Natural disasters are inherently a phenomenon that has adverse consequences for society (Paprotny et al., 2018). It damages the living environment and life of human beings. Flood disasters, as one of the most common and expensive natural disasters, have caused huge losses to human lives and property (Hu et al., 2018; Ao et al., 2020). With the development of social industry and economy, the warming of the atmosphere caused by greenhouse gas emissions may increase the risk of river flooding (Hallegatte et al., 2013; Hirabayashi et al., 2013; Willner et al., 2018; Bloeschl et al., 2019). Therefore, many studies want to build a system for predicting flood risk (Alfieri et al., 2017; Shafizadeh-Moghadam et al., 2018; Choubin et al., 2019; Khosravi et al., 2019), and a variety of machine learning methods are used in these studies. Choubin et al. (2019) used multivariate discriminant analysis (MDA), classification and regression trees (CART), and support vector machine (SVM) algorithms to predict flood risk in Iran’s Khiyav Chai drainage basin. The results show that the residential areas at the outlet of the drainage basin are very susceptible to floods. Khosravi et al. (2019) adopted three Multi-Criteria Decision-Making techniques (VIKOR, TOPSIS, and SAW) and two Machine Learning methods (NBT and NB) to test the flood sensitivity modeling of the Ningdu River Basin in China. Finally, their research shows that the NBT model is a powerful tool for evaluating flood-prone areas, and can properly plan and manage flood disasters. Nevertheless, predicting flood risk cannot substantially reduce the life and economic losses of human society. Afforestation can strengthen the stability of water, soil, and carbon sinks in the forest ecosystem, thereby effectively coordinating the relationship between humans and the natural environment. A considerable number of studies have shown that afforestation can weaken the impact of global warming and effectively reduce the risk of river flooding (Hong et al., 2018, 2020; Liu X. et al., 2018; Forster et al., 2021). Thus, afforestation is widely used to resist flood disasters.

Plants have evolved numerous resistance mechanisms to resist flood disasters, including plant morphological Screening of Candidate Genesristics, metabolic responses, and molecular transcriptional regulation (Loreti et al., 2016; Du et al., 2017; Yin et al., 2017; Zeng et al., 2019; Lukic et al., 2020; Lee et al., 2021). Among the diverse plant populations, poplar has become the main flood-resistant tree varieties in flood-prone areas due to its rapid growth and flood resistance features. Many studies have shown that the root system is the key organ of poplar responding to Flooding stress (Coleman et al., 2000; Major and Constabel, 2007; Berhongaray et al., 2013; Ye et al., 2018; Gerjets et al., 2021). Flooding stress affects the diffusion of oxygen in plant root tissues. At the same time, it limits the mitochondrial respiration of root cells and accumulates toxic substances, which seriously affects its normal physiological activities (Arbona et al., 2008; Voesenek and Bailey-Serres, 2013; Tian et al., 2019). In addition, flooding stress will destroy the photosynthesis performance of plants, which will inhibit plant growth and biomass accumulation (Du et al., 2012; Zhu et al., 2016; Zheng et al., 2017; Xiong et al., 2019; Zhou et al., 2020). Flooding stress not only reduces the chlorophyll content of plants, but also reduces the carotenoid content (Zhou et al., 2017). Kreuzwieser et al. (2009) found that the metabolite changes occurred in leaves and roots of submerged poplar. Du et al. (2012) compared the physiological and morphological adaptability of two poplar clones (hypoxia-resistant and hypoxia-sensitive) to flooding, and Peng et al. (2018) monitored the different response mechanisms of these two clones of poplar to flooding stress. These studies have greatly promoted people’s understanding of the waterlogging resistance mechanism, and to a considerable extent, strengthened people’s resistance to flood risks. Thus far, there are still few studies on the influence of poplar on the waterlogging resistance factors. These factors include the intrinsic features of poplar trees (photosynthesis and chlorophyll fluorescence, etc.) and external environmental features (ambient temperature, humidity, etc.). As a popular research direction, machine learning has recently been gradually introduced into the field of plant science. For the research on the resistance of poplar to waterlogging, Xie and Shen (2021) used poplar photosynthesis features and external environmental factors to predict the waterlogging tolerance of poplar. By using the SVR method in machine learning, they confirmed the feasibility of applying photosynthesis and other characteristic parameters to predict poplar flood resistance. However, previous prediction studies did not consider important parameters such as chlorophyll fluorescence. Additionally, the related forecasting research is not systematic enough, and the corresponding investigation and research are still lacking.

