Method for estimating leaf water use efficiency based on multi-source feature coupling and application

Abstract

The present application relates to the field of agricultural informatization and crop water physiological monitoring technology, solves the technical problems of insufficient estimation accuracy, single response factor, lack of causal (physiological) explanation and engineering application closed loop in the existing leaf water use efficiency estimation method, especially relates to a leaf water use efficiency estimation method based on multi-source feature coupling and application, synchronously collects photosynthetic parameters, environmental factors and electrophysiological parameters to construct a feature vector when the leaf is fully expanded, uses an integrated learning model to predict the photosynthetic parameters and calculate the leaf water use efficiency; SHAP is used to give feature contribution and identify the main control factor; a WUE thermal map is generated by combining coordinate interpolation for zoned irrigation and agronomic optimization, and the sliding time window is used for online updating to improve the timing robustness. Field verification of cotton, corn and wheat shows that the method has high precision, strong adaptability and is easy to deploy and popularize.

Description

Technical Field

[0001] This invention relates to the fields of agricultural informatization and crop water physiology monitoring technology, and in particular to a method and application for estimating leaf water use efficiency based on multi-source feature coupling. Background Technology

[0002] Water is one of the key ecological factors limiting crop yield and quality. How to achieve precise regulation of crop water use and improve leaf water use efficiency (WUE) in field settings has always been a core issue in dryland and water-saving agriculture. Two commonly used indicators at the leaf scale are instantaneous water use efficiency. and intrinsic water use efficiency These values ​​reflect the carbon assimilation capacity per unit of transpiration loss and per unit of stomatal conductance, respectively. Accurate, dynamic, and leaf-scale estimation of WUE helps identify stress responses, implement differentiated irrigation, and conduct variety evaluation, thus having direct production and scientific significance.

[0003] There are two main approaches to existing technologies: one is to rely on portable photosynthesis instruments (such as LI-6400, TPS-2, etc.) to measure the net photosynthetic rate. Pore ​​conductance and transpiration rate Then calculate the instantaneous water use efficiency according to the formula. and intrinsic water use efficiency The first approach offers high measurement accuracy, but its limitations, such as time-consuming single-point operations, sensitivity to microclimate fluctuations, and limited sample throughput, make it difficult to achieve robust monitoring of high frequency, multiple leaves, and long-term time series in real fields. The second approach uses a data-driven method, employing environmental factors such as temperature, humidity, light, and CO2, or indirectly inferred internal CO2 concentrations, as inputs to construct predictive models. While this improves efficiency, the lack of direct characterization of leaf structure and water-conducting / retaining capacity results in insufficient generalization and stability across different crops, time periods, and plots. Furthermore, the "black box" nature of the predictions limits interpretability, making it difficult to support subsequent physiological diagnosis and regulatory decisions.

[0004] It is worth noting that plant electrophysiology and electrical impedance characterization have been explored in areas such as water stress identification and ion flux monitoring, showing potential to reflect tissue water content and structural state. However, in the field estimation of leaf water resistance (WUE), there is still a lack of systematic standards regarding measurement conditions (such as consistency of pressure application and frequency setting), parameter derivation and quantification methods, quality control, and reproducibility. More importantly, there is a lack of a standardized and reusable technical approach in the existing literature on how to effectively couple electrical information with photosynthetic parameters and environmental factors to steadily improve the prediction accuracy and robustness of leaf-scale WUE.

[0005] Furthermore, the link from "numerical estimation" to "management application" remains weak in existing technologies: on the one hand, field data has obvious time-varying and heterogeneous characteristics, and the model lacks an online update mechanism to adapt to the dynamic environment; on the other hand, agricultural management focuses more on spatial distribution rather than single-point values, but spatialization links such as coordinate interpolation for leaf-scale WUE, anisotropic processing, cross-validation selection, and raster resolution have not yet formed common practices, making it difficult to stably support the identification of inefficient areas and the generation of prescription maps.

