Evaporation model parameter calibration method for rice field and reed wetland

Abstract

The invention discloses an evapotranspiration model parameter calibration method for a rice field and a reed wetland, and belongs to the technical field of ecological hydrological simulation. Comprising the following steps: S1, collecting multi-source data of a rice field and a reed wetland observation station, and processing the collected data; s2, performing sensitivity analysis on each parameter of the PML model and the PT-JPL model; s3, dividing growth stages according to different growth periods of the rice and the reed, setting a dynamic parameter optimization range, and calibrating key parameters of the PML model and the PT-JPL model in stages based on multi-year actually measured evapotranspiration data and a Bayesian optimization algorithm; s4, the calibrated optimal parameter combination is substituted into the corresponding PML model or PT-JPL model, and wetland evapotranspiration is simulated; and S5, verifying a model simulation effect through multiple reference indexes, and screening an optimal model and parameter combination. By the adoption of the method, the parameters can be accurately calibrated, the wetland evapotranspiration simulation accuracy and adaptability are improved, and ecological hydrological research is supported.

Description

Technical Field

[0001] This invention relates to the field of eco-hydrological simulation technology, and in particular to a method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands. Background Technology

[0002] Evapotranspiration is a crucial link in the hydrological cycle and energy balance. Accurately simulating the evapotranspiration of wetland vegetation is of great significance for watershed water resource management and ecological protection. The Penman-Monteith-Leuning model (PML model) and the Priestley-Taylor Jet Propulsion model are important models for this purpose. While laboratory models (PT-JPL models) are commonly used tools for estimating regional evapotranspiration, they still have significant limitations in wetland vegetation applications. Existing technologies often rely on general empirical values ​​or calibration results from other regions, failing to optimize parameter ranges for the specific characteristics of paddy fields and reed wetlands, which are characterized by "water saturation and homogeneous vegetation." Furthermore, parameter systems are often constructed based on non-wetland scenarios such as semi-arid watersheds, which do not match the actual environment of these wetland types and their characteristics, such as high vapor pressure differentials and stable vegetation growth cycles, leading to insufficient model adaptability. Simultaneously, most studies employ a single fixed-parameter calibration model, neglecting the physiological activity differences among different wetland vegetation types and growth stages, easily resulting in parameter "averaging" errors. This makes it difficult to accurately reflect the dynamic process of evapotranspiration at each stage. Additionally, some studies only use traditional optimization algorithms for a single model, failing to combine dual models with efficient global optimization algorithms, making it difficult to balance calibration efficiency and simulation accuracy, further limiting the accuracy of wetland vegetation evapotranspiration simulations. Summary of the Invention

[0003] The purpose of this invention is to provide a method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands, thereby solving the aforementioned technical problems.

[0004] To achieve the above objectives, this invention provides a method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands, comprising the following steps: S1. Collect multi-source data from rice paddy and reed wetland observation stations, including meteorological, flux data and leaf area index data, and process the collected data, dividing the data into two groups according to time period: calibration data and validation data. S2. Based on the growth characteristics of rice and reeds, sensitivity analysis was conducted on the parameters of the PML model and PT-JPL model to determine the sensitive parameters and set empirical values ​​for the non-sensitive parameters. S3. Divide the growth stages according to the different growth periods of rice and reeds, set the dynamic parameter optimization range in combination with wetland characteristics, and calibrate the key parameters of the PML model and PT-JPL model in stages based on years of measured evapotranspiration data and Bayesian optimization algorithm. S4. Select other years or periods that were not calibrated, and substitute the optimal parameter combination after calibration into the corresponding PML model or PT-JPL model. Combine the processed data and the inherent physical calculation logic of the model to simulate wetland evapotranspiration. S5. Based on measured evapotranspiration data that were not calibrated, the simulation effect of the model was verified through multiple reference indicators, and the optimal model and parameter combination suitable for the target wetland was selected.

[0005] Preferably, S1 specifically includes: S11. Based on the MOD15A2H product, collect remote sensing data of leaf area index. After extracting the remote sensing data of leaf area index of paddy fields and reed wetlands, compare it with the measured leaf area index data in the field. Optimize the data by linear interpolation to ensure the consistency between the two. Collect observation data of paddy field and reed wetland stations, including daily meteorological driving data and measured evapotranspiration data obtained based on station flux observations. S12. Based on the 3σ criterion, extreme outliers in meteorological elements and measured evapotranspiration data were removed. Erroneous data exceeding the physical reasonable range caused by instrument malfunction and invalid data with negative surface absorbed usable energy were manually deleted. Short-term missing data were filled using linear interpolation, and long-term missing data were estimated by combining the correlation of the data with the meteorological data of the same period. Leaf area index remote sensing data were processed by moving average. S13. Convert all processed data into units required for PML and PT-JPL model calculations, align them in the time dimension using date as the key, and ensure that meteorological data, measured evapotranspiration data and leaf area index data for the same date correspond one-to-one. S14. Divide the data processed by S13 into two groups according to the observation period or year: one group is the calibration data used for calibration, and the other group is the verification data used for verification.

