Agricultural planning grounded in regional rainfall patterns in the Colombian Orinoquia: An essential step for advancing climate-adapted and sustainable agriculture

2.1 Study area

The Orinoquia Region in Colombia covers an area of 346,072 km², accounting for 35% of the total Orinoco basin [20]. This region is characterized by a diverse range of landscapes, including mountains, foothills, and plains (such as alluvial, eolian, peneplains, and highlands). The Eastern Cordillera of the Andes marks the western boundary of the study area, while the La Macarena Mountain range defines its southwestern limits (Figure 1).

Figure 1

Location of the study area and landscape characterization. Map created by the authors based on information from IDEAM (http://www.siac.gov.co/catalogo-de-mapas).

The rainfall pattern in the Colombian Orinoquia is primarily monomodal [21]. The pattern is a response of the influence of the ITCZ which drives a single annual peal in precipitation in the region (Figure 2). However, some researchers suggest variations in local rainfall distribution patterns that create areas with different lengths of dry periods [8,22]. Spatially related to the predominant eastern winds, the highest annual rainfall is found in the mountainous regions and foothills of the eastern cordillera and near the La Macarena Mountain range. Conversely, the areas with the lowest annual rainfall are in the central and northern parts of the Orinoquia (Figure 2). According to the Regional Integrated Climate Change Plan for the Orinoquia (PRICCO) [23], this less rainy area is characterized by a low number of rainy days and a prolonged duration of low rainfall.

Figure 2

Rainfall characterization: intra-annual rainfall regime and annual isohyets. Triangles indicate the location of weather stations, and the number beside each triangle corresponds to the station ID listed in Table 1. The characterization is based on precipitation data from the period 1983–2022. Map created by the authors using information from IDEAM (http://www.siac.gov.co/dhime).

2.2 Data

The daily precipitation data used in this study come from standard meteorological stations operated by IDEAM. Since most of the regional stations were installed between 1972 and 1982, the analysis period was established between 1983 and 2022. Those stations with more than 30% missing monthly data were excluded from the database, resulting in a final selection of 62 stations (Table 1).

The limited number of weather stations throughout the expansive Colombian Orinoquia necessitates reliance on daily precipitation data from CHIRPS [15] (https://www.chc.ucsb.edu/data/chirps/). After evaluating other global precipitation datasets such as GPCC [24], ERA5 [24], TRMM [25], and PERSIANN [26], CHIRPS v2.0 was selected based on several criteria: (i) temporal alignment with our study period (1983–2022); (ii) daily temporal resolution, which was not available in GPCC and limited in TRMM; (iii) higher spatial resolution (0.05°), which is especially important in a region with complex rainfall patterns; and (iv) previous validation studies in Colombia showing good performance. While a full intercomparison is beyond the scope of this study, CHIRPS was the only product that consistently met all operational requirements for this study. Furthermore, its reliability has been reaffirmed across various latitudes, including extensive studies conducted in Colombia (Paredes-Trejo et al., [27]); [18,19,28–33].

Although CHIRPS uses data from in situ stations, the stations and periods included in its interpolation process vary from month to month, as documented on its official platform (https://data.chc.ucsb.edu/products/CHIRPS2.0/diagnostics/list_of_stations_used/monthly/). In our case, the observed dataset was provided by IDEAM and includes a fixed group of stations with complete records from 1983 to 2019, selected based on data availability, continuity, and spatial coverage. These stations are not necessarily the same as those used by CHIRPS in its calibration process. Therefore, even if there is some degree of overlap, the comparison remains relevant for evaluating CHIRPS performance over the region and time period of interest.

2.3 Monthly bias adjustment of CHIRPS rainfall estimates

To correct the bias present in the CHIRPS precipitation data, we applied a monthly multiplicative correction [34] using in situ rainfall observations from 62 meteorological stations. For each station, we identified the nearest CHIRPS grid point based on geographic coordinates (latitude and longitude). We then calculated a monthly bias ratio as the quotient between observed and CHIRPS-estimated precipitation values as equation (1).

where P obs , i , m is the rainfall observed at station i in month m , and P CHIRPS , i , m is the CHIRPS value at the corresponding pixel.

These ratios were computed for each month and station over the entire study period (1983–2022). The resulting bias values were then interpolated spatially using inverse distance weighting [35] to create continuous monthly bias surfaces across the CHIRPS grid. Finally, each CHIRPS estimate was corrected by multiplying it with the interpolated bias ratio for the corresponding location and month and is given as equation (2).

