Reference crop evapotranspiration estimation method based on user data information

Abstract

The invention discloses a reference crop evapotranspiration estimation method based on user data information, and the method comprises the following steps: S1, constructing a national database, and calculating the daily-scale reference crop evapotranspiration; s2, a user uploads meteorological data and geographic information, and necessary information of more than two users needs to be provided according to a data template; s3, calculating a cumulative frequency curve of each meteorological factor; s4, performing quantile sequence comparison, selecting 15-20 sites with the highest distribution matching degree, and obtaining a simplified data set; s5, the Euclidean distance is calculated, and a data group with the minimum distance is selected to enter an alternative group; and S6, constructing a reference crop evapotranspiration model to carry out evapotranspiration estimation. According to the method, the optimal machine learning modeling reference crop evapotranspiration method is customized for the user by using the meteorological data owned by the user, the calculation error is lower, the data information owned by the user is fully mined, and the customized and refined reference crop evapotranspiration model is really realized.

Description

Technical Field

[0001] The invention relates to an estimation method, in particular to a reference crop evapotranspiration estimation method based on user data, and belongs to the field of irrigation management and hydrology and water resources application. Background Art

[0002] Evapotranspiration is a basic element in the field of agricultural water conservancy engineering and hydrological resources, and is also an important criterion for precise irrigation decision-making and water resource management in farmland. However, in most parts of the world, the evapotranspiration process cannot be directly observed, and the accuracy of evapotranspiration calculations mainly depends on the reference crop evapotranspiration. Reference crop evapotranspiration refers to the evapotranspiration of grassland (such as alfalfa) that is about 0.12m high and completely covers the ground under conditions of lush growth and sufficient water. Reference crop evapotranspiration is a key reference standard for crop evapotranspiration calculations. Its calculation accuracy directly determines the estimated results of actual evapotranspiration, which is crucial for terrestrial hydrothermal cycle simulation and precise agricultural resource management.

[0003] Currently, the Penman-Monteith (PM) model recommended by the Food and Agriculture Organization of the United Nations (FAO) is often used in the agricultural field to calculate reference crop evapotranspiration. However, the PM model requires comprehensive meteorological data such as maximum and minimum temperature, humidity, sunshine hours (or radiation), wind speed and atmospheric pressure, and most regions in the world cannot provide all the meteorological information required by the PM model.

[0004] On the other hand, although many scholars have proposed alternative PM estimation methods that use less meteorological data. These methods mainly include empirical model methods, machine learning methods, etc. For example, the Chinese patent application number 202010675336.3 proposed a method for predicting reference evapotranspiration using BP neural network and precipitation information. This type of method relies on PM data crop training target values, usually using local data for training and verification. There is great uncertainty in its ability to be applied in surrounding areas, and users are usually faced with the problem of how to choose the most appropriate method. Summary of the invention

[0005] In order to solve the deficiencies of the above-mentioned technologies, the present invention provides a reference crop evapotranspiration estimation method based on user data information.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a reference crop evapotranspiration estimation method based on user data information, characterized in that the estimation method comprises the following steps:

[0007] S1. Build a national database and calculate the reference crop evapotranspiration of daily-scale data based on the Penman-Monteith model equation;

[0008] S2. Users upload meteorological data and geographic information through online submission, and provide two or more mandatory information among the daily maximum and minimum temperatures, relative humidity, sunshine hours, and wind speed data according to the data template, and provide altitude and latitude information to perform pressure correction and extraterrestrial radiation calculation;

[0009] S3, calculating the cumulative frequency curve of the user according to each meteorological factor;

[0010] S4. Compare the quantile sequence obtained from the cumulative frequency curve of the meteorological factors in the user data with the quantile sequence of the meteorological factors in the database, select 15-20 stations with the highest distribution matching degree, and obtain a streamlined data set;

[0011] S5, comparing the user data with the simplified data set one by one, calculating the Euclidean distance, and taking the data set with the smallest distance into the candidate set;

[0012] S6, using the dataset created in S5 and the XGBoost machine learning method, taking the calculation result of the Penman-Monteith model equation as the objective function and the meteorological factor types provided by the user as input variables, to build a reference crop evapotranspiration model for the user;

[0013] S7, model training and optimization until the optimal reference crop evapotranspiration model is obtained;

[0014] S8. Estimate the reference crop evapotranspiration based on the optimal reference crop evapotranspiration model.

