Potassium (K), one of the most crucial macronutrients for plant growth and metabolism, often varies in terms of its availability in soil based on the texture, pH, mineral composition, and cation exchange capacity (1). In soils, four different forms of K prevail namely, water-soluble (Ws-K), exchangeable (Ex-K), non-exchangeable (NEK), and lattice K (LK) (2, 3). Ws-K is the primary form of K available to plants and is directly absorbed by roots. On the contrary, Ex-K serves as a crucial buffer in soil systems maintaining a dynamic equilibrium by replenishing the Ws-K when depleted (3, 4). Even though being less accessible, non-exchangeable K acts as a reserve pool supplementing the two previous forms on depletion while playing a crucial role in long term fertility of soils and overall buffering (5, 6). Lattice or mineral K is the least accessible form which may convert to other accessible forms when subjected to long term weathering (7–10). The two most significant management factors that affect K equilibrium in soils, aside from these soil characteristics, are fertilization and cropping (11).

Seven distinct types of soil are found in India due to its varied landscapes with unique chemical characteristics and formation history, shaped by the forces of nature (12). Of these type four major soil types were chosen for this study. Among them, alluvial soils, which cover over 75 million hectares in the Brahmaputra Valley and Indo-Gangetic Plains, are generally deficient in organic matter, phosphate, and nitrogen (13). They support a variety of crops and are neutral to slightly alkaline in nature. The parent rock’s potassium-rich minerals, such as mica and feldspar, release K upon weathering, which accounts for their high potash concentration. Calcareous soils, which cover roughly 229 Mha in arid and semi-arid regions, contain a lot of K as they originate from of K-containing minerals and moreover leach less as compared to other soil types. But the available K in these soils is now low due to intensive cultivation and sparse usage of K fertilizer (14). Calcium carbonate (CaCO₃) affects the physical and chemical characteristics of calcareous soils, including nutrient availability, water retention and crusting, which in turn affects plant growth (15). Research shows that increasing fixed potassium levels in soils through calcium carbonate removal improves plant availability of K (16). The third type considered in this study, red soil, makes up around 10.6% of the area, and is mostly found in Jharkhand (17). These soils are generally rich in minerals like aluminum, iron and especially in K. However red soils are often reported to have low in nitrogen, phosphorus, and overall organic matter (18). The largest and most important soil group in India for agriculture is alluvial soils, which are created by river silt deposits. These soils are generally fertile and contain a good balance of essential nutrients such as potassium, nitrogen, phosphorus, thereby supporting crop growth. Lastly, black soils of India, are composed of weathered Deccan basalt alluvium, and are abundant in exchangeable K and smectite clay minerals such as calcium, magnesium and iron (19).

Among others, one of the crucial parameters on which soil K availability varies is water activity of soil (20). Given that rice (one of the most economically important crop in India) is mostly grown in submerged environments, changes to the soil’s electrochemical and chemical characteristics brought about by submergence may be linked to changes in the dynamics of K availability. This in turn may have an impact on the availability of K to rice (21). The effect of submersion and non-submersion on soil’s various K pools and its bioavailability remains insufficiently explored. Developing effective fertilizer recommendations, particularly for lowland rice farming, requires an understanding of the forms and transformations of potassium in soils under submerged settings. Therefore, it is essential to develop a model for predicting the uptake of various K pools into different parts of the rice plant (root, shoot, and grain) under both submerged and non-submerged conditions, as both are crucial for plant growth, development and yield. In order to identify important input parameters and assess their impact on exposure outcome variability, Sobol sensitivity analysis (SSA) was used (22). The relative contributions of each variable to the total variance in model outcomes are measured by Sobol sensitivity indices (SSIs). This approach makes it easier to identify the input variables that have a significant, noteworthy, or insignificant impact on the study. Machine learning approaches such as random forest and Taylor plot also serve as a great tool for such assessment, and thus a broad range of investigations have been carried out using such mathematical models. Thus the four different soil types and rice grown on them in submerged and non-submerged conditions were compared in this study and the basic aim was (a) to measure how different potassium pools change under submerged and non-submerged conditions, considering soil properties and their availability for rice plant uptake., (b) to predict the contribution of different soil potassium pools in various rice plant parts (root, shoot, and grain) using Sobol, Taylor, and Random Forest methods.