[0006] In summary, existing technologies still have significant gaps in the accurate, robust, and interpretable prediction of leaf-scale WUE, as well as in the spatialized expression and operable output for field management. There is an urgent need to develop a systematic technology that takes into account physiological characterization, predictive consistency, and practical application. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method and application for estimating leaf water use efficiency based on multi-source feature coupling. It belongs to the interdisciplinary field of agricultural engineering, plant phenotyping, and intelligent sensing technology, and is particularly suitable for application scenarios such as field management of dryland crops, water use efficiency assessment, and irrigation optimization decision support.

[0008] To address the technical problems existing in current leaf water use efficiency estimation methods, such as insufficient estimation accuracy, single response factor, lack of causal (physiological) explanation and engineering application closed loop, this invention proposes an intelligent estimation method for leaf water use efficiency (WUE) based on multi-source coupling features and interpretable machine learning, and adopts the following technical solution.

[0009] A method for estimating leaf water use efficiency based on multi-source feature coupling, the method comprising the following steps: S1. After the target plant leaves are fully expanded, measure the net photosynthetic rate of the target plant leaves using a portable photosynthesis measurement device. Pore ​​conductance and transpiration rate The photosynthetic parameters were calculated and used as measured values. Instantaneous water use efficiency Compared with intrinsic water use efficiency ; S2. Simultaneously acquire environmental factors reflecting the external regulatory conditions of the plant's photosynthetic-transpiration system during the measurement period, including air temperature (T), relative humidity (RH), light intensity (PAR), and atmospheric conditions. Concentration and wind speed (WS); S3. Using an electrophysiological measurement device, apply at least three levels of clamping force to the target plant leaf, measure its physiological capacitance C and resistance R at a preset measurement frequency, and calculate the leaf capacitive reactance value Z to complete the acquisition of electrophysiological parameters. S4. Establish a multi-source feature fusion dataset including photosynthetic parameters, environmental factors, and electrophysiological parameters. ; S5. Fusion of multi-source feature datasets As input to an interpretable machine learning model, supervised learning is employed using measured values. The model is trained as a supervised target to obtain an estimation model for estimating the photosynthetic parameters of plant leaves. S6. Estimated values ​​based on the output of the estimation model Recalculate the estimated value This enables intelligent estimation of leaf water use efficiency.

[0010] Furthermore, the photosynthetic parameters were measured at a light intensity of 800 μmol·m⁻¹. -2 ·s -1 The experiment was conducted under standard conditions where the blade chamber temperature was 20-25℃.

[0011] Furthermore, the instantaneous water use efficiency Compared with intrinsic water use efficiency The calculation formula is: The measured value It is used to reflect the CO2 fixation capacity under unit water loss and is an important indicator for measuring the water-saving performance of crops.

[0012] Furthermore, the at least three clamping forces include 0.5N, 1.0N, and 1.5N, and the preset measurement frequency is 10kHz.

[0013] Furthermore, the calculation of the leaf volumetric resistivity Z is performed at a preset measurement frequency and is used to characterize the physiological state of the leaf tissue's ability to retain water and conduct water. The calculation formula is as follows: in, The frequency of the electrical signal is the preset measurement frequency.

[0014] Furthermore, the interpretable machine learning model is an ensemble learning model XGBoost with feature importance score output function or a neural network model containing an attention mechanism, and is trained using supervised learning.

[0015] Furthermore, during the training process of the interpretable machine learning model, the measured instantaneous water use efficiency is used. Compared with intrinsic water use efficiency As the target output, backpropagation optimization is performed using mean squared error (MSE) or mean absolute error (MAE) as the loss function. Among them, the multi-source feature fusion dataset The training data consists of measured samples from multiple historical time periods, and a sliding time window is used to extract the training data.

[0016] Furthermore, the estimated value The expression is: The estimated value The calculation formula is: in, These are the net photosynthetic rate, transpiration rate, and stomatal conductance output by the estimation model, respectively. These are the estimated instantaneous water use efficiency and intrinsic water use efficiency, respectively.

[0017] Furthermore, the method for estimating leaf water use efficiency also includes: Output estimated value The system outputs importance scores for identifying the main driving factors among the input variables, and generates WUE spatial heat maps for field water management and zoning decisions based on the sampled coordinates using interpolation methods.