[0015] As a preferred embodiment, in S1, a database is constructed based on the daily maximum and minimum temperatures, relative humidity, sunshine hours, wind speed, altitude and latitude data of 2400 stations of the China Meteorological Data Sharing Network, and the constructed database is placed in the cloud platform.

[0016] Preferably, the formula for calculating the reference crop evapotranspiration using the Penman-Monteith model is as follows:

[0017]

[0018] Where △ is the slope of the saturated water vapor pressure temperature curve, which is calculated from the observed daily maximum and minimum temperatures; Rn is the net radiation, in MJ m -2 day -1 , calculated from the sunshine hours; G represents soil heat flux, in MJ m - 2 day -1 ; γ is the psychrometric constant, unit is kPa℃ -1 , calculated from atmospheric pressure; e s Indicates the saturated water vapor pressure in kPa, ea is the actual water vapor pressure in kPa, which is calculated from the maximum temperature Tmax, the minimum temperature Tmin and the relative humidity RH; u2 is the wind speed at 2m above the ground.

[0019] Preferably, in S2, the air pressure correction calculation formula is as follows:

[0020]

[0021] Where Pa is the air pressure in kPa and Z is the altitude in m.

[0022] Preferably, in S3, the cumulative frequency curve calculation method of each meteorological factor is:

[0023]

[0024] Among them, P is the cumulative frequency, m is the number less than a certain value, and n is the total number of data; the values ​​of 100 different meteorological factors at 1%-100% quantiles are determined according to the cumulative frequency curve.

[0025] Preferably, in S4, the root mean square error RMSE is used to measure the difference between the user data and the quantile sequence of each site; the smaller the error, the higher the distribution match between the site and the user data, thereby selecting 10-15 sites with the highest distribution match.

[0026] Preferably, in S5, the Euclidean distance is calculated as follows:

[0027]

[0028] Among them, j is the number of meteorological factors, x is the meteorological factor of the user data, and y is the corresponding meteorological factor in the streamlined data set; this process is repeated more than 5 times to obtain a training set with more than 5 times the amount of user data.

[0029] Preferably, in S8, the specific process of applying the reference crop evapotranspiration model to perform modeling is:

[0030] 1) Prepare the input meteorological factor data

[0031] Collect real-time meteorological factor data of the target area, including temperature, relative humidity, sunshine hours, wind speed and air pressure; ensure that the temporal resolution of meteorological factors is consistent with the model training data;

[0032] 2) Data preprocessing

[0033] Preprocess the meteorological factor data, including checking data integrity, filling missing values, and removing outliers; and make the meteorological factor data consistent with the training data format;

[0034] 3) Reference crop evapotranspiration model loading

[0035] Use the trained XGBoost model to load into the system and confirm that the model is loaded successfully and the input and output match;

[0036] 4) Reference crop evapotranspiration ET0 prediction

[0037] The preprocessed meteorological factor data is input into the XGBoost model, and the predicted value of ET0 is output, which is the reference crop evapotranspiration prediction of the target area;

[0038] 5) Post-process and verify the prediction results.

[0039] The present invention discloses a reference crop evapotranspiration estimation method based on user data information. The method utilizes the meteorological data owned by the user to tailor the most suitable machine learning modeling reference crop evapotranspiration method for the user, which has lower calculation error. Meanwhile, compared with the traditional method of building a simplified model for the user to use, the method can more fully mine the data owned by the user, and truly realize the customized and refined reference crop evapotranspiration model. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of the method of the present invention;

[0041] Figure 2 is the root mean square error of the method of the present invention and the control method under 4 different input combinations;

[0042] Figure 3 Comparison between the simulated values ​​of Beijing Station calculated by the control method and the values ​​calculated by the PM model;

[0043] Figure 4 Comparison between the simulated value of Beijing Station calculated by the present invention and the value calculated by the PM model; DETAILED DESCRIPTION

[0044] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0045] The present invention tailors a suitable model for the user, which can more accurately estimate the reference crop evapotranspiration in areas with insufficient data. With the help of machine learning and meteorological database, the meteorological data closest to the user's data distribution is screened out to train the model, thereby reducing the model uncertainty caused by data distribution differences when the model is applied.

[0046] For the reference crop evapotranspiration estimation method based on user data disclosed in the present invention, the process of the estimation method is as follows: Figure 1 As shown, the following steps are included:

[0047] S1. Construct a national database: Based on the daily maximum and minimum temperatures, relative humidity, sunshine hours, wind speed, altitude, and latitude data of 2,400 stations on the China Meteorological Data Sharing Network, a database is constructed, and the reference crop evapotranspiration of daily scale data is calculated according to the Penman-Monteith model equation. The constructed database is placed in the cloud platform.