The study highlights the diverse physical and chemical properties of the analyzed soils, categorized into red, alluvial, calcareous, and black soil groups. The lowest pH values were recorded in S1 (red soil) and S5 (alluvial soil), collected from Giridih, Jharkhand, and Darjeeling, West Bengal, India, respectively. The acidic nature of Giridih soil is attributed to the presence of mineral-rich mica mines (17), while Darjeeling soil acidity is primarily due to high rainfall in the Eastern Himalayan region, which leaches basic cations like calcium, magnesium, potassium, and sodium, leading to an increase in hydrogen (H⁺) and aluminum (Al³⁺) ion concentrations (32). Electrical conductivity (EC), indicative of soil salinity, ranged from 0.13 dS/m in S5 (alluvial) to 0.73 dS/m in S8 (black soil). While these values suggest generally non-saline conditions, the relatively higher EC in black soil points to a greater concentration of soluble salts, which could influence water uptake and nutrient mobility (33). CEC and clay content were highest in S8 (black soil), with CEC reflecting the soil’s capacity to retain essential cations like potassium. This high CEC, combined with the significant clay content (41.2%), suggests a strong capacity for nutrient retention and water holding, enhancing fertility (17). In contrast, sandy soils or those with lower clay content typically show reduced nutrient retention, necessitating more frequent nutrient supplementation (34). Organic carbon (OC) content, a vital indicator of soil organic matter and microbial activity, was comparatively higher in alluvial (S5, 1.74%) and black (S8, 1.12%) soils. Higher OC content promotes soil structural stability, water retention, and nutrient cycling, which are essential for sustainable agricultural productivity (35).

Analyzing soil K forms revealed notable variations across soil samples under natural (non-submerged) and submerged conditions. Among all the collected soil samples, S1, representing red soil, exhibited the highest potassium content across all forms of soil potassium under both submerged and natural conditions. This outcome can be attributed to the proximity of S1’s collection site to agricultural soils near mica mines. Mica is a significant source of various K forms, and Giridih, where this soil was sourced, is recognized as one of the largest mica producers globally (17). The submerged condition consistently enhanced the availability of all K forms across the soil samples compared to the natural condition. This improvement under submerged conditions may be attributed to changes in redox potential and soil microbial activity, which facilitate the release of potassium from less available pools (36). It was previously reported by Ghosh et al. (2023b) that rhizospheric microorganisms have the potential to convert non-exchangeable K into available forms by producing organic acids which could also be a potent role player (37). A strong negative correlation was observed between K forms and soil physicochemical properties (e.g., pH, EC, and clay content) under both natural and submerged conditions. This suggests that soil properties significantly influence K dynamics, potentially affecting its bioavailability for plant uptake. These findings emphasize the importance of soil type and management practices in optimizing potassium availability, especially in agricultural contexts involving submerged cultivation systems like rice.

S1 (red soil) outperformed all other samples in submerged conditions, suggesting that this soil is more favorable for potassium availability under submerged cultivation. A comparable trend was reported by Ghosh et al. (2023a) in their studies (17). Interestingly, when comparing the two methods, natural soil demonstrated a more sequential release of potassium than submerged soil, which could be attributed to differences in soil redox conditions and microbial properties and activities between the two environments (38). This sequential release pattern in natural soils may be linked to the slower mobilization of K from less available pools, whereas submerged conditions likely facilitate a more rapid release due to changes in soil chemistry and microbial processes. The CR-K, indicative of interlayer K with limited solubility, serves as a practical measure of the soil’s long-term K supply potential (23). This parameter showed minimal variation across the studied soils (17). When compared to the critical threshold of 0.2 cmol(p⁺)/kg proposed by (39), the experimental soils demonstrate an ample capacity to supply K for plant growth. These findings highlight the importance of both soil type and environmental conditions in determining the bioavailability of potassium, which is crucial for efficient nutrient management in agricultural practices (40).