[0006] Preferably, S2 specifically includes: PML model parameters include the extinction coefficient of available energy. k A Extinction coefficient of shortwave radiation k Q Water vapor pressure difference when the pore conductance is half of its maximum value D 50 Visible radiation flux when porosity is half its maximum value Q 50 Soil moisture stress factor f and maximum stomatal conductance of leaves g sx ,in, g sx For sensitive parameters, k A , k Q ,D 50 , Q 50 As a non-sensitive parameter, and set as an empirical value, based on the long-term supersaturation characteristics of soil moisture in the target wetland, f Set to 1; the PT-JPL model parameters include the light energy utilization coefficient. k 1. Moisture stress coefficient k 2. Evapotranspiration regulation coefficient β and optimal growth temperature of vegetation T opt ,in, k 1. k 2. β , T opt All of these are sensitive parameters. Based on the similarity between wetland and shrubland environments, empirical values ​​of shrubland were selected as the initial values ​​for the PT-JPL model parameters.

[0007] Preferably, S3 specifically includes: S31. The growth stages of natural wetland vegetation reeds are divided into budding stage, vegetative growth stage, reproductive growth stage, and maturity-early yellowing stage; the growth stages of artificial wetland vegetation paddy fields are divided into greening stage, tillering and jointing stage, booting and heading stage, and milk-maturity stage. S32. Based on the growth characteristics of rice and reeds, a dynamic parameter optimization range is set, whereby the growth stages are defined based on S31. g sx The values ​​were set to 0.001–0.10 ms for each stage of the reproductive period. -1 0.001~0.15ms -1 0.001~0.20ms -1 0.001~0.25ms -1 , k 1 is set to 0.55~0.85. k 2 is set to 0.75~0.95. β Set to 1.43~2.2, T opt Set to 20–28℃; S33. Based on the data processed in S1 and the optimization range of dynamic parameters for vegetation in S32, perform phased Bayesian optimization calibration, specifically including: S331, with g sx , k 1. k 2. β , T opt To optimize the variables, a parameter search space is constructed based on the parameter optimization range set in S32; S332. Using the error between the model simulation value and the measured value at the site as the evaluation standard, the root mean square error is selected. RMSE or negative Nash coefficient NSE Construct the objective function f(x) The optimization objective is to minimize f(x) ; S333. Use the Latin hypercube sampling method to select 5 to 10 sets of initial parameter combinations within the parameter search space. x i Run the PML model and the PT-JPL model respectively, and calculate the objective function values ​​corresponding to the models. y i To form the initial dataset D= { x i , y i}; S334, Based on the initial dataset D A Gaussian process surrogate model is fitted, which outputs the predicted mean and predicted variance for any parameter point. By optimizing the acquisition function, the next parameter combination point to be evaluated is determined. x next ; S335, in x next Run the PML model and the PT-JPL model respectively to obtain the true objective function values. y next ,Will( x next , y next Add to dataset D And update the Gaussian process proxy model; S336. Repeat S334-S335 until the convergence condition is met or the preset maximum number of evaluations is reached, then select the objective function that makes the objective function... f(x) The optimal parameter combination is used as the calibration result for the corresponding PML model or PT-JPL model.

[0008] Preferably, S4 specifically includes: For years or periods not included in the calibration, the optimal parameter combination obtained from S336 was substituted into the PML model and the PT-JPL model, respectively. At the same time, the daily meteorological data and leaf area index data processed by S1 were combined to calculate the daily total evapotranspiration. The PML model is based on the Penman-Monteith principle and combines canopy conductance and soil moisture stress factor to calculate the total evapotranspiration through energy balance. The PT-JPL model is based on the Priestley-Taylor theory and obtains the simulated value of total evapotranspiration by simulating vegetation canopy transpiration, soil evaporation and intercepted evaporation respectively and superimposing them.

