the “San Juan summer” (June–July) [38,39]. During the peak of the rainy season, warm, low-salinity surface waters from the Panama-Colombia basin flow towards La Guajira. Weak winds reduce offshore Ekman transport, causing these warm waters to remain in the coastal area, displacing the thermocline downward and inhibiting coastal upwelling [39]. Conversely, during the dry and windy months, strong northeast trade winds and a robust westward force from the CLLJ intensify the upwelling [38]. The core of the upwelling is bounded to the north by the Caribbean Current and to the west by warmer waters from the Panama-Colombia oceanic region [39]. The Caribbean Counter Current also plays a significant role, with its effect on upwelling still a subject of debate [38].

The frontal zone between the cold, nutrient-rich coastal waters and the warm, low-nutrient open-ocean waters is irregular and oscillates annually [39]. The upwelling generates strong cold anomalies in sea-surface temperature data and is closely linked to high chlorophyll-a concentrations [38]. Moreover, the mesoscale variability, particularly the presence of anticyclonic eddies formed in the eastern Caribbean, further intensifies these upwelling events by steering the advection of cold upwelling filaments westward [40].

The La Guajira upwelling system is distinguished by the perpetual presence of cold, saline waters that surface along the northern coast of the Colombian Caribbean. These events exhibit a marked weakness during the rainy season and a heightened intensity during the dry season and the San Juan summer (June–July), when the northeast trade winds reach their zenith and the CLLJ exerts a robust westward force along the coast of La Guajira [15,23,39,41,42].

2.2. Methods

2.2.1. Computational Domain

The study region spans the La Guajira and Magdalena coasts in the southern Caribbean Sea (Colombia), encompassing the core upwelling corridor from Santa Marta to Punta Gallinas. The model domain extends from 10.5° N to 14.5° N and 77.5° W to 70.5° W (Figure 1), large enough to include the coastal upwelling jet, the shelf–slope exchange, and interaction with the Caribbean Current.

2.2.2. Model Code Base and Configuration

We used ROMS AGRIF [32], complemented with CROCO Tools utilities [32] for grid generation, bathymetry processing, and boundary files. ROMS solves the hydrostatic, Boussinesq, Reynolds-averaged Navier–Stokes equations with split-explicit time stepping for barotropic/baroclinic modes [28]. The baroclinic time step was 600 s and the barotropic (external) time step 10 s, satisfying the CFL criterion for both gravity waves and internal dynamics over the chosen grid.

Horizontal discretization used a rectilinear Arakawa C-grid with uniform resolution of 1/30° (≈2 nautical miles; ≈3.7 km). The vertical grid employed S-coordinates with 32 terrain-following levels. We set the critical depth to h_c = 200 m and used stretching parameters θ_s = 7.0 and θ_b = 2.0 to concentrate resolution near the surface and seabed in shallow areas. Vertical levels are mapped as:

$$z(x,y,\sigma ,t)=S(x,y,\sigma )+\zeta (x,y,t)\left[1+{\displaystyle \frac{S(x,y,\sigma )}{h(x,y)}}\right]$$

(1)

$$S(x,y,\sigma )=h_c\sigma +\left(h\right(x,y)-h_c)C\left(\sigma \right)$$

(2)

where h(x, y) is bathymetric depth, C(σ) the vertical narrowing function, which adjust the level distribution from the free surface (ζ) to the ocean floor [38,39].

2.2.3. Bathymetry and Topographic Smoothing

Bathymetry was extracted from GEBCO (30 arc seconds; GEBCO_2020 grid) [43,44], bilinearly interpolated to the model grid, and smoothed using a Shapiro filter to limit the r-factor (r_x0) and reduce pressure-gradient errors. We enforced r_x0 ≤ 0.20 on the shelf (depth < 200 m) and ≤0.25 elsewhere. Known coastal embayment and narrow shelf segments were inspected to avoid artificial channels after smoothing.