This correction allowed us to adjust the satellite-derived precipitation values to better reflect the observed rainfall patterns in the study region, providing a more reliable dataset for regionalization and climatological analysis. The CHIRPS dataset resulting from the application of the bias correction method is referred to hereafter as “CHIRPS corrected.”

2.4 Validation of CHIRPS

Three statistical metrics: the Spearman’s rank correlation coefficient (ρ) [36], the PBias coefficient [37], and the root mean square error (RMSE) [38] were used to compare the CHIRPS and in situ station precipitation time series. The analysis was performed monthly, allowing the evaluation of the certainty of CHIRPS data at different times of the year.

Looking for a monotonic relationship between the series, a correlation level (ρ) higher than 0.5 was established as an acceptance criterion without establishing causality. The PBias was used to identify whether the CHIRPS data overestimated or underestimated the actual values; in this case, the criterion defined by Moriasi et al. [39], who considered values between −25 and 25% acceptable, was followed. As for the RMSE, low values were desirable; however, it is crucial to consider the natural variability of precipitation throughout the year. Table 2 provides a mathematical description of the statistical metrics used.

2.5 Regionalization of the precipitation

The rainfall seasonality of the Orinoquia was regionalized using the k-means regionalization method, an unsupervised clustering algorithm that conglomerates individuals into k groups based on their characteristics.

The k-means algorithm (Li & Wu, [40]) clusters data by attempting to separate samples into k groups of equal variances, minimizing a criterion known as inertia or sum of squares within the cluster. This clustering algorithm fits a large number of samples well and has been used in a variety of knowledge areas, including meteorology and climatology [41,42]. The clustering classified each pixel based on the type of intra-annual rainfall pattern identified. Prior to applying the clustering algorithm, the monthly precipitation values from the bias-corrected CHIRPS dataset were standardized to eliminate the influence of magnitude and focus the grouping solely on seasonal patterns. The algorithm begins by choosing k initial points, known as centroids, representing the clusters’ initial centers. Each CHIRPS pixel is assigned to the cluster whose centroid is closest in terms of the Euclidean distance metric defined as equation (6).

where d ( x i , c j ) is the distance between the point x i (monthly precipitation) and the centroid c j of cluster j . p is the number of variables or dimensions, in this case it is the monthly precipitation.

Once all the points have been assigned to a cluster, the centroid of each cluster is recalculated by averaging the positions of all the points assigned to that cluster. The new centroid c j is calculated as using equation (7).

where C j is the centroid of the cluster j ; | C j | is the number of points in the cluster j ; x i is the precipitation values in the cluster j .

The assignment and recalculation steps are repeated iteratively until the centroids do not change significantly.

To determine the most appropriate number of clusters, three complementary evaluation methods were applied: the Elbow Method (based on inertia), the Silhouette Score, and the Davies-Bouldin Index. The Silhouette Score [43] measures how similar a point is to its own cluster compared to other clusters, ranging from −1 to 1; higher values indicate well-defined clusters. The Davies-Bouldin Index [44] evaluates the average similarity between each cluster and its most similar one, with lower values indicating better separation between clusters. The main idea is to identify the point on the graph where the reduction in the sum of the squared errors (or inertia) starts to decrease more slowly, which resembles an “elbow.” From this point on, adding more clusters does not significantly improve cluster compactness. The k-means method was applied from k = 1 to k = 20 regions considering the size of the database, the study area, and the variability of the data. For each value of k, the sum of the squared errors or intra-cluster inertia, which measures the dispersion of the points within the clusters, was calculated. The inertia is calculated using equation (8).

where x i is the feature vector of each point; C j is the centroid of the cluster j ; k is the number of clusters.

Finally, the inertia obtained for each value of k is plotted. The x-axis corresponds to the number of clusters (k), and the y-axis represents the inertia. Together with the Silhouette Score and Davies-Bouldin Index, these metrics supported the final decision regarding the optimal number of rainfall regions. MacQueen [45] further describes the k-means method and the elbow method.

2.6 Definition of planting times

Under equatorial conditions, understanding the timing of the dry and rainy seasons, as well as soil conditions, is crucial for deciding when to plant crops [12]. In the Orinoquia region, the primary characteristic of soil is its low moisture retention capacity, particularly in flat areas, along with significant water loss due to surface runoff throughout the territory, which includes plains, piedmont, and mountainous regions [5,46]. The timing of the rainy and dry seasons also accounts for the transition periods that occur between these seasons, as noted by Gallo et al. [9] for locations situated between the foothills and highlands.

