AI-guided investigation of biochar’s efficacy in Pb immobilization for remediation of Pb contaminated agricultural land

Abstract

This study evaluated the lead (Pb) immobilization efficiency of biochar in contaminated agricultural soil. The biochar was produced from a range of major biomass residues and pyrolyzed under well-controlled conditions. Ten different types of standard biochar samples were derived from five different feedstocks (i.e., softwood, miscanthus straw, rice husk, oilseed rape straw, wheat straw) and pyrolyzed at 550 ℃ and 700 ℃. Pb-contaminated soil near an abandoned mine was incubated with 2.5% (w w^− 1) of biochar. Incubation was conducted for various durations at room temperature under both short-term (21 days) and long-term (214 days) conditions. This variation explicitly accounted for the simulated microplastic contamination during the long-term incubation period. A novel framework has been developed to predict the long-term immobilization effect of various biochar types using a machine-learning approach, following the successful identification of optimal biochar implementations. This prediction method utilizes a small on-field dataset by employing a data augmentation approach, showcasing an innovative approach to forecasting the effects of different biochar types over time. After the incubation period, soil samples were analyzed for their chemical properties. As a result, oil seed rape biochar was the highest in pH, EC, exchangeable Ca²⁺, Mg²⁺, and K⁺, total nitrogen content, soil organic matter content, and available phosphate. In return, OSR 700 treated soils showed the highest content of exchangeable cations and the lowest content of available Pb after the incubation period. The most efficient biochar for immobilizing lead (Pb) in soil appears to be OSR 700, based on the available evidence.

Introduction

Abandoned mines, covering approximately 49 million km² globally [30], represent a substantial challenge to land use management and biodiversity conservation. The remediation of these sites is crucial not only for mitigating the ecological impacts of past mining activities but also for enabling sustainable land use changes, potentially transforming these areas into valuable resources for agriculture, forestry, or recreation. However, the journey towards effective rehabilitation is fraught with challenges, particularly the widespread contamination of soil with toxic substances like lead (Pb). The severity of Pb contamination in abandoned mine areas, noting its persistence and the complex issues it poses for remediation efforts. Pb in soil is a silent threat that affects both ecological systems and human health. With its affinity to plant uptake, dissolved Pb from the soil is accumulated in the human body through the food chain. Its presence is a considerable threat to food safety and is linked to developmental disorders in children and various health complications [24]. Previous study draws a direct connection between effective soil Pb remediation and the advancement of the United Nations Sustainable Development Goals (UN SDGs), illustrating the broad, positive impact of successful clean-up efforts on global health and sustainability [8]. Addressing the legacy of abandoned mines, especially those contaminated by Pb contamination, is therefore not just an environmental imperative but a crucial step towards a healthier, more sustainable future.

Biochar, a carbon-rich material produced through the pyrolysis of biomass, plays a pivotal role in immobilizing heavy metals, including lead (Pb), as a promising substance for soil remediation. The mechanism behind this immobilization is multifaceted, primarily involving physical and chemical processes. Firstly, the porous nature of biochar provides a vast surface area that facilitates the adsorption of heavy metals. This adsorption occurs due to the strong affinity of these metals for the negatively charged surface groups on biochar, such as carboxyl and hydroxyl groups. Additionally, the high pH of biochar can induce the precipitation of some heavy metals as hydroxides, further reducing their mobility in soil.

Furthermore, biochar can alter the soil’s cation exchange capacity (CEC), enhancing its ability to retain positively charged metal ions and thus reducing leaching into groundwater. This is particularly relevant for Pb, which can exist in soil as a positively charged ion. Also, it is derived from various wastes generated from agriculture, industry, or waste management facilities. The majority of biochar has been produced using biomass debris that is discharged after cultivation. Biochar gives key solutions to the valorization of wastes. It can be widely used as carbon sequestration materials, adsorbent of gas and water contaminants [9, 26, 27], or amendment to soils [10].

Immobilization of biochar in both soil and water has been evaluated not only with pristine biochar but also with engineered or designed biochar upon the contamination site [26, 33]. Apart from chemical or physical activation, the inoculation of bacteria on the biochar surface also enhances heavy metal immobilization [31]. Especially the biochar application on Pb immobilization has been studied with pristine biomass-based biochar [4], magnesium oxide-coated biochar [29], cement-combined biochar [21], or iron-modified biochar [35].

However, it is difficult to define the preponderant mechanisms of biochar or standardize their properties. There are numerous feedstocks subjected to biochar, and hundreds of combinations in its production can be designed. Thus, biochar can be designed and additionally activated depending on the application purpose [2]. Furthermore, due to the reliability of the feedstock condition of biochar, heterogeneousness is unavoidable in biochar production. It is essential to establish and provide the general and standardized features of representative biochar. The standardized feedstocks and production conditions can homogenize the biochar properties. It will bring out the reliable comparison of research results in different study areas or the root-sharing investigation on mechanisms of biochar in various aspects.

Despite its promise as a soil amendment for remediating contaminated lands, biochar application faces challenges rooted in the variability of its interactions with diverse soil constituents, particularly in environments burdened with heavy metals like lead (Pb). The heterogeneity of biochar, influenced by its feedstock and pyrolysis conditions, leads to inconsistent performance in contaminant immobilization and soil fertility enhancement. Various uncertainties that could happen on-field (i.e., the presence of microplastic) introduce additional layers of complexity that traditional remediation strategies struggle to address effectively. Advanced computational methods present a viable pathway to surmount these obstacles, offering sophisticated tools to model, optimize, and apply artificial intelligence (AI) to the field of biochar.

