CN118297222A - Remote sensing prediction method and device for vegetation net primary productivity - Google Patents

Abstract

The present invention provides a remote sensing prediction method and device for vegetation net primary productivity, which relates to the field of remote sensing prediction technology. The method comprises: obtaining the longitude and latitude coordinates corresponding to the aboveground vegetation net primary productivity and the underground vegetation net primary productivity data sets respectively, and extracting the global multi-year average vegetation net primary productivity, the multi-year average precipitation and the annual precipitation anomaly rate at the observation point from the global data set; constructing and training a machine learning model according to the global multi-year average vegetation net primary productivity, the multi-year average precipitation and the annual precipitation anomaly rate; according to the machine learning model, inputting the global multi-year average vegetation net primary productivity, the multi-year average precipitation and the annual precipitation anomaly rate data sets of the global range to generate the predicted aboveground vegetation net primary productivity and underground vegetation net primary productivity global data sets. The present invention solves the mismatch problem of the time dimension in the existing scheme and greatly improves the model performance.

Description

A remote sensing prediction method and device for vegetation net primary productivity

Technical Field

The invention relates to the technical field of remote sensing prediction, in particular to a remote sensing prediction method and device for vegetation net primary productivity.

Background technique

Net primary productivity (NPP) refers to the carbon fixed by plants through photosynthesis minus the carbon consumed by their own respiration, also known as net primary productivity. NPP is a basic indicator of plant dynamics and an important part of the global carbon cycle. Aboveground net primary productivity (ANPP) is the energy source for heterotrophic organisms including humans and livestock, and belowground net primary productivity (BNPP) is the key carbon input for soil carbon. Accurately estimating global ANPP and BNPP is of great significance for a deeper understanding of the global carbon balance and the comprehensive realization of sustainable development goals.

Currently, model methods are used to estimate NPP on a global scale, including empirical model methods (such as the Miami model), parameter model methods (such as the light energy utilization model), process model methods (such as the BIOME_BGC model), etc. However, these methods are difficult to achieve the above-ground and underground division of NPP, and it is still a challenge to accurately estimate ANPP and BNPP on a global scale.

In the current existing technology, a global ANPP and BNPP estimation scheme based on field measured data and machine learning methods has been proposed. However, this scheme has the following defects:

(1) The environmental factors used are time-invariant variables, such as multi-year averages or the results of a single survey. However, ANPP and BNPP in the wild are measured at different time points. The mismatch in the temporal dimension results in limited reliability of the estimated results.

Summary of the invention

The technical problem to be solved by the present invention is to provide a remote sensing prediction method and device for net primary productivity of vegetation, which solves the mismatch problem of time dimension in existing solutions and greatly improves model performance.

In order to solve the above technical problems, the technical solution of the present invention is as follows:

In a first aspect, a remote sensing prediction method for vegetation net primary productivity is provided, the method comprising:

Obtain the global multi-year average vegetation net primary productivity product generated based on the parameter model;

According to the global multi-year average vegetation net primary productivity product, environmental covariates are obtained, including global annual precipitation data and multi-year average precipitation data;

Based on the global annual precipitation data, the annual precipitation anomaly rate is calculated;

Obtain ground-truthed net primary productivity of above-ground vegetation and net primary productivity of below-ground vegetation;

Obtain the longitude and latitude coordinates corresponding to the aboveground net primary productivity and underground net primary productivity datasets, and extract the global multi-year average vegetation net primary productivity, multi-year average precipitation, and annual precipitation anomaly rate at the observation point in the global dataset;

Build and train a machine learning model based on the global multi-year average vegetation net primary productivity, multi-year average precipitation, and annual precipitation anomaly rate;

According to the machine learning model, the global multi-year average vegetation net primary productivity, multi-year average precipitation, and annual precipitation anomaly rate datasets at the global scale are input to generate the predicted global datasets of aboveground vegetation net primary productivity and belowground vegetation net primary productivity.

