CN116502050A - Dynamic interpolation method and system for missing evapotranspiration observations at global flux sites - Google Patents

Abstract

The invention belongs to the technical field of flux observation data processing, and relates to a dynamic interpolation method and system for missing evapotranspiration observations at global flux sites. The method includes: obtaining observational data of global flux stations and MODIS remote sensing data; constructing a dynamic interpolation database of missing evapotranspiration observations of global flux stations; obtaining available variables and the quantity of each available variable station by station; sorting the importance of available variables; Variable combinations of various available variable quantities; sort the variable combinations according to the interpolation accuracy; use the random forest method to establish a dynamic interpolation model for the missing evapotranspiration observations of each variable combination site by site; dynamically interpolate the evapotranspiration observation data at the missing time, and continuously Update the imputation rate until the imputation rate reaches 100%. The invention can realize the high-precision dynamic interpolation of flux observation evapotranspiration deficiency, and helps to improve the practical value of flux observation data.

Description

Dynamic interpolation method and system for missing evapotranspiration observations at global flux sites

technical field

The invention relates to the technical field of flux data processing, in particular to a dynamic interpolation method and system for missing evapotranspiration observations at global flux sites.

Background technique

EC (Eddy covariance, referred to as EC) is a method for real-time measurement of turbulent flux between ecosystems and the atmosphere based on micrometeorology methods and eddy related technologies. Long-term EC flux observation data are of great significance for water cycle and energy budget analysis, short-term weather forecast, long-term climate forecast, and agricultural irrigation management. Due to external interference such as instrument failure, system failure, poor management, or weather and data quality control, etc., there are often a large number of missing time series data obtained from observations, and the realization of high-quality interpolation of missing data is crucial for in-depth research on land-atmosphere interpolation. Carbon, water and energy fluxes are critical.

Existing data imputation methods usually perform observation missing imputation with a fixed number of variables. For example, the flux observation vacancy data interpolation method officially adopted by the FLUXNET2015 flux observation data set is the marginal distribution sampling (MDS) method, which uses three variables of incident shortwave radiation, air temperature and saturated water vapor pressure deficit to perform flux observations Imputation of missing data.

The disadvantage of the MDS interpolation method is that the accuracy of the interpolation is limited by the quantity and quality of the three variables, and the duration of the interpolation is also limited, making it difficult to interpolate a time gap of more than 60 days. Other imputation methods have similar limitations, and the imputation of flux missing data largely depends on the choice of variables by researchers. Since all stations use limited and identical variables, the imputation accuracy will be limited by the selected fixed variables. The method with high imputation accuracy has a low imputation rate, making it difficult to achieve an ideal balance between imputation accuracy and imputation rate. .

In addition, in practice, the number of variables that can be obtained from different stations is uneven, and it cannot be guaranteed that all stations have the input variables required by the imputation method, so it is difficult to realize the imputation of missing data at all stations, so that the valuable data obtained from some stations cannot be obtained. Get the most out of it.

Contents of the invention

In order to solve the problems existing in the prior art that the time length of interpolation is limited by the quantity and quality of data, the interpolation accuracy is limited, the interpolation rate is insufficient, and it is difficult to achieve a balance between interpolation accuracy and interpolation rate, the present invention provides a global flux A dynamic imputation method for missing evapotranspiration observations at a site.

In the first aspect, the present invention provides a dynamic interpolation method for flux observation missing data, including:

Obtain the observation data and MODIS remote sensing data of the global flux site; the observation data of the flux site include meteorological observation data and flux observation data;

According to the meteorological observation data, the flux observation data and the MODIS remote sensing data, construct a dynamic interpolation database for missing evapotranspiration observations at the global flux site;

Preprocessing the data in the dynamic interpolation database for the lack of evapotranspiration observations, and constructing input data sets site by site;

Based on the input data set, obtain the available variables of each of the flux sites and the quantity of each of the available variables site by site; the available variables are used to characterize the observation data of the flux sites and the MODIS remote sensing data;