Through modeling, it is possible to simulate the multifaceted biochar-soil interactions, facilitating a deeper understanding of the mechanisms that govern biochar’s performance across different environmental conditions [3]. Optimization techniques can then refine biochar formulations and application protocols [22], ensuring maximum efficacy in specific soil types and contamination scenarios. For instance, Xu et al. [36] identified the ideal production condition of biochar from kitchen waste focusing on remediating Cd (II) in polluted water by optimizing between tradeoff of environmental risk and its remediation performance. Apart from the microscopic focus on the synthesis of biochar, a study from a larger perspective that discusses the potentiality of biochar in attaining circular agriculture via the water-food-energy-carbon nexus is proposed [36]. By integrating AI into the biochar application process, we can transcend the limitations of static remediation protocols. AI enables the dynamic understanding of biochar in the production (from organic waste [18] and six different biomass types [16] and application (remediation of pesticide in agricultural land [23] and water treatment [38]). Predictive analytics powered by AI can inform stakeholders about the optimal conditions for biochar use, including the selection of feedstock, pyrolysis conditions, and application rates that are most likely to succeed in specific soil contexts. This intelligent approach facilitates a targeted, data-driven strategy for biochar deployment, optimizing its environmental benefits and ensuring a more effective response to soil contamination issues.

Few research on biochar amendment in contaminated soil moved forward to adjust the AI technology in various aspects. The interactions of heavy metals and biochar were predicted with the use of machine learning. By collecting the existing experimental data on biochar adsorption, feasible heavy metal immobilization capacity was predicted [34]. Not only focusing on the varieties of heavy metals but the adsorption capacity of different types of biochar was predicted with machine learning technology. The immobilization of heavy metals including Pb was examined with the adsorbents produced in distinct pyrolysis conditions such as by charcoal, biochar, hydrochar, modified biochar [14], and carbon nanotubes [19]. Moreover, focusing on the mechanism of biochar adsorption of Pb, the most effective soil and biochar physicochemical properties [25]. However, not much machine learning-adjusted research was conducted in compliance with the parallel incubation experiment.

The advent of biochar as a soil amendment has opened a new frontier in environmental engineering, promising an innovative solution to the pervasive issue of heavy metal contamination in agricultural soils. Biochar’s porous structure and high surface area confer an exceptional ability to immobilize contaminants like lead (Pb), mitigating their bioavailability and fostering soil health. However, the identification of ideal biochar for environmental remediation necessitates a sophisticated understanding of its interaction with diverse soil matrices and contaminants, which vary across temporal scales—from short-term to long-term—and under varying conditions, such as the presence of microplastics. Therefore, an advanced computational method particularly machine learning is implemented which revolutionizes this domain by enabling the understanding of biochar’s efficacy based on the prediction of the expansive datasets of soil characteristics measured post-incubation. Despite the promise of biochar, the existing studies have primarily been limited to static laboratory conditions, and the application of AI in dynamic, real-world scenarios remains underexplored. This research aims to bridge this gap by integrating AI predictions with empirical incubation experiments, offering a holistic view of biochar’s performance under realistic environmental conditions.

Materials and methods

With the understanding of the background for this study which highlights the potentiality and effectiveness of immobilizing lead (Pb) from contaminated sources, a framework is proposed to address such issue. In this study, contaminated soil samples were from an abandoned agricultural land adjacent to the Tancheon mine (36°44′N′05″N, 127°12′12″E) in Chungcheongnam-do, South Korea. The soils were used for a previous study for heavy metal(loid) immobilization [12]. Mining activity in the Tancheon mine was halted in 2008, but nearby agricultural lands were highly contaminated during the operation of the mine. Upon the identification of serious heavy metals contamination, agricultural activities in nearby lands were banned. The total Pb found in the heavily contaminated soil is 1,445 mg Pb/kg soil; details of the other characteristics of the soil sample can be referred to Table A1.

As shown in Fig. 1, the first step covers the preliminary experimental setups including the obtain enough contaminated soil and the biochar sample which originated from different sources of biomass (see Sect. 2.1 for more details). Then, an incubation sample (mixture of biochar with contaminated soil) is prepared and incubated at the greenhouse for different periods which is then evaluated with its characteristics (see Sect. 2.2). The results of on-field experimentation are then analyzed and compared accordingly with different types of biochar sets (including control: no biochar applied) based on the measured characteristics. Lastly, an AI-guided prediction on the Pb immobilization effect of long-term incubation was conducted covering several approaches which include: (a) data augmentation with the relatively small amount of data obtained from experimentation, (b) development of prediction model including the hyperparameter optimization, and (c) results of the prediction approach (see Sect. 2.3).

Fig. 1

An overall framework proposed to address the objective of determining the effectiveness of Pb immobilization with biochar

Microcosm experimentation

Contaminated soil was obtained and air-dried for storage to retain the origin of the sample from the abandoned agricultural land. Then, the soil samples were sieved through 2 mm size before the incubation. The 600mL HDPE bottle with an aeration lid was used. In each bottle, 100 g of soil was thoroughly mixed with 2.5 g of biochar (2.5% w w − 1). Deionized water was added to maintain 70% of the water holding capacity which was calculated as 30 mL. The incubation was performed inside the dark container which was maintained at room temperature.

Two major scenarios: short-term (ST) which incubated for 21 days and long-term (LT) incubation which lasted for 214 days. An extra scenario in which variation was induced which microplastic is added to the incubation sample at day 90 to identify the effect of changes in Pb immobilization. Microplastic is frequently reported with its existence in land which involves human activities (i.e., fertilizer input to the agricultural land) and is prone to public health concerns. A total of 99 sample sets were prepared (33 for each scenario) which included 11 different biochar samples and the control (no biochar was applied); each incubation set was prepared in 3 repetitive attempts. Once the incubation period was achieved, soil samples were removed from bottles and air-dried for 2–3 days. All the samples were stored in sampling bags for further soil analysis.