[0018] Furthermore, the model output includes, in addition to, the estimated instantaneous water use efficiency. Compared with intrinsic water use efficiency In addition, it includes contribution scores, sensitivity analysis results, and visualization ranking information for each input feature variable, which are used to assist in the diagnosis and regulation analysis of water use efficiency.

[0019] Furthermore, this method is applicable to estimating the water use efficiency of field leaves of dryland crops, including crops such as cotton, corn, and wheat, and can be used to support agricultural application scenarios such as irrigation optimization decision-making, cultivation management regulation, and variety selection.

[0020] By employing the above technical solution, the present invention provides a method and application for estimating leaf water use efficiency based on multi-source feature coupling, which has at least the following beneficial effects: This invention improves the accuracy and completeness of leaf-scale WUE estimation. In addition to photosynthetic parameters and environmental factors, it introduces electrophysiological impedance features, enabling the input features to simultaneously cover the three physiological pathways of assimilation, transpiration, and structural / hydrothermal processes. This overcomes the dimensional limitations of relying solely on external factors or single physiological quantities, and facilitates more stable estimation performance under heterogeneous field conditions.

[0021] This invention constructs a machine learning framework with interpretable outputs, facilitating factor tracing and mechanism interpretation. It employs XGBoost or attention mechanism networks to directly output importance scores / sensitivity, enabling the ranking and directionality determination of input features based on their contribution, such as identifying the net photosynthetic rate of WUE. Pore ​​conductance transpiration rate The dominant influence of factors such as leaf bulk resistance value Z is used to transform "numerical results" into "diagnostic information" to support subsequent physiological analysis and control design.

[0022] Enhanced time-series robustness to adapt to dynamic field environments. This invention constructs training / prediction samples through a sliding time window and updates parameters on a rolling basis, mitigating the impact of short-term weather fluctuations and equipment drift on the model. This ensures that the estimation results remain continuous, stable, and usable under different weather conditions and growth stages, facilitating long-term online deployment and maintenance.

[0023] This invention enables spatial representation from point to surface, directly serving zone management. It is based on sampled coordinate pairs. , Interpolation mapping is performed (ordinary kriging or inverse distance weighting can be selected as needed, and the interpolation strategy and raster resolution can be determined through cross-validation), and a WUE spatial heat map is output. This can intuitively identify areas with inefficient water use and differentiated regions, providing a quantitative basis for zoned irrigation, crop layout, and field operation and maintenance.

[0024] It is highly feasible and reproducible, and easy to integrate with existing field equipment. This invention only requires a conventional portable photosynthesis meter and electrophysiological measurement device to complete data acquisition; the key measurement conditions (at least three levels of clamping force, preset frequency, and Z-calculation formula) are clearly defined and reproducible; the data format and training process are compatible with mainstream platforms, making it easy to promote and apply across multiple plots, crops, and teams.

[0025] It has a wide range of applications and is easy to integrate into systems to form a closed-loop engineering system. It is suitable for estimating leaf WUE of dryland crops such as cotton, corn, and wheat. It can be integrated with field phenotypic monitoring platforms, mobile diagnostic devices, farmland IoT nodes, and intelligent irrigation control systems to form a closed-loop path of "online estimation - spatial mapping - zoning decision-making", thereby improving the operability and timeliness of water-saving management. Attached Figure Description

[0026] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of the leaf water use efficiency estimation method in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the measurement of the blade static electrical impedance parameters in Embodiment 2 of the present invention; Figure 3 This is a thermal map showing the spatial distribution of WUE (wax-elemental geometry) of cotton leaves in the field in Embodiment 2 of the present invention. Figure 4 The estimation model for cotton leaves in Embodiment 2 of the present invention and The prediction fitting results are shown in the figure; Figure 5 This is a visualization of the feature influence based on SHAP analysis in Embodiment 2 of the present invention; Figure 6 The estimation model in Embodiment 3 of this invention is used for corn and wheat leaves. and The predicted-measured fitting results are shown in the figure. Figure 7 The corn and wheat in Example 3 of this invention The main feature contribution comparison results are shown in the figure. Detailed Implementation

[0027] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This will allow for a full understanding of how the present application uses technical means to solve technical problems and achieve technical effects, and to facilitate its implementation.

