Site description

This study was conducted in the Montes de María (MM), also known as Serranía de San Jacinto, a low-elevation mountain range located near the Caribbean coast, north of the Colombian Andes. The MM region is the primary avocado-producing area in the Colombian Caribbean, particularly for the West Indian avocado (Persea americana var. americana). The agricultural landscape in MM is characterized by agroforestry systems where avocado and cacao orchards are frequently intercropped with yam (Dioscorea sp.) (^{Burbano-Figueroa, 2019}). These agroforestry systems resemble secondary forests with high trees (>15 m) and old avocado trees that are randomly distributed throughout the landscape. Producers intervene minimally in the crop, primarily acting as gatherers rather than active farmers. In these systems, avocado trees serve multiple purposes, including providing shade for other crops, fruit for consumption, and a secure source of income. Farming practices in MM are typified by non-technified subsistence agricultural economies (^{Yabrudy, 2012}).

The MM region falls within the tropical dry forest biome, characterized by prolonged dry seasons and periodic droughts. This region experiences a tropical savanna climate with dry winters (Aw), influenced by the elevational variations of the MM massif. The area exhibits a bimodal rainfall distribution, with a distinct dry season and a mean annual temperature of 27°C. The geographical range of the farms studied lies between latitudes 9.64º and 9.97º and longitudes -75.31º and -75.15º, with altitudes ranging from 175 to 566 m a.s.l., spanning the departments of Bolivar and Sucre (Fig. 1).

Figure 1. Spatial distribution of sampling sites (red dots) across avocado farms in the Montes de María region, Colombia. Black dots and labels indicate the locations of urban centers within the municipalities of Montes de María (Colombia). 

Sample collection

Soil samples were collected from 32 avocado orchards in the MM region between 2013 and 2015. These farms were selected based on the presence of avocado decline symptoms, with the aim of investigating the relationship between soil properties and disease severity. For each farm, six soil samples were collected from different trees, representing a range of avocado decline symptoms. The sampling procedure involved collecting soil from two specific locations around each tree: the middle and the outer edge (border) of the root zone. The middle refers to the area near the trunk, where root density is typically higher, and the border represents the outer extent of the root system, near the canopy edge. This approach ensured that soil conditions influencing tree health were captured accurately.

Soil samples were collected at a depth of 0-20 cm from both the middle and the border of the root zone. These samples were placed in plastic bags, properly labeled, and stored under cool conditions (10-12°C) during transport to the site of analysis. A range of chemical and physical analyses were performed on these soil samples (Tab. 1), including pH, calcium content, phosphorus levels, organic matter, and texture (Tab. 1S; see supplementary tables https://osf.io/pa8bs/).

For hydraulic conductivity measurements, specific soil samples were collected separately using metal rings to preserve the soil structure. These intact soil cores were carefully extracted from the field to minimize disturbance. The samples were then transported to the site of analysis while still inside the metal rings to maintain the integrity of the soil structure. In the laboratory, hydraulic conductivity was assessed using a constant-head permeameter, a method that measures the soil's ability to transmit water when it is saturated. This technique provides key insights about water movement through the soil, which is important for understanding its impact on tree health, especially in relation to root diseases like avocado decline.

In addition to soil analysis, the severity and incidence of avocado decline were recorded for each farm. The study determined the prevailing level of disease using the highest recorded data on the incidence and severity of avocado decline from at least three different transects per farm.

Assessment of avocado decline

The severity of avocado decline on each farm was visually estimated along 100-m transects using a progressive symptoms scale. These transects were placed in areas reported by farm owners as having the highest levels of tree decline. Soil samples were collected from the same areas where decline symptoms were observed. The transect observation method was chosen due to the random distribution of trees on highly rugged terrain.

The severity scale included three levels:

The severity index (SI) was calculated by summing the frequencies of each symptom level, with scores ranging from 1 to 3, and assigning 3 to the most severe level (SDB).

Exploratory data analysis

Correlations between 19 soil properties and the occurrences of SI and SDB were estimated using Pearson's product-moment correlation coefficients. SDB occurrence was selected for further analysis over SI because it showed less variation related to visual estimation by humans (based on field assistant calibration before data collection) and was strongly correlated with overall decline severity. The data revealed a high correlation (r² = 0.86) between SDB and SI. A previous study in MM with a larger sample size showed a similar trend (^{Osorio-Almanza et al., 2017}).

