Spatial variability of edaphic properties is influenced by land use type, topography, formation characteristics, depth, human activities, and time ^[75]. Assessing spatial variability through sampling is an essential step in precision agriculture processes that helps farmers make informed decisions regarding the distribution of agricultural inputs ^[110]. Spatial analysis of soil properties can be performed using various statistical methods, geostatistical methods, and geographic information system (GIS) approaches that facilitate interpolation of geographically located data to improve interpretation accuracy and digital mapping of the resource ^[111].

The analysis of spatial variability in soil utilization employs pattern-based statistics, a type of geographic analysis also known as location analysis ^[111]. This approach utilizes geostatistical and geometrical techniques to understand spatial patterns ^[111]. It involves applying statistical and data manipulation techniques to information that can be stored in a local geodatabase ^[111]-[113]. Spatial soil analysis reveals characteristics of the sample location, such as whether the samples are scattered or clustered. The spatial information itself pertains to the position, area, shape, and size of the defined area, and this data is typically stored as coordinates and topology ^{[111], [113], [114]}. One key outcome of this spatial analysis of soil properties is the creation of digital maps.

Digital mapping aims to identify and define soil units with some level of uniformity, delineated by clear boundaries. Nonetheless, soil characteristics are seldom entirely consistent; even within the same mapped units, there can be considerable random fluctuations in properties, complicating the detection of changes in average values between different mapping units ^{[115], [116]}.

According to ^{[115], [117]}, properties change continuously due to the impacts of climate variability and the effects of different soil management practices. Soil spatial variation can present in different patterns. For example, it might show relatively stable changes within certain map units but encounter sharp shifts in average values at soil boundaries. Alternatively, soil property variations might be smooth and continuous, with minimal and consistent fluctuations. Another possibility is having slight, non-random changes within soil units, along with either smooth or abrupt transitions in average values at boundaries. Lastly, soil properties might display sudden changes in variance at boundaries, whether or not there are changes in the average value ^[115].

The properties of soil are not static; they change over time ^{[92]-[101], [118]}. This variability stems from both intrinsic processes, which are linked to natural phenomena, and extrinsic processes, which are related to human management and cultivation practices ^[119]. Geological, hydrological, and biological factors are the primary intrinsic sources that influence soil formation and lead to variability. In contrast, extrinsic variability often results from differences in cultivation techniques and management, including variations in tillage methods, irrigation and drainage practices, and the management of crops and crop residues ^[119]. The temporal variability of soil can be analyzed in relation to plant age and through annual, seasonal, or daily cycles ^[115].

Temporal variability in soil can manifest during the growing season, and it can also occur from year to year, month to month, or even day to day, often influenced by weather and climate conditions. The physical properties of soil, such as moisture content, bulk density, aggregate stability, and penetration resistance, are interconnected and significantly affected by both climatic variability and climate change ^[119]. Similarly, chemical properties like pH, salinity, and soil organic carbon content are also impacted by these climatic factors ^{[92], [99] -[101]}.

GIS technology utilizes a combination of information management tools and methods that have enabled the administration, editing, and analysis of geospatial data on a global, regional, and local scale ^[120]. In the agricultural context, remote sensing technology uses data from crops and images taken from satellites, aerial remote sensors, and ground equipment ^{[57], [121]}. The processed data can be deconstructed into spatial layers that can then be processed and analyzed in a GIS in multiple ways to reveal crop conditions and monitor pests and diseases ^[44].

Several studies have applied GIS approaches for the detection of Black Sigatoka in banana crops, using images taken by hyperspectral sensors to characterize the severity of the disease ^{[17], [28], [50], [122]-[124]}. These studies have demonstrated that the implementation of hyperspectral sensors is an effective tool for early disease identification, capturing detailed data on leaf reflectance, which allows for the detection of subtle differences in crop health that are not visible to the human eye. Previous research has leveraged these capabilities by correlating the data with disease severity indicators, thus improving decision-making in the management of Black Sigatoka ^{[16], [45], [87], [125]}.

These studies show that geostatistical methods have been fundamental in describing the spatial and temporal spread of Sigatoka in banana crops. For example, the use of kriging in specific studies ^{[16], [68], [87], [88]} allowed for modeling the distribution of the disease and generating predictive maps of the most affected areas. However, these studies demonstrate that the accuracy of these models depends on the density of the sampling points, the experimental setup, which includes the type of banana and the extent of the area to be studied. These factors ensure the effectiveness of the geostatistical tools. Table 7presents the cited studies, detailing the geostatistical methods and the experimental setup used to evaluate Sigatoka infection in banana crops.

Table 7

Source: own elaboration.

