*3.2. Farmers' Willingness to Remediate Soil*

The results of the SEM showed that the five exogenous latent variables explained 95% of the variability in farmers' willingness to remediate soil and that participation status and perceived benefits were the main factors influencing the strength of farmers' willingness to participate in soil remediation (Figure 5). The direct path coefficient of participation status on farmers' willingness to remediate was 0.42, and it indirectly affected farmers' willingness by affecting perceived benefits, with a total path coefficient as high as 0.86. There was a significant positive correlation between the two (*p* < 0.05), indicating that the more actively farmers participated in remediation, the stronger their willingness to remediate soil in the future. In terms of each observable variable of participation status, PS1 (farmland area for remediation) had the largest contribution, with a standardized coefficient of 0.98, followed by PS2 (project labor income) with a standardized coefficient of 0.78.

**Figure 5.** SEM of farmers' willingness to participate in soil remediation. Note: From the unstandardized regression weights, there were only two insignificant paths (*p* > 0.05): household endowments to farmers' willingness and technology characteristics to farmers' willingness. The reliability analysis and validity tests of the model are presented in Supplementary Tables S4–S6.

The path coefficient for the effect of perceived benefits on farmers' willingness to remediate was 0.49, with a significant positive correlation (*p* < 0.05), indicating that the higher the farmers' perceived benefits of participating in remediation, the stronger their willingness to remediate soil in the future. In terms of each observable variable of perceived benefits, the contribution of each observable variable was close, with a PB2 (subsidies for participation in remediation can cover losses) standardized coefficient of 0.78, followed by a PB1 (remediation does not affect food supply) standardized coefficient of 0.77, and a PB3 (participating in remediation can improve income) standardized coefficient of 0.75. This indicated that the farmers' perceived benefits were sourced more from the non-loss of contaminated farmland; that is, it was necessary to ensure that at least the farmland involved in soil remediation had an income comparable to that of farmland with normal production functions.

According to rational behavior theory, perceived value is the comparison between the benefits and risks that farmers experience from their behavior. Individual farmers' behaviors follows the paradigm of "cognitive trade-off–perceived value–willingness to act–behavioral response" in action logic [20,45]. Perceived benefit refers to the result of subjective evaluation made by individuals through product benefits, service quality, emotional satisfaction, and so on [46]. In the study of farmers' economic behavior, perceived benefits are considered an important basis [20]. The higher the level of farmers' perceived benefits, the higher the level of comprehensive assessment. A high level of perceived benefits reflects farmers' positive attitudes towards agricultural policies [23]. Similar results were obtained in this study, where farmers' perceived benefits from participation in current soil remediation projects to obtain land rent and labor income directly contribute to future willingness to remediate. The perceived risks came mainly from the possible risks associated with the technical characteristics of remediation, which were much smaller than the health risks of soil heavy metal stress, and had little impact on farmers' willingness to participate in remediation (path coefficient of 0.04, technical characteristics).

#### *3.3. Farmers' Technology Preference*

The RF results showed that technical characteristics were the most important factors in dichotomizing farmers' choice of phytoremediation or passivation; they followed the order: soil quality > secondary contamination > remediation period (Figure 6a). Although farmers' attention to technical characteristics had little effect on their willingness to participate in soil remediation, it directly affected their preference for remediation technologies.

**Figure 6.** Factors influencing farmers' technology preference. (**a**) The relative importance of the variables of farmers' technology preference. Frequency distribution of respondents' scores on (**b**) soil quality, (**c**) secondary pollution and (**d**) remediation period. Note: (**a**) was based on the results of the MeanDecreaseGini of the RF, and the sum of all factors was normalized to 100%. Then, the importance values of the independent variables were readjusted. The blue, yellow, green, and gray circles are technical characteristic variables, household characteristic variables, current participation variables, and personal characteristic variables, respectively. There were significant differences (*p* < 0.05) in the scores of soil quality, secondary contamination, and remediation period between the sample farmers who chose phytoremediation and passivation after the Mann–Whitney U test (data not normally distributed).

Specifically, the relative importance of soil quality was 17.1%. The scores of farmers who chose phytoremediation and passivation on soil quality were 3.57 ± 1.21 and 2.49 ± 1.04, respectively, indicating that farmers who paid more attention to the possible impact of remediation technology on soil quality were more inclined to choose phytoremediation (Figure 6b). The relative importance of secondary contamination was 16.8%. The scores of farmers who chose phytoremediation and passivation on secondary contamination were 3.57 ± 1.21 and 2.49 ± 1.04, respectively, indicating that farmers who were more concerned about possible secondary contamination from remediation technologies were more inclined to choose phytoremediation (Figure 6c). Passivation remediation has

been proven to be an effective, convenient, and low-cost remediation method [16,47], and many passivators have been widely used in remediation practices for metal-contaminated soils [48]. However, the durability of the stabilization effect of passivation under dynamic environmental conditions during long-term remediation and the possible negative effects of passivator application (secondary contamination) remain well-known bottlenecks [16,28,49]. At the same time, these technical barriers to passivation are the main factors limiting its application in agricultural soil remediation.

The relative importance of the remediation period was 11.5%. The scores of farmers who chose phytoremediation and passivation on remediation period were 4.11 ± 0.96 and 4.62 ± 0.97, respectively, suggesting that farmers who placed more value on the remediation period had a higher propensity for passivation (Figure 6d). Phytoremediation is recognized as an eco-friendly, green, and sustainable approach to soil remediation, but it also faces important performance and efficiency issues. For most heavy metals, remediation of contaminated soils by phytoremediation alone spans decades [50,51]. That is, agricultural production needs to be interrupted for a long period, which is also the main factor restricting farmers' choices of phytoremediation. From another perspective, a higher score (full score of 5) indicated that the remediation period was a factor that farmers care about when participating in soil remediation. The desire for healthy soil quality was evident in the fact that some farmers undertake a longer remediation period to bring their farmland back to full health.

