The use of soil modelling techniques is becoming more and more relevant for multiple disciplines related to soil sciences such as environmental science, soil geography, agronomy, ecology, and land management (Flathers and Gessler 2018; Rodrigo-Comino et al. 2018; van Leeuwen et al. 2019). This is because collecting a representative number of soil samples at a large scale is time-consuming and costly.
Soil nutrients are key factors for crop production and play important roles in plant growth but can limit production where insufficient amounts prevail (Pittman et al. 2021; Sharma et al. 2022). However, they are susceptible to modification by human activities, and are therefore difficult to be assessed, particularly at the catchment scale or even at the hillslope scale (Wang et al. 2008; An et al. 2019). Therefore, the determination of soil nutrient contents, both micronutrients and macronutrients, are indispensable as sustainable agricultural production requires optimum levels. The essential elements, without which plant growth will not be possible, come from the mineral components of the soil. Depending on the number of elements needed by the plant, the distinction can usually be made between macronutrients and micronutrients (Blume et al. 2016a). Micronutrients, including iron (Fe), copper (Cu), manganese (Mn) and zinc (Zn), are found in low concentrations (5–100 mg kg− 1 or less) in agro-ecosystems (He et al. 2005; Alloway 2013).
The spatial variability of micronutrient levels in the topsoil is influenced by the complex interactions between climate, time, parent material, topography, vegetation, soil physicochemical properties, fertiliser application, land use, and possibly exogenous inputs from industry (Srisomkiew et al. 2022). Therefore, a better understanding of the dynamic variation of micronutrients and the factors contributing to their levels in the soil is paramount (Zhu et al. 2016; Zhu et al. 2021). Micronutrient deficiency is one of the most important plant nutrition phenomena that have significant impact on the growth and development of plants (Atucha et al. 2013; Wani et al. 2017; Miran et al. 2021). Regarding human health, the levels of micronutrients should be at the optimum as both deficiency or excess have negative effects. While deficiency in soil micronutrients limits crop productivity in many regions around the world, deficiency of micronutrients in humans can cause serious diseases such as anemia and neutropenia. Approximately two billion people worldwide are believed to suffer from at least one disease as a result of micronutrient deficiency (Jones et al. 2013; Krasilnikov et al. 2021; Denton-Thompson and Sayer 2022). Chronic micronutrient deficiency leads to deterioration of children's health and reduced mental development. It also limits the reproductivity of humans and can trigger chronic diseases (Krasilnikov et al. 2021). The presence of essential elements in food largely comes from the soil as up to 95% of food is produced by agriculture. This can be affected by multiple factors such as elevated soil erosion and runoff rates (Bashagaluke et al. 2018; Cerdà et al. 2022). Therefore, the detection of the levels of the micronutrients require efficient solutions (Keesstra et al. 2021).
Sustainable farming and effective land management require the understanding of the spatial distribution of the soil properties (Taghizadeh-Mehrjardi et al. 2022). Developing adequate agricultural management strategies and policies, along with minimising the environmental impact on farming revenue, requires detailed site-specific soil knowledge (Snapp 2022). While measuring the factors affecting soil micronutrient distribution in agricultural areas in arid and semi-arid regions is often difficult, it is crucial to understand the factors influencing their availability and concentration for proper management of soil fertility and land use. This is particularly important for arid land management (Naimi et al. 2021).
Although it is well known that nutrient deficiencies limit yield, information on the spatial distribution of soil nutrient content and nutrient availability is limited (Hengl et al. 2017; Hengl et al. 2021). At the local scale, the spatial variability of soil properties can be mapped using an interpolation technique (Panday et al. 2018). Digitisation of the contents and heterogeneity of soil micronutrients is considered extremely important to assess soil fertility for efficient agricultural practices, particularly for farmers aiming to increase crop productivity (Dong et al. 2019; Dad and Shafiq 2021; Suleymanov et al. 2021). However, accurately estimating micronutrients at any scale requires a wide range of methods capable of monitoring the soil surface. Traditionally, soil spatial information has been presented as polygonal maps at coarse scales (Hengl et al. 2017). Digital soil mapping (DSM) methodology, a sub-science of pedometry, has been recently used to map soil characteristics (Lark et al. 2014; Hong et al. 2015). DSM uses data science and predictive modeling approaches to establish the relationships between environmental variables and soil properties (Ließ et al. 2021; Wadoux et al. 2021a). The successful integration of these innovative disciplines has led to active work on a global (FAO and ITPS 2018), continental (Hengl et al. 2017) and regional (Ali and Morghanm 2013; Pelegrino et al. 2019; Zhang et al. 2020; Vasu et al. 2021; Zhu et al. 2021; Kaya and Başayiğit 2022) scales for the diagnosis of spatial distribution of soil properties associated with soil fertility. The DSM methods are cost-effective and allow the creation of high spatial resolution maps characterised by a high accuracy. Remote sensing and digital soil mapping are being employed to study the spatial variability of soil properties, and soil nutrients for that matter (Wang et al. 2022). However, there is limited information on the potential use of remote sensing-based estimators of soil nutrients in arid and semi-arid regions (Akbari et al. 2021). Soil data for spatial analysis is lacking in arid and semi-arid regions, making it difficult for establishing land use management and policies (Smith et al. 2019). Given the advances in technology during the last 25 years or so, there are expectations of significant improvements in the resolution, accuracy, and coverage of spatial data in the future (Burke et al. 2021; Droz et al. 2021).
Despite the increasing number of environmental covariates used as input to DSM, covariate selection to limit the number of the input covariates has been highlighted (Liang et al. 2020; Wadoux et al. 2021b). Without a doubt, increasing the number of the environmental variables improves machine learning model accuracy, but this approach increases uncertainty in the input data that makes the results difficult to interpret in soil science. Thus, there is a need to balance covariate parsimony and model performance through an appropriate covariate selection on case-based reasoning and pedological relevance (Liang et al. 2021; Wadoux et al. 2020). The abundance of data and the use of knowledge as a primary driver in the various areas of soil science have several implications in the methods that are yet to be documented. As a matter of fact, “scenario” and “grouping” strategies have been studied effectively in evaluating environmental variables, especially in digital soil mapping (Wadoux et al. 2021b). When used as a scientific methodology involving iterations and revisions, scenarios can help to challenge existing assumptions that could lead to new research topics (Ramirez et al. 2015; Keesstra et al. 2016). However, a “scenario” may have a different definition depending on the field of scientific application and may not serve the same purpose nor include an additional mode of production (Ramirez et al. 2015). In spatial modelling studies, the term scenario can be associated with the “future”.
Despite these methodological approaches and different types of machine learning algorithms integrated into the DSM methodology, the estimation of micronutrient contents of soils has been rarely studied (Hengl et al. 2017; Pelegrino et al. 2019; Hengl et al. 2021, Zhu et al. 2021).
This study aims to investigate the factors affecting micronutrient distribution in the topsoil, and to predict micronutrient distributions by integrating different sets of environmental variables with machine learning models in the piedmont plain of north-eastern Iran, an arid area where agriculture is the main human activity.