Potassium (K) concentration in rice roots, shoots, and grains was highest in S1 (red soil) among all soil types, highlighting its superior bioavailable K levels and suitability for rice cultivation. In terms of K distribution within the rice plant, the grain had the highest K concentration, followed by the shoot, and then the root. The ranking of K content across the rice plant parts was consistent: rice grain > rice shoot > rice root, irrespective of the soil type. This pattern is in line with the plant’s physiological needs, as K plays a key role in grain formation and overall plant health (41). A strong positive correlation was observed between the different forms of soil K (natural and submerged conditions soil) and the K concentration in rice roots, shoots, and grains. This correlation suggests that the availability of soil K directly influences its uptake by rice plants, with higher potassium concentrations in the soil leading to higher potassium concentration in the plant tissues. These findings underscore the importance of understanding soil K dynamics and their impact on rice plant growth, which is essential for optimizing fertilization strategies and improving crop yield (42).

Sobol sensitivity analysis was conducted to identify the key input factors influencing K uptake by various rice plant parts (root, shoot, and grain) under different extraction methods (natural and submerged). Among these, Ws-K and Ex-K from natural soil conditions were found to be the most favorable forms for K uptake, as compared to the submerged condition. This indicates that natural soil conditions provide more readily available K for plant absorption, which is crucial for optimal growth. Potassium uptake by rice roots was highest from Ws-K in natural soil. Similarly, Ws-K and Ex-K from natural soil were superior for shoot K, while Ex-K from natural soil contributed most to grain potassium. These findings emphasize the critical role of soil K forms in rice nutrition and optimizing fertilization for better yield and quality (43).

Machine learning (ML) techniques, including seven models like Random Forest (RF), XGBoost, SVM, RTree, LR, and KNN were used to predict soil K availability and its accumulation in the soil-rice system. Evaluated via solubility tests and Taylor diagrams, RF emerged as the most accurate model, with strong correlation coefficients and low RMSE values, demonstrating its reliability for forecasting K dynamics and optimizing crop yields. The RF model accurately predicted K availability and accumulation in both natural and submerged soils, proving valuable for soil fertility management and crop nutrition. Its effectiveness in different soil conditions highlights the potential of machine learning in optimizing K uptake and advancing precision agriculture (44).

The Variable Importance plot from the Random Forest (RF) models highlights the critical role of Ws-K and Ex-K from natural soils in potassium uptake by rice plants, particularly in grains, roots, and shoots. Metrics such as %IncMSE and IncNode purity confirm that these K forms significantly enhance model accuracy and decision tree purity, emphasizing their importance over submerged soil conditions. The three-dimensional partial dependence plot further illustrates how Ws-K and Ex-K from natural soils significantly influence K content across rice plant parts, outperforming submerged soil conditions. PDPs offer valuable insights into the relationships between K forms and plant uptake, demonstrating the superiority of natural soils for optimizing potassium availability and rice yield.

The study provides valuable insights into the influence of different soil types and conditions (natural and submerged) on K availability and its uptake by rice plants. The soil samples, classified as red, alluvial, calcareous, and black soils, displayed varying physical and chemical properties, which significantly impacted potassium dynamics. Red soil (S1) demonstrated the highest potassium content, particularly under submerged conditions, highlighting its suitability for rice cultivation. The study also revealed that water-soluble potassium (WsK) and exchangeable potassium (ExK) from natural soils were the most favorable for potassium uptake by rice plants, with the grain containing the highest K concentration. The sequential extraction methods used in this study, along with machine learning models, provided a comprehensive understanding of K availability and its mobilization under different soil conditions. The Random Forest model proved to be most reliable for predicting K dynamics, offering a promising tool for precision agriculture and nutrient management. Sobol sensitivity analysis further reinforced the critical role of WsK and ExK in enhancing K uptake, emphasizing their importance in improving rice growth and yield. Overall, the findings underline the significance of soil management practices, including optimizing K availability through appropriate fertilization strategies and understanding the underlying soil properties, to enhance rice productivity. The use of advanced techniques such as machine learning and sensitivity analysis presents an innovative approach to soil fertility management, paving the way for more efficient and sustainable agricultural practices.