The relationships between soil parameters and SDB occurrence were analyzed using principal component factor (PCF) analysis with Varimax rotation. Loading values with an absolute value of 0.35 or higher were considered significant for intercorrelation among variables within a principal component. We chose a threshold of 0.35 for the factor loadings in our PCF analysis to strike a balance between statistical rigor and biological relevance. A loading of 0.35 explains roughly 12% of the variance, which is meaningful in ecological studies where variability is high due to complex environmental interactions. In similar research contexts, thresholds ranging from 0.30 to 0.40 have been widely used to capture key associations without being overly restrictive (^{Hair et al., 1998}; ^{Tabachnick and Fidell, 2013}).

Given the complexity of soil-plant-pathogen interactions, a more stringent threshold could have excluded important variables from our analysis, potentially omitting meaningful insights into avocado decline. This value allows us to identify relevant soil parameters, such as Ca and pH, that contribute to disease dynamics, while ensuring that our findings remain interpretable within the scope of the study (^{Pituch and Stevens, 2015}).

Soil properties are known to influence the interactions between pathogens, pests, and their hosts, potentially explaining the local variations in tree decline and mortality (^{Sterne et al., 1977}; ^{Sterne et al., 1978}; ^{Larkin, 2015}; ^{Ababa, 2024}; ^{Cao et al., 2024}; ^{Chaudhary et al., 2024}). However, the extensive variability in soil properties between locations complicates our understanding of their effects on soil-borne diseases and their specific interrelations. To address this complexity, PCF analysis is a valuable tool for identifying general trends and correlations between biologically relevant soil properties, distilling them into a smaller number of factor components (^{Sano et al., 2015}; ^{Lima et al., 2017}; ^{Klimkowicz-Pawlas et al., 2019}; ^{Buss et al., 2019}; ^{Vaca et al., 2023}; ^{Yadav et al., 2023}). By applying Varimax rotation, these components can be orthogonally transformed to align the principal components with observed data, enhancing the interpretation of results by maximizing the correlation between underlying latent constructs and observed variables (^{Vasu et al., 2016}).

PCF analysis has proven effective in understanding soil conditions related to plant diseases, both at the experimental-field level and the farm level (^{Khaledian et al., 2016}; ^{Li et al., 2017}; ^{Naseri and Tabande, 2017}; ^{Rangel-Peraza et al., 2017}; ^{Arcoverde et al., 2019}; ^{Buss et al., 2019}; ^{Indarto et al., 2022}). It is also widely used to identify underlying soil parameter structures that define soil quality (^{Mukherjee and Lal, 2014}; ^{Vasu et al., 2016}) and to assess the impact of land-use changes on soil health (^{Cherubin et al., 2016}; ^{Raiesi, 2017}).

Predictive models for explaining the effect of soil parameters over SDB

To determine the relationship between soil parameters and shoot dieback (SDB) severity, we conducted a stepwise multiple regression analysis. Stepwise selection, an iterative method, was utilized to select the most important predictors. The analysis started with all variables, and the best subset of variables was chosen by minimizing the Akaike Information Criterion (AIC), which balances model fit and complexity. The final model was established after eliminating insignificant variables based on their AIC values. This model aims to identify the most significant soil properties contributing to SDB. The resulting regression coefficients serve to quantify the impact of each variable on SDB.

In addition to stepwise regression, we used ridge regression to address multicollinearity among soil variables, which can lead to unstable coefficient estimates in ordinary least squares regression. Ridge regression incorporates a penalty to the regression coefficients, shrinking them toward zero but never completely eliminating them. This method allows for the inclusion of all predictor variables, regardless of multicollinearity, and provides a more robust model for predicting SDB severity. Before applying ridge regression, all soil parameters and the SDB index were standardized to have a mean of zero and a standard deviation of one. The optimal penalty term (lambda) for the ridge regression model was determined through cross-validation, ensuring the best fit for the data. The final model included coefficients for all soil parameters, providing insights into their relative influence on SDB severity while accounting for the presence of multicollinearity.

