Machine learning based vegetation transpiration simulation method and system

Abstract

This application relates to a method and system for simulating vegetation transpiration based on machine learning. The method includes: acquiring real-time vegetation monitoring data, and based on the real-time vegetation monitoring data, quantifying the stomatal aperture level of leaves to obtain a stomatal conductance function; based on the stomatal conductance function and the real-time vegetation monitoring data, solving for stem water potential, stem water potential change rate, and preliminary transpiration rate through coupled equations; simulating stem water storage changes based on the stem water potential and the stem water potential change rate to obtain a stem water storage sequence; inputting the preliminary transpiration rate and the stem water storage sequence into a residual prediction model to predict the preliminary transpiration residual value, obtaining a residual prediction value; and adding the preliminary transpiration rate and the residual prediction value to obtain a transpiration rate estimate. This method can introduce dynamic plant water capacity, simulate the dynamic process of stem water filling and releasing, and improve the reliability of vegetation simulation.

Description

Technical Field

[0001] This invention belongs to the field of machine learning simulation and prediction, and in particular relates to a method and system for simulating vegetation transpiration based on machine learning. Background Technology

[0002] With the development of ecohydrology and terrestrial hydrological simulation technologies, mechanistic models of vegetation transpiration based on the soil-plant-atmosphere continuum theory have emerged. These technologies attempt to quantitatively describe the complete transport process of water from the soil through plant roots, stems, and leaves, ultimately dissipating into the atmosphere, using physical equations. This provides a theoretical framework for the accurate simulation and prediction of water and carbon fluxes in ecosystems. Traditional methods for simulating vegetation transpiration primarily rely on empirical or semi-empirical parameterization of the relationships between key physiological processes and environmental factors.

[0003] In traditional techniques, stomatal behavior is typically characterized by treating atmospheric aridity and soil moisture stress as independent influencing factors, expressed as simple combination functions such as multiplication. The crucial buffering mechanism of water storage and release in plant stems is often handled using simplified assumptions of constant capacity or complete neglect. Furthermore, when faced with complex nonlinear interactions, models often exhibit limitations in scenarios outside the training data range.

[0004] However, current traditional approaches have significant problems. Oversimplification or neglect of stem water storage dynamics prevents models from reflecting the time-lag effects between water absorption, transport, and consumption, resulting in a mismatch between simulation results and measured data in terms of phase. Mechanistic models based on fixed parameters and forms often experience a sharp decline in simulation accuracy and generalization ability when dealing with boundary conditions such as extreme climates or water stress. These shortcomings collectively limit the practical application value of traditional methods in fields such as precision agriculture, water resource management, and climate change research. Summary of the Invention

[0005] Therefore, it is necessary to provide a machine learning-based vegetation transpiration simulation method and system that can introduce dynamic plant water capacity, simulate the dynamic process of stem filling and releasing water, and improve the reliability of the simulation, in order to address the above-mentioned technical problems.

[0006] Firstly, this application provides a method for simulating vegetation transpiration based on machine learning, including:

[0007] Real-time vegetation monitoring data was acquired, and based on this data, the stomatal aperture level of leaves was quantified to obtain the stomatal conductance function. The real-time vegetation monitoring data included vegetation environment data, soil data, and plant water potential parameters.

[0008] Based on stomatal conductance function and real-time vegetation monitoring data, stem water potential, stem water potential change rate and preliminary transpiration rate are solved by coupled equations.

[0009] Based on the stem water potential and the rate of change of stem water potential, the stem water storage change was simulated to obtain the stem water storage sequence;

[0010] The preliminary transpiration rate and stem water storage sequence are input into the residual prediction model to predict the preliminary transpiration residual value and obtain the residual prediction value.

[0011] The preliminary transpiration rate and the residual prediction value are added together to obtain the transpiration rate estimate; the transpiration rate estimate is used to characterize the simulation results of vegetation transpiration.

[0012] Furthermore, based on stem water potential and the rate of change of stem water potential, a stem water storage change simulation was performed to obtain a stem water storage sequence, including:

[0013] The instantaneous plant water capacity is calculated based on the stem water potential and the maximum water capacity.

[0014] The formula for calculating the instantaneous plant water capacity is as follows:

[0015] ;

[0016] In the formula, This represents the instantaneous water volume of the plant. To ensure the maximum water capacity under full water supply conditions, The attenuation coefficient is... The stem's water potential;

[0017] Based on the instantaneous plant water capacity, stem water potential change rate and simulation time step, the net change in stem water storage is calculated to obtain the stem water storage change.

[0018] The real-time stem water storage is obtained by summing the changes in stem water storage and the historical stem water storage, and then by integrating the real-time stem water storage and the historical stem water storage to obtain the stem water storage sequence.

[0019] Furthermore, the formula for calculating the coupling equation is as follows:

[0020] ;

[0021] in, For root hydraulic conductivity, For soil water potential, The stem's water potential, This represents the initial transpiration rate. This represents the instantaneous water volume of the plant. The stem water potential change rate is denoted as .

[0022] Furthermore, the formula for calculating the initial transpiration rate is:

[0023] ;

[0024] In the formula, This represents the initial transpiration rate. It is the reciprocal of the porosity conductivity value. For aerodynamic drag, The slope of the saturated water vapor pressure-temperature curve. Net radiation, For soil heat flux, air density, The specific heat of air at constant pressure. is the dry / wet gauge constant, and VPD is the saturated vapor pressure difference.

[0025] Furthermore, the residual prediction model was trained using the following method:

[0026] Time-synchronized data was obtained by extracting and aligning preliminary transpiration simulation values, environmental driving factor data, simulated stem water storage, and measured transpiration values ​​from historical datasets.

[0027] At each aligned simulation time point, preliminary transpiration simulation values, environmental driving factor data, and simulated stem water storage at the previous simulation time point are extracted from the time synchronization data to obtain a feature vector;

[0028] For each simulated time point, the target variable is obtained by subtracting the preliminary simulated transpiration value from the measured transpiration value.

[0029] By integrating feature vectors and target variables, an initial sample set is obtained, which is then divided into a training set, a validation set, and a test set to obtain the training dataset.

[0030] Based on the training dataset, the XGBoost model framework is trained to obtain the residual prediction model.