[0009] Preferably, S5 specifically includes: S51. Based on measured evapotranspiration data that were not calibrated, the simulation effect of the model is verified by calculating reference indicators, including the coefficient of determination. R 2 Nash coefficient NSE Root mean square error RMSE ,in, R 2 The formula representing the correlation between simulated and measured values ​​is shown below: ; in, For the number of samples, For sample index, , For the first The measured evapotranspiration of each sample This is the average of all measured evapotranspiration. For the first Simulated evapotranspiration for each sample This is the average of all simulated evapotranspiration. NSE The efficiency of the model simulation is expressed by the following formula: ; RMSE The formula for expressing the deviation between simulated and measured values ​​is shown below: ; S52, Filter out R 2 ≥ A, NSE ≥ B and RMSE ≤ C The model and its corresponding parameter combination are used as the candidate set. The consistency of the time series trend is determined by calculating the deviation rate of the linear regression slope between the simulated values ​​and the measured values. The models with a deviation rate ≤ 0.5% are selected. M Furthermore, the model and parameter combination with a trend fit exceeding any other fit is considered the optimal result.A for R 2 The critical value is dimensionless. B for NSE The critical value is dimensionless. C for RMSE The critical value, in mm.d -1 , M This is the critical value for the deviation rate.

[0010] Preferably, the daily-scale meteorological driving data include near-surface air temperature, relative humidity, and saturated vapor pressure difference. VPD Net radiation R n Soil heat flux G air pressure P .

[0011] Preferably, the dynamic adjustment of the parameter optimization range is adapted to the differences in physiological activity of rice and reeds at different growth stages, responds to the increased photosynthetic demand and transpiration cooling mechanism after the temperature rises, and matches the transpiration characteristics of wetland vegetation at each growth stage.

[0012] Preferably, the convergence condition for Bayesian optimization is that the objective function value is continuous. R The magnitude of change in each iteration < q Or preset a maximum number of evaluations not less than S The iteration stops when any one of the following conditions is met: q This represents the critical value for the magnitude of change in the objective function value.

[0013] Therefore, the present invention employs the above-mentioned method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands, which has the following beneficial effects: 1. High adaptability: Based on the unique characteristics of paddy fields and reed wetlands, namely "water saturation and uniform vegetation" and the physiological differences of vegetation at different growth stages, the parameter optimization range is dynamically set and calibrated in stages to avoid parameter averaging errors and solve the problem of mismatch between the existing technical parameter system and the wetland environment.

[0014] 2. High efficiency and accuracy in calibration: The Bayesian optimization algorithm combined with the Gaussian process surrogate model is used to quickly search for the optimal parameter combination with fewer sample points, balancing calibration efficiency and accuracy.

[0015] 3. High simulation accuracy: Through parallel calibration of PML and PT-JPL dual models and verification with multiple indicators, the optimal model and parameter combination suitable for the target wetland are selected. After optimization, the simulated values ​​of the model are more correlated with the measured values ​​and the deviation is smaller.

[0016] 4. High data reliability: Through data cleaning, completion, and standardization processes, the quality of input data is guaranteed, laying the foundation for accurate model simulation.

[0017] 5. Wide applicability: The results can accurately support wetland watershed water resource management and ecological protection, and provide reliable data support for relevant decision-making.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] Figure 1 This is a flowchart of a method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands according to the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0021] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0022] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0023] like Figure 1As shown, this invention provides a method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands, comprising the following steps: S1, collecting multi-source data from paddy field and reed wetland observation stations, including meteorological, flux, and leaf area index data, and processing the collected data, dividing the data into calibration data and validation data according to time periods to ensure the reliability and consistency of the input data; S2, combining the growth characteristics of rice and reeds, performing sensitivity analysis on each parameter of the PML model and PT-JPL model, determining sensitive parameters, setting empirical values ​​for non-sensitive parameters, and clarifying the parameter functions and initial setting basis; S3, dividing the growth stages according to different growth periods of rice and reeds, and combining wetland... The dynamic parameter optimization range is set based on the characteristics of "water saturation and uniform vegetation". The key parameters of the PML model and PT-JPL model are calibrated in stages based on multi-year measured evapotranspiration data and Bayesian optimization algorithm to avoid averaging errors in parameters; S4, other years or periods that were not calibrated are selected, and the optimal parameter combination after calibration is substituted into the corresponding PML model or PT-JPL model respectively. Combined with the processed data and the inherent physical calculation logic of the model, wetland evapotranspiration is simulated; S5, based on the measured evapotranspiration data that were not calibrated, the simulation effect of the model is verified through multiple reference indicators, and the optimal model and parameter combination suitable for the target wetland is selected to improve the credibility and applicability of the simulation results.

