Long-Term Spatiotemporal Analysis of Precipitation Trends with Implications of ENSO-Driven Variability in the Department of Magdalena, Colombia

Abstract

The Magdalena department, influenced by southern trade winds and ocean currents from the Atlantic and Pacific, is a climatically vulnerable region. This study assesses the Magdalena Department’s precipitation trends and stationary patterns by analyzing multi-year monthly records from 55 monitoring stations from 1990 to 2022. To achieve this, the following methods were used: (i) homogeneous regions were established by an unsupervised clustering approach, (ii) temporal trends were quantified using non-parametric tests, (iii) stationarity was identified through Morlet wavelet decomposition, and (iv) Sea Surface Temperature (SST) in four Niño regions was correlated with stationarity cycles. Silhouette’s results yielded five homogeneous regions, consistent with the National Meteorological Institute (IDEAM) proposal. The Department displayed decreasing annual trends (−32–−100 mm/decade) but exhibited increasing monthly trends (>20 mm/decade) during the wettest season. The wavelet decomposition analysis revealed quasi-bimodal stationarity, with significant semiannual cycles (~4.1 to 5.6 months) observed only in the eastern region. Other regions showed mixed behavior: non-stationary in the year’s first half and stationary in the latter half. Correlation analysis showed a significant relationship between SST in the El Niño 3 region (which accounted for 50.5% of the coefficients), indicating that strong phases of El Niño anticipated precipitation responses for up to six months. This confirms distinct rainfall patterns and precipitation trends influenced by the El Niño–Southern Oscillation (ENSO), highlighting the need for further hydrometeorological research in the area.

1. Introduction

Precipitation is a meteorological phenomenon that significantly affects a region’s water availability [1]. Understanding precipitation’s spatial distributions and cycles contributes to territorial planning, water management, and agricultural productivity within watersheds [2]. Climate change, driven by anthropogenic activities, has led to more extreme weather events, making long-term precipitation data essential for gaining insights into climatic patterns [3]. This knowledge supports the development of sustainable strategies to comprehend the hydrological cycle dynamics and ensure the availability of ecosystem services [4]. Over the past two decades, research has increasingly focused on rapid changes in water balance due to regional climate phenomena, particularly in tropical regions, which account for 32% of global precipitation [5]. Large-scale atmospheric circulations are responsible for these precipitation patterns, each uniquely influencing the hydrometeorological dynamics of these zones [6,7,8].

El Niño–Southern Oscillation (ENSO) is the primary driver of interannual rainfall variability, with its impact largely determined by fluctuations in the mean Sea Surface Temperature (SST) of the Tropical Pacific Ocean (TPO) [9]. The Oceanic Niño Index (ONI) measures the magnitude and persistence of these fluctuations, allowing for the identification of La Niña or El Niño events [10]. Changes in the ONI are reflected in long-term hydrometeorological trends [11]. For example, using non-parametric trend tests and time-frequency decomposition, Zhang et al. [12] evaluated spatiotemporal precipitation changes in the Yarlung Zangbo River in China. Similarly, Sharma and Goyal [13] employed wavelet transforms to examine drought across India. Both studies indicate a decline in rainfall and streamflow, alongside an increase in aridity, particularly during El Niño events. Other studies have achieved comparable results using analogous methods [14,15,16].

The long-term changes in precipitation and their correlation with ENSO demonstrate that this oscillation can explain the interannual rainfall variability in Colombia, which affects river discharge, soil moisture, and vegetation activity [17]. Esquivel et al. [18] used the Kendall correlation at various SST lead times to evaluate predictability limits. They found significant negative correlations between interannual precipitation and average SST in the El Niño 3 region, specifically with a 6-month lag in the Pacific basin. These findings align with Navarro-Monterroza et al. [19], who reported the most significant changes during dry periods. These teleconnections are related to long-term precipitation trends in the southwest and northeastern regions. Ávila et al. [20] found similar results when analyzing extreme precipitation trends in the High Basin of the Cauca River. They observed positive trends in both 1-day and 5-day maximum rainfall amounts and an increase in the frequency of very wet days in the northern and southern regions. These regions strongly correlated with La Niña events occurring 2–3 months earlier, suggesting that decreases in SST may precede flooding in the basin by approximately three months. Likewise, da Motta Paca et al. [21] studied the Amazon River basin and showed that the annual precipitation trends have a wide spatial distribution in the Colombian sub-basins. Two sub-basins exhibited decreasing trends, and one sub-basin (western) showed an increasing trend but a maximum of 17.6 mm/yr [21].