Standard biochar samples were produced by the United Kingdom Biochar Research Center (UKBRC). A total of 10 kinds of biochar were selected, which consisted of five kinds of feedstocks and two levels of pyrolysis temperatures. The feedstocks include miscanthus straw (MSP), oilseed rape straw (OSR), rice husk (RH), softwood pellet (SWP), and wheat straw pellet (WSP). Pyrolysis temperature is distinguished as 550℃ and 700℃ representing medium- and high-temperature conditions. The combination of feedstocks and temperatures were abbreviated with MSP 550, MSP 700, OSR 550, OSR 700, RH 550, RH 700, SWP 550, SWP 700, WSP 550, and WSP 700.

The characteristics of standard biochar were verified by UKBRC (Table 1) and details on the production can be found in the literature [20]. All types of standard biochar are basic in pH ranging from 7.91 to 10.4. This feature acts as a promising factor for soil remediation from Pb contamination since most heavy metals reduce soil pH in acidic conditions. The majority of standard biochar is carbon, and the maximum content of total carbon was 90.2% in SWP700. However, in the case of rice husk, the carbon content was not more than half of the materials. Standard biochar contains nitrogen sources up to 1.39%. In the case of softwood pellet-derived biochar, the total nitrogen contents were undetectable.

Table 1 Physicochemical properties of standard biochar provided by UKBRC

Soil and biochar analyses

The soils were characterized by their physicochemical properties at the incubation stage. Soil texture was analyzed in a previous study by the pipette method [11]. Both soil pH and electrical conductivity (EC) were determined with a ratio of 1:5 (soil to deionized water suspension). The exchangeable cations and available Pb content were analyzed by the 1 M ammonium acetate (NH₄OAc at pH 7) solution. K⁺, Ca²⁺, Mg²⁺, Na⁺, and Pb²⁺ were extracted and measured using an inductively coupled plasma optical emission spectrophotometer (ICP-OES). Total nitrogen content and total carbon content were determined using Element Analyzer. The morphology and microstructure of the standard biochar were analyzed using scanning electron microscopy (SEM, Hitachi, SU 70, Japan) equipped with embedded electronic dispersive x-ray spectroscopy at an accelerating voltage of 15 kV. The samples for SEM were prepared on conductive carbon tape and platinum-coated, respectively.

Prediction model development

As introduced earlier, this study focused on the effect of immobilization effect of Pb with various types of biochar samples prepared which were incubated alongside the contaminated soil sample obtained in the short- and long-term incubation period. Yet, the incubation period is relatively long (214 days) and several uncertainties are bound within (i.e., unpredictive immobilization effect after a long period of incubation). This is then overcome with artificial intelligence (AI) methods to predict and identify the long-term effect of the Pb immobilization while applying the biochar sample. However, low data availability is a strong concern as there is no luxury while preparing multiple experimental setups in real-case scenarios (in this case, three repetitive attempts were made for each biochar type only). A data augmentation approach is first implemented where its projectiles more datasets mimicking the real data obtained from the field-site experiments with the aid of the Gaussian Mixture model (except the sum of basic exchangeable cations which is not augmented but a summation of the augmented cation of K⁺, Ca²⁺, Mg²⁺, and Na⁺). The overall explanation of data augmentation can be found in Appendix A1. A total of 1,000 data sets for each experimental setup is projected in a single planar with the aid of t-distributed Stochastic Neighbor Embedding (t-SNE) which reduces the data dimensions. The t-SNE is an unsupervised learning high-dimensional reduction method that is commonly used for multi-feature datasets. The governing equation for t-SNE is to minimize Kullback-Leibler Divergence, $\:KL\:\left(P\right|\left|Q\right)$ with the focus on probability distributions P in high-dimensional space and Q in low dimensional space. which minimizes mismatching between $\:{p}_{ij}$ and $\:{q}_{ij}$ as shown in Equ. 1. There have been a series of research articles implemented the t-SNE algorithm on wide ranges of reporting such as wastewater treatment plant, renewable energy, and organic contaminants by crops from soil [7, 32]; Lee and Hwangbo, [15]; Gao et al., [6].

$$\:KL\:\left(P\right|\left|Q\right)=\:\sum\:_{i\ne\:j}{p}_{ij}log\frac{{p}_{ij}}{{q}_{ij}}$$

(1)

Upon the preparation of the augmented dataset, all the 11 augmented features from field experiment including pH, EC, TN, TC, K⁺, Ca²⁺, Mg²⁺, Na⁺, Summation of cation, P₂O₅, and available Pb (see Sect. 2.2) are input to the subsequent prediction approach. Four different prediction models considered in this study are (a) Random Forest, (b) Extra Gradient Boosting (XGB), (c) Light Gradient Boosting (LGB), and (d) Catboost. A detailed explanation can be found in Appendix A1, which covers the prediction model approach and configuration (i.e., train test split, overfitting prevention, etc.). All the models are fine-tuned to achieve better performance with hyperparameter optimization which is selected based on their respective performance metrics. The hyperparameter for each prediction model is optimized with a randomized search on a given range of hyperparameters considered (see Table A2). The hyperparameters are selected according to the theoretical importance of each prediction algorithm. On top of that, the data is discriminated by dropping randomly and inducing noise to the dataset before the prediction model as the augmented data are biased towards the output of three repetitive experimental setups only which has no variation that aligns with real-case scenarios. The prediction model is then conducted to predict the Pb effect at the end of long-term incubation by inducing multiple inputs with the identified incubated sample characteristics as explained earlier including pH, EC, TN, TC, various exchangeable cations, and P₂O₅ (the remaining Pb after immobilizing at short term is one of the inputs) for both short-term and long-term incubation period.