[0028] Example 1: To address the technical problems of insufficient estimation accuracy, single response factors, lack of causal (physiological) explanation and engineering application closure in existing leaf water use efficiency estimation methods, this embodiment proposes a leaf water use efficiency estimation method based on multi-source feature coupling. In addition to photosynthetic parameters and environmental factors, electrophysiological impedance features are introduced, enabling the input features to simultaneously cover the three physiological pathways of "assimilation-transpiration-structure / hydroconduction." This overcomes the dimensionality deficiency of relying solely on external factors or single physiological quantities, and is beneficial for maintaining more stable estimation performance under heterogeneous field environments. Figure 1 As shown, the method includes the following steps: S1. Construct a multi-source input feature system at the blade scale.

[0029] This step aims to comprehensively acquire multidimensional information on leaf photosynthetic physiological activities, environmental stress states, and structural-functional responses, serving as the foundational data source for subsequent estimation models. Furthermore, to improve the timeliness and representativeness of the data, the following four types of parameters are used as input: S11. Acquisition of photosynthetic parameters: Using a TPS-2 portable photosynthesis system, at a light intensity of 800 μmol·m⁻¹, photosynthetic parameters were obtained. -2 ·s -1 Net photosynthetic rate was measured under standard conditions of leaf chamber temperature 20–25℃. Pore ​​conductance and transpiration rate Therefore, the instantaneous water use efficiency can be calculated. Compared with intrinsic water use efficiency The formula is as follows: This parameter reflects the CO2 fixation capacity per unit of water loss and is an important indicator for measuring the water-saving performance of crops.

[0030] S12. Environmental Factor Acquisition: Air temperature (T), relative humidity (RH), light intensity (PAR), and atmospheric temperature are acquired via environmental sensors. concentration Parameters such as wind speed (WS) reflect the external regulatory conditions of the plant's photosynthetic-transpiration system.

[0031] S13. Electrophysiological parameter acquisition: Using an electrophysiological measuring device under at least three clamping forces (e.g., 0.5N, 1.0N, 1.5N), at a preset measurement frequency. (Preferably at approximately 10kHz) Measure the static capacitance C (pF) and static resistance R (kΩ) of the blade respectively, and calculate the capacitive reactance Z of the blade. The calculation formula is as follows: in, The frequency of the electrical signal is denoted as Z. The leaf volumetric impedance value Z is used to quantify the degree to which cell structure inhibits water channels and is an indicator reflecting the plant's water conductivity and structural integrity.

[0032] By combining the inputs of the above multi-source factors, a high-dimensional feature space can be constructed, providing comprehensive representational support for WUE modeling and estimation.

[0033] S14. Feature Summarization: Forming a multi-source feature fusion dataset for modeling. ,Right now: Simultaneously, the sampling coordinates (latitude and longitude / in-field coordinates) and timestamps are retained to support the spatialization and serialization processing in step S5.

[0034] S2. Build and train an interpretable machine learning model.

[0035] This step, while maintaining interpretable output, learns and models multi-source features to directly predict... Then calculate the estimated value. .

[0036] S21. Model and Input: An interpretable ensemble learning model (such as XGBoost) or a neural network model incorporating an attention mechanism can be used, respectively denoted as... , The input is from step S14. Its predicted output for: S22. Estimated Value Calculation: Calculate the predicted value... Substituting into the formula in step S11, we get: S23. Training and Validation: Based on actual measurements , The value is used as a monitoring signal, and the mean squared error (MSE) or mean absolute error (MAE) is selected as the loss function. A regularization term is added to control the model complexity. The optimization objective is as follows: or, in, For the first The measured label values ​​of each sample (which can be...) , or one of), For the corresponding predicted value, For the sample size, For the set of model parameters, This is the regularization coefficient. Training / validation / test splits or cross-validation (such as 5-fold) can be used to evaluate the generalization ability of the estimation model.

[0037] S24. Sliding Time Window Update (connected to step S4): To adapt to time-varying conditions in the field, training data is updated according to a sliding time window (e.g., ...). The model is constructed and updated on a rolling basis to ensure its stability under different meteorological stages and fertility processes.

[0038] S3. Construct a feature importance interpretation module.