Table 7 shows that the evolution of statistical and geostatistical methods has advanced over time, highlighting geostatistics as key in the spatial modeling of Sigatoka infection. The transition has moved from basic approaches of correlation and autocorrelation analysis to more complex techniques, such as semivariograms and kriging, which allow for an accurate representation of the disease's spread. Additionally, the use of software has shifted from GS+, LCOR2, Surfer, among others ^{[16], [88]}, to R ^{[68], [87]}, facilitating the integration of advanced packages to perform robust analyses, thereby improving inferences and predictions in the management of Sigatoka.

In this regard, geostatistical analysis has proven to be an important tool for understanding the distribution and severity of Sigatoka in banana crops, as the spatial characterization of the disease is essential for its management. However, ^[88] emphasized the need to consider climatic variability and soil conditions that influence the incidence and severity of the disease. This led to studies such as ^[16], which applied geostatistical techniques to analyze the relationship between soil fertility and Black Sigatoka severity, using spatial distribution maps to identify critical areas within plantations. Similarly, ^[68], through geostatistical techniques, evaluated the correlation between Sigatoka severity and soil properties, considering climatic variables. As a result, their study employed advanced spatial analyses to map the disease and relate it to soil nutritional factors, highlighting the need for data-driven agronomic approaches to improve disease management. Likewise, ^[87] employed spatial modeling to study Black Sigatoka in Cavendish banana cv. Gran Enano, revealing distribution patterns that can guide decision-making in precision agriculture.

Nevertheless, despite these advances, the integration of new technologies, such as hyperspectral drones, satellite sensors, and/or machine learning techniques, remains a challenge for new research. These tools, combined with geostatistical analysis, would allow farmers to adopt more precise and efficient strategies for Sigatoka control, such as more targeted fungicide applications and improved monitoring programs ^{[9], [32], [33]}. Previous studies ^{[17], [28], [50], [124]} have shown advances in machine learning techniques and hyperspectral imaging applied to disease management or detection in bananas. However, none of those reported so far have associated these techniques with soil properties data and climatic variables.

For example, ^[17] evaluated the use of hyperspectral images to detect the early stages of Sigatoka through a penalized logistic regression model (PLS-PLR), achieving 98 % accuracy. This advancement allows for disease identification before it becomes visible to the naked eye, although it still does not incorporate climatic or soil variables into its analysis. ^[50] also developed machine learning-based models, such as SVM and neural networks, for early Sigatoka detection, demonstrating the potential of artificial intelligence techniques in precision agriculture, though it still does not account for environmental factors. Similarly, ^[124] designed a hyperspectral imaging system to detect the disease presymptomatically, using advanced optical techniques, but without yet integrating geostatistical or soil data. Finally, ^[28] employed drones and machine learning algorithms to monitor Sigatoka, demonstrating that remote sensing combined with artificial intelligence can surpass traditional methodologies, although it did not consider the influence of climatic and soil variables on the spread of the disease.

Climate variability poses challenges to water resources, diminishes crop yields, and raises the prevalence of pests and diseases, significantly affecting agriculture, particularly in tropical areas. With the imminent climate risks to agricultural output and food security, there is a growing emphasis on global and national programs and policies to prioritize adaptation strategies for agricultural production in response to climate change ^[126].

The FAO forecasts that food demand will double by 2050, posing a significant challenge for the scientific community to boost agricultural productivity ^[126]. Therefore, studying the influence of climate variability and the relationship between black Sigatoka and edaphoclimatic conditions requires an appropriate methodology. In response to the spatial dependence between crops and plant diseases, GIS, remote sensing, and spatial analysis techniques have been employed in epidemiological research in the last two decades. These techniques have enhanced the collection, storage, retrieval, analysis, and visualization of spatial data, leading to a deeper understanding of the factors that affect the emergence of epidemics ^{[87], [125], [127]-[129]}. This method produces more reliable results concerning the area's plant disease epidemiology and soil variability ^[125].

Spatial analysis through geostatistics has been commonly used to relate edaphic and climatic conditions to pest and disease severity ^[71]; some authors have studied its application in banana crops to describe Sigatoka behavior, obtaining patterns of disease behaviors explained by edaphic conditions ^{[16], [45], [68], [87]}. Other research on GIS and remote sensing approaches have oriented their studies on the detection of black Sigatoka associated with climatic conditions using simulation models of the disease using physiological responses to infection given optimal climatic conditions of temperature, precipitation, relative humidity, evapotranspiration, solar radiation, plant leaf wetness, among others ^{[9], [12], [15], [18], [49], [52], [61]-[64], [125]}.

So far, research that has used and applied GIS approaches, remote sensing, and spatial analysis in banana crops to detect black Sigatoka has separately studied the edaphic and climatic conditions associated with the disease. Few investigations have studied the relationship between edaphoclimatic conditions and black Sigatoka, and those that have been conducted have not used GIS, remote sensing, or spatial analysis approaches for disease detection on a spatiotemporal scale ^{[24], [36], [54], [65]}.