[0048] The formula for calculating the reference crop evapotranspiration using the Penman-Monteith model is as follows:

[0049]

[0050] Where △ is the slope of the saturated water vapor pressure temperature curve, which is calculated from the observed daily maximum and minimum temperatures; Rn is the net radiation (MJ m -2 day -1 ), calculated from the sunshine hours; G represents soil heat flux (MJ m -2 day -1 ), its influence can be ignored; γ is the psychrometric constant (kPa℃ -1 ), which can be calculated from the atmospheric pressure; e s represents the saturated water vapor pressure (kPa), e a is the actual water vapor pressure (kPa), which can be calculated from the maximum temperature Tmax, the minimum temperature Tmin and the relative humidity RH; u2 is the wind speed at 2m above the ground.

[0051] S2. Upload meteorological data and geographic information: Users upload meteorological data and geographic information through online submission. According to the data template, users provide two or more mandatory information of daily maximum and minimum temperatures, relative humidity, sunshine hours, and wind speed data, and provide altitude and latitude information to perform air pressure correction and extraterrestrial radiation calculation.

[0052] The barometric pressure correction calculation formula is as follows:

[0053]

[0054] Where Pa is the air pressure (kPa) and Z is the altitude (m). This formula adjusts the air pressure by comparing the current altitude with standard sea level conditions.

[0055] S3. Calculate the cumulative frequency curve of each meteorological factor for the user. The calculation method is:

[0056]

[0057] Wherein, P is the cumulative frequency, m is the number of values ​​less than a certain value, and n is the total number of data. According to the cumulative frequency curve, the values ​​of different meteorological factors of 1%-100% quantiles (a total of 100) are determined respectively.

[0058] S4. The quantile sequence obtained based on the cumulative frequency curve of the meteorological factors of the user data is compared with the quantile sequence of meteorological factors in the database, and 10-15 stations with the highest distribution matching are selected to obtain a streamlined data set.

[0059] Specifically, the root mean square error (RMSE) is used to measure the difference between the user data and the quantile series of each site. The smaller the error, the higher the distribution match between the site and the user data.

[0060] S5. Compare the user data with the simplified data set one by one, calculate the Euclidean distance, and select the data set with the smallest distance to enter the candidate set. The calculation method is:

[0061]

[0062] Where j is the number of meteorological factors, x is the meteorological factor of the user data, and y is the corresponding meteorological factor in the simplified data set. This process is repeated more than 5 times to obtain a training set with more than 5 times the amount of user data.

[0063] S6, using the dataset created in S5 and the XGBoost machine learning method, taking the calculation result of the Penman-Monteith model equation as the objective function and the meteorological factor types provided by the user as input variables, to build a reference crop evapotranspiration model for the user;

[0064] S7, model training and optimization until the optimal reference crop evapotranspiration model is obtained;

[0065] S8. Estimate the reference crop evapotranspiration based on the optimal reference crop evapotranspiration model.

[0066] The specific process of subsequent application of the model is as follows:

[0067] 1) Prepare the input meteorological factor data

[0068] Collect real-time meteorological factor data of the target area, including temperature (maximum, minimum and average), relative humidity, sunshine hours (or solar radiation), wind speed and air pressure, etc. According to data requirements, ensure that the temporal resolution of meteorological factors (such as day, hour) is consistent with the model training data.

[0069] 2) Data preprocessing

[0070] Check data integrity, fill missing values, and remove outliers to maintain data quality. If the input data unit or format is different from the training data, it needs to be converted to ensure consistency.

[0071] 3) Reference crop evapotranspiration model loading

[0072] Use the trained XGBoost model to load it into the system and confirm that the model is loaded successfully and the input and output match.

[0073] 4) Reference crop evapotranspiration ET0 prediction

[0074] The preprocessed meteorological factor data is input into the loaded XGBoost model, and the model performs hierarchical weighted calculation of the decision tree based on the input features and outputs the predicted value of ET0. The predicted value of ET0 output by the model is the reference crop evapotranspiration prediction for the target area.

[0075] 5) Post-processing and verification

[0076] Perform necessary post-processing on the forecast results, including unit conversion and error correction. If there is real historical data, a comparative analysis between the model forecast value and the actual value can be performed to verify the accuracy of the model.