According to the available variables, the random forest method is used to rank the importance of each of the available variables site by site;

According to the results of the quantity and importance ranking of each of the available variables, determine a variable combination that includes the quantity of each of the available variables;

Calculating the interpolation accuracy and imputation rate of all the variable combinations, and sorting each of the variable combinations according to the interpolation accuracy from high to low;

Based on the variable combinations that include various available variable quantities, the random forest method is used to establish dynamic interpolation models for missing evapotranspiration observations site by site for each combination of variables, and a set of dynamic interpolation models for missing evapotranspiration observations is obtained; The missing dynamic interpolation model of evapotranspiration observation is used to characterize the functional relationship between evapotranspiration observation data and the combination of variables;

According to the meteorological observation data and the MODIS remote sensing data at the time when the evapotranspiration observation data is missing, sequentially interpolate the evapotranspiration observation data at the missing time, and continuously update the interpolation rate until the interpolation rate reaches 100%, and Label the combination of variables used to impute each gap.

In the second aspect, the present invention provides a dynamic interpolation system for lack of evapotranspiration observations at global flux sites, including a first acquisition unit, a database construction unit, a preprocessing unit, a second acquisition unit, a first sorting unit, a determination unit, a second Sorting unit, model building unit and imputation update unit:

The first acquisition unit is used to acquire the observation data and MODIS remote sensing data of the global flux site; the observation data of the global flux site includes meteorological observation data and flux observation data;

A database construction unit, configured to construct a dynamic interpolation database for missing evapotranspiration observations at each flux site according to the meteorological observation data, the flux observation data, and the MODIS remote sensing data;

A preprocessing unit, configured to preprocess the data in the dynamic interpolation database for missing evapotranspiration observations, and construct input data sets site by site;

The second acquisition unit is configured to acquire the available variables of each of the flux sites and the quantity of each of the available variables site by site based on the input data set; the available variables are used to characterize the observation data of the flux sites with the MODIS remote sensing data;

The first sorting unit is configured to sort the importance of each of the available variables site by site using a random forest method according to the available variables;

A determination unit, configured to determine a variable combination including the quantity of various available variables according to the quantity and importance ranking of each available variable;

The second sorting unit is used to calculate the interpolation accuracy and imputation rate of all the variable combinations, and sort each of the variable combinations according to the interpolation accuracy from high to low;

The model building unit is used to establish a dynamic interpolation model of missing evapotranspiration observations for each of the variable combinations site by site based on the variable combinations including various available variable quantities, and obtain a dynamic interpolation model for missing evapotranspiration observations. A collection of complementary models; the dynamic interpolation model for missing evapotranspiration observations is used to characterize the functional relationship between evapotranspiration observation data and the combination of variables;

The interpolation update unit is used to interpolate the meteorological observation data and the MODIS remote sensing data at the missing time in turn, and constantly update the interpolation rate until the interpolation rate reaches 100%, and the variables used to interpolate each gap combination to label.

The beneficial effects of the present invention are: the present invention adopts the random forest algorithm in machine learning, can obtain the maximum number of variable combinations based on the available variables of different sites, and train a random forest algorithm with high robustness and strong predictability based on different variable combinations. Forest model, and based on the test set to obtain the imputation accuracy of different variable combinations, and finally according to the accuracy of the random forest model results, sort first and then interpolate to achieve high-precision and large-scale interpolation.

On the basis of the above technical solutions, the present invention can also be improved as follows.

Further, the weather observation data includes observation date, observation time, air temperature, wind speed, atmospheric pressure, air humidity, incident short-wave radiation and net radiation; the flux observation data includes the evapotranspiration observation data; the MODIS remote sensing The data include normalized difference vegetation index data and leaf area index data.

Further, the data in the dynamic interpolation database for missing evapotranspiration observations are preprocessed, and input data sets are constructed site by site, including:

Obtain the observation date and observation time, calculate the order of the days of the observation data of the flux site in a year one by one, and convert the observation time to the half-hour time of the observation data of the flux site at 48 days a day The order in the half-hour time is obtained by the order of time; the order of days, the order of time, and the meteorological observation data are combined with the MODIS data of the pixel corresponding to the flux station extracted according to the latitude and longitude, and constructed site by site Input dataset.