[0031] Secondly, this application also provides a vegetation transpiration simulation system based on machine learning, comprising:

[0032] The conductance module is used to acquire real-time vegetation monitoring data and, based on the real-time vegetation monitoring data, quantify the stomatal aperture level of leaves to obtain the stomatal conductance function; wherein, the real-time vegetation monitoring data includes vegetation environment data, soil data and plant water potential parameters;

[0033] The solver module is used to solve stem water potential, stem water potential change rate and preliminary transpiration rate based on stomatal conductance function and real-time vegetation monitoring data through coupled equations.

[0034] The simulation module is used to simulate changes in stem water storage based on stem water potential and stem water potential change rate, and obtain stem water storage sequence;

[0035] The prediction module is used to input the preliminary transpiration rate and stem water storage sequence into the residual prediction model to predict the preliminary transpiration residual value and obtain the residual prediction value.

[0036] The correction module is used to add the preliminary transpiration rate and the residual prediction value to obtain the transpiration rate estimate; the transpiration rate estimate is used to characterize the simulation results of vegetation transpiration.

[0037] Thirdly, this application also provides a computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement any step of the method provided in the first aspect of this application.

[0038] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any step of the method provided in the first aspect of this application.

[0039] The aforementioned machine learning-based vegetation transpiration simulation method and system acquires real-time vegetation monitoring data and, based on this data, quantifies leaf stomatal aperture levels to obtain a stomatal conductance function. The real-time vegetation monitoring data includes vegetation environment data, soil data, and plant water potential parameters. Based on the stomatal conductance function and the real-time vegetation monitoring data, stem water potential, stem water potential change rate, and preliminary transpiration rate are solved using coupled equations. Based on the stem water potential and stem water potential change rate, stem water storage changes are simulated to obtain a stem water storage sequence. The preliminary transpiration rate and stem water storage sequence are input into a residual prediction model to predict preliminary transpiration residual values, obtaining residual prediction values. The preliminary transpiration rate and residual prediction values ​​are summed to obtain an estimated transpiration rate. This estimated transpiration rate is used to characterize the simulation results of vegetation transpiration. This approach introduces a dynamic plant water capacity model, expressing stem water storage capacity as a function of stem water potential. Based on this, it calculates the dynamic changes in stem water storage, replacing the traditional assumption of constant water capacity. It quantitatively simulates the dynamic process of stem water release and filling on a diurnal scale, as well as the nonlinear decay of water capacity under drought stress. This effectively corrects the temporal phase deviation between simulated transpiration and measured sap flow data, more realistically reflecting the plant's water buffering mechanism. The XGBoost machine learning algorithm is introduced to perform residual correction on the initial transpiration simulation values. Combining the physical reliability of the mechanistic model under normal conditions with the powerful ability of machine learning to capture complex nonlinear relationships, especially in fitting extreme values, it can focus on learning the simulation bias of the mechanistic model under boundary conditions such as severe drought, sufficient water supply, or extreme high temperature. This significantly improves the model's generalization performance and extreme value simulation accuracy outside the training data distribution range. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 A schematic diagram illustrating the process of a machine learning-based vegetation transpiration simulation method provided in an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of the structure of a vegetation transpiration simulation system based on machine learning, provided in an embodiment of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0044] In one embodiment, such as Figure 1 As shown, a method for simulating vegetation transpiration based on machine learning is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0045] Step 101: Obtain real-time vegetation monitoring data, and based on the real-time vegetation monitoring data, quantify the stomatal aperture level of leaves to obtain the stomatal conductance function; wherein, the real-time vegetation monitoring data includes vegetation environment data, soil data and plant water potential parameters.

[0046] Among them, real-time vegetation monitoring data refers to the set of data collected in real time through sensor networks to describe the state of the vegetation-soil-atmosphere system.

[0047] Vegetation environment data are physical conditions of the atmosphere above the canopy, such as net solar radiation, air temperature, air humidity, and wind speed, which are used to quantify the atmosphere's ability to demand water.

[0048] Soil data refers to the moisture state of the soil in the root zone of plants, with soil water potential being the core, used to characterize the soil's water supply capacity.

[0049] Plant water potential parameters are relevant parameters used to determine or estimate leaf water potential. Leaf water potential is the energy state of water within the leaf and is a core internal physiological variable connecting soil water supply and atmospheric demand, playing a pivotal role in the stomatal conductance model of this embodiment.

[0050] The quantitative level of stomatal aperture refers to obtaining a quantitative index through calculation to characterize the degree of stomatal opening on the blade.

[0051] The stomatal conductance function is a specific mathematical formula used to calculate stomatal conductance, which is a quantitative index of the stomatal aperture level of a leaf. Its value directly determines the rate at which water vapor is lost from the leaf. This function is a model with a pre-defined form based on the theory of plant hydraulic processes, and its key parameters are closely related to leaf water potential.

[0052] The terminal synchronously receives and aggregates real-time vegetation monitoring data streams from atmospheric sensors, soil sensors, and plant physiological sensors. It invokes a pre-defined stomatal conductance model framework based on leaf hydraulic processes. This framework features a fixed structure but contains several coefficients or variables that need to be determined. Real-time monitoring data is used to determine the model's internal parameters. The terminal uses soil data, plant water potential parameters, and environmental data, following the principle of continuous water transport, to derive or correlate the leaf water potential-related states required by the model, thereby determining all undetermined parameters in the model. After completing the parameter determination process, an executable calculation formula is obtained, with all parameters having specific values ​​specific to the current vegetation state and environment. This specifies how to calculate the corresponding stomatal conductance output value from a future leaf water potential input value. At this point, a specific stomatal conductance value is not output, because the calculation of this function also requires leaf water potential, which will be solved synchronously as one of the unknowns in the entire coupled system.

[0053] For example, the formula for calculating the output value of porosity conductance is:

[0054]

[0055]

[0056]

[0057]

[0058] In the formula, This is the output value of porosity conductivity. For reference porosity conductance; The curvature parameter of the vulnerability curve; This refers to the atmospheric carbon dioxide concentration. This represents the diffusivity of water vapor relative to carbon dioxide. This is a correction factor; To determine marginal water use efficiency, it can be based on leaf water potential ( Perform the calculation:

[0059]

[0060] In the formula, Marginal water use efficiency under conditions of sufficient water supply and CO2 concentration. for The average value over the previous 24 hours; yes The sensitivity parameters. It can be seen that in this porosity conductivity model, and All with Related. By simultaneously linking VPD and SM, a model based on vegetation hydraulic processes is expected to better describe the regulatory effects of VPD and SM on stomatal conductance, where leaf water potential can be approximated as stem water potential.

