CN111833202A - A short-term forecasting method of farmland evapotranspiration considering dynamic changes of crop coefficients and rainfall - Google Patents

Abstract

The invention discloses a short-term forecasting method for farmland evapotranspiration considering dynamic changes of crop coefficients and rainfall. The method includes obtaining meteorological data of the growing environment of farmland crops; calculating a forecasting benchmark according to the reference crop evapotranspiration on the forecasting base day and the measured farmland evapotranspiration. daily crop coefficients; build training sets and test sets respectively, and perform preprocessing; establish a feedforward neural network model considering the dynamic changes of crop coefficients and the influence of rainfall, and perform training optimization; use the optimized feedforward neural network model to test Set data for short-term prediction of crop evapotranspiration. The present invention takes into account the influence of crop coefficient changes and rainfall on crop evapotranspiration, and effectively constructs a nonlinear relationship between the reference crop evapotranspiration and its driving factors, and accordingly, the crop evapotranspiration that is more in line with the actual growth conditions of crops can be obtained. Provide scientific basis for future water management of farmland underlying surface.

Description

A short-term forecasting method of farmland evapotranspiration considering dynamic changes of crop coefficients and rainfall

technical field

The invention belongs to the technical field of crop transpiration prediction, and in particular relates to a short-term prediction method for farmland evapotranspiration considering the dynamic change of crop coefficients and rainfall.

Background technique

Evapotranspiration is an important part of the terrestrial hydrological cycle, and evapotranspiration (ET _c ) has important guiding significance for irrigation planning and regional water resources allocation. Accurate prediction of evapotranspiration can save irrigation water to a certain extent. Therefore, in order to better manage crop irrigation water consumption and improve crop water use efficiency, it is urgent to accurately predict crop evapotranspiration.

At present, the prediction methods of evapotranspiration are mainly divided into four categories: time series method, grey model method, empirical formula method and neural network model method. Due to the single data used by the time series method (only the historical data of evapotranspiration is used), the ultra-historical changes under the influence of other factors cannot be fully considered, so there is uncertainty in the prediction accuracy. The grey forecasting method is essentially an exponential model. When the objective function has zero growth, the systematic error is serious, and the more the forecast period is, the more serious the error is. The empirical formula method requires parameter correction for different study areas, and requires a lot of meteorological data, making the calculation more complicated. Artificial neural network is a nonlinear theory developed in recent years. It does not need to understand the specific structural conditions of the nonlinear system. It has the functions of self-organization, self-adaptation and self-learning. It is very suitable for simulating and processing many influencing factors and complex relationships. The system provides an effective way for the prediction and evaluation of time series with highly nonlinear dynamic relationship. There is a complex nonlinear relationship between evapotranspiration and its driving factors. In view of the large blindness in the prediction of the conventional water consumption prediction model, the low fitting accuracy and the easy distortion of the prediction, the neural network calculation method is introduced to establish a non-linear model. The prediction model of linear artificial neural network can consider the influence of many factors on evapotranspiration. The established prediction model has high prediction accuracy, is simple and easy to implement, and has good application and promotion value.

At present, the BP neural network model is used to predict evapotranspiration mainly by training and testing conventional variables, and most of the crop coefficients recommended by FAO are used, or the historical measured crop coefficients are used to predict evapotranspiration. However, with the difference between the crop itself and the external conditions, the crop coefficient is also constantly changing, and there are obvious regional and time series differences. Studies have shown that the crop coefficient values recommended by FAO are suitable for process calculations with large time steps, but cannot reflect the daily dynamic changes of crops. When predicting evapotranspiration, there is a slight deviation between the predicted value and the measured value. Therefore, for the acquisition of crop coefficients, it is necessary to consider the influence of crop growth stages on its dynamic changes. In addition, rainfall is an important factor affecting the prediction accuracy of evapotranspiration. At present, no scholars have directly considered rainfall into the prediction of evapotranspiration.