Analyzing Colombia’s precipitation dynamics is challenging due to its complex topography and variable climate. This complexity has resulted in various climate regionalization approaches, most of which depend on a traditional definition of climate types. Early studies utilized the Köppen classification system [22]; currently, IDEAM employs a Caldas–Lang classification [23]. However, the index is incomplete because longitude and latitude are excluded, and predefined altitude values are used [24]. This provokes a lack of accuracy when analyzing localized hydrological processes. According to Canchala et al. [25], clustering methods enable detailed hydroclimatological analysis in Colombia by regionalizing rainfall by identifying areas with similar precipitation. This approach benefits local water resource management and environmental risk assessment [26,27,28].

The Magdalena department is in northern Colombia and presents a moderate vulnerability index regarding water scarcity [29]. However, the forecast for the Water Availability Index (IDH) 2040 indicates a high to very high drought threat in eight out of thirty municipalities, representing 21.6% of the area [30]. This situation could worsen due to the uneven distribution of water usage, with only 12% allocated for human consumption and a significant 88% used for agriculture, livestock, and manufacturing [31]. During the El Niño 2014 to 2016, 26 out of the 29 municipalities in the department, including the district of Santa Marta, declared a state of public calamity due to severe water shortage. This crisis affected the domestic water supply and reduced fishing productivity, compounded by issues like wetlands desiccation and soil erosion [29]. Recent studies show that 60.4% of the territory was highly vulnerable to drought conditions [32]. Municipalities located in the central zone of the department, particularly those that have already experienced significant impacts from past drought events, are of particular concern.

Studies on precipitation trends in Magdalena are limited, with most relying on the Climate Hazard InfraRed Precipitation Station (CHIRPS) dataset, which has been validated against surface station data. Arregocés et al. [33] identified varying precipitation trends across Caribbean departments. Magdalena showed decreasing precipitation trends early in the year, followed by a significant increase in rainfall during November. Also, it experienced the fewest months with significant trends, a variation likely influenced by local topography and land–sea interactions. This paper aimed to analyze long-term precipitation trends in the department of Magdalena, Colombia, with a focus on spatiotemporal variability driven by the ENSO. Multi-year precipitation records were analyzed to (a) propose a hydrological regionalization using a clustering approach of meteorological stations, (b) identify long-term trends in annual and monthly precipitation, (c) determine predominant stationary periods within the precipitation regime, and (d) link precipitation patterns to regional factors influencing local variability.

2. Materials and Methods

2.1. Study Area

The department of Magdalena is located on the northern coast of Colombia, between latitudes 11°36′58″ N and 8°56′25″ N and longitudes 73°32′50″ W and 74°56′45″ W (see Figure 1). It covers a total area of 23,188 km² and has a diverse altitude range [34]. The department features a complex geographical layout that can be divided into three main physiographic units: (1) the Sierra Nevada de Santa Marta (SNSM), (2) the Magdalena River valley, and (3) the central valley along the Ariguaní River. The Magdalena Valley, situated near the flood-prone Momposina depression, provides the best conditions for water supply, carbon sequestration, and soil fertility. In contrast, the Ariguaní Valley is crucial for the department’s agriculture [35].