2.2.4. Numerical Schemes and Parameterizations

Momentum and tracer advection employ a third-order upstream-biased (UP3) scheme with multidimensional flux limiting to curb spurious extrema, while lateral viscosity uses harmonic viscosity with a weak Smagorinsky-like deformation dependence (nondimensional coefficient 0.2), yielding effective A_h ≈ O(50–200) m² s⁻¹ in energetic shear zones, and tracer lateral diffusivity is set to Aht ≈ 50 m² s⁻¹ on the shelf, tapering offshore with grid metrics. Vertical mixing is parameterized through a Generic Length Scale (GLS) turbulence closure with k–ε stability functions (Kantha–Clayson), background vertical viscosity A_v0 = 1 × 10⁻⁴ m² s⁻¹, and tracer diffusivity K_v0 = 1 × 10⁻⁵ m² s⁻¹, with surface and bottom boundary layers evolving prognostically under GLS. Bottom stress applies quadratic drag with C_d = 2.5 × 10⁻³ uniformly and roughness length z₀ = 1.0 × 10⁻³ m, though sensitivity tests with Cd in [2.0–3.0] × 10⁻³ did not alter seasonal cycle conclusions. The pressure-gradient scheme utilizes a density-Jacobian formulation with spline vertical averaging to minimize truncation errors over steep topography, with these choices collectively prioritizing robust climatological spins with limited artificial diffusion while maintaining stability along the steep shelf break.

2.2.5. Surface and Lateral Boundary Conditions

Surface forcing: Monthly COADS (COADS05) climatology [45,46] provided wind stress components (τ_x, τ_y), net heat flux, shortwave radiation, and freshwater flux (E–P). We prescribed fluxes directly (no bulk formulation), consistent with the climatological approach, and did not include atmospheric pressure loading or diurnal cycles.

Initial and open boundary conditions: GLORYS global reanalysis monthly means (2000–2015) [27] provided temperature, salinity, sea level, and 3D velocities. Open-boundary schemes followed common ROMS practice: Chapman for free surface, Flather for depth-mean velocity, and radiation–nudging (Orlanski-type) for 3D baroclinic velocities and tracers. Tracer and baroclinic velocity nudging time scales were 5 days within the first two grid cells, relaxing to 15 days by the fifth cell. A 10-gridpoint sponge layer increased lateral viscosity and tracer diffusivity by a factor of 3 to damp boundary reflections.

Tides and rivers: Barotropic tides were not included

Sea state and waves: No wave coupling was applied.

River’s runoff. It is important to note that this model configuration does not explicitly include freshwater forcing from riverine discharge or coastal lagoon systems. The most significant sources of freshwater input in the study region are the Magdalena River and the Ciénaga Grande de Santa Marta estuarine complex, both located in the southwestern sector of the domain (approximately 74–75° W). The Magdalena River, with a mean annual discharge of approximately 7000 m³·s⁻¹, exhibits strong seasonal variability with maximum flow during the rainy season (September–November) and minimum flow during the dry season (December–April), when upwelling is most intense. While this omission may affect nearshore salinity representation, the primary focus on wind-driven upwelling dynamics justifies this simplification for the climatological scope of the study. The implications and potential effects of this limitation are addressed in the Section 4.

2.2.6. Spin-Up, Integration Length, and Diagnostics

The integration consisted of a 5-year spin-up from a quiescent state (“cold start”), followed by 5 additional years under identical climatological forcing to accumulate a stable seasonal cycle. Unless stated otherwise, all maps and sections use monthly means averaged over the final 5 years. Model output was archived daily and postprocessed to monthly fields. Diagnostics included:

Vertical velocity w at ρ-points (converted to m day⁻¹ for readability).

Mixed-layer depth by a 0.2 °C threshold relative to 10 m temperature.

Upwelling indicators: surface isotherm 25 °C (SST < 25 °C), surface isohaline 36.5 (SSS > 36.5), and sea-level anomaly η < −0.15 m; the “upwelling area” corresponds to grid cells simultaneously meeting all three criteria.

Sensitivity: thresholds were varied by ±0.5 °C, ±0.2 in salinity, and ±0.05 m in sea level; the qualitative seasonality and spatial footprint were invariant to these perturbations.