[0061] Step 102: Based on the stomatal conductance function and real-time vegetation monitoring data, the stem water potential, stem water potential change rate, and preliminary transpiration rate are solved by coupling equations.

[0062] Specifically, the coupling equation is a mathematical relationship that describes the instantaneous balance of water intake and expenditure within a plant. At any given moment, the difference between the amount of water absorbed by the plant roots from the soil and the amount of water lost through transpiration from the leaves to the atmosphere must equal the rate of change in the amount of water stored in the stem.

[0063] Stem water potential refers to the energy state of water within the plant stem, and is usually a negative value. Its magnitude reflects the degree of water stress within the plant and is the core driving force behind water transport and distribution within the plant.

[0064] The stem water potential change rate refers to the instantaneous speed at which the stem water potential changes over time, i.e., whether it is decreasing at an accelerating rate, decreasing at a slower rate, or tending to stabilize.

[0065] The preliminary transpiration rate is a vegetation water transpiration rate initially calculated based on the current stomatal aperture and atmospheric environmental conditions using classical calculation methods based on energy balance and aerodynamics. This preliminary calculation is the baseline value for correction.

[0066] The terminal establishes two core computational relationships and links them together. The terminal calls the parameterized stomatal conductance function. The stomatal conductance function requires stem water potential, which is currently unknown. Therefore, the terminal prepares this function as a module to be calculated. The terminal reads vegetation environment data from the real-time vegetation monitoring data. A computational relationship is prepared; knowing the stem water potential, the stomatal conductance can be calculated using the stomatal conductance function. Then, combined with the environmental data, the corresponding preliminary transpiration rate is calculated using physical formulas, resulting in the first computational relationship. The terminal reads soil data from the real-time vegetation monitoring data and calls the plant's root hydraulic conductance parameters. Simultaneously, the terminal incorporates an instantaneous plant water capacity related to stem water potential, describing how stem water storage capacity changes with water potential. Substituting these terms into the coupling equation yields the second computational relationship, resulting in a combined system containing multiple equations. Stem water potential, stem water potential change rate, and preliminary transpiration rate are three interrelated unknowns that must simultaneously satisfy the above two computational relationships. The terminal employs numerical solution algorithms, such as iterative methods like the Newton-Raphson method or the Euler method with time-step integration. The solver provides initial guesses for the unknowns and then iteratively calculates, substituting the guesses into the two equations to check if the water balance conditions required by the coupled equations are met. If not, it automatically adjusts the guesses for the unknowns and recalculates until a self-consistent numerical solution that makes all equations true simultaneously is found. Once the numerical solution process converges, the terminal obtains a uniquely determined set of solutions for the current time point and simultaneously outputs the values ​​of stem water potential, stem water potential change rate, and preliminary transpiration rate.

[0067] Step 103: Based on the stem water potential and the stem water potential change rate, simulate the stem water storage change to obtain the stem water storage sequence.

[0068] Specifically, stem water storage change simulation refers to the calculation to track the dynamic process of the increase or decrease of water in the stem over time.

[0069] Instantaneous plant water capacity is a variable describing the water storage capacity of a plant stem, which changes dynamically with changes in stem water potential. When a plant is short of water and the water potential is very low, its water storage capacity will decrease accordingly.

[0070] The change in stem water storage refers to the specific amount by which the net increase or decrease in the water stored in the stem occurs within a very short calculation time interval.

[0071] The stem water storage sequence refers to a series of data points arranged in chronological order. Each data point records the total amount of water stored in the stem at the corresponding simulation time. The sequence fully demonstrates the continuous change of water storage from the start of the simulation to the current time.

[0072] Based on the current stem water potential value, the terminal calculates the instantaneous plant water capacity of the stem at the current moment according to the dynamic rules describing the change in water storage capacity with water potential. The terminal multiplies this instantaneous plant water capacity value by the stem water potential change rate, and then by a preset, very short calculation time step to obtain the change in stem water storage during the time interval from the previous time step to the current time step. The terminal reads the stem water storage at the end of the previous calculation time point from storage, adds the historical cumulative value to the newly calculated stem water storage change, and obtains the real-time total stem water storage at the end of the current time point. The newly calculated real-time total stem water storage is saved and added to a continuously growing record list in chronological order.

[0073] Step 104: Input the preliminary transpiration rate and stem water storage sequence into the residual prediction model to perform preliminary transpiration residual value prediction and obtain the residual prediction value.

[0074] The residual prediction model is a machine learning program that is pre-trained using a large amount of historical data. Its function is to learn the deviation pattern between the simulation results of the physical model and the actual observations, and to predict the deviation for new data.

[0075] The preliminary transpiration residual value refers to the difference between the vegetation transpiration rate actually measured by precision instruments and the preliminary transpiration rate calculated by the physical model, representing the simulation error of the physical model.

[0076] The residual prediction value refers to a predicted value of the magnitude of the simulation error of the physical model at the current moment, calculated and output by the residual prediction model after the current data is input into the residual prediction model.

[0077] The terminal extracts information from the available data at the current moment and combines it into a data packet with a specific format. This packet includes the initial transpiration rate value, the values ​​of various environmental driving factors monitored at the current moment, and the stem water storage value from the previous simulation step. The combined feature vector is then input into a pre-trained residual prediction model. This model contains a series of complex judgment rules learned from historical data. After analyzing and calculating the input features according to its built-in rules, the model outputs the residual prediction value.

[0078] Step 105: Add the preliminary transpiration rate and the residual prediction value to obtain the transpiration rate estimate; wherein, the transpiration rate estimate is used to characterize the simulation results of vegetation transpiration.

[0079] Among them, addition refers to the most basic arithmetic addition operation, which adds two values ​​together.

[0080] The transpiration rate estimate is the optimal estimate of vegetation transpiration rate obtained after the machine learning model corrects the errors in the initial results of the physical model. It represents the entire simulation system's judgment on the vegetation transpiration process.