The Magdalena and Ariguaní basins share morphometric characteristics and are often considered a single sedimentary mega-basin. Meanwhile, the SNSM’s geological complexity and tropical location shape local wind and precipitation patterns. Northeastern trade winds increase cloudiness and rainfall in the northern windward areas, resulting in more rainy days than other regions [36]. Conversely, the Foehn effect in the eastern regions creates a drier climate, known as the “rain shadow of the SNSM”, which stretches from the Magdalena River to the department’s center [37]. As a result, flat areas receive an annual rainfall of between 1000 and 1500 mm/year, while southern regions and areas near the SNSM exceed 2000 mm/year [23]. Additionally, the influence of the Intertropical Convergence Zone (ITCZ) leads to a bimodal rainfall pattern: two rainy seasons from March to May and September to November, and two dry seasons from December to February and June to August [38].

2.2. Database

Ground meteorological data were obtained from the network of meteorological stations operated by the Instituto de Hidrología, Metereología y Estudios Ambientales (IDEAM), at http://dhime.ideam.gov.co/webgis/home/ (accessed on 16 February 2023). This network comprises 55 stations that have recorded data from 1990 to 2022. The compiled data underwent quality control based on three criteria: (a) individual data series should have at least 30 years of availability, (b) the years included must have records from the driest and wettest seasons, and (c) the final filtered time series should not contain more than 10% missing information [39]. The analyzed variables include rainfall and rainy days, reported in monthly and annual resolutions. Rainy days are defined as calendar days where more than 0.1 mm of rainfall is recorded, according to IDEAM. A summary of the cleaned data for each station can be found in Table S1.1 of Supplementary Materials S1.

Figure 1 (right side) presents the annual precipitation data that met the quality criteria, including total rainfall and rainy days. The map shows that the wettest zones are mainly located in northern Magdalena, where precipitation values range from 1900 to 2700 mm. In contrast, the rest of the department records an average annual rainfall of 1500 to 1900 mm. Rainy days were distributed relatively evenly across the center (<100 d/year), while the northern and southern zones presented a higher average (>100 d/year). Furthermore, the data show that all stations registered positive rainfall anomalies from 2010 to 2011, with precipitation values ranging from 1450 to 6000 mm. Likewise, rainy days were in the same period (56–239 d/year). Negative anomalies were less frequent, with annual rainfall levels ranging from 50 to 700 mm and rainy days between 1 and 26 d/year. These findings illustrate significant variability in the annual precipitation regime across the department, highlighting substantial changes in the amount and duration of rainfall from 1990 to 2022. Further data distribution can be found in Supplementary Materials S1, Figure S1.1.

SST data from four Niño regions were obtained from the US National Oceanic and Atmospheric Administration (NOAA) database at https://www.cpc.ncep.noaa.gov/data/indices/sstoi.indices (accessed on 25 May 2023). The data are available at the same resolution as the IDEAM data. Brunner et al. [40] validated the temporal scale and resolution, demonstrating that a monthly resolution provides sufficient information about seasonal behavior. The Niño SSTs were selected to assess the influence of regional drivers on local precipitation variability in the Magdalena department. Previous studies have shown that Niño SST is more suitable for analyzing localized extreme precipitation events than broader indices like the Multivariate ENSO Index (MEI) [41,42,43].

The SST data show an extended cooling phase in the Equatorial Pacific from October to January, affecting the El Niño 1 + 2 and El Niño 3 regions, followed by a warming phase from February to June. In contrast, the El Niño 3.4 and El Niño 4 regions exhibited relative stability, with minimal seasonal variability throughout the study period (Supplementary Materials S1, Figure S1.2). Distinct patterns of anomalies are observed in both La Niña and El Niño events. These include precursor anomalies that show conditions opposite to those at the peak during February, March, and April (FMA), peak anomalies occurring in October, November, and December (OND), and decay anomalies appearing in June, July, and August (JJA) after the peak [44]. The geographic coordinates of each Niño region are listed in Table S1.2 of Supplementary Materials S1.