2.2.7. Validation Datasets and Skill Metrics

We validated the ROMS climatological monthly mean SST against climatological monthly composites derived from MODIS Aqua Level-3 mapped products (4 km resolution) spanning 2003–2021. For each calendar month, we computed multi-year averages from the MODIS time series, yielding 12 climatological monthly fields directly comparable to the model’s climatological forcing framework. MODIS fields were quality controlled using standard cloud and stray-light masks, and bilinearly interpolated to the ROMS grid. Comparisons are better in the oceanic portion of the domain.

For each month, we computed spatial difference maps (MODIS − ROMS) and calculated validation metrics—root mean square error (RMSE), mean absolute error (MAE), and the Willmott index of agreement d [47,48]—using all valid grid points within that month. P represents modelled SST, O represents observed (MODIS) SST, O denotes the spatial-mean of observations, and N is the number of valid pixels. The Willmott index quantifies model–observation correspondence and ranges from 0 (no agreement) to 1 (perfect agreement). The Willmott index is a standardized measure of error in a model’s prediction, which allows one to quantify the correspondence between variables. It ranges from 0 (no concordance) to 1 (perfect concordance).

$$RMSE=\sqrt{{\displaystyle \frac{1}{N}}\sum (P-O{)}^2}$$

(3)

$$MAE={\displaystyle \frac{1}{N}}\sum \left|P-O\right|$$

(4)

$$d=1-\left({\displaystyle \frac{\sqrt{\sum (P-O{)}^2}}{\sum (\left|P-\overline{O}\right|+\left|O-\overline{O}\right|{)}^2}}\right);0\le d\le 1$$

(5)

We also examined spatial error structure by mapping monthly SST differences (ROMS − MODIS) and variance fields for both products over 2003–2007 to assess the reproduction of mesoscale and coastal variability.

2.2.8. Reproducibility and Preprocessing

Grid and forcing preparation: CROCO Tools [23] was used to generate the grid and vertical stretching, interpolate GEBCO bathymetry, compute r_x0. diagnostics, and build boundary/initial files from GLORYS (2000–2015 monthly means). COADS05 fluxes were remapped conservatively to the model grid.

Stability checks: Maximum barotropic Courant numbers remained <0.7; no lateral viscosity “blow-up” flags were encountered after year 2 of spin-up.

Postprocessing: Sections P1–P3 were extracted along fixed coast-normal transects at Cabo de la Vela, Riohacha, and Parque Tayrona (Figure 1), and composited for peak (March) and relaxation (October) months.

Notes on limitations (also discussed in Section 4): the exclusion of river discharge and tides, the use of monthly climatological fluxes (no synoptic variability), and the coarse boundary relaxation inevitably damp nearshore SST variance and can introduce a slight cold bias during peak upwelling. These modelling choices are consistent with the objective of building a stable, interpretable climatological baseline for La Guajira and do not affect the main seasonal behaviour.

3. Results

The ROMS model demonstrated a satisfactory response to the atmospheric forcing imposed by climatological data, adequately reproducing the oceanic conditions characteristic of each season of the year. This response indicates that the model reaches a quasi-balanced energy state, without presenting abrupt transitions or numerical instabilities. This is essential to avoid problems such as excessive artificial diffusion or the generation of unrealistic values in key system variables.

The modelling demonstrated that the prevailing atmospheric conditions are conducive to the occurrence of upwelling events during a significant portion of the year, beginning in December and extending until early August, with a weakening during the wet season as a result of the southern migration of the Intertropical Convergence Zone (ITCZ), which reduces the intensity of the trade winds responsible for Ekman suction, the primary mechanism generating upwelling [17]. The results, based on sea surface temperature (SST) fields, demonstrate that the upwelling system in La Guajira remains active for most of the year. It begins its intensification phase in December, reaches maturity during the February–March–April (FMA) quarter, and enters a relaxation phase from September to the end of November (SON) (Figure 2).

The initial phase of the upwelling phenomenon is characterized by the rise in the 25 °C isotherm toward the surface. This is a reliable indicator of upwelling. During the peak season (February–April), surface temperatures between 23 and 24 °C were observed in the coastal area between Santa Marta and Punta Gallinas. It is important to note that this model did not account for the influence of warm water from the Magdalena River and the Ciénaga Grande de Santa Marta on the sea surface temperature (SST). The 25 °C isotherm reaches latitudes close to 13° N, and vertical velocities up to 8.5 m/day are recorded in the upwelling zone during this phase.