Based on the above considerations, the main purpose of this article is to consider more comprehensive feature parameters and use a variety of machine learning methods to predict the flood resistance of poplar. At the same time, it aims to supplement and improve the key content and procedures of poplar flood resistance prediction. First of all, the evaluation indicators of waterlogging tolerance were well defined and explained. Then, 26 internal characteristics and external environmental factors of poplars were screened by using feature selection algorithms such as significance test and stepwise regression. Finally, four machine learning methods were used to establish the flood resistance models of poplar, and the results were comprehensively evaluated in detail. Compared with previous studies, this study supplements the evaluation index and prediction system of poplar waterlogging tolerance. The main contribution is that the definition and analysis of evaluation indicators for waterlogging tolerance have been improved, and more comprehensive characteristic parameters have been considered. Moreover, the feature selection, prediction methods, and evaluation indicators were adjusted, and more machine learning methods and results have been considered and analyzed. This research has enriched the prediction of poplar’s flood resistance, which is of great significance to poplar’s accurate flood resistance, intelligent selection of seedlings, and cultivation of high-quality saplings. Furthermore, to a considerable extent, it promotes the research of flood resistance mechanisms, which have great theoretical and practical value.

The methodology is divided into data processing, feature selection, model establishment and evaluation. The main procedures are shown in Figure 1, and the specific implementation steps will be introduced one by one below.

Machine learning is a field of artificial intelligence (AI). Compared with traditional statistical models, machine learning has higher performance, and at the same time, its complexity is relatively lower (Mekanik et al., 2013). In fact, before establishing the regression model of machine learning, we performed multiple linear regression (MLR) on the five variables (Fv, qN_Fo, Fm, H₂OS/H₂OR, and RH_S/RH_R) selected by the significance testing and correlation analysis. However, the results show that the coefficients of determination (R²) of MLR on the training set and test set are 0.5616 and 0.5172, respectively. The regression results are shown in Appendix Figure A1. Many studies have compared machine learning models with traditional statistical models (Aertsen et al., 2010; Rezaeianzadeh et al., 2014; Idowu et al., 2016; Johnson et al., 2016; Wang and Srinivasan, 2017). In most cases, machine learning models are better than traditional statistical models, such as linear regression. The model and the variables are not linearly related in most situations, and the variables involved are also multivariate. Therefore, more and more fields have begun to use machine learning algorithms. Even so, while machine learning has many advantages, it also has limitations. For example, many machine learning models lack interpretability and are prone to overfitting. For this reason, these problems still need to be considered in practical applications.

The risk of resisting flood disasters can be mainly divided into two aspects. One is to directly predict flood disasters in the risk areas, and take preventive measures before the disaster occurs, such as transferring personnel and valuable finances. Generally speaking, the key variables that need to be considered in flood forecasting include 25 factors such as water level, river flood, soil moisture, and rainfall (Maier et al., 2010). Among these key variables affecting flood forecasting, rainfall and the spatial examination of the hydrologic cycle have the most significant effects (Nourani and Komasi, 2013). Although many studies have predicted the risk of flooded areas (Sampson et al., 2015; Tehrany et al., 2015; Wang et al., 2015; Darabi et al., 2019), this method still cannot essentially eliminate the impact of flood disasters. At the same time, it is relatively difficult to predict flood disasters. Thus, people have to consider another method to resist flood disasters. Another way to resist the impact of flood disasters is mainly through building dams and afforestation. The key to afforestation is to understand the waterlogging resistance mechanism of plants. The research on the waterlogging resistance mechanism of plants mainly focuses on exploring the ways for plants to resist flood stress (Wang et al., 2013, 2021; Najeeb et al., 2015; Duy et al., 2016). These studies have analyzed the waterlogging resistance mechanism from the molecular level of proteins and metal ions. However, there are few studies on waterlogging resistance prediction, and a complete system is still lacking. Xie and Shen (2021) used the SVR method in machine learning to predict the waterlogging resistance of poplar. However, there are still some limitations in their studies, such as chlorophyll fluorescence features that have not been considered. Compared with the previous research, we considered more accurate feature parameters and more kinds of machine learning methods. Additionally, we improved the prediction system of poplar resistance to waterlogging and added two quantitative definitions of waterlogging resistance evaluation indexes, which has made considerable improvements.