It is logical to define seasonal zones based on monthly weather patterns, particularly the amount of accumulated rainfall and the likelihood that evapotranspiration will exceed the average monthly reference evapotranspiration (Peña et al., [47]). For example, if there is a noticeable increase in rainfall in March compared to February, but March’s total is below 150 mm, then April will be designated as the first month of the rainy season. However, to officially mark April as the beginning of the rainy season, the amount of rainfall must also be considered. Suppose April’s total is just above 150 mm. In that case, it will indicate a transition period occurring in mid-April, suggesting that the majority of rainfall will likely happen in the last week of the month. Conversely, if April’s rainfall surpasses 150 mm, this will indicate a short transition period early in the month, marking mid-April as the start of the rainy season. The threshold of 150 mm was established based on agronomic studies conducted in the Colombian Orinoquía [47], which indicate that cumulative monthly precipitation below this value is generally insufficient to maintain adequate soil moisture for crop establishment. This threshold is consistent with recommendations for rainfed cropping systems in soils with low water retention capacity.

Optimal planting times for both transitional and perennial crops in each region were established by considering intra-annual rainfall patterns and local biophysical conditions. Since rainfall is crucial for agricultural productivity under rainfed conditions, planting dates were determined based on their impact on soil moisture [9]. For transitional crops like corn, soybean, and rice, which support regional agroindustry and national food security, planting dates were chosen to reduce the risk of water shortages during critical growth stages (Peña et al., [47]). In contrast, determining optimal planting dates for perennial crops, including grasses, palm, cocoa, or coffee, products of significant importance in Colombia’s international trade largely depends on water availability, which is essential for their growth and development. Young plants are especially vulnerable to water deficits; therefore, planting during the rainy season promotes their establishment and strengthens them for the upcoming dry season [12].

3.1 Validation of the precipitation data

The monthly cumulative precipitation data generated by CHIRPS modified in the flat zone of the Orinoquia (plain, alluvial plain, aeolian plain, peneplain, high plateau; Figure 1) show a strong positive correlation (ρ) with surface precipitation measurements (Figure 3). This suggests that CHIRPS modified effectively represents the monthly surface measurements, highlighting its potential usefulness for agricultural decision-making. In contrast, the performance of CHIRPS modified in the mountainous zones, certain piedmont areas of the western Orinoquia (Mountain, Foothills; Figure 1), and the southwestern region (La Macarena Mountain range) was less satisfactory, exhibiting predominantly positive but low correlations. From a temporal perspective, CHIRPS modified performed better during the dry months of December, January, February, and March, with correlation coefficients exceeding 0.9. However, during the rainy season (April to November), its performance was suboptimal, with Spearman correlation coefficients mostly around 0.5 in five rain gauges. CHIRPS is particularly effective during dry seasons, as it was specifically designed to monitor drought conditions. It demonstrates a higher accuracy in detecting subtle variations in precipitation deficits and can identify extended patterns of water scarcity. Other researchers in Colombia have also reported the strong performance of CHIRPS in dry seasons, notably in the geographic valley of the Cauca River [19] and southwestern Colombia [18].

Figure 3

Spatial representation of Spearman’s correlation coefficient (ρ) obtained by relating monthly CHIRPS modified precipitation data and surface measured precipitation. Map created by the authors.

Considering the bias correction applied to the CHIRPS dataset, all stations reported PBias values between −20% and 20% (Figure 4), with the exception of three stations located in the La Macarena Mountain range (southwestern region), which showed PBias values close to −40% during the rainy months (May to August).

Figure 4

Spatial representation of the PBias obtained by relating monthly CHIRPS modified precipitation data to surface measured precipitation. Map created by the authors.

RMSE values remain below 100 mm during the December to March period in five rain gauges (Figure 5). This suggests that during a time when little or no rain is expected (a relatively normal scenario), the total rainfall reported by CHIRPS modified can inaccurately reach up to 300 mm. Conversely, from April to September, the RMSE indicates errors close to 200 mm in four rain gauges, which is linked to the significantly higher rainfall during this time. Overall, the RMSE remains relatively low (under 50 mm) when considering the natural variability of rainfall data (standard deviation) in this country region.

Figure 5

Spatial representation of the RMSE obtained by relating monthly CHIRPS modified precipitation data and surface measured precipitation. Map created by the authors.

The analysis results reflected in the PBias, Correlation (ρ), and RMSE values obtained when comparing CHIRPS modified with data from precipitation stations indicate that using CHIRPS data is a viable and reliable option for agricultural decision-making at any time of the year. The certainty associated with the data is considered medium to high, which supports its usefulness in planning and managing agrarian activities.