Furthermore, the global multi-year average vegetation net primary productivity products generated based on the parameter model are obtained, including:

Determine the data source for the global multi-year average vegetation net primary productivity product;

According to the data source, determine the relationship between vegetation growth and environmental factors, and build a parameter model;

Obtaining input data, and inputting the input data into the parameter model, so that the parameter model calculates the net primary productivity of vegetation year by year;

The net primary productivity of vegetation over many years was screened to obtain screening data;

The screening data were aligned and matched in time and space to calculate multi-year averages;

Based on the multi-year average, the multi-year average vegetation net primary productivity product is generated.

Furthermore, environmental covariates are obtained based on the global multi-year average vegetation net primary productivity product. Environmental covariates include global annual precipitation data and multi-year average precipitation data, including:

According to the known time point data, the annual precipitation values of the missing time points are estimated by linear interpolation method to obtain aligned data;

Determine the geographical unit and time unit for matching according to the alignment data;

From the aligned data, annual precipitation values corresponding to each selected geographic unit and time unit are extracted, as well as multi-year average precipitation values corresponding to each geographic unit are extracted;

The annual precipitation value and the multi-year average precipitation value are matched with the global multi-year average vegetation net primary productivity value of the corresponding geographical unit and time unit to obtain the matching results, wherein the matching results include an annual precipitation value, a multi-year average precipitation value and a global multi-year average vegetation net primary productivity value uniquely corresponding to each geographical location and time point.

Furthermore, based on the global annual precipitation data, the annual precipitation anomaly rate is calculated, including:

pass Calculate long-term average annual precipitation Where w _i is the weight of the i-th year, β is the decay factor, P _i is the annual precipitation in the i-th year, and N is the number of years used to calculate the long-term average;

According to the long-term average annual precipitation, Calculate the standard deviation of long-term annual precipitation s _P , where γ _i is the robustness weight of the i-th year;

According to the standard deviation of long-term annual precipitation, Calculate the precipitation anomaly rate Rate _i , where is the absolute deviation of the precipitation in the ith year from the weighted long-term average annual precipitation, and δ is the adjustment parameter.

Furthermore, we obtained the ground-measured net primary productivity of above-ground vegetation and net primary productivity of below-ground vegetation, including:

Identify study area and target vegetation;

According to the actual situation of the study area, estimate the feasibility and potential risks of field measurement to obtain the evaluation results; determine the field measurement plan based on the evaluation results, and determine the schedule based on the field measurement plan;

According to the field measurement plan, the above-ground vegetation and underground vegetation biomass of each sample plot are measured to obtain measurement data;

The measured data are converted into global multi-year average vegetation net primary productivity as needed.

Furthermore, the latitude and longitude coordinates corresponding to the aboveground vegetation net primary productivity and underground vegetation net primary productivity datasets are obtained, and the global multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate at the observation point are extracted from the global dataset, including:

Get the aboveground vegetation net primary productivity and belowground vegetation net primary productivity datasets and their latitude and longitude coordinates

Obtain observational datasets of net primary productivity of above-ground vegetation and net primary productivity of below-ground vegetation;

Formatting the observation data set to obtain formatted data;

Extract and organize the latitude and longitude coordinates of each observation point in the formatted data;

The global dataset is obtained and spatially matched with the global dataset according to the latitude and longitude coordinates of the observation point. For each observation point, the multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate values corresponding to the observation point are extracted.

Furthermore, after inputting the global multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate datasets at the global scale according to the machine learning model to generate the predicted global datasets of aboveground vegetation net primary productivity and belowground vegetation net primary productivity, it also includes:

The predicted global datasets of aboveground net primary productivity and belowground net primary productivity were compared with the ground-measured datasets of aboveground net primary productivity and belowground net primary productivity to obtain comparative results.

Based on the comparison results, the prediction performance of the machine learning model is evaluated to obtain an evaluation result.

Second, a remote sensing prediction device for vegetation net primary productivity, including:

The acquisition module is used to obtain the global multi-year average vegetation net primary productivity product generated based on the parameter model; obtain environmental covariates based on the global multi-year average vegetation net primary productivity product, which includes global annual precipitation data and multi-year average precipitation data; calculate the annual precipitation anomaly rate based on the global annual precipitation data; obtain the ground-measured net primary productivity of vegetation above ground and net primary productivity of vegetation below ground data sets;

The processing module is used to obtain the longitude and latitude coordinates corresponding to the aboveground vegetation net primary productivity and underground vegetation net primary productivity datasets, and extract the global multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate at the observation point in the global dataset; build and train a machine learning model based on the global multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate; according to the machine learning model, input the global multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate datasets on a global scale to generate predicted global datasets of aboveground vegetation net primary productivity and underground vegetation net primary productivity.