2.3. Statistical Methods

2.3.1. Identification and Validation of Homogeneous Regions

Various studies have applied an unsupervised clustering approach to identify homogeneous hydrometeorological regions [45,46,47,48]. Kaufman and Rousseeuw [49] developed the Partition Around Medoids (PAM) clustering method, which has been employed to regionalize precipitation variables in the Magdalena department. This method enhances the performance concerning outliers and noise, making it particularly effective for clustering observed hydrometeorological data [50]. PAM was implemented using two distance measures: (i) Euclidean distance, known for accurately measuring the distance between two points, and (ii) Dynamic Time Warping (DTW), which effectively manages gaps in data by aligning time series of varying lengths [49]. Distance matrices were computed using non-scaled and z-score standardized annual data, following the methodologies outlined by Gómez et al. [51] and Hosking and Wallis [52] for the unsupervised clustering of precipitation variables (see Supplementary Materials S2, Section S2.1).

The Silhouette coefficient is used to assess clusters’ similarity and dissimilarity [53]. The average Silhouette coefficient was calculated to evaluate the compactness of each cluster formed under various scenarios [54]. Stations with null or negative individual Silhouette coefficients were manually reassigned. At the same time, geographic attributes (longitude, latitude, altitude) were incorporated using the spdep package in R (v4.4.1) [55]. A total of 17 runs for each scenario were conducted to evaluate the performance of between two and eighteen clusters. The final delineation of homogeneous regions was achieved through the Thiessen polygons interpolation method and compared with the climatic classification proposed by IDEAM.

The Thornthwaite moisture index is valuable for assessing precipitation variability while minimizing the impact of other climate variables [23]. It has been adapted for use in Colombia and includes a global moisture index—comprising moisture and aridity indices relative to annual PET (Potential EvapoTranspiration). Other classification systems, such as PET, have limitations in the Tropics or do not provide better performance. Additionally, the Caldas–Lang classification system considers temperature as a key variable; however, its altitude ranges are not relevant for the study area, as nearly 86% of the region is situated below 1000 masl [23].

2.3.2. Trend Assessment

Monotonic trend detection was evaluated using Spearman’s Rho (SR) and Mann–Kendall (MK) non-parametric tests, as these methods do not require adjustments to meet normal distribution assumptions [56]. These tests are essential in hydrometeorological studies due to their unbiasedness and consistency [57]. They are highly suitable for common hydrological probability distributions [58]. Previous studies have shown that the database aligns with these distributions [59]. The Theil–Sen (TS) estimator was also used to quantify the magnitude of change. This median-based approach is less sensitive to outliers than classical least squares regression [60].

MK and SR tests were used to estimate annual and monthly trends for rainfall and the number of rainy days, which helped determine global changes in precipitation. At the same time, the TS estimator was utilized to assess the magnitude of these trends. Literature suggests no significant differences exist between Mann–Kendall and Spearman’s Rho outcomes [61]. However, McNemar’s test was performed to assess trend consistency. This test categorizes paired dichotomous responses by both methods [62]. Supplementary Materials S2, Section S2.2, provides detailed mathematical formulations for these tests based on the work of Teegavarapu [56] and WMO [63], which have been applied in previous studies [64,65].

2.3.3. Stationarity Analysis and Correlations with ENSO Indices

Spectral analysis is a widely used method for decomposing time-series data, allowing researchers to examine the underlying structure of these data by treating them as analog signals. The Continuous Wavelet Transform (CWT) was used to identify the cyclic behavior and determine the contribution of each cycle to the overall variance [66]. This approach is suitable for analyzing time series with non-stationary patterns that vary in frequency [67], such as hydrometeorological datasets. Recent studies have demonstrated the effectiveness of this method [42,68].

The Morlet wavelet was chosen for its accuracy in frequency analysis, precision in frequency analysis, and capability to effectively handle edge effects. This wavelet can analyze negative and positive oscillations, making it ideal for investigating the non-linear relationship between the ENSO and local climate conditions [69]. The CWT computation used parameters proposed by Díaz et al. [43], except for the upper Fourier period, which was set to one-third of the series length. Data preprocessing included decomposition using Loess to account for trend effects [70]. Finally, Wavelet Coherence (WC) analysis was applied to assess non-linear relationships between local precipitation and ENSO indices [71]. The signification of stationary or non-stationary behavior in precipitation patterns for each homogeneous region was determined according to the test proposed by Torrence and Compo [72]. Supplementary Material S2, Section S2.3, provides more details of these methods. Figure 2 presents a framework summarizing the methodological aspects proposed in this study.