To sum up, a lot of research on evapotranspiration prediction has been carried out, but there are two main problems: (1) The fixed or simple difference crop coefficient recommended by FAO-56 is suitable for long-term evapotranspiration prediction, while The prediction of short-term evapotranspiration has a low applicable value, and the basic crop coefficient curve in the crop coefficient method is only linearly differentiated by the determined three nodes, which simplifies the processing of the crop growth process, which will cause a large deviation; ( 2) Due to the uncertainty of rainfall, there are few studies on evapotranspiration prediction models considering rainfall. Only considering deterministic meteorological factors will inevitably lead to deviations in evapotranspiration prediction under typical weather, and the applicability is weak.

SUMMARY OF THE INVENTION

In view of the above deficiencies in the prior art, the present invention provides a short-term forecasting method for farmland evapotranspiration considering the dynamic change of crop coefficients and rainfall.

In order to achieve the above-mentioned purpose of the invention, the technical scheme adopted in the present invention is:

A short-term prediction method of farmland evapotranspiration considering the dynamic changes of crop coefficients and rainfall, including the following steps:

S1, obtain the meteorological data of the growing environment of farmland crops, the meteorological data includes the highest temperature, the lowest temperature, the number of sunshine hours and the rainfall;

S2. Calculate the crop coefficient on the forecast base day according to the reference crop evapotranspiration on the forecast base day and the farmland measured evapotranspiration;

S3, according to the meteorological data obtained in step S1 and the measured evapotranspiration of the farmland and the calculated crop coefficient in step S2, respectively construct a training set and a test set, and preprocess the data of the training set and the test set;

S4. Establish a feedforward neural network model considering the dynamic changes of crop coefficients and the influence of rainfall, and use the training set data to train and optimize the model;

S5, using the feedforward neural network model optimized in step S4 to predict the evapotranspiration of farmland crops according to the test set data.

Further, the step S2 specifically adopts the Penman algorithm to calculate the reference crop evapotranspiration on the forecast base day, and the calculation formula is:

Among them, ET ₀ is the reference crop evapotranspiration, Δ is the slope of the saturated water vapor pressure curve, Rn is the surface net radiation, G is the soil heat flux, γ is the dry and wet constant, T _mean is the daily average temperature, and u ₂ is the set height position wind speed, _es is the saturated water vapor pressure, and _ea is the actual water vapor pressure.

Further, the step S2 specifically adopts the vorticity correlation method to calculate the measured evapotranspiration of the farmland on the prediction base day, and the calculation formula is:

Among them, w' is the vertical wind speed fluctuation, and q' is the water vapor density fluctuation value.

Further, the calculation formula of the crop coefficient of the prediction base day in the step S2 is:

Among them, K _c is the crop coefficient on the forecast base day, and ET _c-EC is the measured value of the eddy correlation system;

And set the crop coefficient of the field crops in the future set time to be the same as the crop coefficient of the forecast base day.

Further, the preprocessing of the training set and the test set data in the step S3 is specifically:

Using the hyperbolic tangent transformation function, the sample data measurement values are standardized according to the weights of the maximum and minimum values of the sample data measurement values in the training set and the test set, which are expressed as:

Wherein, X' is the measured value of the sample data after standardization, X is the measured value of the sample data, and X _max and X _min are the maximum and minimum values of the measured value of the sample data, respectively.

Further, the step S4 is specifically:

Construct a feedforward neural network model with three-layer topology including input layer, hidden layer and output layer, and set the input layer to contain 4 neurons, the hidden layer to contain 10 neurons, and the output layer to contain 1 neuron in the BP neural network. Yuan.