All the hyperparameter-optimized models are then evaluated and compared among each other with the performance metrics considered in this study including mean absolute error (MAE), mean squared error (MSE), and coefficient of termination, R². MAE (see Equ. 2) measures the average absolute difference between the predicted values and the actual values. It provides a straightforward indication of how far off the predictions are from the actual values, with lower values indicating better performance. Where n represents several data points, $\:{y}_{i}$ is the actual (observed) value for the i^th data point and $\:\widehat{{y}_{i}}$ is the predicted value for the i^th data point. On the other hand, mean squared error (see Equ. 3) measured the average of squared differences between predicted and the actual values which penalized larger errors to provide insights into the model’s ability to capture data variation. Lastly, the R² value (see Equ. 4) measures the proportion of variance in actual values that can be explained with predicted values ranging between 0 and 1; which is the mean of actual value. The models are evaluated with the favor towards a smaller value for the MAE and MSE, but the higher the value is for R².

$$\:MAE=\frac{1}{n}\text{}{\sum\:}_{i=1}^{n}\left|{y}_{i}-\:\widehat{{y}_{i}}\right|$$

(2)

$$\:MSE=\frac{1}{n}\text{}{\sum\:}_{i=1}^{n}{\left({y}_{i}-\:\widehat{{y}_{i}}\right)}^{2}$$

(3)

$$\:{R}^{2}=1-\text{}\frac{{\sum\:}_{i=1}^{n}{\left({y}_{i}-\:\widehat{{y}_{i}}\right)}^{2}}{{\sum\:}_{i=1}^{n}{\left({y}_{i}-\:\stackrel{-}{y}\right)}^{2}}$$

(4)

Results and discussions

Effect of biochar on soil chemical properties

Standard biochar amendment increased pH and electrical conductivity (EC) in Pb-contaminated soils (refer to Figure B1 (a) and (b)). Through the soil incubation, the standard biochar derived from oil seed rape and wheat straw pellet in both pyrolysis temperatures (550 ℃ and 700 ℃) increased the soil pH at a statistically significant level. Among them, the pH of OSR 700 treated soils rose to the highest at 7.03. It is parallel with the pristine property of standard biochar, that the OSR 700 possesses the highest pH value as shown in Table 1. The amendment of alkaline OSR 700 biochar changed the soil pH from acidic to neutral and fostered more favorable conditions for Pb immobilization.

In most standard biochar, higher pyrolysis temperature (700 ℃) led to higher and more basic pH values of biochar itself. Some studies reported the rising temperature of pyrolysis led to an increase in pH, which results supported the properties of standard biochar [17, 37]. Though the fundamental pH of standard biochar is basic, it does not always change the soil pH as well. Excluding WSP and OSR biochar, the standard biochar pH value ranges from 7.91 to 9.81 and did not alter the soil pH compared with control soil. As an example, the soil pH of the OSR 550 amendment was statistically higher than that of the RH 700 amendment, while the fundamental pH was found to be higher in RH 700. Thus, the pH of biochar itself may not directly affect the soil pH through the incubation.

In the case of soil EC, it rose in most of the treatments under the incubation. After the incubation, the EC of OSR 700 amended soils was 3.36 times higher than the control soil. The oil seed rape biochar rose the EC the highest followed by wheat straw and rice husk biochar in both pyrolysis temperatures. The increment of EC ranged from 0.023 dS/m in WSP 550 to 0.190 dS/m in OSR 700. Conversely, the softwood pellet biochar did not affect soil EC as well as pH. Two levels of pyrolysis temperature were not varied in EC values, which was not significantly varied with control soils. SWP 700 amendment even slightly decreased the soil EC less than the control. The soil pH and EC resulted in a positive correlation with Spearman correlation analysis under a significant level of 0.05 (Fig. 2). The addition of alkaline standard biochar affected pH and EC in a parallel way.

Fig. 2

Spearman correlation heatmap of soil chemical properties post-incubation with biochar amendment. The figure presents the interrelationships among soil properties with red and blue indicating positive and negative correlations, respectively. Values marked with an asterisk indicate statistical significance. *EC: Electrical Conductivity, TN: Total Nitrogen, TC: Total Carbon, Sum: Sum of exchangeable cations

The sum of basic exchangeable cation content including Ca²⁺, Mg²⁺, K⁺, and Na⁺ tends to follow the trends of soil pH and EC (see Figure B1(c)). The oil seed rape-derived biochar amended soils contained the highest exchangeable cations, especially OSR 700 rendered the concentration the highest at 9.96 cmol⁺/kg, after the wheat straw pellet and miscanthus straw pellet biochar. Among the four major cations in soil, the concentration of K⁺ has drastically increased the biochar produced from the three types of feedstocks. Consistently, the total K content (wt%) of OSR, MSP, and WSP biochar was the largest. The naturally contained cation in standard biochar directly affected soil cation contents. However, the effect was not linear.

Available phosphate in biochar-amended soils ranged from 189.0 mg/kg to 241.3 mg/kg in short-term incubation, and 163.1 mg/kg to 211.3 mg/kg in long-term slightly decreased except OSR 550 and MSP 700 (see Figure B1(d)). Among all the standard biochar types, only OSR 550 treated samples contained significantly higher concentrations in long-term incubated soils. In the aspect of available phosphate, no noticeable alteration by biochar addition or different incubation time was found in either short- or long-term incubation.

Comparing the content of four exchangeable cations (Ca²⁺, Mg²⁺, K⁺, and Na⁺) in short- and long-term incubation, the extended incubation increased cation contents such as Ca²⁺, Mg²⁺, and K⁺ in every standard biochar treated soil (see Figure B2), which correlated with the previous results [13]. However, Na⁺ content is oppositely decreased in most soils. The content of exchangeable Ca²⁺ was significantly decreased even during the extended incubation period (see Figure B2(d)). OSR 700 biochar contained a distinguishably large amount of ammonium acetate extractable Ca²⁺ up to 5.316 cmol⁺/kg, while other biochar was not significantly different from the control soils. Similarly, in long-term incubated soils, only oil seed rape-derived biochars significantly increased exchangeable Ca²⁺. According to the previous finding, the higher temperature of pyrolysis brings out increased CEC as a result of abundant minerals such as Ca that exist on biochar [1, 17].