Both models were implemented in R using the MASS package for stepwise regression and the glmnet package for ridge regression. Model performance was assessed by calculating the mean squared error (MSE), and model coefficients were interpreted in the context of their biological relevance to avocado decline in the MM region.

Diversity of soils across the Montes de María

The avocado farms in the MM region exhibited wide variability in soil texture, ranging from heavy clay to lighter loam textures. Several farms presented intermediate textures, such as clay loam and sandy clay loam. Figure 2 illustrates these variations using a soil texture triangle, showing that the majority of soil samples were classified as clay and clay loam.

The pH levels of the soils also varied widely, spanning from moderately acidic (5.82) to moderately alkaline (8.19). This broad pH spectrum underscores the diverse soil conditions across the sampled farms. In addition to pH, the analysis revealed significant variability in the levels of calcium (Ca) and other exchangeable cations, measured by the effective cation exchange capacity (ECEC). The calcium concentrations ranged from 2.29 to 43.80 cmol₍₊₎ kg^-1, while the ECEC values varied from 11.66 to 46.68 cmol₍₊₎ kg^-1.

Figure 2. Soil types of avocado farms at Montes de María, Colombia. 

Multivariate approaches to identifying key soil factors influencing shoot dieback

Given the limitations of linear regression in handling multidimensional and correlated datasets, we turned to alternative multivariate techniques, such as principal component factor (PCF) analysis and hierarchical clustering, to better identify the key soil parameters contributing to avocado decline. By applying these complementary methods, we were able to gain deeper insights into the complex relationships between soil properties and avocado decline in the MM region. Together, these techniques provided a more comprehensive understanding of the factors driving disease severity, highlighting the soil parameters most strongly linked to avocado tree health.

PCF analysis, a dimensionality reduction technique, was used to simplify the dataset by identifying the key soil variables that account for the most variance within the system. This approach is particularly useful in ecological studies where numerous interacting variables may obscure relationships. By grouping related variables into principal components, PCF analysis allows us to pinpoint the soil properties that are most strongly associated with avocado decline (^{Fabrigar and Wegener, 2012}). The five principal components identified in our analysis explained 76.81% of the total variance, which is considered substantial in ecological studies, allowing us to focus on the most relevant factors (Tab. 3).

Among these, PCF1, primarily associated with calcium (Ca) and pH, demonstrated a significant positive correlation with shoot dieback (SDB), with a loading value of 0.42. This suggests that higher Ca levels and alkaline soils may contribute to increased avocado decline severity. In contrast, PCF5, which was associated with phosphorus (P), exhibited a notable negative correlation with SDB, with a loading value of -0.62. This indicates that elevated phosphorus levels may play a protective role against shoot dieback, potentially alleviating the impact of the disease. While PCF1 and PCF5 were most strongly correlated with SDB, the remaining components (PCF2, PCF3, and PCF4) displayed weak or insignificant relationships with avocado decline in this analysis.

In addition to the PCF analysis, we used hierarchical clustering to group the soil variables based on their similarities and contributions to avocado decline. Hierarchical clustering generates a dendrogram, a tree-like diagram that visually represents how variables are grouped together step by step, either merging or dividing, based on their proximity in the dataset (^{Everitt et al., 2011}). As shown in figure 3, the clustering process identified a group of soil properties, including Ca, pH, and effective cation exchange capacity (ECEC), which are positively correlated with shoot dieback (SDB). This clustering pattern aligns with the results from PCF1, reinforcing the strong association between these soil characteristics and avocado decline severity.

While PCF analysis simplifies the complexity of the dataset by grouping soil variables into principal components, hierarchical clustering provides a visual tool for understanding the relationships among these variables. The two methods together provide complementary insights into the key drivers of avocado decline. By using both PCF analysis and hierarchical clustering, we ensured that our results were not biased by any single method and that the identified associations between soil properties and avocado decline were robust and reliable.

Figure 3. Correlations between soil properties of Montes de María orchards. Hierarchical clustering revealed a strong cluster comprising pH, calcium (Ca), and effective cation exchange capacity (ECEC). This cluster is significantly associated with higher values of avocado decline, as indicated by the Severity Index and Shoot Dieback (SDB) correlations shown on the upper side of the figure. 