[0039] To achieve "interpretable output," variable importance and sensitivity analysis is performed on the trained estimation model. Methods such as Shapley values ​​(SHAP) or feature gains can be used to output the importance and sensitivity of each input variable. , The average contribution and directionality (positive / negative impact) of the data were visualized using ordinal plots, bar charts, or heatmaps. These results were used to identify key driving factors to aid in physiological diagnosis and management decisions.

[0040] S4. Introduce a sliding time window mechanism to enhance timing robustness.

[0041] Set window length (e.g., 7 days), with As a training set, As a forecast set, model parameters are updated daily (or weekly). This mechanism reduces the impact of short-term extreme weather or operational disturbances on the estimation results, improving robustness and usability over long-term series.

[0042] S5. Output the estimation results and provide spatial visualization and control prompts.

[0043] In output , Simultaneously, importance scores and sensitivity results are provided to facilitate the identification of key driving factors among the input variables. Based on the estimation results carrying spatial coordinates, an interpolation method is used to generate a WUE spatial heat map, realizing the distribution expression from point to field. This WUE spatial heat map is used to visually identify water use inefficient areas and differentiated regions, providing support for zoned irrigation and cultivation management. When necessary, a "prioritize intervention in low-value areas" strategy can be provided in conjunction with management objectives (e.g., labeled according to quantile thresholds).

[0044] S6. Scope of application and system integration path.

[0045] This method is applicable to estimating the water use efficiency of leaves in dryland crops (such as cotton, corn, and wheat). It can be integrated with field phenotypic monitoring platforms, mobile diagnostic devices, farmland IoT nodes, and intelligent irrigation control systems to achieve a closed-loop application of "online estimation - spatial mapping - zoning decision-making," and can be expanded to more scenarios as needed.

[0046] Example 2: This embodiment aims to fully verify the application effect and accuracy of the intelligent estimation method for leaf water use efficiency under actual field conditions, with particular attention to the model's predictive ability, the explanatory power of factor contributions, and its guiding value for differentiated water management in the field.

[0047] I. Experimental conditions and material selection.

[0048] The experiment was conducted in a cotton field in Shihezi, Xinjiang, using the widely cultivated local cotton variety "Xinluzhong 42" as the research subject. The experimental field covered an area of ​​approximately 200m × 100m, with a cotton planting density of approximately 100,000 plants / hm². 2 The experiment was conducted during the budding to boll-forming stage, following standard field management practices. Twenty sampling sites were set up, with five plants randomly selected from each site as the subjects of measurement to minimize individual differences.

[0049] II. Experimental Equipment and Measurement Methods.

[0050] To ensure measurement accuracy and reproducibility, a standardized process is adopted: (1) Measurement of photosynthetic parameters: Using a portable photosynthesis measurement system (TPS-2 model) from PP-SYSTEMS (USA), under clear weather conditions from 9:00 to 11:00, at a leaf chamber temperature of 22℃ and a light intensity of 800 μmol·m⁻², measurements were taken. -2 ·s -1 Under these conditions, the net photosynthetic rate of the upper functional leaves of each plant was measured. (μmol·m -2 ·s -1 ), transpiration rate (mmol·m -2 ·s -1 and porosity (mol·m -2 ·s -1 Two leaves were measured on each plant, and the average value was taken from three consecutive measurements on each leaf.

[0051] (2) Measurement of environmental factors: Temperature is recorded using an integrated multi-parameter weather sensor. (°C), relative humidity (%), photosynthetically active radiation (μmol·m -2 ·s -1 CO2 concentration (ppm) and wind speed (m·s) -1 The sampling interval was 10 minutes, and the average value of the photosynthesis measurement period was used as the model input.

[0052] (3) Measurement of electrophysiological parameters: A portable plant physiological electrical impedance analyzer was used, and the measurement frequency was set. The static resistance of the central part of the blade was measured under three clamping forces (0.5N, 1.0N, and 1.5N). (kΩ) and static capacitance (pF), the average of 3 values ​​for each range. Calculate the blade bulk modulus using the following formula. (kΩ): in, For electrophysiological measurements, the frequency (Hz) is used.