In the case of Mg²⁺, the notable difference was found in oil seed rape biochar as well, but in OSR550 that the pyrolysis temperature was low (see Figure B2(b)). Including all the standard biochar addition in both short- and long-term incubation, OSR550 was the only treatment that significantly increased the exchangeable Mg²⁺ up to 1.105 cmol⁺/kg and 1.008 cmol⁺/kg respectively. However, a notable decrease from short-term to long-term was found, while wheat straw pellet biochar maintained the same.

Most standard biochar amendments increased exchangeable K⁺ content in soils (see Figure B2(c)). The dramatic change was found in OSR and MSP-originated biochars in short-term incubation. The OSR700 treatment contained the highest exchangeable K⁺ at 3.4240 cmol⁺/kg which is more than five times of control. However, the softwood biochars had no significant differences from the control. The exchangeable K⁺ content in long-term incubation was slightly different in the order of standard biochars, which was ranked as oil seed rape > wheat straw pellet > miscanthus straw pellet > rice husk. MSP 550 decreased 1.093 cmol⁺/kg of exchangeable K⁺ from short-term to long-term incubation which is the largest gap. The results are approximately parallel with fundamental biochar properties, that oil seed rape biochar contains the largest total K (wt%) and softwood biochar is the lowest (see Table 1).

Extension of the incubation period led to increasing exchangeable Na⁺ contents in standard biochar-amended soils except for OSR 700 treatment (see Figure B2(a)). In short-term incubation, the standard biochars that were pyrolyzed at higher temperatures (700 ℃) contained more ammonium acetate exchangeable Na⁺ than 500 ℃ pyrolyzed biochar. The significant differences between biochar types were diminished in long-term incubation. Softwood pellet-derived biochar in both pyrolysis temperatures exceptionally maintained negligible change of Na⁺ contents.

Only a few standard biochar altered the total nitrogen contents and the difference between short and long-term incubation was valid in the total nitrogen content of rice husk, softwood pellet, and wheat straw pellet added soils. The nitrogen contents were higher in long-term incubation while that of control soils was the opposite (see Figure B3(a)). In short-term incubated soils, all the nitrogen contents were not significantly different except OSR550 and SWP550 biochar in the convert way. While the long-term incubated soils showed similar trends in oil seed rape and wheat straw pellet biochar including two temperatures by increasing the nitrogen contents. In some biochar that contains high nitrogen from its feedstock, the alteration of soil nitrogen happens once the biochar is mixed with soils. Thus, it is parallel with the native standard biochar total nitrogen contents that ranged from 1.29 at OSR 550 to < 0.1 at SWP 550 and SWP 700. However, even with the highest contents of nitrogen, OSR 700 significantly affected soil nitrogen content only in long-term incubation.

Since biochar mainly consists of carbon derived from biomass feedstock, the addition of biochar directly leads to an increase in total carbon contents in soils (see Figure B3(b)). Given the similarity of biochar with carbon sources to soils that act as soil organic matter once it is applied to the soils, the soil organic matter content in treatments was all distinguishably increased. The high soil organic matter implies higher water capacity with enhanced fertility. The soil organic matter contents of treatments varied during the incubation period. Rice husk, Softwood pellet, and wheat straw biochar increased soil organic matter significantly over a longer period. While the biomass feedstocks maintained similar trends without significant differences in the pyrolysis temperatures.

The addition of standard biochar to Pb-contaminated soils instigated a discernible modulation of the soil’s microbial ecosystem, as evidenced by the differential relative abundances of microbial phyla observed (Fig. 3A). The control sample, devoid of biochar, displayed a microbial community with a substantial representation of Proteobacteria, Actinobacteria, and Chloroflexi, phyla commonly associated with nutrient cycling and soil structural integrity. In contrast, biochar-amended soils demonstrated a proliferation of Acidobacteria and Verrucomicrobia in treatments such as MSP 550 and MSP 700, taxa often linked to the degradation of complex organic compounds and prevalent in nutrient-limited environments, suggesting an enhanced capacity for organic matter stabilization and potential Pb immobilization.

Fig. 3

A. Relative abundance of microbial phyla in Pb-contaminated soil amended with biochar. The biochars were derived from Miscanthus Straw (MSP), Oilseed Rape Straw (OSR), Rice Husk (RH), Softwood Pellets (SWP), and Wheat Straw Pellets (WSP), at pyrolysis temperatures of 550℃ and 700℃, labeled as MSP 550, MSP 700, OSR 550, OSR 700, RH 550, RH 700, SWP 550, SWP 700, WSP 550, WSP 700 respectively, alongside one untreated control. B. SEM-EDX images of biochar which were used in Pb sorption experiments: (a) OSR 550 (b) OSR 700 (c) SWP 700 (d) WSP 700. The orange squares indicate the spots where the EDX peak was obtained

Fig. 4

Effectiveness of Pb immobilization by applying biochar to the contaminated soil. The biochars were derived from Miscanthus Straw (MSP), Oilseed Rape Straw (OSR), Rice Husk (RH), Softwood Pellets (SWP), and Wheat Straw Pellets (WSP), at pyrolysis temperatures of 550℃ and 700℃, labeled as MSP 550, MSP 700, OSR 550, OSR 700, RH 550, RH 700, SWP 550, SWP 700, WSP 550, WSP 700 respectively, alongside one untreated control. The immobilization capacity was based on the untreated control for each short-term (ST) and long-term (LT) incubation. The star marks denote the biochar amendment with the increase of immobilization capacity in time

The soil environment, rich in diverse media, serves as a crucial determinant in the proliferation and function of different microbial taxa. For instance, biochars derived from high-temperature pyrolysis, such as OSR 700, favored an increase in Firmicutes, a phylum known for its resilience to stress and its members’ role in metal resistance and reduction. Similarly, the augmentation of Gemmatimonadetes in RH 550-amended soils underscores the interaction between biochar’s physicochemical properties and microbial niche specialization. These bacteria, along with other observed phyla, could contribute to the immobilization of Pb through various mechanisms, including sorption processes, biofilm formation, and direct uptake by microbial biomass. This intricate synergy between biochar amendments and microbial dynamics underscores the potential for biochar to create a more hospitable soil environment that not only mitigates metal toxicity but also harnesses microbial processes for soil remediation.