Assessment of predictive models for explaining the effect of soil parameters on SDB

To assess the relationship between soil parameters and shoot dieback (SDB) severity, both stepwise multiple regression and ridge regression models were applied. The stepwise regression model identified a combination of soil variables that were significantly related to SDB, including potassium (K), phosphorus (P), and silt content. The final model demonstrated a good fit, with an adjusted R² of 0.42, indicating that these variables explain 42% of the variation in SDB severity (Tab. 4). This model suggested that potassium and silt content were positively associated with SDB, indicating that higher levels of these factors may contribute to increased dieback severity. Conversely, phosphorus exhibited a negative association with SDB, implying that higher phosphorus levels may mitigate the impact of the disease.

In contrast, the ridge regression model, which accounts for multicollinearity by shrinking the regression coefficients of less significant variables, provided a broader perspective by including all soil parameters, reflecting their collective impact on SDB. In this model (Tab. 5), the ridge coefficients showed that pH (estimate: 0.0617) and calcium (estimate: 0.0397) had the highest positive contributions to SDB severity, indicating that higher pH and calcium levels were associated with an increased risk of dieback. These findings are consistent with the earlier discussions about the effect of high pH and calcium in causing nutrient imbalances, particularly iron deficiency, which could exacerbate tree vulnerability. Furthermore, exchangeable cations (ECEC) showed a positive contribution (estimate: 0.0325), reinforcing the notion that soil conditions with high base saturation may negatively impact avocado health in MM. On the other hand, phosphorus (estimate: -0.0365) and organic matter (OM) (estimate: -0.0291) had negative coefficients, corroborating their protective role against SDB, as also noted in the stepwise model (Tab. 5).

While the stepwise regression model focused on identifying key variables that independently influence SDB, the ridge regression provided a more comprehensive view of how multiple factors, even those with smaller individual effects, interact to contribute to the decline. The consistency between the two models regarding the protective role of phosphorus, and the importance of soil pH and calcium, highlights these as critical areas for future management strategies in mitigating avocado decline. The divergence in other variables, such as the role of organic matter and exchangeable cations, underscores the complexity of soil-plant-pathogen interactions in avocado orchards and points to the need for multifaceted approaches in addressing tree decline in MM.

Principal component factor (PCF) analysis and hierarchical clustering reveal soil parameters contributing to avocado decline

The findings from the PCF analysis, particularly PCF1 (dominated by calcium, pH, and exchangeable cations), are further validated by clustering analysis and the ridge regression model, reinforcing the significance of these soil parameters in avocado decline. Clustering analysis grouped farms with high pH and calcium levels, which consistently correlated with higher SDB severity, a pattern also highlighted by PCF1. Similarly, ridge regression identified pH and calcium as significant contributors to SDB severity, aligning with the results from the PCF analysis. This convergence of independent analytical methods underscores the robustness of PCF1 in explaining the key drivers of avocado decline in the Montes de María region.

Furthermore, PCF5, primarily associated with phosphorus (P) and organic matter (OM), also emerged as a significant factor across multiple models. Both the PCF analysis and the stepwise regression model revealed that phosphorus played a protective role, with a negative correlation to SDB severity. This consistency across methods strengthens the case for phosphorus and organic matter as crucial elements in mitigating tree vulnerability. The identification of PCF1 and PCF5 as significant factors by different analytical techniques highlights their importance in explaining avocado decline, providing strong evidence of the role soil properties play in the disease's progression.

Environmental stressors and the "decline disease spiral" in avocado trees

Our findings align with previous studies that have emphasized the significance of environmental factors in tree decline and mortality (^{de Toledo et al., 2011}; ^{Whyte et al., 2016}; ^{Gazol et al., 2018}; ^{Scarlett et al., 2021}). The decline observed in MM can be understood through the lens of the "decline disease spiral" model proposed by ^{Manion (1991}). This model describes tree mortality as a progressive and cumulative process, where a series of interacting factors (predisposing, inciting, and contributing), rather than a single stressor, lead to tree decline and eventual death. The process begins with predisposing factors-such as unfavorable soil conditions, advanced tree age, or prolonged nutrient deficiencies-that gradually weaken the tree over time. These underlying vulnerabilities make trees more susceptible to inciting factors, such as prolonged drought, extreme temperatures, or pest infestations (Manion, 1991; ^{Das et al., 2016}; ^{Etzold et al., 2019}).