The present invention has the following beneficial effects:

The present invention considers the influence of the dynamic changes of crop coefficients and rainfall factors on the short-term prediction of crop evapotranspiration, establishes a feedforward neural network model considering the dynamic changes of crop coefficients and the influence of rainfall, and uses the evapotranspiration measured by the vorticity correlation method as the measured value pair. The model is trained and optimized to effectively construct the nonlinear relationship between the reference crop evapotranspiration and its driving factors. Based on this, the crop evapotranspiration that is more in line with the actual growth conditions of the crop can be obtained, providing a scientific basis for the future water management of the underlying surface of the farmland. . .

Description of drawings

Fig. 1 is the schematic flow chart of the short-term prediction method of farmland evapotranspiration considering the dynamic change of crop coefficient and rainfall in the present invention;

Fig. 2 is the comparison chart of the predicted value and the measured value of the present invention and the multiple linear regression model in the embodiment of the present invention;

FIG. 3 is a graph showing the verification result of the predicted value and the measured value of the present invention and the multiple linear regression model in the embodiment of the present invention.

Detailed ways

The specific embodiments of the present invention are described below to facilitate those skilled in the art to understand the present invention, but it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes Such changes are obvious within the spirit and scope of the present invention as defined and determined by the appended claims, and all inventions and creations utilizing the inventive concept are within the scope of protection.

As shown in FIG. 1 , an embodiment of the present invention provides a short-term prediction method for farmland evapotranspiration considering dynamic changes of crop coefficients and rainfall, including the following steps S1 to S5:

S1, obtain the meteorological data of the growing environment of farmland crops, the meteorological data includes the highest temperature, the lowest temperature, the number of sunshine hours and the rainfall;

In this embodiment, the present invention takes Daxing District of Beijing as a research area to describe the relevant data. Daxing District, Beijing (39°26′-39°51′N, 116°13′-116°43′E) is located in the Yongding River Shock Plain in the northern part of the North China Plain, with a total area of 1031km2. It belongs to a temperate semi-humid monsoon climate with an annual average temperature is 12.1°C. The average annual rainfall for many years is 540mm, with more rainfall in July and September, accounting for more than 80% of the total annual rainfall. The underlying surface of Beijing Daxing Experiment Station is mainly farmland, including corn/wheat and soybean, among which corn and wheat are the main ones. The whole growth period of winter wheat is about 260 days (October 1st to June 30th of the following year). In normal years, winter wheat needs supplemental irrigation to ensure the water demand of crops. The growth period of summer corn is about 90 days (July 1st to September 30th), and no irrigation is required during the whole growth period of summer corn. There was no water stress in the growing stage of crops in the study area. From the comparative analysis of the climate and underlying surface conditions of the experimental station and the Daxing area, the experimental station has a good typicality.

The present invention collects historical meteorological data from 2015 to 2017 on the China Meteorological Science Shared Service Network, and collects daily weather forecast data for the forecast period from 1d to 7d from 2018 to 2019 on the Weather Network. Historical meteorological data include: air pressure P _a , wind speed U, maximum temperature T _max , minimum temperature T _min , average relative humidity RH, sunshine hours n, rainfall P and so on. Weather forecast data and information include: maximum temperature T _max , minimum temperature T _min , weather conditions, and the like.

According to the various phenomena caused by the influence of rainfall on the ground, the present invention estimates the rainfall to be divided into 11 grades according to the rainfall grade table. The rainfall value is then determined according to the rainfall forecast information of the weather forecast. The rainfall grade table is shown in Table 1.

Table 1 Rainfall grade table

In the present invention, sunshine hours refer to the length of time during which the radiation intensity of the sun on a plane perpendicular to its rays exceeds or equals to 120w/m2 every day. According to the geographical location parameters of the region, the daily radiation _Ra in a year can be calculated from the solar constant, the solar inclination angle, etc. The calculation formula is:

Among them, _Ra is the radiation, c is the speed of light, G _sc is the solar constant, d _r is the relative distance between the sun and the earth, W is the inclination of the sun, h is the local latitude, in radians; k _s is the sunset time angle.