3. Results

3.1. Homogeneous Regions by Clustering

Figure 3 compares the outcomes of four clustering scenarios for identifying homogeneous precipitation regions. All scenarios resulted in two clusters, except for the Euclidean + standardized scenario (Figure 3B), which revealed four clusters. Notably, this scenario showed a distribution of stations that aligned more closely with the department’s topography, even though geographic information was deliberately excluded from the analysis. Among the scenarios, the Euclidean + non-standardization scenario (Figure 3A) performed the best overall, successfully grouping 72% of the stations with a cluster Silhouette score of 0.34, the highest among all outcomes (the average cluster performance of each scenario is detailed in Table S3.1 of Supplementary Materials S3). Therefore, the homogeneous precipitation regions were defined based on scenario B. It is worth mentioning that Cluster 1 needed adjustments for seven stations located towards the south, as these had lower individual Silhouette scores. Cluster 3 exhibited the lowest affinity with its stations, while Cluster 2 showed good overall affinity, except for the Minca station, which had a negative Silhouette score. Table S3.2 of Supplementary Materials S3 provides detailed statistical results for the clusters.

The final homogenous precipitation regions were analyzed using Euclidean distance combined with z-scores (Figure 3B), with k = 5. This approach aimed to enhance the performance of Cluster 3 and address anti-correlations. Figure 4 illustrates the distribution of total annual rainfall data across each cluster. Table S3.3 displays a complete list of stations for each region, and the statistical results are in Table S3.4 of Supplementary Materials S3. The clustering results indicate that rainfall in the Magdalena department is primarily influenced by orographic factors. The valley physiographic unit accounts for 75% of the Central region, whereas 60% of the Western and 71% of the Southern regions correspond to the department’s plains. The Northern region is predominantly aligned with the SNSM massif, which comprises nearly 97% of its area. In contrast, the Eastern region has approximately 4% of its area designated as plains, with only two stations located within this physiographic unit.

3.2. Trend Behavior

Figure 5 shows the results of total annual rainfall trends. Only ~15% of the stations (eight in total) exhibited a significative trend in either of the two tests (MK or SR), with decreasing trends being the predominant type observed. The magnitude of these changes varied between −32 and −100 mm/decade. Also, Figure 5 depicts the monthly trends in total rainfall. The results show that April is the only month without a significant trend. At the same time, the first half of the year presents a few stations with significant trends (zero–three stations per month), except February. The year’s second half shows more significant trends (three–eight stations per month). November shows the highest number of significant trends (eight stations), with half of the same stations reported in the annual assessment. During this month, seven stations showed increasing trends, with magnitudes ranging from 9 to 30 mm/decade. Cluster region comparison, after obtaining the trends per station in each region versus the total region stations, revealed the following trends. The South exhibited the highest relative distribution for significant monthly trends with a value of 1.42, followed by the North or West at 1.14 and Central at 0.84. For annual cumulative trends, the North showed the highest relative distribution (2.7), followed by the South (1.35), Central (0.92), and West (0.68). Despite the last two regions having more stations, they showed the lowest standardization trend values per station. The Eastern region did not contribute to the annual or monthly significant trends. Detailed results of the significant monotonic trends in monthly and annual rainfall can be found in Table S3.7 of Supplementary Materials S3.

The annual analysis of rainy day trends reveals that only ~16% of the stations (specifically, nine stations) show significant trends in one of the two tests conducted. Among these, increasing trends are more prevalent (Figure 6). The magnitude of these trends varied between 0.5 and 2 days/decade. In terms of monthly trends, overall negative trends are observed from December to February; however, only ~23% of stations (three) have a non-null Theil–Sen value during this period. From August to November, increasing trends are noted, with magnitudes ranging from 0.07 to 0.3 days/decade. October has the highest number of significant trends (ten stations), followed closely by November. Also, these months exhibited the highest Theil–Sen values in the monthly assessment. For detailed results on significant monotonic trends in monthly and annual rainy days, refer to Table S3.8 in Supplementary Materials S3. Regarding regional trends, similar to accumulated precipitation, the Eastern region presented the highest annual (2.4) and monthly trend values (2.1), with many decreasing trends occurring during half of the year. The Central region showed a consistent trend value of 1.1 on both temporal scales (annual and monthly). Conversely, the Western region exhibited a relative distribution of only 0.55. Meanwhile, the Northern and Southern regions did not contribute significantly to annual and monthly trends.