In analyzing the SEM-EDX results, the biochars exhibit a porous structure conducive to adsorption processes (Fig. 3B). Although the SEM images did not show a clear change in the pore structure with the fabrication temperature, the previous study has shown that the pyrolysis temperature not only dictates the physical transformation of biochar but also its chemical functionality, crucial for adsorption processes [28]. As the temperature increases, the biochar undergoes extensive morphological changes, developing a more pronounced porous structure, thereby enhancing its surface area. Concurrently, this thermal process modulates the surface chemistry, augmenting the presence of functional groups that are instrumental in binding heavy metals. These transformations are consequential in defining the adsorption potential of the biochar, with higher temperatures typically promoting a more favorable adsorption profile due to the synergetic effect of increased surface complexity and chemical reactivity.

Pb immobilization in soil

The immobilization of Pb in soils with standard biochar amendment occurred distinctly in each short- and long-term incubation. However, overall standard biochar amended soils eluted less available Pb ions compared with the control. The immobilization of standard biochar successfully occurred, but differently by the incubation time, pyrolysis temperatures, and the biochar feedstocks. The concentration of available Pb was decreased by the incubation period. Following the concentration in control soils, all the standard biochar-amended soils immobilized Pb significantly. The difference between the two incubation periods ranged from 10.51 mg/kg at control to 2.095 at OSR 700. If separated by the incubation period, OSR 700 decreased the largest up to 13.10 mg/kg of available Pb in the short-term and 4.68 mg/kg in the long-term incubation. Thus, oil seed rape-derived biochar that was pyrolyzed at 700 ℃ immobilized Pb in soil with the highest capacity.

The pyrolysis temperature also varied the immobilization capacity of standard biochar. The higher capacity was revealed at 700℃ by OSR700, RH700, and WSP 700 in the short term while only OSR550 maintained the statistically analogous value. Among them, OSR700 ranked with the highest immobilization capacity followed by OSR550. The averages of standard biochar in the same pyrolysis temperature in each incubation period also revealed identical trends. The available Pb was extracted from the standard biochars in short-term incubation with an average of 9.02 mg/kg and 8.46 mg/kg and long-term with 4.19 mg/kg and 3.86 mg/kg that pyrolyzed at 550 ℃ and 700℃ respectively. A more detailed visualization can be referred to in Figures B4 and B5.

However, the most notable difference in immobilization capacity occurred in the types of feedstocks. The highest available Pb reduction was rendered in oil seed rape-derived biochar regardless of the incubation period or pyrolysis temperature, with wheat straw pellet biochar followed. On the opposite side, softwood pellet-derived biochar was not valid in Pb immobilization by trapping similar amounts of available Pb with the control that has no additional biochar applied. The previous study investigated a similar result that the biochar application decreased Pb availability with the aspect of dissolved organic carbon [5].

Effect of microplastic in pb immobilization progress

A comparison between the short- and long-term incubation effect on Pb immobilization was discussed in the previous section. As mentioned in Sect. 2, there is a variation of this experimental setup which induces the microplastic at Day 90 of long-term incubation as a resemblance of the real case scenario where heavy human activity was involved. A visualization of an overall comparison between ST, LT, and LT-microplastic is shown in Fig. 4. In most cases, the results agreed with the findings earlier where OSR-type biochar (OSR 700 in specific) is the most effective in immobilizing Pb which the available Pb reported on average is 2.374 mg Pb/kg soil. Interestingly, most of the incubated samples (including the control set) resulted in a further immobilizing effect on Pb. The further immobilization percentage with the induced microplastic for the control, MSP 550, MSP 700, OSR 700, RH 550, RH 700, SWP 550, and SWP 700 are 7%, 4%, 8%, 3%, 10%, 9% 5%, and 3%, respectively. This is common where the induced microplastic could improve the immobilization effect with its surface functional group. However, only the incubated sample for OSR 550, WSP 550, and WSP 700 does not show any improvement in immobilizing the Pb (marked with a red asterisk in Fig. 5) as compared to the long-term incubation scenario without microplastic.

Fig. 5

Comparison between the incubation effect on Pb immobilization for short- and long-term alongside the induce of microplastic among different types of biochar application on contaminated soil. The biochars were derived from Miscanthus Straw (MSP), Oilseed Rape Straw (OSR), Rice Husk (RH), Softwood Pellets (SWP), and Wheat Straw Pellets (WSP), at pyrolysis temperatures of 550℃ and 700℃, labeled as MSP 550, MSP 700, OSR 550, OSR 700, RH 550, RH 700, SWP 550, SWP 700, WSP 550, WSP 700 respectively, alongside one untreated control

Detail interpretation across different incubation periods

As deduced from previous sections OSR 700 performed the best in terms of immobilizing Pb at all three incubation scenarios. Yet, there is a lack of understanding of such a phenomenon. This section will cover the interpretation from different perspectives including (1) the relation between the characteristic of each biochar type (as in Table 1) towards the Pb immobilization effect and (2) the overall effect identification on the incubated soil characteristics with different biochar type across all incubation scenarios. Figure 6 (a) shows a correlation plot with the Spearman correlation (significant level of 0.05) approach between biochar’s characteristics with the available Pb after incubation including short-term (ST_Pb), long-term (LT_Pb), and long-term with the presence of microplastic (LT (MP)_Pb). Whereas any biochar characteristics that responded with a relation index larger than 0.5 (regardless of the magnitude) are considered with a significant correlation. Among all the features, the moisture, Total N, Total P, Total K, and pH act similarly where the higher the specific featured value, the less Pb immobilization effect. As the total ash, Total C, Total H, and total surface area only show a strong relation during the short incubation period; this is likely due to the immobilization capacity of biochar already occupied in the long term (including the presence of microplastic) incubation. Interestingly, the relation of the Pb and other features of biochar improved with the presence of microplastic at long-term incubation which shows the potentiality of microplastic in immobilizing heavy metal by surface functional group.