[0024] Specifically, the measured evapotranspiration data and meteorological data used in this embodiment were obtained from observations and remote sensing data from meteorological and flux monitoring equipment deployed in paddy fields and reed wetlands in the study area. S1 specifically includes: S11, collecting leaf area index remote sensing data based on the MOD15A2H product, extracting the leaf area index remote sensing data of paddy fields and reed wetlands, comparing it with the measured leaf area index data in the field, and optimizing it through linear interpolation to ensure consistency between the two; collecting observation data from paddy field and reed wetland stations, including diurnal meteorological driving data and station-based data. The measured evapotranspiration data obtained from flux observations specifically refers to the sum of surface water evaporation and vegetation transpiration. This measured evapotranspiration data originates from long-term continuous observations at wetland flux observation stations. S12. Data deviating from the mean by three times the standard deviation under a normal distribution are identified as outliers. Extreme outliers in meteorological elements and measured evapotranspiration data are removed based on the 3σ criterion. Erroneous data exceeding the physically reasonable range due to instrument malfunctions and invalid data with negative surface absorbed usable energy are manually deleted. Surface absorbed usable energy is... R n -G ,in, R n Net radiation refers to the difference between the total solar radiation received by the Earth's surface and the reflected and emitted radiation. GSoil heat flux refers to the amount of heat exchange between soil layers. Linear interpolation is used to complete short-term missing data, where "short-term" refers to a single day or 2-3 consecutive days. Long-term missing data is estimated by combining the correlation with contemporaneous meteorological data, such as estimation based on multiple regression relationships of temperature and radiation. A 10-day sliding window is used to process leaf area index remote sensing data through a moving average to eliminate the interference of short-term fluctuations on vegetation growth trends. S13: All processed data are uniformly converted to the units required for calculation by the PML and PT-JPL models, aligned in the time dimension using the date as the key to ensure a one-to-one correspondence between meteorological data, measured evapotranspiration data, and leaf area index data for the same date, avoiding simulation errors caused by time misalignment. S14: The data processed in S13 are divided into two groups according to the observation period or year: one group is calibration data for participation in calibration, and the other group is validation data for participation in validation.