The relaxation phase (Figure 2) coincides with the rainy season. During this period, the trade winds weaken, reducing the momentum necessary for Ekman transport. An influx of warm subsurface waters associated with the Panama–Colombia gyre (Caribbean Coastal Undercurrent, CaCU) is also observed, as described by Andrade et al. [15], and Correa-Ramirez et al. [25]. During this phase, surface temperatures exceed 27 °C, and the 25 °C isotherm descends to depths between 80 and 100 m off the coast; see Figure 3.

Like the use of the 25 °C isotherm as an indicator of upwelling, salinity concentrations were evaluated, revealing that the 36.5 isohaline behaves similarly during upwelling events. This demonstrates the ocean’s response to atmospheric forcing, which varies by season. The synchrony between the 25 °C isotherm and the 36.5 isohaline suggests that they delineate the same body of water with specific properties corresponding to the Subtropical Underwater (SUW). The SUW is responsible for the maximum surface salinity during upwelling in La Guajira, as reported by Correa-Ramírez et al. [25]. During the relaxation phase (SON), maximum surface salinity values close to 36 are observed, with a variation in only 0.5 units between high-intensity events and relaxation periods. This behaviour is partially modulated by the CaCU, which flows along the continental slope (at a depth of ~100 m), transporting water from the Panama–Colombia gyre.

Figure 4 shows the salinity behaviour in cross sections P1 (Cabo de la Vela), P2 (Riohacha), and P3 (Tayrona Park) in March (maximum upwelling) and October (maximum relaxation). It illustrates a core of maximum salinity (~37) between a depth of 100 and 150 m, which is flanked above and below by 36.5 isohaline.

During upwelling events, only the 36.5 layer reaches the surface. It rises vertically in a similar way to the 25 °C isotherm, allowing upwelling water masses to be delimited between depths of 80 and 160 m. In May, June, and July, the core of maximum salinity migrates to depths of less than 50 m near the coast without reaching the surface. This is possibly due to turbulent processes that favour salinity diffusion, as well as the phenomenon of double diffusion. This is because the mass has salinity gradients above and below its vertical position.

Sea level also serves as an indicator of upwelling, since the ascent of deep waters—characterized by low temperature and high salinity—results in a denser water mass due to its physicochemical properties. Combined with the wind-driven displacement of surface layers, this process produces local variations in sea level, causing the upwelling region of La Guajira to exhibit lower sea levels (−0.15 m, according to ROMS model estimates) compared to the mean sea level.

During the mature upwelling phase (FMA), model-predicted sea level depressions range from −0.15 to −0.20 m in the coastal zone between Cabo de la Vela and Santa Marta, coinciding spatially with the coldest SST values and highest surface salinities. During relaxation (SON), sea level anomalies diminish to values between −0.05 and +0.05 m, reflecting the reduced wind-driven divergence and the influence of the CaCU.

Recognizing that upwelling involves multiple processes occurring simultaneously, it can be identified at the surface through the 25 °C isotherm (SST < 25 °C), the 36.5 isohaline (Surface Salinity > 36.5), and the −0.15 m sea level isoline (η < −0.15 m). Accordingly, we propose integrating these three variables into a single upwelling indicator to understand the phenomenon, acknowledging that the behaviour of each variable depends on the others. Thus, areas that meet the overlap of the three criteria are considered upwelling regions (Figure 5).

The seasonal evolution of the integrated upwelling area shows a clear annual cycle, with maximum extent between February and late May (>20,000 km²), reaching its peak in mid-March and exceeding 60,000 km² (Figure 6). Secondary maxima, although considerably smaller than those in March, occur in mid-July (~15,000 km²). From September to November, the suggested indicators are not observed.

When the modelling results are compared with sea surface temperature (SST) satellite data obtained from the MODIS Aqua sensor, the main features of coastal upwelling in the northern Colombian Caribbean are adequately represented, as is ocean circulation. The predominant flow of the Caribbean Current toward the northwest is highlighted.