Fig. 6

Detail interpretation of the immobilization effect of biochar at different incubation periods: (a) correlation between remaining Pb with the characteristics of biochar and (b) overview of the soil characteristics after incubation with PCA. The biochars were derived from Miscanthus Straw (MSP), Oilseed Rape Straw (OSR), Rice Husk (RH), Softwood Pellets (SWP), and Wheat Straw Pellets (WSP), at pyrolysis temperatures of 550℃ and 700℃, labeled as MSP 550, MSP 700, OSR 550, OSR 700, RH 550, RH 700, SWP 550, SWP 700, WSP 550, WSP 700 respectively, alongside one untreated control

With a successful understanding of the biochar feature on the Pb immobilization, further exploration of the effect of Pb immobilization effect with other soil incubation characteristics can be found in Fig. 6(b) with the aid of principle component analysis (PCA). This approach reduces the dimensionality of all incubated soil characteristics with biochar applied into two planar of PC1 and PC2 across all incubation periods that ellipses with 95% confidence interval (ST: red, LT: blue, and LT w/ MP: Green). The first principal component (PC1) appears to demonstrate samples along the axis influenced by pH, TC, and EC which play a significant role in the impact of biochar treatment; the second PC (PC2) shows a strong association towards available Pb. The treatments grouped under ‘Short-term’ indicate that certain biochars may rapidly influence soil pH, nutrient availability, and Pb immobilization. In contrast, the ‘Long-term’ and ‘Long-term w/ microplastics’ clusters suggest that the enduring effects of biochar on soil conditions evolve. The presence of microplastics might modulate these long-term effects, potentially by affecting the soil’s physical structure, biota, or the biochar’s stability and interaction with soil contaminants. Treatments near the ‘Short-term’ label that are closely clustered suggest a consistent and immediate effect on soil conditions, whereas widely dispersed treatments in the ‘Long-term’ region demonstrate diverse outcomes, possibly due to varying degrees of biochar decomposition, interaction with soil matrices, or changes in bioavailability of pollutants. A broad ellipse in the ‘Long-term w/ microplastics’ zone implies that the interaction between biochar and microplastics yields a more variable modification of soil properties over time. A clearer representation of the comparison for both the long-term and long-term with microplastic can be referred to Figure C12.

Prediction of long-term incubation

The prediction approach on the long-term incubation effect based on the description earlier (also in Appendix A1) is successfully conducted. Data augmentation is conducted in which each experimental setup with 1,000 datasets in both short-term and long-term. The dataset of an overall augmentation approach is shown in Fig. 7(a) visualized in t-SNE methods that reduce the dimensions of the overall dataset; a different projection of the t-SNE for short- and long-term incubation can be referred to in Figures C1 and C2. The distribution of each characteristic for both short- and long-term incubation is shown in Fig. 7(b) which is demonstrated with OSR 700 biochar sample experimental setup. The distribution of other biochar samples in both the short- and long-term can be referred to in Figures C3 to C12. Then, the respective augmented data is input to the prediction model that has undergone hyperparameter optimization.

Fig. 7

Data augmentation approach of 1,000 different samples with Gaussian Mixture model: (a) a projection of the overall data generated covering short- and long-term incubation across different biochar samples (including control) and (b) demonstration of the distribution in terms of short- and long-term incubation of OSR 700 biochar sample

Each of the hyperparameter-optimized models is analyzed and compared with the performance metrics in terms of MAE, MSE, and R². An overall result for the predicted outcome is shown in Table 2 (the best model is highlighted accordingly). In most cases of the prediction approaches for each experimental setup, light gradient boosting outperformed the other model with better and more competitive performance metrics. Notably, only RH 550 and WSP 700 performed better in the hyperparameter-optimized Random Forest model, whereas MSP 550, RH 700, and WSP 550 performed better in the Categorical Boosting model. Upon identifying the best of each prediction model, the predicted available Pb is compared with the original available Pb as shown in Fig. 8(a) and (b) which are categorized with the pyrolyzed temperature of 550℃ and 700℃, respectively. In which, the distribution for the available Pb on both the predicted and original are displayed as histogram on both axial. This prediction approach successfully addresses and provides an understanding of the long-term incubation effect for Pb immobilization with the originally limited data.

Table 2 Performance metrics of different prediction models implemented on biochar

Fig. 8

Comparison of the predicted available Pb with the original augmented data based on the best prediction model obtained for each biochar type: (a) biochar sample pyrolyzed at 550℃ and (b) 700℃

Interpretation with SHAP analysis

An extensive interpretation of the model prediction approach is required as it is commonly recognized as a black box model with lack of understanding towards the outcome. Herein, SHAP analysis is used as a detailed interpretation of how various soil characteristics influence the predicted availability of lead (Pb) in soils treated with biochar at both short and long-term incubations. The SHAP analysis is based on the augmented datasets for each biochar sample (including absence of biochar: Control) MSP, OSR, RH, SWP, and WSP. The best prediction model for each biochar sample identified earlier is used for the SHAP analysis. Which, most biochar samples showed the best performance with light gradient boosting; except MSP 550 & RH 700 with Catboost, and RH 550 &WSP 700 with Random Forest. A demonstration with biochar sample OSR 700 in terms of global and local interpretation can be referred to in Fig. 9.