S2. Calculate the crop coefficient on the forecast base day according to the reference crop evapotranspiration on the forecast base day and the farmland measured evapotranspiration;

In the present embodiment, in order to obtain a more accurate reference crop evapotranspiration on the forecast base day, the present invention uses the (Penman-Monteith) PM method to calculate the reference crop evapotranspiration on the forecast base day based on meteorological data, and the calculation formula is:

Among them, ET ₀ is the reference crop evapotranspiration, Δ is the slope of the saturated water vapor pressure curve, Rn is the surface net radiation, G is the soil heat flux, γ is the dry and wet constant, T _mean is the daily average temperature, and u ₂ is the set height Location wind speed, specifically the wind speed at a height of 2 meters, _es is the saturated water vapor pressure, and _ea is the actual water vapor pressure.

The present invention adopts the eddy correlation system (Campbell Scientific Inc., USA) to measure the measured evapotranspiration of the farmland on the forecast base day, and the calculation formula is:

Among them, w' is the vertical wind speed fluctuation, and q' is the water vapor density fluctuation value.

The above eddy correlation system consists of CSAT3 three-dimensional ultrasonic anemometer, LI7500 CO2/H2O open-circuit gas analyzer, HMP45C air temperature and humidity sensor and CR5000 data collector. The net radiation Rn is measured by the CNR4 net radiation sensor, and the soil heat flux G is measured by two HFP01 soil heat flux panels located 2cm below the surface. Accumulated from 24h data. In the actual data processing process of vorticity correlation, the abnormal data is eliminated according to the following principles: (1) the data of the precipitation period and 1 hour before and after; (2) the data that obviously exceeds the physical meaning; (3) the abnormal data of the sensor state. In addition, the latent heat flux is corrected by calculating the intraday Bowen ratio to eliminate errors caused by energy non-closure.

The present invention calculates the crop coefficient of the predicted base day according to the reference crop evapotranspiration and the farmland measured evapotranspiration obtained by the above-mentioned method, and is expressed as:

Among them, K _c is the crop coefficient on the forecast base day, and ET _c-EC is the measured value of the eddy correlation system;

And set the crop coefficient of farmland crops in the short-term set in the future to be the same as the crop coefficient of the forecast base day, specifically set the crop coefficient in the next 1 to 7 days to be the same as the dynamic crop coefficient of the forecast base day, that is, K′ _{c- var1} = _Kc =K'c _-var2 ...= _K'c-var7 .

Since the crop coefficient on the forecast base day can reflect the influence of meteorological data on evapotranspiration, as well as the influence of crop types, soil water and fertilizer conditions and field management level on evapotranspiration, the present invention considers the crop coefficient on the forecast base day to affect the crop in the field. The influence of short-term forecast of evapotranspiration, the construction of short-term forecast of crop evapotranspiration can more accurately predict the evapotranspiration of crops, so as to achieve short-term accurate forecast of crop evapotranspiration.

S3, according to the meteorological data obtained in step S1 and the measured evapotranspiration of the farmland and the calculated crop coefficient in step S2, respectively construct a training set and a test set, and preprocess the data of the training set and the test set;

In this embodiment, since a part of the meteorological data from 2015 to 2019 in the meteorological data obtained in step S1 may not exist or be abnormal due to environmental interference and human operation, the present invention finally obtains the obtained meteorological data after screening the obtained meteorological data. 900 datasets. Each training sample randomly selects 100 sets of data from the historical meteorological data from March to September in 2015-2017 as the training set, and randomly selects 100 sets of data from the historical meteorological forecast data from March to September in 2018-2019 as the training set. validation set.

Since the dimensions of various data in the data set are different and the magnitudes are relatively large, the present invention needs to preprocess the data of the training set and the test set; specifically, the data magnitudes are standardized, so as to avoid over-training and improve certain The convergence degree and calculation speed of the number of neural nodes in some layers are improved, and the calculation accuracy is improved.