Further, based on the variable combinations containing various available variable quantities, the random forest method is used to establish a dynamic interpolation model for missing evapotranspiration observations of each variable combination site by site, including:

set up represents the evapotranspiration observation data in the input data set, and the functional relationship is /> ,but:

;

in, The order of the days of the year representing the date of the observed data for the flux site, /> Represents the order of the half-hour time where the observation data of the flux site is located in the 48 half-hour time of the day, /> Represents the air temperature at the time when evapotranspiration does not appear missing in the input data set, /> represents the incident shortwave radiation at the time when evapotranspiration does not appear missing in the input data set, /> represents the net radiance at which no missing moments in evapotranspiration appear in the input dataset, /> Represents the wind speed at the time when evapotranspiration does not appear missing in the input data set, Represents the normalized difference vegetation index at the time when evapotranspiration does not appear missing in the input dataset.

Further, according to the meteorological observation data and the MODIS remote sensing data at the time when the evapotranspiration observation data is missing, sequentially interpolate the evapotranspiration observation data at the missing time, including:

;

In the formula, represents the imputation result of the observed evapotranspiration data, /> Represents the air temperature observation data corresponding to the missing time of evapotranspiration, /> Represents the incident shortwave radiation observation data corresponding to the missing time of evapotranspiration, Represents the net radiation observation data corresponding to the missing time of evapotranspiration, /> Represents the wind speed observation data corresponding to the missing time of evapotranspiration, /> Represents the normalized difference vegetation index remote sensing data corresponding to the missing time of evapotranspiration.

Further, among the variable combinations with the same number of available variables, if the available variables included in the variable combinations are different, the data interpolation rates after interpolation using the random forest method are different.

Further, according to the meteorological observation data and the MODIS remote sensing data at the time when the evapotranspiration observation data is missing, the evapotranspiration observation data at the missing time are sequentially interpolated by means of ergodic interpolation and updating.

Description of drawings

FIG. 1 is a flow chart of the dynamic interpolation method for flux data gaps provided by Embodiment 1 of the present invention;

Figure 2 is a schematic diagram of the principle of dynamic interpolation of flux data gaps;

Fig. 3 is a schematic diagram of the flux data gap dynamic interpolation system provided by Embodiment 2 of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. The components of the embodiments of the invention generally described and illustrated in the figures herein may be arranged and designed in a variety of different configurations.

Example 1

As an example, as shown in Figure 1, in order to solve the above-mentioned technical problems, this example provides a dynamic interpolation method for missing evapotranspiration observations at global flux sites, including:

Obtain observation data and MODIS remote sensing data of global flux stations; observation data of flux stations include meteorological observation data and flux observation data;

Based on meteorological observation data, flux observation data and MODIS remote sensing data, construct a dynamic interpolation database for missing evapotranspiration observations at global flux sites;

Preprocess the data in the dynamic interpolation database for missing evapotranspiration observations, and construct input data sets site by site;

Based on the input data set, the available variables of global flux stations and the quantity of each available variable are obtained station by station; the available variables are used to represent the observation data of flux stations and MODIS remote sensing data;

According to the available variables, the random forest method is used to rank the importance of each available variable site by site;

According to the number and importance ranking results of each available variable, determine the variable combination that includes the number of available variables;

Calculate the imputation accuracy and imputation rate of all variable combinations, and sort each variable combination according to the imputation accuracy from high to low;

Based on variable combinations that contain various available variable quantities, the random forest method is used to establish dynamic interpolation models for missing ET observations site by site, and a set of dynamic interpolation models for missing ET observations is obtained; dynamic interpolation models for missing ET observations are obtained; The model is used to characterize the functional relationship between evapotranspiration observation data and variable combinations;

According to the meteorological observation data and MODIS remote sensing data at the time when the evapotranspiration observation data is missing, the evapotranspiration observation data at the missing time are sequentially interpolated, and the interpolation rate is continuously updated until the interpolation rate reaches 100%. combination to label.