[0053] For example: a sample of a plant at a certain sampling point was measured. , Substituting into the above formula, we can obtain .like Figure 2 The diagram shown illustrates the principle of static impedance acquisition.

[0054] III. Feature Construction and Model Training.

[0055] Construct the input vector for modeling: .

[0056] Instantaneous water use efficiency based on actual measurements Compared with intrinsic water use efficiency As supervisory labels, a total of 300 valid samples were obtained, which were divided into training and test sets in an 8:2 ratio. The calculation formula is as follows: The XGBoost regression model was selected for training, and its parameters (learning rate 0.1, maximum depth 5, regularization parameter) were determined through multiple rounds of optimization. To adapt to the time-varying environment in the field, the training data uses a sliding time window (e.g., (Daily) Updated on a rolling basis.

[0057] IV. Model Prediction Results and Performance Evaluation.

[0058] The test set results are shown in Table 1, namely: Table 1. Results of the estimation model on the test set. The results show that the estimation model has high prediction accuracy for both WUE indices. A value greater than 0.85 indicates that the error is controlled within a small range, demonstrating that the estimation model of this invention is stable and reliable under actual field conditions. Figure 4 To estimate the model for cotton leaves and The predicted fitting results are shown in the figure, which verifies the regression accuracy of this method on the test set.

[0059] V. Feature contribution and factor explanatory analysis.

[0060] The SHAP feature contribution analysis module was used to interpret the model and obtain the ranking of the variables' contributions to WUE prediction. Examples are shown in Table 2.

[0061] Table 2. Ranking of Variables' Contribution to WUE Prediction Visible porosity Factors such as nitrogen, water content (Z), and temperature (T) are important influencing factors, providing clear scientific guidance for field water management. Furthermore, to more intuitively demonstrate the specific direction and magnitude of the impact of each input variable on the estimation model output, Figure 5 A visualization of the feature influence based on SHAP analysis is provided to represent the distribution of SHAP contribution values ​​of different input variables to the estimation model output. Each bar in the figure represents the distribution of a feature's SHAP value across different samples, with color representing the magnitude of the feature value (red for high values, blue for low values), and length representing the degree of influence on the prediction result. It can be seen that stomatal conductance... Not only does it have the highest average contribution, but it also has high porosity. Values ​​generally The model produces a positive impact, while low values ​​show a negative contribution; variables such as capacitance Z and temperature T also show a similar trend, further illustrating that the model has good interpretability for the direction of influence of the dominant factors.

[0062] VI. Spatial Distribution of WUE and Recommendations for Field Management Generate results based on estimations with spatial coordinates, such as... Figure 3 The heatmap shows the spatial distribution of WUE. The interpolation method used was either ordinary kriging (OK) or inverse distance weighting (IDW), with leave-one-out cross-validation used to select the optimal method. The grid resolution was selected based on the plot size and sample density (e.g., 1–5 m). The heatmap shows significant spatial heterogeneity within the experimental area: high-efficiency areas… The average soil moisture content was generally greater than 3.2 μmol / mmol, while it was lower than 2.4 μmol / mmol in inefficient areas. Combined with field surveys, soil moisture content in inefficient areas was generally below 45% of field capacity. Based on this, it is recommended to prioritize differentiated supplementary irrigation and monitor its effectiveness.

[0063] Example 3: This embodiment aims to verify the universality, robustness, and application potential of the intelligent estimation method for leaf water use efficiency under different crop types and field environmental conditions. Maize and wheat were selected as research subjects for systematic verification.

[0064] I. Experimental Site and Crop Selection.

[0065] The experiments were conducted from May to July 2024 in a maize experimental field in Bijie City, Guizhou Province, and from March to May 2024 in a wheat experimental field in Nanjing City, Jiangsu Province. The main local varieties selected were "Zhengdan 958" for maize and "Yangmai 16" for wheat. Conventional production management measures were adopted in both experimental sites, and the experiments were carried out during the key growth stages (jointing-tasseling stage for maize and jointing-flowering stage for wheat).

[0066] Fifteen sampling points were randomly set at each experimental site. Five plants were randomly selected from each sampling point as the test subjects, and two fully expanded functional leaves of each plant were measured.