[0081] In one embodiment, based on stem water potential and the rate of change of stem water potential, a stem water storage change simulation is performed to obtain a stem water storage sequence, including:

[0082] Step 201: Calculate the instantaneous plant water capacity based on stem water potential and maximum water capacity.

[0083] The formula for calculating the instantaneous plant water capacity is as follows:

[0084] ;

[0085] In the formula, This represents the instantaneous water volume of the plant. To ensure the maximum water capacity under full water supply conditions, The attenuation coefficient is... This refers to the stem's water potential.

[0086] Specifically, stem water potential refers to the water potential in the xylem of a plant stem, which is a negative value. Its magnitude reflects the degree of water stress in the plant and is a core variable that drives water transport and determines the water state of tissues.

[0087] Maximum water capacity refers to the maximum water storage capacity that a plant's stem tissue can achieve under ideal conditions where it is adequately watered and free from water stress. It is an inherent parameter of a species or individual and represents the upper limit of the stem's water storage potential.

[0088] The attenuation coefficient is an empirical parameter that controls the rate at which the stem water capacity decreases as the stem water potential decreases. The larger the value, the more sensitive the water capacity is to changes in water potential, and the faster the water storage capacity will decrease during drought.

[0089] Instantaneous plant water capacity refers to the actual, instantaneous water storage capacity of plant stem tissue under the current stem water potential conditions. It is a variable that changes dynamically with the real-time water status of the plant. Physically, it means the amount of water that the stem can absorb or release per unit land area when the stem water potential changes by 1 MPa.

[0090] The terminal reads the current stem water potential. It then retrieves the pre-set maximum water capacity and attenuation coefficient, representing the plant under adequate water supply, from the model parameter library. These three values ​​are substituted into the calculation formula. After calculation, the terminal obtains a specific value, the instantaneous plant water capacity. The instantaneous plant water capacity quantifies the stem's immediate buffer capacity as an internal reservoir under the current stem water potential. If the plant has sufficient water, its water storage capacity is strong; if the plant is under drought stress, the exponential term will cause the instantaneous plant water capacity to be significantly lower than the maximum water capacity, indicating a severe decline in water storage capacity.

[0091] Step 202: Based on the instantaneous plant water capacity, stem water potential change rate and simulation time step, calculate the net change in stem water storage to obtain the stem water storage change.

[0092] Specifically, the stem water potential change rate refers to the rate at which the stem water potential changes over time. It is obtained by synchronous coupling and describes whether the stem water potential is rising, falling, or stable, and how fast it changes.

[0093] The simulation time step refers to the time interval used by the model when performing numerical calculations. It is a preset constant that determines the time resolution of the model's calculations.

[0094] The change in stem water storage refers to the net increase or decrease in water stored in the stem tissue due to changes in stem water potential within a simulated time step. A positive value indicates an increase in stem water storage, while a negative value indicates a decrease in stem water storage.

[0095] The terminal performs a multiplication operation based on the definition of plant water capacity, which is the change in water storage corresponding to a unit change in water potential. It multiplies the current water storage capacity by the rate of change in water potential, and then by the duration of the change, to obtain the net change in the total water storage in the stem during that time period. After completing the multiplication operation, the terminal outputs the change in stem water storage, directly reflecting how much water was stored in or released from the stem during the previous simulation step, used to balance the plant's water balance. This is a key increment for updating the total water storage in the stem.

[0096] Step 203: Sum the changes in stem water storage and the historical stem water storage to obtain the real-time stem water storage, and integrate the real-time stem water storage and the historical stem water storage to obtain the stem water storage sequence.

[0097] Specifically, historical stem water storage refers to the total amount of water stored in the stem calculated and saved at the end of the previous simulation time step, representing the cumulative water volume of the stem up to the previous moment.

[0098] Real-time stem water storage refers to the total amount of water stored in the stem at the end of the current simulation time step. It is the latest estimate of stem water storage based on the latest water balance.

[0099] The stem water storage sequence refers to an ordered data set consisting of stem water storage values ​​at all time points arranged in simulated time sequence, which fully records the dynamic process of stem water storage evolution over time.

[0100] The terminal reads the historical stem water storage from its internal memory, representing the total water storage at the end of the previous time step. Simultaneously, it retrieves the stem water storage change representing the latest change and performs algebraic addition, adding the historical stem water storage to the stem water storage change. If the stem water storage change is positive, the total increases; if it is negative, the total decreases, completing the time-step update of the total stem water storage. After calculating the real-time stem water storage, the terminal outputs this real-time stem water storage as the final result for this time step. For subsequent calculations, the terminal saves this value as a new historical stem water storage for use in the next simulation time step. Simultaneously, the terminal adds this newly calculated real-time stem water storage value, according to its corresponding timestamp, to the list of water storage values ​​stored in all previous time steps. As the simulation process continuously cycles through steps 201-203, the record list grows continuously, recording the stem water storage volume at each simulation moment from the beginning to the present in chronological order. The ordered list is the stem water storage volume sequence, which serves as an important state variable. It is used as input for residual prediction and also intuitively demonstrates the regulatory effect of stem water storage dynamics on the transpiration process.

[0101] In one embodiment, the formula for calculating the coupling equation is:

[0102] ;

[0103] in, For root hydraulic conductivity, For soil water potential, The stem's water potential, This represents the initial transpiration rate. This represents the instantaneous water volume of the plant. The stem water potential change rate is denoted as .

[0104] Specifically, root hydraulic conductivity represents the ease with which water enters the plant root system from the soil and is transported upwards within it. It is a quantitative indicator of the overall water conduction capacity of the root system. Its value is affected by factors such as soil moisture, root distribution, and physiological state. The higher the value, the stronger the root system's water absorption capacity.

[0105] Soil water potential refers to the energy state of water in the soil, which is usually a negative value. Its magnitude reflects the dryness of the soil and its ability to supply water to plants. It is the driving force for the movement of water from the soil to the roots.

[0106] Stem water potential refers to the energy state of water in the xylem of a plant stem. It is also a negative value and represents the level of water stress in the plant. It is the key potential energy that drives water to be transported upward in the stem and reach the leaves. In the process of solving the equation, it is a core variable that needs to be determined and reflects the real-time water state of the plant.

[0107] Preliminary transpiration rate refers to the preliminary estimate of the rate at which plants lose water to the atmosphere through their leaves, calculated using the Penman-Monteith formula based on current stomatal aperture and atmospheric conditions.