Table 1. Model validation results. Root Mean Square Error (RMSE, °C), mean absolute error (MAE, °C), and concordance index (d, dimensionless).

Month	RMSE	MAE	d
January	0.497	0.397	0.997
February	0.645	0.497	0.997
March	0.939	0.633	0.997
April	1.250	0.951	0.996
May	1.463	1.256	0.996
June	1.688	1.433	0.996
July	1.301	1.130	0.997
August	0.835	0.667	0.995
September	0.692	0.571	0.992
October	0.510	0.442	0.990
November	0.443	0.384	0.992
December	0.407	0.303	0.997

presents the statistical validation, showing the mean square error, mean absolute error, and concordance index calculated monthly. Overall, the model performed well compared to the satellite data, with a maximum root mean square error (RMSE) of 1.68 °C in June and a minimum of 0.40 °C in December. The concordance index yielded values close to 1 for all months, reaching 0.99 in several periods, indicating excellent correspondence between simulated and observed values.

Validation showed that the relative error (point-to-point) between August and February remains below 1 °C. During upwelling maturity events (FMAs), however, the model tends to simulate slightly cooler surface temperatures, with a difference of between 1 and 3 °C (Figure 6).

The error pattern shows systematic cold biases during peak upwelling months (February–April) in the nearshore zone, where ROMS underestimates SST by 1–3 °C relative to MODIS observations. In contrast, during the relaxation phase (September–November), errors are minimal (<0.5 °C) and randomly distributed. The largest absolute errors occur in the transition zone between upwelled and non-upwelled waters, suggesting limitations in resolving sharp frontal gradients at the model’s horizontal resolution.

The analysis of sea surface temperature (SST) variance reveals the temporal and spatial structure of thermal fluctuations in the study region. Monthly multi-year differences demonstrate recurrent cooling along the Guajira coast from December to August, reaching maximum intensity between February and April (up to −3 °C relative to the surrounding waters), corresponding to the intensification of the Caribbean Low-Level Jet (CLLJ). During September–November, the system enters a relaxation phase characterized by weakened winds and SST exceeding 27 °C.

The observed SST variance fields exhibit localized maxima (0.8–2.0 °C²) near Riohacha and Cabo de la Vela, particularly during peak upwelling months (February–April) (Figure 7). These high-variance cores are concentrated within 50 km of the coast and extend along the continental shelf. Variance decreases sharply offshore, dropping below 0.4 beyond the 200 m isobath. A secondary variance maximum appears during June–July (0.6–1.2 °C²), associated with the San Juan summer upwelling event. Minimum variance (<0.3 °C²) occurs during September–November, coinciding with the relaxation phase.

The model-reproduced variance fields capture the dominant spatial and temporal patterns observed by satellite, with maxima collocated near the major upwelling centers and consistent seasonality (Figure 8). However, systematic differences are evident:

(a)

Variance amplitude. The model underestimates the coastal variance, with peak values below 1.6 °C², compared to satellite maxima of 2.0 °C². This suggests enhanced numerical dissipation and limited capacity to resolve fine-scale turbulence.

(b)

Spatial extent. High-variance regions extend farther offshore in the model (up to 100 km) than in the satellite data, where variability remains confined to the nearshore zone (<50 km). This difference reflects the coarser structure of climatological forcing, which smooths the small-scale wind variability driving coastal SST fluctuations.

(c)

Temporal persistence. Both datasets identify a primary maximum during February–April and a secondary enhancement in June–July. However, the model maintains elevated variance into late boreal summer (August–September), while the satellite record shows a more abrupt decline after July.

(d)

Seasonal evolution. The model reproduces the general seasonal cycle but with smoother transitions between phases and less pronounced monthly variability than observations.

(e)

Quantitative skill. The discrepancies align with statistical validation (Table 1), where RMSE values remain below 1.7 °C and the concordance index (d) exceeds 0.99. The systematic underestimation of variance is consistent with the slight cold bias (−1 to −3 °C) identified during peak upwelling (Figure 6).

(f)

Physical interpretation. The absence of explicit freshwater forcing (e.g., Magdalena River discharge), and synoptic wind events likely contribute to the weaker modeled variance. These factors are well documented in previous observational studies as key modulators of thermal variability in the region.