[0025] S2 specifically includes: PML model parameters including the extinction coefficient of available energy. k A Extinction coefficient of shortwave radiation k Q Water vapor pressure difference when the pore conductance is half of its maximum value D 50 Visible radiation flux when porosity is half its maximum value Q 50 Soil moisture stress factors f and the maximum porosity of the blade g sx ,in, g sx As a sensitive parameter, the original model will include g sx Set to 0.008. k A , k Q , D 50 , Q 50 This is a non-sensitive parameter with low sensitivity to simulation results, and its value is set as an empirical value. k A =0.6、 k Q =0.6、 D 50 =30W.m -2 , Q 50 =0.8kPa, f This refers to the degree to which soil moisture conditions limit vegetation transpiration. Based on the long-term supersaturation characteristics of the target wetland soil, fSetting it to 1 indicates no soil moisture limitation, which aligns with the wetland hydrological environment, but because... k A Its contribution to the relative change rate of evapotranspiration is very small, therefore it is ultimately counted as 0 in this embodiment. The PT-JPL model parameters include the light energy utilization coefficient. k 1. Moisture stress coefficient k 2. Evapotranspiration regulation coefficient β and optimal growth temperature of vegetation T opt ,in, k 1. k 2. β , T opt All are sensitive parameters. k 1 refers to the efficiency of vegetation in converting light energy, which affects the photosynthetic and transpiration processes of the canopy. k 2 refers to the degree to which soil moisture limits vegetation evapotranspiration. β It also features a humidity response coefficient, making it suitable for the transpiration regulation needs under high water vapor pressure differentials in wetlands. T opt The optimal temperature threshold for vegetation growth and development determines the peak range of evapotranspiration efficiency. Since wetlands are more similar to shrubland environments, existing empirical values ​​for shrubland were chosen as the initial parameter values ​​for the original PT-JPL model. These initial values ​​were set to... k 1 = 0.56 k 2 = 0.91 β =1.17、 T opt =24℃. Note that this needs to be verified in conjunction with the characteristics of the study area. Single-parameter sensitivity analysis shows that all four core parameters have high sensitivity in evapotranspiration simulation, and their optimal value range is... k Take 0.77 to 0.89. k 2. Take 0.77~0.95, β Take a value of 2.10 to 2.55. T opt Take 22-30℃, and k 1. k 2. β , T optPhased calibration is required to match vegetation growth dynamics. Specifically, S3 includes: S31, dividing the growth stages of natural wetland vegetation (reed) into budding stage (April), vegetative growth stage (May to mid-July), reproductive growth stage (late July to August), and maturity-early yellowing stage (September to October); and dividing the growth stages of artificial wetland vegetation (rice paddy) into greening stage (mid-May to early June), tillering and jointing stage (June to late July), booting and heading stage (late July to late August), and milk-ripe to maturity stage (late August to September), aligning with the phenological characteristics of both types of vegetation; S32, setting the dynamic parameter optimization range based on the growth characteristics of rice and reeds, where the growth stages defined in S31... g sx The values ​​are dynamically set to 0.001–0.10 ms according to the progression of the reproductive period. -1 0.001~0.15ms -1 0.001~0.20ms -1 0.001~0.25ms -1 To adapt to the pattern of physiological activity gradually increasing and then decreasing, k The value is set to 0.55–0.85, slightly higher than that of forests or farmland, to match the higher light capture efficiency of wetland vegetation. k The value of 2 is set at 0.75–0.95, which is close to that of shrubland but slightly lower, reflecting the perennially moist environment of wetlands. β The value is set at 1.43–2.2, higher than that of grassland, to meet the transpiration regulation requirements of wetlands under high vapor pressure differentials. T opt The temperature was set to 20–28℃, covering the typical temperature range of wetland vegetation during the growing season; S33, based on the data processed in S1 and the optimization range of dynamic parameters for rice and reeds in S32, a phased calibration of Bayesian optimization was performed to obtain an approximate optimal solution with fewer sample points, thereby improving calibration efficiency. Specifically, this included: S331, using… g sx , k 1. k 2. β , T opt To optimize the variables, a parameter search space with clear boundaries is constructed based on the parameter optimization range set in S32; S332 uses the error between the model simulation value and the measured value at the site as the evaluation criterion, and selects the root mean square error. RMSE or negative Nash coefficient NSE Construct the objective function f (x) The optimization objective is to minimize f(x) This makes the simulated values ​​closer to the measured values; S333, using the Latin hypercube sampling method, select 5 to 10 representative initial parameter combinations within the parameter search space. xi Run the PML model and the PT-JPL model respectively, and calculate the objective function values ​​corresponding to the models. y i To form the initial dataset D= { x i ,y i This provides the foundational data for constructing the Gaussian process surrogate model; the S334 Gaussian process surrogate model is a model that fits existing data and replaces the real model to quickly predict parameter performance and uncertainty, based on the initial dataset. D A Gaussian process surrogate model is fitted, which outputs the predicted mean (simulation performance estimate) and predicted variance (estimated uncertainty measure) for any parameter point. The next parameter combination point to be evaluated is determined using the optimized acquisition function in Bayesian optimization. x next , x next The most worthwhile parameter combination point to evaluate; S335, in x next Run the PML model and the PT-JPL model respectively to obtain the true objective function values. y next ,Will( x next , y next Add to dataset D The Gaussian process surrogate model is updated to improve prediction accuracy; S336, S334-S335 are repeated until the convergence condition is met or the preset maximum number of evaluations is reached, and the objective function is selected. f(x) The optimal parameter combination is used as the calibration result for the corresponding PML model or PT-JPL model.

[0026] S4 specifically includes: selecting years or periods not involved in calibration, substituting the optimal parameter combination obtained from S336 into the PML and PT-JPL models respectively, and combining daily meteorological data processed by S1, such as near-surface air temperature, relative humidity, and leaf area index data, to calculate daily total evapotranspiration. The PML model is based on the Penman-Monteith principle, combining canopy conductance and soil moisture stress factors to calculate total evapotranspiration through energy balance, accurately characterizing the interaction between vegetation and the atmosphere. The PT-JPL model is based on the Priestley-Taylor theory, simulating vegetation canopy transpiration, soil evaporation, and intercepted evaporation separately and superimposing them to obtain the simulated value of total evapotranspiration, adapting to the simulation of regional-scale evapotranspiration and its components.