[0067] II. Parameter Measurement Methods and Equipment

[0068] The experimental equipment and measurement methods are the same as in Example 2, including: (1) Measurement of photosynthetic parameters: A portable photosynthesis meter, model TPS-2, from PP-SYSTEMS (USA), was used. The observation period was from 9:00 AM to 11:00 AM, with the leaf chamber temperature set at 22°C and the light intensity controlled at 800 μmol·m⁻². -2 ·s -1 The net photosynthetic rate of the leaves at each sample point was measured. transpiration rate and pore conductance Each leaf was measured three times and the average value was taken.

[0069] (2) Measurement of environmental factors: Environmental parameters of the experimental field were collected in real time using an automatic weather station, including air temperature (T), relative humidity (RH), photosynthetically active radiation (PAR), carbon dioxide concentration (CO2), and wind speed (WS). The average value recorded every 10 minutes was taken as the data for the corresponding time period.

[0070] (3) Measurement of electrophysiological parameters: Using a portable plant physiological electrical impedance analyzer, set the measurement frequency. Three clamping pressures (0.5N, 1.0N, 1.5N) were used to measure the resistance R and capacitance C of the blade at each sample point. The average value of each pressure was taken to calculate the capacitive reactance of the blade. ,Right now: III. Data Construction and Model Training.

[0071] The same feature construction method as in Example 2 is used to form the feature input vector. .

[0072] Instantaneous water use efficiency was measured. and intrinsic water use efficiency The prediction target of the estimation model is: 200 valid samples were collected for each crop, and randomly divided into training and testing sets at an 8:2 ratio. XGBoost models were then built for training and validation on each set. To adapt to the time-varying environment in the field, a sliding time window (e.g., ...) was used, consistent with Example 2. (Days) Build training data and continuously update model parameters.

[0073] IV. Prediction accuracy and model performance evaluation.

[0074] The statistical results of the prediction performance of the estimation model on the two crop test sets are shown in Table 3.

[0075] Table 3. Predictive performance statistics of the estimation model on two crop test sets. The results show that the method of the present invention exhibits stable and good prediction accuracy under different crop types of maize and wheat. All values ​​were above 0.85, and MAE was below 0.02, indicating high generalization ability and cross-crop adaptability. The prediction-measurement fit for each crop and indicator is as follows: Figure 6 As shown.

[0076] V. Key Factor Contribution Analysis and Difference Analysis.

[0077] SHAP feature contribution analysis was performed on the estimation models to obtain the ranking of key influencing factors for different crops, as shown in the table below: Table 4 Ranking of key influencing factors for different crops The results showed that both were based on porosity. With leaf bulk resistance As the primary driver, but in CO2 concentration, , Differences exist in secondary factors, reflecting variations in environmental response and structure-hydraulic coupling among different crops. (Two crops) The main feature contribution to, for example Figure 7 As shown, a direct comparison can be made. Z, T, CO2 concentration Differences in the importance of factors such as [missing information].

[0078] VI. Field verification and application potential.

[0079] Further comparison was made between the spatial distribution of WUE predicted by the estimation model and the measured soil moisture content: the soil moisture content in high WUE areas of maize was generally greater than or equal to 70% of field capacity, while it was ≤50% in low-efficiency areas; for wheat, the high-efficiency areas had approximately ≥65%, while the low-efficiency areas had approximately ≤40%. The consistency between crop WUE and soil moisture content was good in both areas (correlation coefficients were both >0.75, p<0.01), verifying the accuracy of the prediction and its field diagnostic capability. For ease of management application, the WUE spatial heat map can be generated by interpolation (such as OK / IDW, selected through cross-validation) using the method in Example 2, for zoning identification and differentiated irrigation decisions.

[0080] The core of this invention lies in constructing a high-dimensional feature system that integrates multi-source information from photosynthesis, physiology, electrical impedance, and environment. Based on an interpretable machine learning model, it achieves accurate estimation of leaf water use efficiency and identification of key factors. It has high precision, strong generalization ability, and field applicability, providing new tools and theoretical support for precision irrigation management and efficient crop cultivation in agriculture.