Optionally, meteorological observation data include but not limited to observation date, observation time, air temperature, wind speed, atmospheric pressure, air humidity, incident shortwave radiation and net radiation; flux observation data include but not limited to evapotranspiration observation data; MODIS remote sensing The data include normalized difference vegetation index data and leaf area index data.

Optionally, preprocess the data in the dynamic imputation database for missing evapotranspiration observations, and construct input data sets site by site, including:

Obtain the observation date and observation time, calculate the order of the days of the observation data of the flux site one by one in a year, and convert the observation time to the half-hour time where the observation data of the flux site is located in the 48 half-hour time of the day The order of time is obtained; the order of days, time, and meteorological observation data are combined with the MODIS data of pixels corresponding to the flux stations extracted according to latitude and longitude, and the input data set is constructed station by station.

Optionally, based on variable combinations that contain various available variable numbers, the random forest method is used to establish a dynamic imputation model for missing evapotranspiration observations of each variable combination site by site, including:

set up Represents the evapotranspiration observation data in the input data set, and the functional relationship between the evapotranspiration observation data and the variable combination is /> ,but:

;

in, Represents the order of the days of the year for the observation data of the flux site, /> Represents the order of the half-hour time where the observation data of the flux site is located in the 48 half-hour time of the day, /> Represents the air temperature at the moment when there is no missing evapotranspiration observation data in the input data set, /> Represents the incident short-wave radiation at the time when there is no missing evapotranspiration observation data in the input data set, /> Represents the net radiance at the time when no evapotranspiration observation data is missing in the input data set, /> Represents the wind speed at the time when evapotranspiration does not appear missing in the input data set, The normalized difference vegetation index representing the moment when evapotranspiration does not appear missing in the input dataset.

Optionally, according to the meteorological observation data and the MODIS remote sensing data at the time when the evapotranspiration observation data is missing, sequentially interpolate the evapotranspiration observation data at the missing time, including:

;

In the formula, represents the interpolation result of evapotranspiration observation data, /> Represents the air temperature observation data corresponding to the missing time of evapotranspiration, /> Represents the incident shortwave radiation observation data corresponding to the missing time of evapotranspiration, /> Represents the net radiation observation data corresponding to the missing time of evapotranspiration, /> Represents the wind speed observation data corresponding to the missing time of evapotranspiration, /> Represents the normalized difference vegetation index remote sensing data corresponding to the missing time of evapotranspiration.

Optionally, among variable combinations with the same number of available variables, if the variable combinations contain different available variables, the imputation rate of the data after imputation using the random forest method is different.

Optionally, according to the meteorological observation data and the MODIS remote sensing data at the time when the evapotranspiration observation data is missing, the evapotranspiration observation data at the missing time are sequentially interpolated, and the evapotranspiration observation data at the missing time are interpolated by traversal interpolation and updating. Perform imputation.

In the actual application process, as shown in Figure 2, the input data set such as FLUXNET2015 contains different types of variables in each site, which increases the complexity of feature selection. By obtaining the site A, site B, site C and site D The maximum available variable N of , and decrease in sequence on the basis of this value, construct N-3 variable combinations (at least 3 variable combinations).

For a single site, different data gaps correspond to different variable categories. If several variables are selected for modeling and prediction, only gaps that partially meet the variable conditions can be predicted, and gaps with missing variables cannot be predicted. The number of variable combinations can meet the variable conditions required to predict different gaps, greatly increase the imputation rate of site data, and realize the complete reconstruction of the data set.

In the combination of the same number of variables, the data interpolation rates of different variable combinations are different after modeling. In order to maximize the interpolation rate of the input data set and calculate the variable combination imputation accuracy of the number of available variables, select The variable combination with the highest interpolation accuracy was used to train the random forest model separately, and the maximum number of variables was the number of variables available at each site.