#### *3.4. Farmers Feature Extraction*

Based on the results of the PCA (Supplementary Tables S8–S11) of the communality, annual income, non-farm income, and education were the factors that best summarized the characteristics of the farmers who selected phytoremediation (Figure 7a). Taking the range of obviously higher frequency distribution of the factors as the characteristics of the main farmers (households) who chose this technology, it can be seen that the main characteristics of farmers who chose phytoremediation were annual household income of 0−16,000 CNY ¥/y (Figure 7b), non-farm income of 0−4000 CNY ¥/y (Figure 7c), and education level of primary school (Figure 7d).

**Figure 7.** Characteristics of farmers intended for phytoremediation. (**a**) communality of characteristics of farmers intended for phytoremediation extracted by PCA. The dark blue, communality > 0.7. Frequency distribution of (**b**) income, (**c**) non-farm income, and (**d**) education level of farmers who chose phytoremediation.

Farm income, annual income, agricultural labor, and farmland were the factors that best summarized the characteristics of the farmers who chose passivation (Figure 8a). From the frequency distribution plot of each factor, the main farmer (household) characteristics for selecting passivation were farm income ≤ 0 (Figure 8b), annual income of 4000−8000 CNY ¥/y (Figure 8c), agricultural laborers of 3 (Figure 8d), and farmland area of 0.13−0.4 hm<sup>2</sup> (Figure 8e).

**Figure 8.** Characteristics of farmers intended for passivation. (**a**) communality of characteristics of farmers intended for passivation extracted by PCA. The dark yellow, communality > 0.7. Frequency distribution of (**b**) farm income, (**c**) income, (**d**) agricultural labor, and (**e**) farmland of farmers who chose passivation.

When formulating soil remediation policies, the farmers' characteristics archived by the local government can be used to preliminarily identify their possible soil remediation technology preference based on the above results, which can be included as one of the important considerations in the comparison of technology options to further ensure the successful implementation of soil remediation projects from the aspect of farmers' willingness to participate.

#### **4. Conclusions**

Based on 553 farmers' face-to-face questionnaire data from four heavy metal-contaminated agricultural soil remediation project sites in China, this study explored farmers' willingness and technology preference for agricultural soil remediation and their key influencing factors using SEM, RF, and other methods.


#### **5. Recommendations and Limitations**

The main conclusions of this study have the following implications for the promotion of soil remediation technology and the formulation of policies to improve farmer satisfaction. For scholars: first, further develop efficient and green passivators to overcome the sustainability and possible secondary pollution problems of passivation; second, further develop the corresponding activators with hyperaccumulator applications to improve the efficiency of phytoremediation. For companies: strengthen the long-term supervision after remediation to promote the safe application of passivators. For government: when applying phytoremediation, consider intercropping or crop rotation remediation modes of low-accumulation crops with hyperaccumulators without interrupting production; in areas where farmers' willingness to remediate soil is low, increase their satisfaction and willingness by raising the level of their perceived benefits. These suggestions for government and companies should be carefully considered when carrying out soil remediation to solve food safety problems, especially in agricultural countries with limited per capita arable land.

This study investigates farmers' willingness and technology preference to participate in soil remediation and the important influencing factors to provide a scientific basis to promote the remediation of heavy metal-contaminated farmland from farmers' perspectives. However, there are some limitations in this study. First, we only surveyed 553 farmers in 4 different regions of China, although the study areas have different pollution levels and cropping structures, the limited sample size may bias the results, and the generalizability to other countries and regions needs to be further verified. Second, we only investigated phytoremediation and passivation, the two most commonly used techniques for heavy metal contaminated farmland, without considering other techniques, such as alternative planting and deep plowing, which deserve further study. Third, our questionnaire did not involve subjective indicators, such as government implementation perceptions and subjective norms that may affect farmers' participation behaviors, which should be considered more comprehensively in future studies. Therefore, future study will consider more study areas, more technologies applied in remediation practices of heavy metal contaminated farmland, and it will construct a theoretical framework of farmers' participation behaviors toward different soil remediation technologies.

**Supplementary Materials:** The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/land11101821/s1, S1: Table S1 Survey area; S2: Survey questionnaire; S3: Table S2 Descriptive statistics of sample farmers; S4: SEM, Table S3 Variable description and data statistics of the SEM, Table S4 Latent variable reliability test, Table S5 KMO and Bartlett's test, Table S6 Model structure validity (model fitness); S5: Random forest, Table S7 Random forest variable descriptions and data statistics; S6: Principal component analysis, Table S8 KMO and Bartlett's test (phytoremediation), Table S9 Total variance explanation of PCA (phytoremediation), Table S10 KMO and Bartlett's test (passivation), Table S11 Total variance explanation of PCA (passivation).

**Author Contributions:** Conceptualization, J.Y.; data curation, J.Y.; formal analysis, Y.Y.; funding acquisition, J.Y.; investigation, Y.Y.; methodology, Y.Y.; project administration, J.Y.; resources, J.Y.; software, Y.Y.; supervision, L.W.; validation, L.W.; visualization, Y.Y.; writing—original draft, Y.Y.; writing—review and editing, L.W. and J.Y. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the National Natural Science Foundation of China (grant number 42077134) and the National Key Research and Development Project of China (grant number 2018YFC1802604).

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author.

**Acknowledgments:** Thanks to the residents in the study area for their support of the investigation. The authors are grateful to the three anonymous referees for very helpful comments and suggestions. **Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.