According to a third aspect, a computing device includes:

one or more processors;

The storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the above method.

In a fourth aspect, a computer-readable storage medium stores a program, and the program implements the above method when executed by a processor.

The above solution of the present invention includes at least the following beneficial effects:

By using annual precipitation data and calculating the annual precipitation anomaly rate, we can better capture the temporal changes in vegetation net primary productivity, solve the time dimension mismatch problem caused by the use of environmental factors that do not change over time in existing technologies, and thus improve the reliability of the estimation results.

Through a machine learning model, the present invention uses environmental covariates such as the global multi-year average vegetation net primary productivity, multi-year average precipitation, and annual precipitation anomaly rate to successfully realize the division of aboveground vegetation net primary productivity (ANPP) and belowground vegetation net primary productivity (BNPP) on a global scale.

By combining the global multi-year average vegetation net primary productivity product generated by the parameter model and ground-based measured data, the present invention can more accurately predict ANPP and BNPP on a global scale. In addition, the reliability and stability of the prediction are further enhanced through the machine learning model.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic flow chart of a remote sensing prediction method for net primary productivity of vegetation provided in an embodiment of the present invention.

FIG. 2 is a schematic diagram of a remote sensing prediction device for net primary productivity of vegetation provided by an embodiment of the present invention.

Detailed ways

The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

As shown in FIG1 , an embodiment of the present invention provides a remote sensing prediction method for net primary productivity of vegetation, the method comprising:

Step 11, obtaining the global multi-year average vegetation net primary productivity product generated based on the parameter model;

Step 12, obtaining environmental covariates based on the global multi-year average vegetation net primary productivity product, the environmental covariates include global annual precipitation data and multi-year average precipitation data;

Step 13, calculating the annual precipitation anomaly rate based on global annual precipitation data;

Step 14, obtaining the ground-measured net primary productivity of vegetation above ground and net primary productivity of vegetation below ground;

Step 15, obtaining the latitude and longitude coordinates corresponding to the aboveground vegetation net primary productivity and underground vegetation net primary productivity datasets, and extracting the global multi-year average vegetation net primary productivity, multi-year average precipitation, and annual precipitation anomaly rate at the observation point in the global dataset;

Step 16, constructing and training a machine learning model based on the global multi-year average vegetation net primary productivity, multi-year average precipitation, and annual precipitation anomaly rate;

Step 17, according to the machine learning model, input the global multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate datasets on a global scale to generate a global dataset of predicted aboveground vegetation net primary productivity and belowground vegetation net primary productivity.

In an embodiment of the present invention, the change of vegetation net primary productivity over time can be better captured by annual precipitation data and calculating the annual precipitation anomaly rate, solving the time dimension mismatch problem caused by the use of environmental factors that do not change over time in the prior art, thereby improving the reliability of the estimation results. The present invention uses a machine learning model and environmental covariates such as the global multi-year average vegetation net primary productivity, multi-year average precipitation, and annual precipitation anomaly rate to successfully realize the division of aboveground vegetation net primary productivity (ANPP) and underground vegetation net primary productivity (BNPP) on a global scale. By combining the global multi-year average vegetation net primary productivity product generated by the parameter model and the ground measured data, the present invention can more accurately predict ANPP and BNPP on a global scale. In addition, the reliability and stability of the prediction are further enhanced by the machine learning model.

In a preferred embodiment of the present invention, the above step 11 may include:

Step 111, determining the data source of the global multi-year average vegetation net primary productivity product;

Step 112, determining the relationship between vegetation growth and environmental factors according to the data source, and constructing a parameter model;

Step 113, obtaining input data, and inputting the input data into the parameter model, so that the parameter model calculates the net primary productivity of vegetation year by year;

Step 114, screening the net primary productivity of vegetation for multiple years to obtain screening data;

Step 115, aligning and temporally matching the screened data to calculate multi-year averages;

Step 116, generating a multi-year average vegetation net primary productivity product based on the multi-year average.