Optionally, according to the meteorological observation data and MODIS remote sensing data at the time when the evapotranspiration observation data is missing, the evapotranspiration observation data at the missing time are sequentially interpolated by means of traversal interpolation and updating.

As shown in Figure 2, for each site, the specific implementation steps of building a random forest model are as follows:

Identify all available variables for this site ;/> It is the sequence of days in a year where the observation data of flux stations are located /> , /> It is the order of the half-hours in the 48 half-hours of a day where the observation data of the flux site is located at half-hours/> , /> , /> ;/> Available variants for day order and half hour order exceptions.

Using a random forest model to divide and /> All available variables other than /> Perform importance sorting to get a sequence of variables sorted by importance: /> ;

Sort the results by importance, in and /> Based on this variable sequence, add the available variables in sequence in the variable combination (each adding /> At least one variable in the variable, preferably, the random forest model to be constructed after adding the variable has at least three variables as the input of the model), and constructs all possible variable combinations (/> , /> is the number of variable combinations, /> is the number of variables, /> That is, the number of variables that do not include the order of days and the order of half hours), so that the amount of data used to build the random forest model in all variable combinations reaches the set ratio of the total data amount.

;

;

;

...

;

;

;

;

...

;

;

;

.

The random forest model is constructed separately for the variable combinations of various available variable numbers as follows:

;

;

;

;

...

;

;

;

;

...

;

;

;

.

In the formula, represents evapotranspiration observation data, /> Represents the functional relationship between observed evapotranspiration data and imputation input variables. where, /> is as a supplement to guarantee the order in no addition to the number of days /> with half hour order /> When other variables are available, the imputation rate of all missing data can still reach 100%. Optionally, use the out-of-bag data that is not involved in the establishment of the random forest model for parameter tuning, and calculate the root mean square error of the imputation rate of the out-of-bag data. When the root mean square error is less than the set value, the parameter tuning is complete.

Test the model interpolation accuracy and obtain the interpolation accuracy of m variable combinations and their respective imputation rates/> .

Sort the imputation accuracy of m variable combinations, and the sorted variable combinations are The corresponding imputation rate is /> .

From the combination of variables corresponding to the highest precision At the beginning, use the sorted variables to combine the corresponding evapotranspiration observation data to obtain evapotranspiration data, and use each evapotranspiration data to interpolate the evapotranspiration observation data at the missing time one by one.

For the random forest model with similar interpolation accuracy, the ET data with high interpolation rate is selected as the ET observation data. If the interpolation rate is similar, the random forest model with less available variables in the variable combination is selected to obtain the ET data; During the imputation process, for each piece of interpolated data, mark the combination of variables used to imput the vacant data, and update the current imputation rate of the flux station until the imputation rate of the flux station reaches 100%. The specific process is as follows:

, /> ;

, /> ;

...

,

;

In the formula, On behalf of No. /> group imputation data, /> number of variables for each imputation, /> is the total number of site data, /> Represents the dynamic interpolation model of missing evapotranspiration observations with different precision from high to low, /> On behalf of the completion of the /> The latest imputation rate after group imputation.

If the interpolation rate reaches 100%, the interpolation ends; if the interpolation rate is less than 100%, use The remaining vacancies are imputed until the imputation rate reaches 100%.

Compared with the traditional method of using fixed variable combinations to interpolate data gaps through machine learning, this method is not limited to the type and quantity of variables corresponding to the time of data gaps, and can select different combinations of variables for different gaps, while ensuring the accuracy of interpolation. Based on this, the data imputation rate is maximized. Compared with the Marginal Distribution Sampling (MDS) method used in the FLUXNET2015 data set, the present invention adopts the random forest algorithm in machine learning, which can obtain the maximum number of variable combinations based on the available variables of different sites, based on different The variable combination trains a random forest model with high robustness and strong predictability, and obtains the accuracy of different variable combinations based on the test set. Finally, according to the modeling accuracy, sort first and then interpolate to achieve high-precision and large-scale interpolation.