[0077] 6) Application decision support

[0078] The predicted ET0 results are applied to irrigation management decisions, and combined with the local crop water demand patterns, scientific irrigation plans are provided for farmers or managers. The irrigation amount or frequency is adjusted according to the predicted ET0 value to ensure that the water demand of crops under different climatic conditions is met, thereby optimizing the efficiency of water resource utilization.

[0079] Example

[0080] The reference crop evapotranspiration estimation method based on user data information of the present invention is further described below in conjunction with specific application examples.

[0081] In this embodiment, a meteorological database from 1956 to 2021 is first constructed for 2,400 stations across the country. The meteorological data includes daily maximum and minimum temperatures, wind speed, relative humidity, sunshine hours, station longitude and latitude, and altitude.

[0082] like Figure 1 , a reference crop evapotranspiration estimation method based on user data information, comprising the following steps:

[0083] (1) Basic data collection: Collect data on meteorological elements such as the highest and lowest temperatures, relative humidity, sunshine hours (or radiation), wind speed, and atmospheric pressure at stations with data in the region, as well as data on the highest and lowest temperatures and longitude and latitude of stations where data is missing.

[0084] (2) User data information: Assuming that the user has reference crop evapotranspiration calculation data from 2400 stations across the country, the cumulative frequency distribution curve of meteorological factors at each station is calculated separately.

[0085] (3) For each station, construct 15-20 alternative stations that are closest to the distribution of meteorological factors of the station.

[0086] The user data is compared with the data of the candidate data set one by one, and the Euclidean distance between each data and the candidate data set is calculated. The data group with the smallest distance is selected to enter the model training group. This process is repeated about 8 times.

[0087] The reference crop evapotranspiration is calculated based on the Penman-Monteith model, and the formula is as follows:

[0088]

[0089] Where △ is the slope of the saturated water vapor pressure temperature curve, which is calculated from the observed daily maximum and minimum temperatures; Rn is the net radiation (MJ m -2 day -1 ), calculated from the sunshine hours; G represents soil heat flux (MJ m -2 day -1 ), its influence can be ignored; γ is the psychrometric constant (kPa℃ -1 ), which can be calculated from the atmospheric pressure; e s represents the saturated water vapor pressure (kPa), e a is the actual water vapor pressure (kPa), which can be calculated from the maximum temperature Tmax, the minimum temperature Tmin and the relative humidity RH; u2 is the wind speed at 2m above the ground.

[0090] The machine learning model uses XGBoost. In the machine learning modeling of this embodiment, a 5-fold cross validation method is used to train the machine learning model using candidate data. The model is applied to user site data, and the root mean square error is used for accuracy evaluation.

[0091] In order to fully demonstrate the superiority of the model, four groups of inputs were set to evaluate the model. The model input combinations are: 1) extraterrestrial radiation, maximum temperature and minimum temperature; 2) total solar radiation, maximum temperature and minimum temperature; 3) extraterrestrial radiation, relative humidity, maximum temperature and minimum temperature; 4) extraterrestrial radiation, wind speed, maximum temperature and minimum temperature.

[0092] The comparison method divides the 2,400 stations in the country into 7 regions according to climate type, and then uses the Kmeans++ clustering method to further subdivide the regional stations into 100-200 stations per group. The above 100-200 stations are randomly divided into 5 groups, and 4 of them are used to train the model each time, and the remaining group is used as an independent sample test. This process is repeated 5 times, so that all stations participate in one test.

[0093] like Figure 2As shown, the root mean square error comparison between the method of this embodiment and the control method under 4 different input combinations is shown. It can be seen that the RMSE of the method of this study is reduced by an average of 13% compared with the control method.

[0094] like Figure 3 In the figure, taking Beijing as an example, the comparison between the simulated values ​​of Beijing Station calculated by the control method and the calculated values ​​by the PM model is shown. It can be seen that the average determination coefficient of the control method under 4 different input combinations is 0.839.

[0095] like Figure 4 In the paper, the simulation value of Beijing Station calculated by the method of the present invention is compared with the value calculated by the PM model, taking Beijing as an example. It can be seen that the determination coefficient of the method of the present invention under 4 different input combinations is 0.902 on average. The determination coefficient of the research method is increased by 7.51% on average compared with the control method, and the effect is significant.

[0096] The above implementation modes are not limitations of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the scope of the technical solution of the present invention also belong to the protection scope of the present invention.