[0027] S5 specifically includes: S51. This embodiment conducts parameter calibration based on observation data from the Liaohe River Delta paddy field station and reed station from 2018 to 2022, and verifies it using 2023 data. Based on the above-mentioned measured evapotranspiration data that were not involved in the calibration, the simulation effect of the model is verified by calculating reference indicators, including the coefficient of determination. R 2 Nash coefficient NSE Root mean square error RMSE ,in, R 2 This represents the correlation between simulated and measured values, and its value is between [0, 1]. R 2 The closer the value is to 1, the higher the correlation between the simulated value and the measured value. The formula is as follows: ; in, For the number of samples, For sample index, , For the first The measured evapotranspiration of each sample This is the average of all measured evapotranspiration. For the first Simulated evapotranspiration of a sample This represents the average value of all simulated evapotranspiration. NSE This represents the efficiency of the model simulation, with a value ranging from (-∞, 1]. NSE The closer the value is to 1, the better the applicability of the model. The formula is shown below: ; RMSE This represents the deviation between the simulated value and the measured value. The smaller the value, the better the simulation effect. The formula is as follows: ; S52, Filter out R 2 ≥ A, NSE ≥ B and RMSE ≤ C The model and its corresponding parameter combination are used as the candidate set. The consistency of the time series trend is determined by calculating the deviation rate of the linear regression slope between the simulated values ​​and the measured values. The models with a deviation rate ≤ 0.5% are selected. M Furthermore, the model and parameter combination with a trend fit exceeding any other fit is considered the optimal result. A, B, C All of these are constant thresholds set based on the actual needs of wetland evapotranspiration simulation and the optimization effects of existing technologies. A for R 2 The critical value is dimensionless.B for NSE The critical value is dimensionless. C for RMSE The critical value, in mm.d -1 , M This is the critical value for the deviation rate, specifically set as follows: A =0.7, B =0.7, C =1.0mm.d -1 , M= A 10% margin was added to ensure the simulation results closely match actual evapotranspiration dynamics. This embodiment was calibrated in stages based on measured evapotranspiration data from 2018 to 2022, and the PML model parameter calibration results and verification were performed using 2023 data, as shown in Table 1: Table 1. PML Model Parameter Calibration Results and Validation

[0028] The results showed that different growth stages g sx The values ​​are consistent with the physiological characteristics of rice and reeds. The optimized PML values ​​used in this study simulate the evapotranspiration of rice paddies and reeds. R 2 Increased to 0.83–0.88, RMSE The diameter is 0.44–0.70 mm.d -1 The Pearson correlation coefficients all exceeded 0.90. This improvement is related to the homogeneous vegetation structure (paddy fields and reeds) and relatively stable evapotranspiration processes in the study area (e.g., irrigation regulation and sufficient water), which reduced the interference of complex terrain or heterogeneous vegetation on the model, thus confirming the advantages of the PML model in areas with high evapotranspiration and homogeneous vegetation. The parameter calibration results and validation of the PT-JPL model are shown in Table 2. Table 2. PT-JPL Model Parameter Calibration Results and Validation

[0029] The optimized PT-JPL model in this study R 2 Increased to 0.76, growing season dayscale RMSE Reduced to 0.67–0.81 mm.d -1 The reasons are twofold: first, evapotranspiration is usually low during the non-growing season, and its contribution to the annual simulation error is limited. Growing season error is more representative of the model's simulation performance of the core evapotranspiration process in paddy fields and reed wetlands; second, this study focuses on paddy field and reed wetland ecosystems with sufficient water and uniform vegetation. The lower environmental heterogeneity significantly reduces the difficulty of parameter calibration and makes it easier for the optimization algorithm to play its role.

[0030] In addition, diurnal meteorological driving data include near-surface air temperature, which characterizes the ambient thermal conditions; relative humidity, which characterizes the degree of air humidity; and saturated vapor pressure difference, which characterizes the difference between air saturated vapor pressure and actual vapor pressure. VPD Net radiation, a core indicator of surface energy input. R n Soil heat flux, a quantitative indicator of energy exchange in the soil layer. G Air pressure, which characterizes the atmospheric pressure at the ground. P The dynamic adjustment of parameter optimization range adapts to the differences in physiological activity of vegetation at different growth stages, responds to the increased photosynthetic demand and transpiration cooling mechanism after temperature rise, matches the evapotranspiration characteristics of wetland vegetation at different growth stages, and improves the model's simulation realism of wetland ecological processes. The convergence condition of Bayesian optimization is that the objective function value is continuous. R The magnitude of change in each iteration < q The preset maximum number of evaluations is no less than S Next, among them R The threshold for the number of consecutive iterations is set to 3. q The critical value for the change in the objective function value is set at 1%. S The maximum number of evaluations threshold is set to 50. Iteration stops when any condition is met, balancing calibration efficiency and optimization accuracy. Wetland vegetation includes natural wetland vegetation (reeds) and artificial wetland vegetation (paddy fields), representing the main types of wetland ecosystems and the application scenarios required for the adaptation method.