[0108] Instantaneous plant water capacity refers to the ability of plant stem tissue to store or release water under a specific stem water potential. It is a dynamic variable that changes with water potential, characterizing the instantaneous buffer capacity of the stem as an internal reservoir.

[0109] The stem water potential change rate refers to the instantaneous rate at which stem water potential changes over time. It describes whether stem water potential is rapidly decreasing, slowly increasing, or remaining stable, and quantifies the dynamic trend of plant water status changes.

[0110] This represents the rate at which the roots absorb water from the soil, driven by the water potential difference between the soil and the stem. Subtracting the rate of water loss through leaf transpiration yields the theoretically available surplus or deficit water that can be used to alter the stem's water storage capacity. The equation represents the actual rate of change in stem water storage. It mandates that at any given moment in the simulation, the net water uptake by the roots must equal the change in stem water storage, ensuring that the model strictly adheres to the law of conservation of mass.

[0111] The root system absorbs water, the stem stores water, and the leaves transpire within a unified mathematical framework. These three processes must be coordinated and simultaneously satisfy the equilibrium condition.

[0112] The stem water potential, rate of change, and transpiration rate in the equation are interdependent unknowns. This equation, together with the stomatal conductance calculation equation and the transpiration calculation formula, constitutes a closed system of equations. The model solves this system of equations simultaneously using numerical methods, thereby obtaining a self-consistent plant water state and physiological output under the current environmental conditions in one go. This allows the plant's water status to make a dynamic and intrinsic response to soil drought and atmospheric demand.

[0113] By introducing dynamic water capacity, the buffering effect of stem water storage on instantaneous imbalances in water supply and demand is clearly quantified. When transpiration demand suddenly exceeds root water uptake capacity, the equation allows for... A negative value indicates that transpiration is partially satisfied by consuming water stored in the stem, thus maintaining balance on both sides of the equation. Conversely, at night, transpiration decreases, and root water absorption may lead to replenishment of stored water, allowing the model to simulate key dynamics such as the lag of transpiration relative to environmental driving forces.

[0114] The overall effect of the coupled equation is to serve as a mandatory physical constraint and mathematical solution core, realizing the dynamic linkage and balance of key processes such as plant water absorption, transport, storage and loss. By simultaneously solving variables such as stem water potential and transpiration rate, the model can simulate the dynamic and self-consistent physiological response of plant water status to changes in the internal and external environment.

[0115] In one embodiment, the formula for calculating the initial transpiration rate is:

[0116] ;

[0117] In the formula, This represents the initial transpiration rate. It is the reciprocal of the porosity conductivity value. For aerodynamic drag, The slope of the saturated water vapor pressure-temperature curve. Net radiation, For soil heat flux, air density, The specific heat of air at constant pressure. is the dry / wet gauge constant, and VPD is the saturated vapor pressure difference.

[0118] Preliminary transpiration rate refers to the theoretical estimate of the rate at which vegetation loses water to the atmosphere through leaf stomata, calculated using physical laws based on current meteorological conditions and plant stomatal aperture. It serves as the basis for subsequent machine learning residual correction.

[0119] The slope of the saturated vapor pressure-temperature curve represents the change in saturated vapor pressure when the air temperature changes by 1 degree Celsius. It reflects the sensitivity of temperature to the air's water-holding capacity and is a key meteorological parameter connecting energy balance and moisture transport.

[0120] Net radiation refers to the difference between the total solar radiation absorbed by the surface of the vegetation canopy and the long-wave radiation emitted outward. It is the net energy income available to the earth's surface and the most fundamental energy source driving the transpiration process.

[0121] Soil heat flux refers to the portion of net surface radiation that enters the soil and is used to raise the soil temperature; in energy balance, it is considered an energy expenditure.

[0122] Air density refers to the mass of air per unit volume and is a fundamental atmospheric physical property involved in calculating sensible and latent heat fluxes.

[0123] The specific heat of air at constant pressure represents the amount of heat required to raise the temperature of a unit mass of air by 1 degree Celsius under constant air pressure. Together with air density, it characterizes the heat capacity of air.

[0124] The saturated vapor pressure difference refers to the difference between the saturated vapor pressure at a certain air temperature and the actual air vapor pressure. It is the main atmospheric driving force that drives the diffusion of moisture from the intercellular spaces of moist leaves to the dry atmosphere and is a direct indicator of the degree of atmospheric dryness.

[0125] Aerodynamic drag refers to the resistance encountered by water vapor as it diffuses turbulently from the vegetation canopy into the atmosphere. It is mainly affected by wind speed and canopy roughness, and reflects the transport efficiency of water vapor by atmospheric turbulence.

[0126] The humidometer constant is a physical constant that combines atmospheric pressure, specific heat of air, and latent heat of vaporization of water, and plays a role in conversion and balancing in humidity calculations.

[0127] Stomatal resistance is the reciprocal of stomatal conductance. It quantifies the physiological resistance encountered by water vapor as it diffuses from the inside of the leaf through the stomata, directly characterizing the stomatal opening level. Resistance is high when stomata are closed and low when they are open. It is the only variable directly controlled by the plant's own physiological state and serves as a bridge connecting plant physiology and atmospheric physics.

[0128] By combining the energy factors that drive transpiration with the transport resistance factors that limit transpiration using precise physical laws, the theoretical transpiration rate of vegetation can be quantitatively solved.

[0129] Achieving the physical coupling of energy balance and mass transport, the numerator of the formula integrates the proportion of available net radiative energy used to drive transpiration, reflecting the energy-driven process, the diffusion-convection process directly driven by air dryness and atmospheric turbulent transport efficiency, and unifying the two core physical mechanisms for calculation within a single framework.

[0130] The formula quantifies the synergistic control of transpiration by vegetation and the atmosphere, with a regulating coefficient term in the denominator. This reflects the relative magnitude of the physiological resistance within the plant and the atmospheric transport resistance. When stomata are closed, transpiration is strongly inhibited even in dry and turbulent conditions, clearly quantifying the synergistic control of transpiration rate by plant physiological regulation and atmospheric environmental conditions.