Predisposing factors set the stage for tree vulnerability by weakening the trees before any major inciting events occur (^{Manion, 1991}; ^{Ciesla and Donaubauer, 1994}). Avocado trees in MM are often grown in soils with suboptimal pH levels and limited nutrient availability (which are not supplemented by fertilization), which reduces their resilience to further stressors. Over time, such conditions diminish the trees' capacity to absorb essential nutrients, further contributing to their weakened state. The advanced age of many avocado trees in MM adds to their susceptibility, as older trees generally have reduced vigor and greater difficulty in coping with environmental changes.

Inciting factors act as triggers, causing acute stress that worsens the condition of already vulnerable trees (^{Manion, 1991}). In MM, the tropical dry forest biome presents a particularly challenging environment for tree health. Prolonged droughts and extreme temperatures are common in this region, creating conditions where water scarcity becomes a major limiting factor for tree recovery and growth. Poor soil fertility exacerbates these challenges by preventing trees from accessing the nutrients they need to survive prolonged stress events. These inciting factors, when combined with the predisposing vulnerabilities, significantly reduce the trees' ability to withstand environmental pressures, pushing them further into decline. Climate change will exacerbate these inciting factors, intensifying the environmental stress experienced by avocado trees in the MM region. Rising temperatures and more frequent and prolonged droughts are expected to become increasingly common as a result of global climate change. These shifts not only increase water scarcity but also alter soil moisture dynamics, making it more difficult for trees to recover from stress events (^{Camarero et al., 2015}; ^{Bauman et al., 2022}; ^{Hammond et al., 2022}). As climate-related stressors amplify, they will contribute to a feedback loop of decline, accelerating tree mortality and threatening the sustainability of avocado production in the region.

Once trees are weakened by predisposing and inciting factors, they become more susceptible to opportunistic pathogens and pests (^{Manion, 1991}; ^{Ciesla and Donaubauer, 1994}; ^{Oliva et al., 2014}). In MM, secondary pathogens like P. cinnamomi and pests such as termites and ambrosia beetles take advantage of the compromised state of the avocado trees, accelerating their decline. These biotic agents act as amplifiers within the "decline disease spiral," hastening the transition from stressed trees to tree mortality. The compounding effect of multiple stress agents-both biotic and abiotic-creates a scenario where each additional factor increases the tree's vulnerability to further damage (^{Manion, 1991}).

The relevance of the "decline disease spiral" to the MM region lies in its ability to provide a theoretical framework for understanding the complex, long-term process of avocado decline in this ecosystem. The dry tropical forest biome of MM presents unique challenges, where avocado trees are especially vulnerable to climate-induced stressors. The region’s pronounced dry season exacerbates water scarcity, compounding the effects of nutrient-poor soils, which increase the likelihood of deficiencies in essential elements like iron and phosphorus. These stressors interact over decades, creating the ideal conditions for the spiral of decline, where weakened trees are further compromised by biotic agents like pests and pathogens.

The multifaceted nature of the avocado decline highlights the need for comprehensive modeling techniques. Predictive models, including those developed through Geographic Information Systems - GIS, can serve as valuable tools to simulate the interactions between environmental factors, soil conditions, and tree health over time. Such models can identify critical thresholds and prioritize areas for intervention, offering a more dynamic approach to managing the decline. These models can provide early warnings of vulnerable regions, allowing stakeholders to take preventive measures before widespread tree mortality occurs. By integrating soil, climate, and pest management strategies, future efforts can better address the long-term challenges posed by the decline disease spiral in MM and work towards the sustainability of avocado production in the region. However, to fully grasp the wide-reaching impact of this process, it is crucial to incorporate modeling and GIS. These tools provide a broader perspective, allowing us to track changes over time and across landscapes, enabling a better understanding of how predisposing, inciting, and contributing factors have interacted over long periods in MM. Without this broader, system-level approach, it is challenging to appreciate the full extent of the long-term avocado decline occurring in this region.