Figure 8. Mean annual cycle of upwelling area extent off La Guajira based on multi-annual model simulations.

Figure 8. Mean annual cycle of upwelling area extent off La Guajira based on multi-annual model simulations.

A detailed examination of the upwelling dynamics reveals the quantitative contribution of forcing mechanisms and water-mass transformations. The intensification of the Caribbean Low-Level Jet during boreal winter–spring produces wind stress values exceeding 0.15 N·m⁻² along the Guajira coast, which translate into offshore Ekman transport on the order of 0.5–0.8 m²·s⁻¹.

This transport is sufficient to induce vertical advection rates up to 8.5 m·day⁻¹ in the nearshore zone (within 30 km of the coast), as reproduced by the model. Maximum vertical velocities occur in February–March at depths between 20 and 40 m, coinciding with the shoaling of the thermocline and halocline.

The simultaneous shoaling of the 25 °C isotherm and the 36.5 isohaline indicates that the upwelled waters predominantly originate from the Subtropical Underwater (SUW) mass located at depths between 80 and 160 m. The thermohaline properties of this water mass (T = 22–25 °C; S = 36.5–37.0) explain the observed cooling of the surface layer and the enhancement of salinity during mature upwelling conditions.

During the relaxation phase (SON), vertical velocities decrease to <2 m·day⁻¹, wind stress drops below 0.08 N·m⁻², and Ekman transport is reduced to <0.3 m²·s⁻¹. The weakening of easterly winds is accompanied by the intensification of the Caribbean Coastal Undercurrent, which introduces relatively warmer subsurface waters (>26 °C) into the shelf break, displacing the SUW downward to depths of 80–100 m.

4. Discussion

The validation metrics obtained (d > 0.99; RMSE < 1.7 °C) are comparable to or exceed those reported for ROMS applications in other Eastern Boundary Upwelling Systems (EBUS). Studies in the California Current System [7,30] and Peru-Chile system [8,32] reported similar accuracy levels under climatological forcing. This suggests that ROMS-AGRIF captures the essential dynamics of the Guajira upwelling with skill comparable to that achieved in better-studied EBUS regions, despite the distinct forcing mechanisms. Unlike classical EBUS dominated by meridional winds and planetary effects [2,3], La Guajira is primarily driven by zonal winds and the Caribbean Low-Level Jet, making these results particularly encouraging for tropical upwelling system modelling.

The model captures the timing, spatial structure, and persistence of the primary (FMA) and the weaker mid-year (JJA) upwelling events in agreement with satellite observations (e.g., [22,39]; Figure 2 and Figure 5). Nonetheless, nearshore SST variance is underestimated (~20–40% within 30 km of the coast; Figure 6 and Figure 7), indicating that synoptic atmospheric fluctuations and mesoscale/submesoscale processes—absent in climatological forcing—modulate upwelling intensity at event scales [31]. Point-to-point errors are small during August–February (<1 °C), whereas during mature events (FMA) the model exhibits a cold bias of 1–3 °C along the Guajira peninsula (Figure 6). We hypothesize three non-exclusive contributors: (i) omission of buoyancy input from the Magdalena River and Ciénaga Grande, which enhances nearshore stratification and modifies mixing; (ii) unresolved submesoscale turbulence controlling cross-shelf heat fluxes; and (iii) wind biases in COADS climatology near steep orography that can misrepresent along-shore stress and stress curl. These factors will be testable with high-frequency reanalysis forcing (e.g., ERA5/CCMP) and inclusion of river discharge.

The simultaneous shoaling of the 25 °C isotherm and the 36.5 isohaline indicates that the upwelled waters predominantly originate from the Subtropical Underwater (SUW) mass located at depths between 80 and 160 m. The thermohaline properties of this water mass (T = 22–25 °C; S = 36.5–37.0) explain the observed cooling of the surface layer and the enhancement of salinity during mature upwelling conditions. This finding confirms previous observational work [15,25] and provides new quantitative estimates of SUW vertical migration rates that were not available from sparse in situ measurements.