[0131] The stomatal resistance value is derived from leaf hydraulic processes through a coupled equation. The plant physiological component of the formula already incorporates the coupling effects of soil moisture stress and atmospheric aridity through the plant's hydraulic structure. Therefore, this formula receives stomatal conductance values ​​from the coupled solution system and real-time monitoring data, outputting a preliminary transpiration rate with clear physical meaning. This forms the physical basis for subsequent simulations of stem water storage changes and for fusion correction of machine learning residual predictions.

[0132] Specifically, as a physical module based on rigorous micrometeorology and plant physiology, this formula can accurately quantify environmental driving forces such as energy, atmospheric aridity, and turbulent transport. Together with the core physiological regulator of plant stomatal aperture, it works to determine the vegetation transpiration rate and is the cornerstone of ensuring the physical interpretability of the results in the entire hybrid simulation framework.

[0133] In one embodiment, the residual prediction model is trained using the following method:

[0134] Step 501: Extract and align the preliminary transpiration simulation values, environmental driving factor data, simulated stem water storage, and measured transpiration values ​​from the historical dataset to obtain time-synchronized data.

[0135] Historical datasets refer to a collection of multi-source data accumulated through observation and simulation over a period of time, including both measured and simulated data.

[0136] Preliminary transpiration simulation values ​​refer to the sequence of simulated vegetation transpiration rates calculated using real-time vegetation monitoring data from historical periods and by running a coupled physical model. These values ​​correspond to preliminary transpiration rates, but were calculated in historical periods.

[0137] Environmental driving factor data refers to key atmospheric and soil environmental variables that drive transpiration processes, obtained from historical monitoring during the same period, including net radiation, saturated vapor pressure difference, air temperature, and soil moisture.

[0138] Simulated stem water storage refers to the dynamic sequence of stem water storage obtained by running a coupled physical model using historical data. It corresponds to the stem water storage sequence, but is generated in a simulated historical period.

[0139] Measured transpiration values ​​are data on vegetation transpiration rates actually observed by precision instruments during the same historical period and are considered true values ​​or reference benchmarks.

[0140] Time-synchronized data refers to a processed data table or dataset in which four types of data—preliminary simulated transpiration values, environmental driving factor data, simulated stem water storage, and measured transpiration values—are strictly aligned in the time dimension to ensure that each row of data corresponds to the exact same observation time point, which is the basis for reliable machine learning training.

[0141] The terminal reads the time series of preliminary transpiration simulation values ​​generated by the physical model operation, the time series of environmental driving factor data recorded by sensors, the time series of simulated stem water storage generated by the physical model operation, and the time series of measured transpiration values ​​observed by the instrument from the storage system. The terminal identifies and unifies the time base of all data. Since the acquisition frequency or recording time of different data sources may vary slightly, time alignment processing is required. The terminal uses a unified time axis as the benchmark to check and match the above four sets of data. For data points with incompletely consistent timestamps, preset rules are used for processing. For example, non-integer data is resampled to the integer time, or records with missing data of any type at a certain time point are deleted. The goal is to ensure that each data entry obtained in the end contains the simulated value, environmental factor, simulated stem water storage, and measured value at the same time. After alignment is completed, the terminal generates a structured dataset. The dataset can be a table, where each row represents an aligned time point, and each column contains the preliminary transpiration simulation value, various environmental driving factor data, simulated stem water storage, and measured transpiration value at that time point.

[0142] Step 502: At each aligned simulation time point, extract the preliminary transpiration simulation value, environmental driving factor data, and simulated stem water storage at the previous simulation time point from the time synchronization data to obtain the feature vector.

[0143] Specifically, the aligned simulation time point refers to the specific, unified moment corresponding to each row of data in the time synchronization data.

[0144] For a given aligned simulation time point, the simulated stem water storage at the previous simulation time point refers to the simulated stem water storage value recorded in the time synchronization data at a time step earlier than the current time point.

[0145] In machine learning, a feature vector is a set of features used to describe a sample, which can be represented as a one-dimensional array or vector.

[0146] The terminal sequentially reads each row of the time-synchronized data, processing each aligned simulation time point. For the currently processed time point, the terminal extracts the preliminary simulated transpiration value and all environmental driving factor data for that time point from the data row. The terminal obtains the simulated stem water storage from the previous simulation time point. In the time-synchronized data, it finds the data row corresponding to the immediately preceding time point and extracts the simulated stem water storage value from that row. For the first time point, no valid sample can be constructed, and this sample may be discarded. The extracted preliminary simulated transpiration value, multiple environmental driving factor data, and simulated stem water storage from the previous time point are merged into a one-dimensional array in a fixed order. This array represents the feature vector for that time point, integrating the preliminary predictions of the physical model, current environmental conditions, and the plant's internal water storage state from the previous time point.

[0147] Step 503: For the simulated time point, subtract the preliminary simulated evaporation value from the measured evaporation value to obtain the target variable.

[0148] Specifically, in supervised learning, the target variable refers to the output value that the model needs to learn and predict. In this embodiment, the target variable is the simulation error of the physical model, i.e., the residual.

[0149] For each aligned simulation time point, the terminal performs a subtraction operation. It reads the measured evapotranspiration value from the row corresponding to that time point t in the time synchronization data, and then reads the preliminary simulated evapotranspiration value from the same row. The difference between the two is calculated; the target variable is the difference between the measured evapotranspiration value and the preliminary simulated evapotranspiration value. This difference represents the simulation residual of the physical model at that time point.

[0150] Step 504: Integrate the feature vectors and target variables to obtain the initial sample set, and divide the initial sample set into the training set, validation set and test set to obtain the training dataset.

[0151] The initial sample set refers to the complete dataset consisting of all aligned simulation time points corresponding to (feature vectors, target variables) pairings. Each sample contains one feature vector and one target variable.

[0152] The training set is a subset of samples that is divided from the initial sample set and used to directly train machine learning models.

[0153] The validation set is a subset of samples that is partitioned from the initial sample set and used to evaluate model performance, perform hyperparameter tuning, and determine when to stop training during the training process. The model does not learn weights from the validation set.

[0154] The test set is a subset of samples partitioned from the initial sample set, used to independently evaluate the model's generalization performance after training and tuning. It is never used during training.

[0155] In this embodiment, the training dataset refers to the overall data resource after it has been partitioned, and explicitly includes three parts: the training set, the validation set, and the test set, with each part having a defined purpose.