Soil properties often display spatial autocorrelation, meaning that the values of these properties at nearby locations tend to be similar. Numerous studies have examined a wide range of soil properties, including morphological, physical, chemical, and biological properties, at the microscale level ^{[16], [75], [78], [92]}. However, our understanding of how these properties vary across both space and time at this detailed level remains limited. This understanding depends on the availability of field data, the application of geostatistical techniques, and the analysis of autocorrelation among these properties. Geostatistics, a branch of applied statistics, has emerged as a valuable tool for understanding and estimating spatial patterns in soil properties ^{[115], [130]-[132]}.

In geostatistics, autocorrelation exists when there are relatively minor variations that are not entirely random in the value of a soil property ^[133]. In this sense, the distance between sampling points is a factor that influences the spatial variability of soil properties, i.e., the closer the sampling points are to each other, the more similar the values are, as opposed to values separated by more distant distances. Beyond the critical distance threshold, where autocorrelation is lost, soil property values no longer exhibit spatially correlated behavior and their relationships become random ^[134]. This type of spatial variation, occurring when the relationship between two sampling points cannot be accurately predicted based solely on their distance, can be quantitatively modeled using various techniques, with semivariogram models being the most common ^{[115], [135]}.

The semivariogram is a technique used to study the spatial distribution of soil properties ^[136]; it helps model how the variance of soil properties changes as the separation distance between sampling points increases ^{[115], [137]}. In geostatistics, the semivariogram is essential for spatial prediction or kriging of a target geographic feature. It is derived by measuring the spatial correlation or covariance between sample pairs at various distances in a dataset to create empirical semivariograms. A graphical representation of the semivariable and the distance or lag corresponding to the separation of the pairs produces the semivariogram ^[115]. The procedure involves plotting the semivariable values against the separation distance and then applying functions-such as linear, spherical, or exponential models-to fit these data points ^[135].

By the end of this century, climate change is projected to have a significant impact on the agricultural sector in developing countries due to associated damages and high adaptation costs, with potential results in 1.30°C to 5.70°C increase in global average temperatures, depending on emission scenarios, whether low or very high ^[138]. Some critical challenges include increasing frequency and intensity of extreme weather events through higher temperatures and more variable precipitation, including heat waves, floods, and droughts, which can significantly affect the agricultural sector ^[139]. Developing countries account for ~70 % of the agricultural potential for climate change mitigation ^[140]. The agricultural sector is a significant contributor to global greenhouse gas (GHG) emissions, directly accounting for approximately 14 % of the total. Furthermore, agriculture indirectly increases emissions through land-use changes, particularly deforestation for agricultural expansion, which contributes an additional 17 % to overall GHG emissions ^{[41], [139], [141]-[143]}.

In this context, banana crops play a key role in carbon sequestration through photosynthesis, contributing to mitigating climate change. However, the effectiveness of this sequestration can be affected by foliar diseases associated with the crop ^[144]-[146]. Studies have shown that CO₂ capture in banana crops reaches 80°t/ha at the production stage ^[144], with biomass contributions of the pseudostem ranging from 75 % to 78 % and leaves from 12°% to 17°% ^{[5], [147]-[149]}.

Developing and implementing effective adaptation and mitigation strategies to minimize the impacts of climate change on crop foliar diseases is a major challenge ^[150]-[153]. Detection of foliar diseases has led researchers to work on image processing for classification and early detection of crop diseases; however, they face multiple challenges because the variability of shapes and colors in the images, together with climate change, pose an additional challenge in identifying multiple foliar diseases, especially when disease patterns are irregular ^[154]. State-of-the-art studies in the area indicate that challenges are related to noise in images caused by electronic devices and lighting effects, out-of-focus imaging, lack of in situ data containing images of infected leaves in crops, variability of climatic conditions, and selection of suitable electronic devices and images ^[155]-157]. With these disadvantages, some researchers have adopted drone cameras, but these also present limitations such as weather and flight time for imaging ^{[28], [157]}. Nevertheless, drone cameras are a convenient tool for small-scale disease monitoring and are useful for validating remote and proximate sensor data for crop health monitoring ^[158].

The detection of black Sigatoka in banana crops faces significant challenges under possible climate change scenarios. As the global climate undergoes alterations, the accurate identification of this disease in crops and the influences of environmental variables on its detection become more complex, forcing geographic information systems to constantly change the algorithms used. Increasing the intensity and frequency of extreme weather events, such as heavy rains or prolonged droughts, could favor the spread and severity of black Sigatoka, making timely detection even more critical. Thus, climate change poses additional challenges in detecting crop diseases, requiring ongoing research and adaptation strategies to ensure food security. Future research on the detection of black Sigatoka in banana crops should prioritize the development of climatically robust algorithms, autonomous detection using artificial intelligence methods, the establishment of real-time monitoring systems, the understanding of the genetic variability of the disease, the implementation of integrated management techniques, and all other measures that humans take to monitor the crop. These guidelines will help address the challenges of climate change, improve detection accuracy, and ensure disease and pest protection in banana crops.