Lastly, Table S3.9 in Supplementary Materials S3 shows no statistical evidence suggesting that the results from the Mann–Kendall and Spearman’s Rho tests differ. Thus, the trend assessment for hydrometeorological variables in the department of Magdalena was found to be statistically significant.

3.3. Stationary Cycles and ENSO Influence

Figure 7 shows the contributions of monthly, annual, and interannual scales to rainfall patterns in each homogeneous region derived from the wavelet power spectrum. All regions exhibit significant annual cycles (~12 months), with power values mainly ranging between 0.3 and 0.6. The Northern region displays the highest annual contribution (69.2% of power values) between 0.6 and 0.9. Statistically significant patterns at the semiannual scale were observed only in the Central and Northern regions. The Eastern region exhibited irregular patterns, alternating significant peaks at the 5% level, while the Western region showed a general decline in semiannual power.

The Central, Northern, and Western regions experienced non-stationary periods at a three-month scale during the year’s first half, with increasing power observed in the latter half. In contrast, the Eastern and Southern regions show erratic power behavior at the same temporal scale. Annual cycles were prominent across all regions, with significant cycles between ~6.7 and ~11.6 months. The Eastern region again displayed peak components between ~4.1 and 5.6 months. Interannual stationarity was associated with ~60 to ~84-month (5 to 7 years) scales in all regions except the South, which had significant cycles at ~32 months (or 2.7 years).

Figure S3.1 displays the monthly mean SST anomalies in El Niño 3.4 region during 1990–2022. The figure marks each event (as either “El Niño”, “La Niña”, or “normal”), along with its magnitude represented by the three-month mean value of the ONI index. Except for the Eastern region, all regions showed power peaks during 2010–2012, aligning with a “La Niña” event. The Northern region had an additional power peak between 2002 and 2004, marking the beginning of an extended “Normal” period. Following this, there was a decline in power after 2005, with a rapid increase from 2008 to 2009, corresponding to another “La Niña” phase.

Wavelet coherence analysis revealed a strong relationship between local precipitation cycles and monthly SST in El Niño regions. The monthly average SST from the El Niño 1 + 2, 3, 3.4, and 4 regions explained approximately 49.5 ± 6.2%, 50.5 ± 5.3%, 49.9 ± 4.8%, and 48.5 ± 4.6% of precipitation variability, respectively, with 95% significance. These results indicate a consistent influence of the ENSO across similar precipitation regions. El Niño 3.4 showed the highest synchronization with local precipitation, particularly at the semiannual (41.8%) and three-month (40.1%) scales, with a phase difference of π/6 or less. Additionally, El Niño 3 exhibited the most statistically significant coherence on an annual scale (21.0% of cycles) and showed robust interannual consistency with SST as a precursor, suggesting a high predictability for this regional factor.

Figure 8 shows the scalogram depicting the correlation between monthly, annual, and interannual cycles of local rainfall and the SST in the El Niño 3 region. The coherence values indicate a strong connection between SST and local precipitation at semiannual and annual scales. However, the lead/lag behavior varies among regions. The Central region shows a complete synchronization at an annual scale. In the Southern and Western regions, precipitation lags by approximately six months during regular periods and has a three months’ lead during La Niña events. The North region shows stronger negative correlations during intense El Niño events. Meanwhile, in the Eastern region, phase synchronization occurs during strong La Niña and El Niño events, with precipitation leading SST by about six months during strong El Niño events while lagging for the same duration during normal conditions. At the semiannual scale, El Niño 3 generally follows precipitation by up to 23 days during El Niño events. This pattern is consistent across all regions except Central, where precipitation leads throughout the period. Complete synchronization is observed only during La Niña events. Significant interannual correlations of around 30 months (2.5 years) were observed, with a lead time of 15 months (1.25 years) in all regions. The Southern region showed greater coherence at time scales of 40 to 70 months (3.5 to 5.8 years), with peaks occurring between 5.3 to 5.5 years. This pattern was particularly evident during strong La Niña events, specifically from 1995 to 2000 and 2012 to 2017, followed by stronger El Niño events before transitioning back to La Niña conditions within about 2.5 years. Additionally, the Eastern region showed a precipitation response during moderate to normal events. For more detailed illustrations, refer to Figures S3.2–S3.4 in Supplementary Materials S3, which display scalograms that show the correlations between monthly, annual, and interannual cycles of local rainfall and mean SST for the El Niño 3.4, El Niño 1 + 2, and El Niño 4 regions.