[0156] The terminal pairs the feature vectors generated for each valid time point with the target variables calculated for the same time point, collecting all pairs to form a complete and structured initial sample set. To scientifically train and evaluate the model, the initial sample set needs to be split into three non-overlapping subsets. Common methods include splitting by time sequence to evaluate the model's performance in unknown time periods, or random splitting. The terminal performs the splitting operation according to a preset ratio, storing the samples in the training set, validation set, and test set respectively, resulting in a training dataset that can be used for subsequent model training and evaluation, containing three data subsets with clearly defined roles.

[0157] Step 505: Based on the training dataset, train the XGBoost model framework to obtain the residual prediction model.

[0158] The XGBoost model framework refers to a specific and efficient gradient boosting decision tree machine learning algorithm framework. It builds a powerful ensemble prediction model by training multiple decision trees sequentially, with each new tree working to correct the prediction residuals of all previous tree combinations.

[0159] The residual prediction model refers to a specific XGBoost model instance that has been trained and is ready for use. Its function is to output a predicted value of the error of the initial evaporation simulation value when a new feature vector is input.

[0160] The terminal takes the feature vectors from the training set as input and the corresponding target variable as the label to initiate the training process of the XGBoost model. Training is an iterative process: building the first decision tree to fit the label; calculating the prediction residual of the first tree; then building the second tree to fit the residual; and so on, continuously adding new trees, each new tree learning to correct the residual error of all previous trees. During training, the terminal uses a validation set to monitor model performance and presets a set of model hyperparameters. The model learns on the training set and evaluates its prediction accuracy on the validation set. By adjusting the hyperparameters and observing the performance on the validation set, the optimal hyperparameter combination is selected. Simultaneously, early stopping is employed: when the model's performance on the validation set no longer improves or even declines after multiple iterations, training is automatically stopped to prevent overfitting to the training data. After the training and tuning process is complete, the model is retrained on all training and validation set data using the optimal hyperparameter configuration, or the best-performing iteration model on the validation set is directly selected as the model. A one-time, independent performance evaluation of the model is performed using a test set to report its generalization ability unbiasedly. An XGBoost model that has been trained and passed evaluation is a residual prediction model.

[0161] To further illustrate the solutions of the embodiments of this application, a specific example is provided below.

[0162] At 9:00 a.m., real-time vegetation monitoring data was acquired from field sensors, including: strong net solar radiation, high air temperature, moderate air humidity, light wind speed, soil water potential in the root zone, and relevant parameters for calculating leaf water potential.

[0163] Using this data, a stomatal conductance calculation rule based on leaf hydraulic processes is instantiated. The core feature of the rule is that it couples the effects of atmospheric aridity and soil drought by using leaf water potential, resulting in a specific and computable stomatal conductance function.

[0164] The stomatal conductance calculation rules, current environmental data, and soil water potential are all incorporated into a simultaneous solution framework. The framework includes two core relationships: a physical relationship in which transpiration is determined by net radiation, water vapor pressure difference, and stomatal aperture; and a coupled equation describing the instantaneous equilibrium between root water absorption, stem water storage changes, and leaf transpiration.

[0165] By using a numerical iterative method, the stem water potential, the rate of change of stem water potential, and the initial estimate of transpiration at 9:00 AM were simultaneously obtained, based on the current atmospheric demand and the calculated stomatal aperture.

[0166] Based on the calculated stem water potential, the preset maximum water storage capacity of the plant and the attenuation coefficient, the instantaneous plant water capacity is calculated, which represents the immediate water storage capacity of the stem tissue under the current water potential.

[0167] The net change in stem water storage during the time interval was calculated by multiplying the instantaneous plant water capacity by the stem water potential change rate and then by a very short calculation time interval.

[0168] The historical stem water storage at 8:55 AM is read, and the net change calculated earlier is added to obtain the real-time stem water storage at 9:00 AM. This value is recorded in the continuously growing stem water storage sequence.

[0169] The preliminary transpiration rate at 9:00 AM, the current environmental driving factor data, and the simulated stem water storage at 8:55 AM are combined into a feature vector and input into a pre-trained XGBoost residual prediction model.

[0170] The learning objective of the model is to predict the simulation error of the physical model. After receiving the feature vector, the model outputs a residual prediction value based on the complex patterns it has learned internally, indicating that it believes the initial simulation value of the physical model may underestimate the actual evaporation.

[0171] The preliminary transpiration rate is added to the residual prediction to obtain the estimated transpiration rate at 9:00 AM.

[0172] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0173] Based on the same inventive concept, this application also provides a machine learning-based vegetation transpiration simulation system for implementing the above-mentioned machine learning-based vegetation transpiration simulation method. The solution provided by this system is similar to the implementation described in the above method. Therefore, the specific limitations of one or more machine learning-based vegetation transpiration simulation system embodiments provided below can be found in the limitations of the machine learning-based vegetation transpiration simulation method described above, and will not be repeated here.

[0174] In one exemplary embodiment, such as Figure 2 As shown, a vegetation transpiration simulation system 600 based on machine learning is provided, comprising:

[0175] The conductance module 601 is used to acquire real-time vegetation monitoring data and, based on the real-time vegetation monitoring data, quantify the stomatal aperture level of leaves to obtain the stomatal conductance function; wherein, the real-time vegetation monitoring data includes vegetation environment data, soil data and plant water potential parameters;

[0176] Solver module 602 is used to solve stem water potential, stem water potential change rate and preliminary transpiration rate based on stomatal conductance function and real-time vegetation monitoring data through coupled equations.

[0177] The simulation module 603 is used to simulate the change in stem water storage based on stem water potential and stem water potential change rate, and obtain the stem water storage sequence.

[0178] Prediction module 604 is used to input the preliminary transpiration rate and stem water storage sequence into the residual prediction model to predict the preliminary transpiration residual value and obtain the residual prediction value.

[0179] The correction module 605 is used to add the preliminary transpiration rate and the residual prediction value to obtain the transpiration rate estimate; wherein the transpiration rate estimate is used to characterize the simulation results of vegetation transpiration.

[0180] Furthermore, the simulation module 603 is also used for:

[0181] The instantaneous plant water capacity is calculated based on the stem water potential and the maximum water capacity.