Fig. 9

Interpretation of the influential soil characteristics for the available Pb after incubation with SHAP analysis with OSR 700 biochar sample: (a) local interpretation at short term incubation, (b) global interpretation at short term incubation, (c) local interpretation at long term incubation, and (d) global interpretation at long term incubation

From the SHAP waterfall plots (Fig. 9a, c), it’s evident that certain features like phosphorus (P₂O₅), total carbon (TC), and pH consistently appear to have significant impacts on the predicted Pb levels. For instance, higher values of P₂O₅ and TC often lead to an increase in the predicted Pb availability, as indicated by positive SHAP values (red bars). This could be due to the role of phosphorus in potentially forming complexes with lead, thereby increasing its availability, while organic carbon from biochar might enhance soil properties that mobilize or stabilize lead. Conversely, features such as potassium (K⁺) and calcium (Ca²⁺) typically exhibit negative SHAP values (blue bars), suggesting their presence might reduce Pb availability. This could be attributed to the competition between these cations and lead for adsorption sites on biochar, potentially reducing the bioavailability of lead.

The global interpretations shown in Fig. 9b and d with dot plots reveal the broader influence of these features across the dataset. The spread of SHAP values across high and low feature values highlights the nonlinear and complex interactions within the soil matrix. For example, higher pH values tend to lower Pb availability, possibly due to increased precipitation of lead compounds at higher pH levels. The SHAP analysis for the other biochar sample can be referred to in Appendix C: Figure C13 to C22.

These insights not only shed light on the specific actions of different soil amendments when combined with biochar but also underscore the complexities of soil chemistry in contaminant dynamics. Understanding these relationships helps in fine-tuning biochar applications to maximize remediation efficacy and can guide future modifications to biochar composition or treatment strategies to target specific soil types and contamination scenarios effectively. This analysis emphasizes the nuanced role of AI in enhancing predictive accuracy and providing a robust scientific basis for soil remediation practices.

Future work

This study developed and validated models using soil from contaminated post-mining sites to reflect common Pb remediation scenarios. The models demonstrated strong predictive abilities during on-field testing within predefined conditions. Preliminary analysis revealed that these models retain predictive reliability even with varied soil parameters, though the range tested was limited. Future research will broaden the models’ applicability by incorporating a more diverse dataset, including soils from different geographical locations and varying contamination levels. This strategy will train our AI systems on a wider data array, enhancing their ability to generalize and perform effectively under real-world conditions.

Future studies will test the data augmentation method across varied datasets and conditions to evaluate its broad application efficacy. This involves altering key soil properties in our training datasets to represent a wide array of environmental scenarios. Such rigorous testing is designed to confirm the robustness of our approach in supporting soil remediation efforts, ensuring that our models are both theoretically sound and practically viable. These efforts are intended to demonstrate the practical utility of the models, affirming AI’s crucial role in environmental engineering, especially in applying biochar for soil remediation.

As a pioneering study in this field, our work details how AI facilitates the strategic application of biochar for Pb remediation. The potential for AI to enhance predictive capabilities and application strategies for biochar in environmental engineering is significant. The AI-driven methodology employed allows for the evaluation of long-term biochar impacts on soil health and Pb immobilization, optimizing biochar application tailored to specific site conditions.

Looking ahead, it is vital to expand AI models to cover more environmental conditions and biochar types, broadening our understanding of biochar’s effectiveness across different ecosystems and pollution levels. By incorporating diverse real-time datasets encompassing various soil types and contamination levels from multiple geographic regions, we aim to enhance the models’ generalization capabilities and ensure their effectiveness under different real-world conditions. This expansion will allow our AI systems to better predict the interactions of biochar with varying soil properties, thus optimizing remediation strategies more effectively.

Further studies will rigorously test these models against new datasets and under different remediation scenarios to validate the robustness and applicability of the data augmentation techniques used. We plan to integrate real-time data from environmental sensors into our models, which will help refine predictions and make the AI tools more responsive to on-the-ground changes in soil conditions. Additionally, developing a user-friendly AI platform that can be easily accessed by environmental engineers in the field will democratize advanced computational resources, making cutting-edge AI analytics a standard part of environmental remediation projects globally.

Moreover, we will explore the economic and ecological impacts of various biochar applications through AI simulations. These insights will aid stakeholders in making informed decisions regarding environmental management practices. Collaborative efforts that combine AI with advancements in nanotechnology could lead to the development of next-generation biochar with enhanced adsorptive and reactive properties, potentially revolutionizing Pb remediation strategies. Emphasizing these interdisciplinary approaches will underscore the transformative potential of AI in biochar technology, advancing both environmental health and biochar application methodologies.

To standardize the properties of biochar in Pb immobilization on soils, the homogeneously produced standard biochar was applied and incubated. In parallel with the original properties that standard biochar possesses, the soil chemical properties and Pb immobilization capacities were varied. Those properties were derived from the type of feedstock or pyrolysis temperatures, while some factors were affected by short- and long-term incubation.

Oil seed rape biochar performed the highest Pb immobilization in all temperature and incubation periods, showing that the features by feedstock most strongly engaged on it. Other than that, OSR 550 or OSR 700 were the highest in pH, EC, exchangeable Ca²⁺, Mg²⁺, and K⁺, total nitrogen content, soil organic matter content, and available phosphate. The result suggests that among the biochar tested, oil seed rape biochar was the most promising substance for soil amendment on heavy metal immobilization. Furthermore, due to the significant difference between oil seed rape-derived biochar in the pyrolysis temperature and incubation period, a more specific and sectionized study would be required to identify detailed properties that may exist.