4. Discussion

The clustering-based approach successfully grouped areas with similar precipitation features, reflecting the topographical complexity even before geographic features were considered. These findings are aligned with recent regionalization studies on rainfall in orographically complex regions [73,74]. Figure 9 compares the Thornthwaite classification with the regionalization proposed in this work. A key difference observed was the underestimation of the eastern limits of the Northern region. This underestimation arises from the geographic distribution of the monitoring stations, particularly in the Moderately Humid zones. This distribution highlighted the need for the better representation of stations in the humid and perhumid zones, which currently contain only three out of fifty-four stations, with just one station situated above 1500 masl. Quiroga-Sánchez [75] confirms this observation, noting the absence of a high-altitude monitoring station in the IDEAM network. The lack of stations restricts the ability to conduct time-series analyses in high-altitude areas and to develop management strategies for critical zones, such as the headwaters that supply lowland basins [35].

The regionalization proposal received validation despite some limitations. The annual cumulative precipitation per region corresponded with the categories outlined by the Thornthwaite index, indicating multiannual rainfall variability across the department. Regions with a higher density of weather stations, specifically the Central region (41%) and the Western region (30%), showed a relatively even rainfall distribution. They accurately identified the shift in climate regime between them. IDEAM [23] states that Dry and Moist subhumid types dominate the central agricultural areas, while Semiarid and Arid types are prevalent along the central Caribbean coast. In the clustering approach, the Western region mainly consists of Semiarid types, while the Central region, found in the Ariguaní Valley, aligns with the Dry and Moist subhumid types. Likewise, Slightly Humid, Moderately Humid, and Humid types were identified in the foothills of the SNSM, which aligns with IDEAM’s classification of these types in the Andean foothills. Bedoya-Soto et al. [76] noted that orographic lifting significantly increases precipitation when moist winds encounter the mountains, leading to considerable variability in rainfall among nearby rain gauges.

The proposed regionalization revealed variability in annual rainfall across five regions, showing a trend of increasing frequent years with below-average rainfall. This trend corresponds with a decline in overall rainfall totals, ranging from −32 to −100 mm/decade, emphasizing the region’s vulnerability to drier conditions [32]. The trend assessment also identified significant long-term changes in annual and interannual precipitation patterns in the Magdalena department, particularly concerning annual behavior. Morales-Acuña et al. [77] found similar results using the CHIRPS v2 monthly database, noting significantly increasing monthly trends in October and November. Additionally, annual increases were mainly observed in the Southern and Western regions, while decreasing trends of −10 mm to −30 mm/year were recorded near the center of the SNSM, consistent with the results in the Eastern and Northern regions. Climate change scenarios predict severe droughts in the SNSM, making these identified trends even more critical [78]. Results from Mesa et al. [79] further confirm the likelihood of droughts, with MK and TS tests showing significantly increasing trends in annual precipitation (79%) and the number of rainy days (80%), surpassing the findings of the present study. The authors also concluded that there is an inverse relationship between total precipitation and the annual number of rainy days [79], suggesting a shift toward fewer but potentially more intense rainfall events in the Magdalena department.