Dynamical interpretation: wind forcing, regional circulation, and sea level. Seasonal relaxation (SON) is characterized by weakened easterlies, reduced Ekman divergence, and the intrusion of the Caribbean Coastal Undercurrent (CaCU) along the shelf break. In the model, warmer subsurface waters (>26 °C) occupy 60–120 m and the SUW core deepens to 80–100 m, consistent with the vertical displacement of the 25 °C isotherm and 36.5 isohaline (Figure 3 and Figure 4). Vertical velocities decrease to <2 m day⁻¹ (domain average over the upwelling wedge). Concurrent sea-level anomalies show coastal depressions of ~−0.15 to −0.20 m relative to the open Caribbean, consistent with satellite altimetry patterns in the region, indicating a dynamically consistent barotropic adjustment to wind-driven divergence modulated by density-driven alongshore flow.

Regional circulation studies in the Caribbean Sea [37,40,49] have emphasised the role of mesoscale eddies in modulating basin-scale transport and variability. Our results confirm that the Caribbean Coastal Undercurrent (CaCU) significantly influences the relaxation phase of upwelling by introducing warmer subsurface waters (>26 °C) that displace the Subtropical Underwater (SUW) to greater depths (80–100 m). This interaction between wind-driven upwelling and geostrophic circulation aligns with recent findings on the importance of the Panama-Colombia gyre in the regional heat budget [38] and suggests that interannual variations in CaCU strength may be a key source of upwelling variability not captured by climatological forcing.

Placing La Guajira within the global context of coastal upwelling systems [2,3] highlights both its distinctive characteristics and its commonalities with better-studied regions. Unlike the “big four” EBUS (California, Humboldt, Canary, Benguela), which are driven by meridional winds and exhibit strong equatorward flow along western continental boundaries, La Guajira is forced by zonal easterly winds associated with the CLLJ and represents a “sideways” configuration where the coastline runs roughly east–west. This geometry produces upwelling dynamics more analogous to equatorial upwelling systems [11] or to secondary upwelling centres in embayed coastlines [50] than to classical EBUS.

These results demonstrate that the Guajira upwelling system is a dynamically consistent response to seasonal atmospheric forcing, modulated by regional subsurface circulation, and maintained by the recurrent vertical advection of SUW into the euphotic zone. The integration of multiple indicators (temperature, salinity, sea level) provides a more complete picture than previous studies based on single variables, whilst the climatological modelling framework establishes a baseline against which interannual variability and long-term trends can be assessed.

This study advances beyond previous modelling efforts in the Colombian Caribbean [26,51] by providing: (1) a longer climatological integration (5-year spin-up plus multi-year analysis) that ensures statistical equilibrium; (2) simultaneous validation against both satellite SST and multiple upwelling indicators (isotherms, isohalines, sea level); and (3) explicit quantification of vertical velocities and their seasonal variability. The multi-indicator approach employed here provides a more robust characterisation of upwelling intensity and extent than single-variable analyses [18,39], whilst the integration of hydrodynamic modelling with operational applications (SAR operations) demonstrates practical societal relevance beyond purely scientific objectives.

5. Conclusions

The implementation of the ROMS-AGRIF model, complemented with CROCO tools, successfully reproduced the main hydrographic and dynamic features of the Colombian Caribbean Sea under climatological forcing. The simulations captured the seasonal cycle of the La Guajira upwelling system with three clearly defined phases: onset (December–January), maturity (February–April), and relaxation (September–November).

During maturity, vertical velocities reach up to ~8.5 m day⁻¹, surface waters cool to 23–24 °C, and surface salinity increases, consistent with the vertical advection of Subtropical Underwater (22–25 °C; 36.5–37.0) sourced from ~80–160 m. During relaxation, weakened easterlies and the intrusion of the Caribbean Coastal Undercurrent along the shelf break deepen the SUW core to ~80–100 m, reduce vertical velocities to <2 m day⁻¹, and restore SST > 27 °C. Modelled coastal sea-level depressions of approximately −0.15 to −0.20 m are dynamically consistent with wind-driven divergence and barotropic adjustment.