The present invention adopts the hyperbolic tangent transformation function to standardize the sample data measurement value according to the weight of the maximum value and the minimum value of the sample data measurement value in the training set and the test set, which is expressed as:

Wherein, X' is the measured value of the sample data after standardization, X is the measured value of the sample data, and X _max and X _min are the maximum and minimum values of the measured value of the sample data, respectively.

By normalizing the sample data measurements, the sample data measurements can be normalized to the range [-1, 1], which shows the maximum nonlinearity.

S4. Establish a feedforward neural network model considering the dynamic changes of crop coefficients and the influence of rainfall, and use the training set data to train and optimize the model;

In this embodiment, BP model modeling includes two stages: preprocessing (including variable selection, data segmentation and data normalization), and training (including architecture and network training process). Each neuron in the upper layer of BP realizes the full connection to each neuron in the next layer through the transfer function, and there is no relationship between the neurons in the same layer. When the learning samples are provided to the neural network, the neural network first performs a forward propagation process. If the error between the output and the target output exceeds expectations, the forward propagation process is transferred to the back propagation process, and the error signal is returned along the original connection path, and the error signal is reduced by modifying the weights of neurons in each layer. As this error back-propagation correction continues, the correct rate of the network's response to the input pattern continues to improve, and finally achieves the applicable accuracy.

The present invention calculates according to the following formula that the setting range of the number of hidden layer units is 3 to 12, and the calculation formula is:

Among them, J is the number of hidden layer units, A is the number of input layer units; B is the number of output layer units; K is a constant, ranging from 1 to 10.

Since the number of input and output neurons is determined by the target and the connection method is fixed, the structure mainly depends on the number of neurons in the hidden layer. Too few hidden nodes will affect the function of the network, while too many hidden nodes will cause the network to over-adapt to the data. Therefore, the present invention sets the optimal BP neural network architecture 4-10-1, that is, the input layer in the BP neural network contains 4 neurons, the hidden layer contains 10 neurons, and the output layer contains 1 neuron; and set the training The learning rate and iterations are 0.1 and 5000, respectively.

The invention uses the training set data as input variables to train a feedforward neural network model considering the dynamic changes of crop coefficients and the influence of rainfall, and uses the eddy correlation method to calculate and predict the measured farmland evapotranspiration on the base day as an output variable, and after multiple iterations of training, Finally, the optimized feedforward neural network model is obtained.

S5, using the feedforward neural network model optimized in step S4 to predict the evapotranspiration of farmland crops according to the test set data.

In order to verify the prediction effect of the feedforward neural network model considering the dynamic change of crop coefficients and the influence of rainfall, the present invention compares the feedforward neural network model considering the dynamic change of crop coefficients and the influence of rainfall with the multiple linear regression model (MLR). , and use the mean absolute error MAE, root mean square error RMSE, coefficient of determination R ² and forecast accuracy ACC to evaluate the prediction accuracy of the model.

The calculation formula of the above evaluation parameters is:

为预测值和期望输出值序列的均值；n为预报值的样本数。ACC可对单个模型的外部预测能力是否达到统计所需精度给出度量，该值越接近于1，表示预测值与实测值越吻合，一般认为ACC>0.6时模型才有实际预测价值。Among them, x _i is the predicted value of crop evapotranspiration, _yi is the expected output value of evapotranspiration of reference crops in the farmland, and i is the predicted sample sequence, i=1,2,...,n;

is the mean of the series of predicted values and expected output values; n is the number of samples of predicted values. ACC can measure whether the external prediction ability of a single model reaches the required statistical accuracy. The closer the value is to 1, the more consistent the predicted value is with the measured value. It is generally believed that the model has actual predictive value when ACC>0.6.

As shown in Figure 2 and Figure 3, it is a comparison chart of the change trend of the predicted value and the measured value of the feedforward neural network model and the multiple linear regression model considering the dynamic change of the crop coefficient and the influence of rainfall in the present invention. It can be seen from the figure that the change trends of the predicted values and the measured values of these two methods are basically the same, but the pre-accuracy rate of the feedforward neural network model considering the dynamic changes of crop coefficients and the influence of rainfall is high, and the errors are all Smaller, it also proves the superiority of the present invention in reflecting complex nonlinear relationships.