This study used machine learning methods to predict the flood resistance of poplar. First, three indicators of flood resistance were defined and evaluated. Then, the data was processed, and feature selection and modeling evaluation were implemented. The whole process is intuitive and specific, which has perfected the research system of waterlogging tolerance prediction to a considerable extent, and at the same time, it has also promoted the research on the mechanism of waterlogging tolerance. This study helps researchers to screen out poplar varieties with strong waterlogging tolerance during the poplar sapling period. It can further cultivate high-quality poplar saplings to achieve the purpose of precise flood resistance. The results of the experiment show that the machine learning algorithm shows high accuracy in predicting the flood resistance of poplar, especially the random forest regression (RFR) and support vector regression (SVR) methods. The final result has certain practical value. In practical applications, these two algorithms can be used first. However, it must be mentioned that although 160 poplar samples were used throughout the experiment, only 20 poplar varieties were actually used for regression analysis. In addition, in the regression analysis, 80 poplar samples from the experimental group were used and averaged according to varieties. Since the waterlogging tolerance of different individuals may be quite different, the final result may deviate from the actual situation. But within the allowable range of error, our research mainly provides a way of predicting waterlogging tolerance and improving the system for predicting waterlogging tolerance. Future research can consider more poplar varieties to improve the universality and stability of the method. In addition, the quantitative relationship of poplar varieties’ impact on flood disasters can be considered. In a word, this research has great theoretical value and practical significance, and the proposed method can meet the actual engineering needs in a considerable range.

To predict the flood resistance of poplar, the author first analyzed the differences between the three evaluation indexes of flood resistance. Then, the final evaluation index of waterlogging tolerance was determined, and outliers were eliminated. For the selection of feature parameters, the first screening was carried out according to the significance test and correlation analysis, and then the three methods of hierarchical clustering, Lasso, and stepwise regression were adopted to screen the features for the second time. The selected features are interpretable and promote the understanding of poplar’s waterlogging resistance mechanism. Finally, four machine learning methods were used to predict and evaluate the flood resistance of poplar. The results show that the random forest regression and support vector regression methods are more precise. Nevertheless, it must be pointed out that there are only four groups of experiments and controls for each variety. Due to sample differences and randomness, the final result may deviate from the actual situation. Future research can consider more poplar species and sample sizes to improve the versatility and stability of the method.

This research has perfected the prediction system of plant resistance to waterlogging, and has important value for accurate flood resistance and scientific seedling selection. Meanwhile, it has also made a great contribution to a better understanding of the mechanism of waterlogging tolerance. The analysis process of this paper is clear and repeatable. When considering the features related to the flood resistance of poplar, the photosynthesis features, chlorophyll fluorescence features, and environmental features are comprehensively considered. After data processing, feature selection, and other operations, the machine learning models were used to predict the flood resistance of poplar. Finally, the regression results show that the random forest regression (RFR) and support vector regression (SVR) methods have high accuracy. On the test set, the coefficients of determination (R²) of the two methods are 0.8351 and 0.6864, respectively, the root mean square errors (RMSE) are 0.2016 and 0.2780, and the mean absolute errors (MAE) are 0.1782 and 0.2031. Based on the above conclusions, our research shows that combining photosynthesis, chlorophyll fluorescence, and environmental variables before flooding experiments, modeling and prediction of machine learning methods against waterlogging can achieve high accuracy, which is suitable for actual engineering problems.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

XX, XZ, JS, and KD participated in the conception and design of this research and revised the manuscript. XZ and KD carried out experiments and organized the database. XX and JS performed the statistical analysis and proposed the methodology. XX implemented visualization and wrote the manuscript. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.