Long-term trends in river discharge in the department show that streamflow directly responds to hydrological fluctuations [69]. Restrepo et al. [80] identified an increasing trend in annual streamflow within the SNSM watersheds from 2000 to 2010, linked to a significant quasi-decadal cycle. This rise in river discharge, combined with the observed decreasing precipitation trends in the Northern (SNSM) region, may signal changes in the area’s water supply, potentially posing risks to agricultural practices [31]. Moreover, Hoyos et al. [81], in their study of the Río Frío watershed, found that rainfall deficits, likely due to these decreasing trends, can lead to reduced water yields. These impacts can lag by up to nine months, with deficit levels ranging between 50% and 80%, depending on factors such as land cover. Furthermore, Damseaux et al. [82] showed that most Regional Climate Models tend to overestimate precipitation in areas with high elevation variability. This study’s trend assessment provides a useful baseline for future drought risk scenarios in Magdalena.

There is currently a lack of consensus among researchers regarding the rainfall regime in the Caribbean region of Colombia. The stationary cycles identified in the present study support the finding of Urrea et al. [83] related to annual rainfall seasonality in northern Colombia. In Magdalena, a robust unimodal regime was observed in both the Northern and Southern regions, while mixed seasonality was noted in the Eastern, Central, and Western regions. Fluctuations between strong and weak rainfall events occur on a three-month scale twice a year, suggesting a quasi-bimodal rainfall pattern consistent with the work of Espinoza et al. [84]. The pattern typically shows lower monthly rainfall in the year’s first half than in the second [69].

WC revealed a six-month cycle advantage of SST in El Niño 3 over local precipitation on an annual scale in the Southern and Western study areas. This finding indicates an inverse correlation between El Niño (La Niña) and the lowest (highest) local precipitation records. These results correspond with Cerón et al. [85], who identified the Andean and Caribbean regions as directly impacted by large positive SST anomalies. However, results show that ENSO-driven changes do not impact rainfall variability annually and semiannually, despite the Eastern and Western regions’ proximity to the Caribbean Sea. Amos and Castelao [86] explained this phenomenon by suggesting that SST distribution is influenced by coastal topography. The Northern region exhibits negative correlations, with anti-phases occurring only during strong El Niño events. This observation is consistent with recent research on ENSO’s influence on local climate variability in other subtropical or tropical altiplanos [87,88].

The interannual coherence between El Niño events and the cluster area is linked to the inherent stability of the ENSO phenomenon. Peak events like La Niña and El Niño typically follow a distinct 2–3-year cycle, returning to the initial phase every 4 to 6 years [44]. Generally, each event signals a transition towards the opposite phase, except in cases where La Niña persists for a second consecutive year [89]. Nevertheless, it is crucial to note that while ENSO significantly influences local climatic variability, it is not the sole driver. Other oceanic regional anomalies, such as the Tropical North Atlantic, are also impacted by ENSO. This interaction can lead to moisture divergence and reduced rainfall, as noted by de Souza et al. [90].

5. Conclusions

The clustering approach effectively identified five distinct rainfall regions within the Magdalena department, which closely align with IDEAM’s Thornthwaite index classification. This alignment is particularly evident in the Central and Western regions, with denser station coverage. The findings underscore the necessity of expanding meteorological networks in the northern and eastern areas, where limited station coverage hampers the accuracy of boundary definitions. Despite these limitations, the regionalization successfully captured the main precipitation patterns characterized by lower annual rainfall averages.

Significant long-term trends were identified, with most stations reporting annual declines in precipitation. Monthly trends indicate a decrease in the number of rainfall events, but an increase in their intensity. These shifts in precipitation totals could increasingly strain regional water resources, especially in the upper SNSM areas.

Annual precipitation patterns were consistent across regions; however, a closer examination of monthly data revealed varying cyclic patterns, particularly during the first half of the year, suggesting a quasi-bimodal rainfall regime.

Lastly, an analysis of regional drivers revealed a strong correlation with ENSO phases, notably with El Niño 3.4 (synchronization) and El Niño 3 (anticipated response). These findings highlight the importance of understanding precipitation trends and climate influences for effective risk management in the Magdalena department, aiding in preparation for potential drought and flood scenarios.