The present invention determines the maximum number of variables according to different stations, so as to determine the variable combination of various variable numbers; determines the specific variables of each combination according to the data gap interpolation rate, and interpolates from high to low according to the modeling accuracy of each variable combination The data gap is enough to realize dynamic data interpolation, taking into account the interpolation rate and interpolation accuracy, so that the interpolation rate reaches 100%.

Example 2

Based on the same principle as the method shown in Embodiment 1 of the present invention, as shown in Figure 3, the embodiment of the present invention also provides a dynamic interpolation system for missing evapotranspiration observations at global flux sites, including the first acquisition Unit, database construction unit, preprocessing unit, second acquisition unit, first sorting unit, determination unit, second sorting unit, model building unit and interpolation update unit:

The first acquisition unit is used to acquire the observation data and MODIS remote sensing data of the global flux site; the observation data of the flux site includes meteorological observation data and flux observation data;

The database construction unit is used to construct a dynamic interpolation database for missing evapotranspiration observations at global flux sites based on meteorological observation data, flux observation data and MODIS remote sensing data;

The preprocessing unit is used to preprocess the data in the dynamic interpolation database for missing evapotranspiration observations, and construct input data sets site by site;

The second acquisition unit is used to obtain the available variables of global flux sites and the quantity of each available variable site by site based on the input data set; the available variables include the sequence of days in a year and half of a day in the observation data of flux sites order of hours;

The first sorting unit is used to sort the importance of each available variable site by site using the random forest method according to the available variables;

A determination unit is used to determine the variable combination including the quantity of various available variables according to the results of ranking the quantity and importance of each available variable;

The second sorting unit is used to calculate the interpolation accuracy and interpolation rate of all variable combinations, and sort each variable combination according to the interpolation accuracy from high to low;

The model building unit is used to establish the dynamic interpolation model of evapotranspiration observation loss for each variable combination site by site based on the variable combination including various available variable quantities, and obtain a set of dynamic interpolation models for evapotranspiration observation loss; evapotranspiration The missing dynamic interpolation model is used to characterize the functional relationship between evapotranspiration observation data and variable combinations;

The interpolation update unit is used to interpolate the evapotranspiration observation data at the missing time according to the meteorological observation data and MODIS remote sensing data at the time when the evapotranspiration observation data is missing, and continuously update the interpolation rate until the interpolation rate reaches 100%. The combination of variables used to imput each gap is marked.

Optionally, meteorological observation data include observation date, observation time, air temperature, wind speed, atmospheric pressure, air humidity, incident shortwave radiation and net radiation; flux observation data include evapotranspiration data; MODIS remote sensing data include normalized difference vegetation index Data with LAI data.

Optionally, preprocess the data in the dynamic imputation database for missing evapotranspiration observations, and construct input data sets site by site, including:

Obtain the observation date and observation time, calculate the order of the days of the observation data of the flux site one by one in a year, and convert the observation time to the half-hour time where the observation data of the flux site is located in the 48 half-hour time of the day The order of time is obtained; the order of days, time, and meteorological observation data are combined with the MODIS data of pixels corresponding to the flux stations extracted according to latitude and longitude, and the input data set is constructed station by station.

Optionally, based on variable combinations that contain various available variable numbers, the random forest method is used to establish a dynamic imputation model for missing evapotranspiration observations of each variable combination site by site, including:

set up Represents the evapotranspiration observation data in the input data set, and the functional relationship is /> ,but:

;

in, Represents the order of the days of the year for the observation data of the flux site, /> Represents the order of the half-hour time where the observation data of the flux site is located in the 48 half-hour time of the day, /> Represents the air temperature at which evapotranspiration does not appear missing in the input data set, /> Represents the incident shortwave radiation at the time when evapotranspiration does not appear missing in the input data set, /> Represents the net radiance at the time when evapotranspiration does not appear missing in the input data set, /> Represents the wind speed at the time when evapotranspiration does not appear missing in the input data set, /> The normalized difference vegetation index representing the moment when evapotranspiration does not appear missing in the input dataset.