[0031] In summary, after optimization by this invention, the simulation accuracy of both models was significantly improved. The PML model further enhanced its advantages by optimizing the core parameter, maximum stomatal conductance of leaves, while the PT-JPL model further enhanced its advantages by optimizing the canopy extinction coefficient, evapotranspiration regulation coefficient, and optimal vegetation growth temperature. Among them, the optimized PML model performed more outstandingly, not only with high overall fit and small error, but also with a higher correlation coefficient between the calibration and validation periods, accurately capturing the trend of evapotranspiration changes; its R 2 higher, RMSE With a smaller root mean square error and more consistent simulation values ​​with observed values, the model exhibits superior error stability. In seasonal dynamic simulations, the model accurately fits the measured sequences at each growth stage, precisely depicting the evapotranspiration distribution characteristics dominated by transpiration (82.2%–89.6%) in paddy fields and reed wetlands, as well as the phenological dynamics of transpiration initially rising and then falling, and soil evaporation decreasing with canopy cover. Furthermore, it demonstrates stable interannual fluctuations and superior bias control and long-term stability in annual cumulative evapotranspiration simulations.

[0032] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands, characterized in that, Includes the following steps: S1. Collect multi-source data from rice paddy and reed wetland observation stations, including meteorological, flux data and leaf area index data, and process the collected data, dividing it into two groups according to time period: calibration data and validation data. S2. Based on the growth characteristics of rice and reeds, sensitivity analysis was conducted on the parameters of the PML model and PT-JPL model to determine the sensitive parameters and set empirical values ​​for the non-sensitive parameters. S3. Divide the growth stages according to the different growth periods of rice and reeds, set the dynamic parameter optimization range in combination with wetland characteristics, and calibrate the key parameters of the PML model and PT-JPL model in stages based on years of measured evapotranspiration data and Bayesian optimization algorithm. S4. Select other years or periods that were not calibrated, and substitute the optimal parameter combination after calibration into the corresponding PML model or PT-JPL model. Combine the processed data and the inherent physical calculation logic of the model to simulate wetland evapotranspiration. S5. Based on measured evapotranspiration data that were not calibrated, the simulation effect of the model was verified through multiple reference indicators, and the optimal model and parameter combination suitable for the target wetland was selected.

2. The method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands according to claim 1, characterized in that, S1 specifically includes: S11. Based on the MOD15A2H product, collect remote sensing data of leaf area index. After extracting the remote sensing data of leaf area index of paddy fields and reed wetlands, compare it with the measured leaf area index data in the field. Optimize the data by linear interpolation to ensure the consistency between the two. Collect observation data of paddy field and reed wetland stations, including daily meteorological driving data and measured evapotranspiration data obtained based on station flux observations. S12. Based on the 3σ criterion, extreme outliers in meteorological elements and measured evapotranspiration data were removed. Erroneous data exceeding the physical reasonable range caused by instrument malfunction and invalid data with negative surface absorbed usable energy were manually deleted. Short-term missing data were filled using linear interpolation, and long-term missing data were estimated by combining the correlation of the data with the meteorological data of the same period. Leaf area index remote sensing data were processed by moving average. S13. Convert all processed data into units required for PML and PT-JPL model calculations, align them in the time dimension using date as the key, and ensure that meteorological data, measured evapotranspiration data and leaf area index data for the same date correspond one-to-one. S14. Divide the data processed by S13 into two groups according to the observation period or year: one group is the calibration data used for calibration, and the other group is the verification data used for verification.

3. The method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands according to claim 2, characterized in that, S2 specifically includes: PML model parameters include the extinction coefficient of available energy. k A Extinction coefficient of shortwave radiation k Q Water vapor pressure difference when the pore conductance is half of its maximum value D 50 Visible radiation flux when porosity is half its maximum value Q 50 Soil moisture stress factors f and the maximum porosity of the blade g sx ,in, g sx For sensitive parameters, k A , k Q , D 50 , Q 50 As a non-sensitive parameter, and set as an empirical value, based on the long-term supersaturation characteristics of soil moisture in the target wetland, f Set to 1; the PT-JPL model parameters include the light energy utilization coefficient. k 1. Moisture stress coefficient k 2. Evapotranspiration regulation coefficient β and optimal growth temperature of vegetation T opt ,in, k 1. k 2. β , T opt All of these are sensitive parameters. Based on the similarity between wetland and shrubland environments, empirical values ​​of shrubland were selected as the initial values ​​for the PT-JPL model parameters.

4. The method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands according to claim 3, characterized in that, S3 specifically includes: S31. The growth stages of natural wetland vegetation reeds are divided into budding stage, vegetative growth stage, reproductive growth stage, and maturity-early yellowing stage; the growth stages of artificial wetland vegetation paddy fields are divided into greening stage, tillering and jointing stage, booting and heading stage, and milk-maturity stage. S32. Based on the growth characteristics of rice and reeds, a dynamic parameter optimization range is set, whereby the growth stages are defined based on S31. g sx The values ​​were set to 0.001–0.10 ms for each stage of the reproductive period. -1 0.001~0.15ms -1 0.001~0.20ms -1 0.001~0.25ms -1 , k 1 is set to 0.55~0.