[0081] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0082] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. Since the above embodiments are substantially similar to the method embodiments, they are described relatively simply; relevant parts can be referred to the descriptions of the method embodiments.

[0083] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for estimating leaf water use efficiency based on multi-source feature coupling, characterized in that, The method includes the following steps: S1. After the target plant leaves are fully expanded, measure the net photosynthetic rate of the target plant leaves using a portable photosynthesis measurement device. Pore ​​conductance and transpiration rate The photosynthetic parameters were calculated and used as measured values. Instantaneous water use efficiency Compared with intrinsic water use efficiency ; S2. Simultaneously acquire environmental factors reflecting the external regulatory conditions of the plant's photosynthetic-transpiration system during the measurement period, including air temperature (T), relative humidity (RH), light intensity (PAR), and atmospheric conditions. Concentration and wind speed (WS); S3. Using an electrophysiological measurement device, apply at least three levels of clamping force to the target plant leaf, measure its physiological capacitance C and resistance R at a preset measurement frequency, and calculate the leaf capacitive reactance value Z to complete the acquisition of electrophysiological parameters. S4. Establish a multi-source feature fusion dataset including photosynthetic parameters, environmental factors, and electrophysiological parameters. ; S5. Fusion of multi-source feature datasets As input to an interpretable machine learning model, supervised learning is employed using measured values. The model is trained as a supervised target to obtain an estimation model for estimating the photosynthetic parameters of plant leaves. S6, estimating value based on the output of the estimation model re-calculate the estimating value , and realize intelligent estimation of leaf water use efficiency.

2. The leaf water use efficiency estimation method according to claim 1, characterized by, The measurement of photosynthetic parameters is carried out under standard conditions of light intensity of 800 μmol·m -2 ·s -1 -25℃ and leaf chamber temperature of 20-25℃.

3. The leaf water use efficiency estimation method according to claim 2, characterized by, The instantaneous water use efficiency The intrinsic water use efficiency The formula for calculating the intrinsic water use efficiency is: The measured values The CO2 fixation capacity per unit water loss is an important index to measure water-saving ability of crops.

4. The leaf water use efficiency estimation method according to claim 1, characterized by, The at least three clamping forces include 0.5N, 1.0N, and 1.5N, and the preset measurement frequency is 10kHz.

5. The method for estimating leaf water use efficiency according to claim 4, characterized in that, The leaf volumetric resistivity Z is calculated at a preset measurement frequency and is used to characterize the physiological state of leaf tissue in terms of water retention and water conduction. The calculation formula is as follows: wherein is the electrical signal frequency, i.e. the preset measurement frequency.

6. The leaf water use efficiency estimation method according to claim 1, characterized by, The interpretable machine learning model is an ensemble learning model XGBoost with feature importance score output function or a neural network model containing an attention mechanism, and is trained using supervised learning.

7. The leaf water use efficiency estimation method according to claim 6, characterized by, In the training process of the explainable machine learning model, the measured instantaneous water use efficiency with intrinsic water use efficiency as the target output, the mean square error MSE or the mean absolute error MAE as the loss function for back propagation optimization; Wherein, the multi-source feature fusion dataset The training data includes measured samples of multiple historical periods, and a sliding time window is used to extract the training data.

8. The leaf water use efficiency estimation method according to claim 1, characterized by, the estimated value The expression for is: The estimated value The calculation formula is: wherein, respectively, are the estimated net photosynthetic rate, transpiration rate and stomatal conductance of the model output; respectively, are the estimated instantaneous water use efficiency and intrinsic water use efficiency.

9. The leaf water use efficiency estimation method according to claim 1, characterized by, Also includes: Output estimated value The system outputs importance scores for identifying the main driving factors among the input variables, and generates WUE spatial heat maps for field water management and zoning decisions based on the sampled coordinates using interpolation methods.

10. An application of a method for estimating leaf water use efficiency, characterized in that, The leaf water use efficiency estimation method as described in any one of claims 1-9 is applied to the estimation of leaf water use efficiency in dryland crops including cotton, corn, and wheat, and is also applied to agricultural application scenarios including supporting irrigation optimization decisions, cultivation management regulation, and variety selection.

Images