Optionally, according to the meteorological observation data and the MODIS remote sensing data at the time when the evapotranspiration observation data is missing, sequentially interpolate the evapotranspiration observation data at the missing time, including:

;

In the formula, represents the interpolation result of evapotranspiration observation data, /> Represents the air temperature observation data corresponding to the missing time of evapotranspiration, /> Represents the incident shortwave radiation observation data corresponding to the missing time of evapotranspiration, /> Represents the net radiation observation data corresponding to the missing time of evapotranspiration, /> Represents the wind speed observation data corresponding to the missing time of evapotranspiration, /> Represents the normalized difference vegetation index remote sensing data corresponding to the missing time of evapotranspiration.

Optionally, among variable combinations with the same number of available variables, if the variable combinations contain different available variables, the imputation rate of the data after imputation using the random forest method is different.

Optionally, according to the meteorological observation data and the MODIS remote sensing data at the time when the evapotranspiration observation data is missing, the evapotranspiration observation data at the missing time are sequentially interpolated, and the evapotranspiration observation data at the missing time are interpolated by traversal interpolation and updating. Perform imputation.

The above are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (8)

1. The dynamic interpolation method for the global flux site evapotranspiration observation missing is characterized by comprising the following steps:

obtaining observation data and MODIS remote sensing data of a global flux site; the observation data of the flux site comprises meteorological observation data and flux observation data;

constructing a dynamic interpolation database of the evapotranspiration observation deletion of the global flux site according to the meteorological observation data, the flux observation data and the MODIS remote sensing data;

preprocessing data in the dynamic interpolation database of the evapotranspiration observation deficiency, and constructing an input data set station by station;

acquiring available variables of each flux site and the number of the available variables from site to site based on the input data set; the usable variables are used for representing the observation data of the flux site and the MODIS remote sensing data;

according to the available variables, carrying out importance ranking on each available variable site by utilizing a random forest method;

determining variable combinations comprising the number of various available variables according to the number of the available variables and the result of the importance ranking;

calculating interpolation precision and interpolation rate of all the variable combinations, and sequencing the variable combinations according to the interpolation precision from high to low;

based on the variable combinations containing various available variable quantities, establishing a dynamic interpolation model of the evapotranspiration observation deficiency of each variable combination station by utilizing a random forest method to obtain a set of the dynamic interpolation models of the evapotranspiration observation deficiency; the evapotranspiration observation missing dynamic interpolation model is used for representing the functional relation between evapotranspiration observation data and the variable combination;

and sequentially interpolating the evapotranspiration observation data at the missing moment according to the meteorological observation data and the MODIS remote sensing data at the missing moment, continuously updating the interpolation rate until the interpolation rate reaches 100%, and marking the variable combination used for interpolating each gap.

2. The dynamic interpolation method of global flux site evapotranspiration observation loss according to claim 1, wherein the meteorological observation data comprises observation date, observation time, air temperature, wind speed, atmospheric pressure, air humidity, incident short wave radiation and net radiation; the flux observation data includes the evapotranspiration observation data; the MODIS remote sensing data comprises normalized vegetation index data and leaf area index data.

3. A method of dynamic interpolation for the loss of transpiration observations at a global flux site according to claim 1 wherein preprocessing the data in the dynamic interpolation database for the loss of transpiration observations, constructing an input dataset site by site, comprises:

acquiring observation dates and observation moments, calculating the number of days sequence of the date of the observation data of the flux station in one year one by one, converting the observation moments into the sequence of the half hour moment of the observation data of the flux station in 48 half hour moments of the day, and obtaining a moment sequence; and combining the number of days sequence, the time sequence and the meteorological observation data with MODIS data of pixels corresponding to the flux sites extracted according to longitude and latitude, and constructing an input data set site by site.