85. k 2 is set to 0.75~0.

95. β Set to 1.43~2.2, T opt Set to 20–28℃; S33. Based on the data processed in S1 and the optimization range of dynamic parameters for vegetation in S32, perform phased Bayesian optimization calibration, specifically including: S331, with g sx , k 1. k 2. β , T opt To optimize the variables, a parameter search space is constructed based on the parameter optimization range set in S32; S332. Using the error between the model simulation value and the measured value at the site as the evaluation standard, the root mean square error is selected. RMSE or negative Nash coefficient NSE Construct the objective function f(x) The optimization objective is to minimize f(x) ; S333. Use the Latin hypercube sampling method to select 5 to 10 sets of initial parameter combinations within the parameter search space. x i Run the PML model and the PT-JPL model respectively, and calculate the objective function values ​​corresponding to the models. y i To form the initial dataset D= { x i ,y i }; S334, Based on the initial dataset D A Gaussian process surrogate model is fitted, which outputs the predicted mean and predicted variance for any parameter point. By optimizing the acquisition function, the next parameter combination point to be evaluated is determined. x next ; S335, in x next Run the PML model and the PT-JPL model respectively to obtain the true objective function values. y next ,Will( x next , y next Add to dataset D And update the Gaussian process proxy model; S336. Repeat S334-S335 until the convergence condition is met or the preset maximum number of evaluations is reached, then select the objective function that makes the objective function... f (x) The optimal parameter combination is used as the calibration result for the corresponding PML model or PT-JPL model.

5. The method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands according to claim 4, characterized in that, S4 specifically includes: For years or periods not included in the calibration, the optimal parameter combination obtained from S336 was substituted into the PML and PT-JPL models respectively. At the same time, the daily meteorological data and leaf area index data processed by S1 were combined to calculate the daily total evapotranspiration. The PML model is based on the Penman-Monteith principle and combines canopy conductance and soil moisture stress factors to calculate the total evapotranspiration through energy balance. The PT-JPL model is based on the Priestley-Taylor theory and obtains the simulated value of total evapotranspiration by simulating vegetation canopy transpiration, soil evaporation and intercepted evaporation respectively and superimposing them.

6. The method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands according to claim 5, characterized in that, S5 specifically includes: S51. Based on measured evapotranspiration data that were not calibrated, the simulation effect of the model is verified by calculating reference indicators, including the coefficient of determination. R 2 Nash coefficient NSE Root mean square error RMSE ,in, R 2 The formula representing the correlation between simulated and measured values ​​is shown below: ； in, For the number of samples, For sample index, , For the first The measured evapotranspiration of each sample This is the average of all measured evapotranspiration. For the first Simulated evapotranspiration for each sample This is the average of all simulated evapotranspiration. NSE The efficiency of the model simulation is expressed by the following formula: ； RMSE The formula for expressing the deviation between simulated and measured values ​​is shown below: ； S52, Filter out R 2 ≥ A, NSE ≥ B and RMSE ≤ C The model and its corresponding parameter combination are used as the candidate set. The consistency of the time series trend is determined by calculating the deviation rate of the linear regression slope between the simulated values ​​and the measured values. The models with a deviation rate ≤ 0.5% are selected. M Furthermore, the model and parameter combination with a trend fit exceeding any other fit is considered the optimal result. A for R 2 The critical value is dimensionless. B for NSE The critical value is dimensionless. C for RMSE The critical value, in mm.d -1 , M This is the critical value for the deviation rate.

7. The method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands according to claim 2, characterized in that: Daily meteorological driving data include near-surface air temperature, relative humidity, and saturated vapor pressure difference. VPD Net radiation R n Soil heat flux G air pressure P .

8. The method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands according to claim 4, characterized in that: The dynamic adjustment of the parameter optimization range is adapted to the differences in physiological activity of rice and reeds at different growth stages, responding to the increased photosynthetic demand and transpiration cooling mechanism after the temperature rises, and matching the evapotranspiration characteristics of wetland vegetation at each growth stage.

9. The method for calibrating evapotranspiration model parameters for paddy fields and reed wetlands according to claim 4, characterized in that: The convergence condition for Bayesian optimization is that the objective function value is continuous. R The magnitude of change in each iteration < q Or preset a maximum number of evaluations not less than S The iteration stops when any one of the following conditions is met: q This represents the critical value for the magnitude of change in the objective function value.

Images