The modified CHIRPS precipitation data are provided in this article as Supplementary Material 1.

3.2 Regionalization of precipitation in the Orinoquia and definition of sowing times

The analysis indicated that five groups (zones) for classification rainfall patterns in the Orinoquia are optimal (Figure 6). This grouping into five categories (zones) allowed for a more accurate classification of the interannual rainfall pattern, reducing inertia and achieving high cohesion within each group. According to the Elbow method, the inertia decreases at a nearly constant rate beyond five clusters. The Silhouette Score shows a significant improvement at k = 5, suggesting better-defined clusters without unnecessarily increasing the number of groups. Similarly, the Davies-Bouldin Index begins to decline from k = 5, indicating improved compactness and separation between clusters, with optimal values occurring between k = 5 and k = 7.

Figure 6

Evaluation of optimal number of clusters using Inertia, Silhouette Score, and Davies-Bouldin Index.

Accordingly, the Orinoquia is divided into five distinct zones, each with a characteristic annual rainfall pattern (Figure 7).

Figure 7

Homogeneous rainfall regions identified and intra-annual regime for each region. Map created by the authors.

Region 1. This extensive area is in the southwest of the region, encompassing the Macarena Mountain range, as well as the southwestern of the study area in the Hilly Terrain and Mountain (Figure 1). Rainfall typically accumulates between April and June, with May being the month that sees the highest levels. In this region, March often receives more than 200 mm of rainfall on average. Therefore, sowing can safely begin in mid-March with a low risk of reduced rainfall later on. However, sowing before March 15 carries a higher risk, as historical water balance data indicates that early March sowings can lead to water deficits during the initial vegetative stages of the crop, as noted by Gallo et al. [9]. Conversely, sowings conducted after the end of June may face dry conditions during critical crop growth stages, a situation that varies significantly from year to year (Figure 8).

Figure 8

Rainfall patterns defined for the five homogeneous zones determined (top) ideal dates for sowing transitory crops based on the risk of having a water deficit at ground level in each zone (middle) and dates for sowing perennial crops based on the risk of not reaching optimal development that allows facing the dry season (bottom).

Region 2. This zone covers alluvial plain, aeolian plain, and a part of the foothills of the region’s northwest. The rainy months are May and June. However, the rainfall in April is also significant, allowing for crop establishment after mid-April. In this region, young crops have a lower risk of experiencing water deficits if they are planted after April 15. This means that sowing earlier, such as in early April, carries a higher risk of moisture-related losses. Therefore, it is considered very risky to plant before the specified date of April 15. Similarly, planting after the first half of June increases the likelihood that the dry days during the transition season will negatively impact productivity. As a result, planting beyond the defined date for this region is also regarded as high risk (Figure 8).

Region 3. This region, located in the northeastern end of the Orinoquía, mainly in the High Plateau (Figure 1). Its pattern starts at the end of April or the beginning of May and ends at the end of November/beginning of December. The rainy months are June and July, with April being a transition month. In this region, the first significant rains typically occur in May, which is why plantings in April, including those made at the end of the month, are not recommended. This pattern is due to the high risk of affecting productivity, given the dry conditions that may occur in some years. In this region, the planting window is shorter than in any other region of the Orinoquia, as plants planted after the second week of June may be subject to dry conditions during the last phenological stages, particularly for annual crops (Figure 8).

Region 4. This region is the smallest and is located in the northwest of the Orinoquia, covering the Foothills area (Figure 1). The agricultural zones located in Region 4 are in stark contrast to those in Region 3, basically because they are subject to high rainfall conditions from April through November. In this sense, it is possible to establish plantings from the beginning of March with a low risk of precipitation reductions after planting, which can occur if planting is done before the indicated date. In addition, the planting window in this Region is wider than in the rest of the Region since it is possible to establish plantings until the first days of July with a low probability of affecting yields of annual species, which are vulnerable to water deficit in late phenological stages prior to crop consolidation (Figure 8).

Region 5. This area encompasses the south-eastern and south-central part of Orinoquía, in the peneplain and hilly terrain, respectively (Figure 1). The weather conditions from May through July have been consistently rainy. Notably, April also experiences significant rainfall, like Region 1. This allows for planting to begin in mid-March, with a low risk of rainfall shortages after planting. As a result, the likelihood of planting failures due to reduced rainfall during critical periods of growth for both annual and perennials is minimized. Given the high rainfall in June and July, along with continued precipitation expected until November – similar to the conditions in Region 4 – planting in Region 5 can occur with a low risk of drought-related losses until the beginning of July (Figure 8).