4. A method of dynamic interpolation of global flux site evapotranspiration observation loss according to claim 1, wherein establishing a dynamic interpolation model of evapotranspiration observation loss for each of said variable combinations site by site using a random forest method based on said variable combinations including the various amounts of said variables available comprises:

is provided withRepresenting said evapotranspiration observation data in said input dataset, said functional relationship being +.>Then:

；

wherein ,the order of days in the year of the date representing the observations of the flux site, +.>The half hour time at which the observations representing the flux site are located is at 48 half hour times of the dayOrder of (1)>Air temperature representing the moment of absence of evaporation in said input dataset,/for>Incident short-wave radiation representing the moment of absence of evaporation in the input dataset,/>A net radiation representing the moment of absence of evaporation in said input dataset,/for>A wind speed representing the moment when no missing is present in the input dataset,and (3) representing the normalized vegetation index of the evaporation non-missing moment in the input data set.

5. The dynamic interpolation method for global flux site evapotranspiration observation loss according to claim 1, wherein sequentially interpolating the evapotranspiration observation data at a loss moment from the meteorological observation data and the MODIS remote sensing data at the loss moment of the evapotranspiration observation data comprises:

；

in the formula ,interpolation result representing said evapotranspiration observation data,/->Represents the air temperature observation data corresponding to the moment of missing of the evapotranspiration,/>incident short-wave radiation observation data corresponding to the moment of absence of evapotranspiration, < >>Net radiation observations corresponding to the moment of absence of evapotranspiration,>wind speed observation data corresponding to the moment of evapotranspiration loss, < >>And representing normalized vegetation index remote sensing data corresponding to the evapotranspiration occurrence deletion moment.

6. A method of dynamic interpolation for global flux site evapotranspiration observation loss according to claim 1, wherein the data interpolation rate after interpolation by a random forest method is different if the variable combinations included in the variable combinations are different among the variable combinations of the same number of the available variables.

7. The dynamic interpolation method for the missing of the evapotranspiration observation of the global flux site according to claim 1, wherein the evapotranspiration observation data at the missing moment is sequentially interpolated by adopting a traversing interpolation and updating mode according to the meteorological observation data and the MODIS remote sensing data at the missing moment of the evapotranspiration observation data.

8. The dynamic interpolation system for the global flux site transpiration observation missing is characterized by comprising a first acquisition unit, a database construction unit, a preprocessing unit, a second acquisition unit, a first ordering unit, a determination unit, a second ordering unit, a model establishment unit and an interpolation updating unit:

the first acquisition unit is used for acquiring the observation data of the global flux site and the MODIS remote sensing data; the observation data of the flux site comprises meteorological observation data and flux observation data;

the database construction unit is used for constructing a dynamic interpolation database of the evapotranspiration observation deletion of the global flux site according to the meteorological observation data, the flux observation data and the MODIS remote sensing data;

the preprocessing unit is used for preprocessing the data in the dynamic interpolation database of the evapotranspiration observation deficiency, and constructing an input data set station by station;

a second obtaining unit, configured to obtain, based on the input data set, available variables of each of the flux sites and the number of the available variables from site to site; the usable variables are used for representing the observation data of the flux site and the MODIS remote sensing data;

the first sorting unit is used for sorting the importance of each available variable site by utilizing a random forest method according to the available variable;

a determining unit configured to determine variable combinations including the number of the various available variables based on the number of the available variables and a result of the importance ranking;

the second sorting unit is used for calculating the interpolation precision and the interpolation rate of all the variable combinations and sorting the variable combinations from high to low according to the interpolation precision;

the model building unit is used for building the evapotranspiration observation missing dynamic interpolation models of the variable combinations station by utilizing a random forest method based on the variable combinations comprising various available variable numbers to obtain a set of the evapotranspiration observation missing dynamic interpolation models; the evapotranspiration observation missing dynamic interpolation model is used for representing the functional relation between evapotranspiration observation data and the variable combination;

and the interpolation updating unit is used for sequentially interpolating the evapotranspiration observation data at the missing moment according to the meteorological observation data and the MODIS remote sensing data at the missing moment, continuously updating the interpolation rate until the interpolation rate reaches 100%, and marking the variable combination used for interpolation of each gap.