Those of ordinary skill in the art will appreciate that the embodiments described herein are intended to assist readers in understanding the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations without departing from the essence of the present invention according to the technical teaching disclosed in the present invention, and these modifications and combinations still fall within the protection scope of the present invention.

Claims (6)

1. A farmland evapotranspiration short-term prediction method considering crop coefficient dynamic change and rainfall is characterized by comprising the following steps:

s1, acquiring meteorological data of the growth environment of the farmland crops according to meteorological forecast data, wherein the meteorological data comprise the highest air temperature, the lowest air temperature, sunshine hours and rainfall;

s2, calculating a crop coefficient of the prediction reference day according to the reference crop evapotranspiration of the prediction reference day and the actual field measurement evapotranspiration;

s3, respectively constructing a training set and a testing set according to the meteorological data obtained in the step S1, the actual field measurement evapotranspiration in the step S2 and the calculated crop coefficient, and preprocessing the data of the training set and the testing set;

s4, establishing a feedforward neural network model considering crop coefficient dynamic change and rainfall influence, and training and optimizing the model by using training set data;

and S5, predicting the evapotranspiration of the farmland crops according to the test set data by using the feedforward neural network model optimized in the step S4.

2. The method for short-term prediction of farmland evapotranspiration considering crop coefficient dynamic changes and rainfall as claimed in claim 1, wherein the step S2 specifically adopts a penman algorithm to calculate the reference crop evapotranspiration of the predicted reference day, and the calculation formula is as follows:

wherein, ET₀For reference to crop evapotranspiration, delta is the slope of the saturated water-vapor pressure curve, Rn is the net surface radiation, G is the soilHeat flux, gamma is the dry-wet constant, T_meanIs the average daily temperature u₂To set the altitude position wind speed, e_sSaturated water vapor pressure, e_aThe actual water vapor pressure.

3. The method for short-term prediction of farmland evapotranspiration considering crop coefficient dynamic changes and rainfall as claimed in claim 2, wherein the step S2 specifically uses vorticity correlation method to calculate the actual farmland evapotranspiration predicted on the reference day by the following formula:

wherein w 'is the pulsating quantity of the vertical wind speed, and q' is the pulsating value of the water vapor density.

4. The method for short-term prediction of agricultural evapotranspiration considering crop coefficient dynamics and rainfall according to claim 3, wherein the calculation formula for predicting the crop coefficient on the reference day in step S2 is:

wherein, K_cTo predict the crop coefficient of the reference day, ET_c-ECIs the measured value of the vorticity correlation system;

and setting the crop coefficient of the farmland crop in the future set time to be the same as the crop coefficient of the prediction reference day.

5. The method for short-term prediction of farmland evapotranspiration considering crop coefficient dynamic changes and rainfall as claimed in claim 1, wherein the preprocessing of the training set and test set data in step S3 is specifically:

and adopting a hyperbolic tangent transformation function to carry out standardization processing on the sample data measurement value according to the weights of the maximum value and the minimum value of the sample data measurement value in the training set and the test set, wherein the weights are expressed as follows:

wherein, X' is the measured value of the sample data after standardization, X is the measured value of the sample data, X_max、X_minRespectively the maximum and minimum values of the sample data measurement.

6. The method for short-term prediction of farmland evapotranspiration considering crop coefficient dynamic changes and rainfall as claimed in claim 1, wherein said step S4 is specifically:

a feedforward neural network model of a three-layer topological structure comprising an input layer, a hidden layer and an output layer is constructed, and the input layer comprises 4 neurons, the hidden layer comprises 10 neurons and the output layer comprises 1 neuron in the BP neural network.

Images