Notwithstanding the overall agreement, the model underestimates nearshore SST variance and exhibits a coastal cold bias of ~1–3 °C during mature events. The most plausible contributors are: omission of buoyancy input from the Magdalena River and the Ciénaga Grande, which modulates stratification and mixing; the absence of synoptic and submesoscale variability under climatological forcing; and potential wind-stress biases near steep orography in the COADS product. These limitations are tractable and point directly to targeted improvements in forcing, physics and boundary conditions.

The model reproduced not only the spatiotemporal variability of upwelling but also the associated vertical velocities (up to 8.5 m·day⁻¹) and the extent of cold, saline waters from Santa Marta to Punta Gallinas. Quantitative skill against MODIS Aqua SST (Willmott d ≈ 0.99; RMSE ≈ 1.4–1.7 °C, monthly scale) supports the model’s ability to capture the phasing, spatial structure and persistence of upwelling under prevailing easterly winds. These results confirm that ROMS is a reliable framework for the study of hydrographic processes in regions with limited in situ observations.

Methodologically, the multi-indicator approach offers two advances for this region: it constrains the provenance and vertical migration of the upwelled water mass using coherent thermal and haline tracers, and it links surface thermohaline signatures with sea-level adjustments indicative of the barotropic response. Together, these diagnostics place La Guajira within the broader spectrum of coastal upwelling systems as a zonally oriented eastern boundary regime forced by the Caribbean Low-Level Jet and modulated by regional subsurface circulation, rather than by persistent equatorward jets typical of the ‘classical’ EBUS.

The validated configuration provides a practical foundation for next steps of high scientific and operational value. Priorities include (i) high-frequency atmospheric forcing (reanalysis and satellite winds) to resolve event-scale dynamics and improve coastal stress and curl; (ii) explicit freshwater inputs to quantify buoyancy-driven controls on nearshore stratification and mixing; (iii) coupling to biogeochemical modules to translate physical variability into ecosystem response metrics; and (iv) incorporation of data assimilation (altimetry, SST and in situ profiles) toward a pre-operational capability for maritime safety, fisheries management and environmental protection. These enhancements will enable rigorous assessment of intra-seasonal variability, interannual modulation and sensitivity to future climate forcing within a unified, process-based framework.

A distinctive contribution of this work, compared with previous studies on the La Guajira upwelling system, lies in the integration of multiple indicators (isotherms, isohalines, and sea-level anomalies) to characterize the seasonal cycle under a consistent climatological framework. Earlier efforts mainly relied on observational datasets, single tracers, or short-term simulations, whereas this study provides a long-term, multi-parameter description of upwelling intensity, extent, and variability. This approach strengthens the understanding of SUW dynamics and their relevance for regional productivity.

Nevertheless, several limitations must be acknowledged. The simulations did not explicitly include freshwater inputs from the Magdalena River or the Ciénaga Grande de Santa Marta, which are known to influence nearshore salinity and circulation patterns. Likewise, the coarse resolution of climatological forcing may underestimate the impact of synoptic or intraseasonal atmospheric events (e.g., trade wind surges and tropical storms). Additionally, the model is purely hydrodynamic and does not incorporate biogeochemical interactions, thereby limiting its ability to directly quantify the implications of upwelling for primary production and fisheries.

Future work should integrate river discharges, high-resolution atmospheric forcing, and coupled physical–biogeochemical models to further resolve coastal–ocean interactions and their ecological consequences. Such extensions would also allow an improved assessment of the response of the Guajira upwelling system to climate variability and long-term change.

In summary, the present study establishes a robust numerical framework to characterize the climatological variability of the La Guajira upwelling system, highlights the importance of multiparametric indicators (temperature and salinity), demonstrates its novelty compared with previous studies, and shows the practical value of hydrodynamic modelling for both scientific understanding and operational applications in the Colombian Caribbean.

Beyond the theoretical scope, the model has proven operational relevance, as illustrated by its application to Search and Rescue (SAR) operations, where the coupling between sea surface temperature, vapor pressure, and atmospheric visibility directly affects mission efficiency. This dual role—scientific and operational—emphasizes the potential of hydrodynamic models as decision-support tools for maritime safety, fisheries management, and environmental protection.