Claims

1. A method for estimating reference crop evapotranspiration based on user data information, characterized in that: The estimation method includes the following steps: S1. Build a national database and calculate the reference crop evapotranspiration on a daily scale; S2. Users upload meteorological data and geographic information through online submission, and provide two or more mandatory information among the daily maximum and minimum temperatures, relative humidity, sunshine hours, and wind speed data according to the data template, and provide altitude and latitude information to perform pressure correction and extraterrestrial radiation calculation; S3, calculating the cumulative frequency curve of the user according to each meteorological factor; S4. Compare the quantile sequence obtained from the cumulative frequency curve of the meteorological factors in the user data with the quantile sequence of the meteorological factors in the database, select 15-20 stations with the highest distribution matching degree, and obtain a streamlined data set; S5, comparing the user data with the simplified data set one by one, calculating the Euclidean distance, and taking the data set with the smallest distance into the candidate set; S6, using the dataset created in S5 and the XGBoost machine learning method, taking the calculation result of the Penman-Monteith model equation as the objective function and the meteorological factor types provided by the user as input variables, to build a reference crop evapotranspiration model for the user; S7, model training and optimization until the optimal reference crop evapotranspiration model is obtained; S8. Estimate the reference crop evapotranspiration based on the optimal reference crop evapotranspiration model.

2. The reference crop evapotranspiration estimation method based on user data information according to claim 1, characterized in that: In S1, a database is constructed based on the daily maximum and minimum temperatures, relative humidity, sunshine hours, wind speed, altitude, and latitude data of 2,400 stations on the China Meteorological Data Sharing Network, and the constructed database is placed in the cloud platform.

3. The reference crop evapotranspiration estimation method based on user data information according to claim 1, characterized in that: In S1, the reference crop evapotranspiration is calculated according to the Penman-Monteith model equation, and the formula is as follows: Where △ is the slope of the saturated water vapor pressure temperature curve, which is calculated from the observed daily maximum and minimum temperatures; Rn is the net radiation, in MJ m -2 day -1 , calculated from the sunshine hours; G represents soil heat flux, in MJ m -2 day -1 ; γ is the psychrometric constant, unit is kPa℃ -1 , calculated from atmospheric pressure; e s Indicates the saturated water vapor pressure in kPa, e a is the actual water vapor pressure in kPa, which is calculated from the maximum temperature Tmax, the minimum temperature Tmin and the relative humidity RH; u2 is the wind speed at 2m above the ground.

4. The method for estimating reference crop evapotranspiration based on user data information according to claim 1, characterized in that: In S2, the air pressure correction calculation formula is as follows: Where Pa is the air pressure in kPa and Z is the altitude in m.

5. The method for estimating reference crop evapotranspiration based on user data information according to claim 1, characterized in that: In S3, the cumulative frequency curve of each meteorological factor is calculated as follows: Among them, P is the cumulative frequency, m is the number less than a certain value, and n is the total number of data; the values ​​of 100 different meteorological factors at 1%-100% quantiles are determined according to the cumulative frequency curve.

6. The method for estimating reference crop evapotranspiration based on user data information according to claim 1, characterized in that: In S4, the root mean square error (RMSE) is used to measure the difference between the user data and the quantile series of each site; the smaller the error, the higher the distribution match between the site and the user data, and thus 10-15 sites with the highest distribution match are selected.

7. The method for estimating reference crop evapotranspiration based on user data information according to claim 1, characterized in that: In S5, the Euclidean distance is calculated as: Among them, j is the number of meteorological factors, x is the meteorological factor of the user data, and y is the corresponding meteorological factor in the streamlined data set; this process is repeated more than 5 times to obtain a training set with more than 5 times the amount of user data.

8. The method for estimating reference crop evapotranspiration based on data information according to claim 1, characterized in that: In S8, the specific process of modeling using the reference crop evapotranspiration model is as follows: 1) Prepare the input meteorological factor data Collect real-time meteorological factor data of the target area, including temperature, relative humidity, sunshine hours, wind speed and air pressure; Ensure that the temporal resolution of meteorological factors is consistent with the model training data; 2) Data preprocessing Preprocess the meteorological factor data, including checking data integrity, filling missing values, and removing outliers; and make the meteorological factor data consistent with the training data format; 3) Reference crop evapotranspiration model loading Use the trained XGBoost model to load into the system and confirm that the model is loaded successfully and the input and output match; 4) Reference crop evapotranspiration ET0 prediction The preprocessed meteorological factor data is input into the XGBoost model, and the predicted value of ET0 is output, which is the reference crop evapotranspiration prediction of the target area; 5) Post-process and verify the prediction results.

Images