[0182] The formula for calculating the instantaneous plant water capacity is as follows:

[0183] ;

[0184] In the formula, This represents the instantaneous water volume of the plant. To ensure the maximum water capacity under full water supply conditions, The attenuation coefficient is... The stem's water potential;

[0185] Based on the instantaneous plant water capacity, stem water potential change rate and simulation time step, the net change in stem water storage is calculated to obtain the stem water storage change.

[0186] The real-time stem water storage is obtained by summing the changes in stem water storage and the historical stem water storage, and then by integrating the real-time stem water storage and the historical stem water storage to obtain the stem water storage sequence.

[0187] The formula for calculating the coupling equation is as follows:

[0188] ;

[0189] in, For root hydraulic conductivity, For soil water potential, The stem's water potential, This represents the initial transpiration rate. This represents the instantaneous water volume of the plant. The stem water potential change rate is denoted as .

[0190] The formula for calculating the initial transpiration rate is as follows:

[0191] ;

[0192] In the formula, This represents the initial transpiration rate. It is the reciprocal of the porosity conductivity value. For aerodynamic drag, The slope of the saturated water vapor pressure-temperature curve. Net radiation, For soil heat flux, air density, The specific heat of air at constant pressure. is the dry / wet gauge constant, and VPD is the saturated vapor pressure difference.

[0193] Furthermore, the machine learning-based vegetation transpiration simulation system 600 also includes a training module for:

[0194] Time-synchronized data was obtained by extracting and aligning preliminary transpiration simulation values, environmental driving factor data, simulated stem water storage, and measured transpiration values ​​from historical datasets.

[0195] At each aligned simulation time point, preliminary transpiration simulation values, environmental driving factor data, and simulated stem water storage at the previous simulation time point are extracted from the time synchronization data to obtain a feature vector;

[0196] For each simulated time point, the target variable is obtained by subtracting the preliminary simulated transpiration value from the measured transpiration value.

[0197] By integrating feature vectors and target variables, an initial sample set is obtained, which is then divided into a training set, a validation set, and a test set to obtain the training dataset.

[0198] Based on the training dataset, the XGBoost model framework is trained to obtain the residual prediction model.

[0199] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a machine learning-based vegetation transpiration simulation method as described above.

[0200] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0201] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0202] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for simulating vegetation transpiration based on machine learning, characterized in that, The method includes: Acquire real-time vegetation monitoring data, and based on the real-time vegetation monitoring data, quantify the stomatal aperture level of leaves to obtain the stomatal conductance function; wherein, the real-time vegetation monitoring data includes vegetation environment data, soil data and plant water potential parameters; Based on the stomatal conductance function and the real-time vegetation monitoring data, the stem water potential, stem water potential change rate and preliminary transpiration rate are solved by coupling equations. Based on the stem water potential and the stem water potential change rate, the stem water storage change is simulated to obtain the stem water storage sequence; The preliminary transpiration rate and the stem water storage sequence are input into the residual prediction model to predict the preliminary transpiration residual value and obtain the residual prediction value. The preliminary transpiration rate and the residual prediction value are added together to obtain the transpiration rate estimate; wherein the transpiration rate estimate is used to characterize the simulation results of vegetation transpiration.

2. The method according to claim 1, characterized in that, The stem water storage change is simulated based on the stem water potential and the rate of change of stem water potential to obtain a stem water storage sequence, including: Based on the stem water potential and maximum water capacity, the instantaneous plant water capacity is calculated. The formula for calculating the instantaneous plant water capacity is as follows: ； In the formula, This represents the instantaneous water volume of the plant. To ensure the maximum water capacity under full water supply conditions, The attenuation coefficient is... The stem's water potential; Based on the instantaneous plant water capacity, the stem water potential change rate and the simulation time step, the net change in stem water storage is calculated to obtain the stem water storage change. The change in stem water storage and the historical stem water storage are summed to obtain the real-time stem water storage. The real-time stem water storage and the historical stem water storage are then integrated to obtain the stem water storage sequence.

3. The method according to claim 2, characterized in that, The formula for calculating the coupling equation is as follows: ； in, For root hydraulic conductivity, For soil water potential, The stem's water potential, This represents the initial transpiration rate. This represents the instantaneous water volume of the plant. The stem water potential change rate is denoted as .

4. The method according to claim 3, characterized in that, The formula for calculating the initial transpiration rate is as follows: ； In the formula, This represents the initial transpiration rate. It is the reciprocal of the porosity conductivity value. For aerodynamic drag, The slope of the saturated water vapor pressure-temperature curve. Net radiation, For soil heat flux, air density, The specific heat of air at constant pressure. is the dry / wet gauge constant, and VPD is the saturated vapor pressure difference.

5. The method according to claim 1, characterized in that, The residual prediction model was trained using the following method: Time-synchronized data was obtained by extracting and aligning preliminary transpiration simulation values, environmental driving factor data, simulated stem water storage, and measured transpiration values ​​from historical datasets. At each aligned simulation time point, the preliminary simulated transpiration value, the environmental driving factor data, and the simulated stem water storage at the previous simulation time point are extracted from the time synchronization data to obtain a feature vector; For the simulated time point, the target variable is obtained by subtracting the preliminary simulated transpiration value from the measured transpiration value; Integrate the feature vectors and the target variables to obtain an initial sample set, and divide the initial sample set into a training set, a validation set, and a test set to obtain a training dataset; Based on the training dataset, the XGBoost model framework is trained to obtain the residual prediction model.

6. A vegetation transpiration simulation system based on machine learning, characterized in that, The system includes: The conductance module is used to acquire real-time vegetation monitoring data and, based on the real-time vegetation monitoring data, quantify the stomatal aperture level of leaves to obtain the stomatal conductance function; wherein, the real-time vegetation monitoring data includes vegetation environment data, soil data, and plant water potential parameters; The solution module is used to solve the stem water potential, stem water potential change rate and preliminary transpiration rate based on the stomatal conductance function and the real-time monitoring data of the vegetation through coupled equations. The simulation module is used to simulate the change in stem water storage based on the stem water potential and the stem water potential change rate, and to obtain the stem water storage sequence. The prediction module is used to input the preliminary transpiration rate and the stem water storage sequence into the residual prediction model to predict the preliminary transpiration residual value and obtain the residual prediction value. The correction module is used to add the preliminary transpiration rate and the residual prediction value to obtain an estimated transpiration rate; wherein the estimated transpiration rate is used to characterize the simulation results of vegetation transpiration.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images