In an embodiment of the present invention, the data source of step 111 includes satellite remote sensing images, data from ground observation sites, meteorological data sets, and other relevant environmental data. The data source helps to ensure the accuracy and consistency of subsequent analysis. The use of the data source can improve the accuracy of NPP estimation, thereby providing more valuable information for global carbon cycle and ecosystem research. Step 112 will analyze the statistical relationship between vegetation growth (represented by NPP) and environmental factors (such as temperature, precipitation, radiation, etc.) based on the selected data source, and build a parameter model based on these relationships. The model can predict vegetation NPP under different environmental conditions. Building a parameter model helps to understand the complex relationship between vegetation growth and environmental factors, and improves the accuracy and generalization ability of NPP prediction. Step 113 collects the input data required for the model, which includes environmental factor data (such as temperature, precipitation, etc.) year by year, and then inputs these data into the previously constructed parameter model to calculate the vegetation NPP year by year. Through year-by-year calculations, vegetation NPP data on a time series can be obtained. Step 114 will filter the calculated multi-year vegetation NPP data. The purpose of the screening is to remove outliers, reduce data noise, or select data points that meet specific conditions. Data screening helps to improve data quality and the reliability of analysis. By removing outliers and noise, their impact on subsequent analysis can be reduced, making the results more accurate and robust. Step 115 will align and match the screened data in time and space to ensure that data from different years and locations are compared and analyzed at the same temporal and spatial scale. By eliminating errors caused by inconsistent data temporal and spatial scales, a more accurate multi-year average vegetation NPP can be obtained, thereby providing more valuable information for global-scale carbon cycle and ecosystem research. The global multi-year average vegetation NPP product generated in step 116 is of great significance for assessing the health of ecosystems, monitoring the impact of climate change, and formulating sustainable environmental management strategies.

In another preferred embodiment of the present invention, the above step 112 may include:

Step 1121, pass Construct the relationship between vegetation growth (NPP) and environmental factors, where β ₀ is the intercept term, b _3j is the coefficient of the j-th basis function corresponding to the third environmental factor R, b _1j is the coefficient of the j-th basis function corresponding to the first environmental factor T, b _2j is the coefficient of the j-th basis function corresponding to the second environmental factor P, B _1j (T) is the j-th basis function corresponding to the first environmental factor T, B _2j (P) is the j-th basis function corresponding to the second environmental factor P, B _3j (R) is the j-th basis function corresponding to the third environmental factor R, k ₁ , k ₂ and k ₃ are the numbers of basis functions used to fit the first environmental factor, the second environmental factor and the third environmental factor, respectively, ∈ is the error term, environmental factor T is temperature, environmental factor P is precipitation, and environmental factor R is solar radiation;

Step 1122, for each prediction variable _Xi (temperature, precipitation and solar radiation), a model matrix _Bi is constructed. Suppose there are n observations and k _i basis functions selected for the prediction variable _Xi , then the model matrix _Bi is an n×k _i matrix, each element _Bijl represents the value of the j j basis function corresponding to the l th observation, where the model matrix Bi _is :

Among them, _xil is the _Xi predictor variable of the l-th observation;

Step 1123, pass The coefficients b _ij and the intercept term β ₀ are estimated, where y _l is the response variable for the lth observation, p is the number of predictor variables (temperature, precipitation, and solar radiation), λ _i is a penalty parameter that controls the smoothness of each predictor variable, and B″ _ij ( _xi ) is the second-order derivative of the basis function.

Step 1124, in order to minimize the objective function G(β ₀ , b), by Iteratively update the parameter estimates to obtain the optimal solution, _αm is the step size of the mth step, H ^-1 is the inverse of the Hessian matrix, is the gradient of the objective function. These quantities are recalculated in each iteration until the convergence condition is met. m represents the current number of iterations, and m+1 represents the number of next iterations. represents the estimated value of the intercept term _β0 at the mth iteration, represents the updated estimate of the intercept term _β0 after the m+1th iteration, b ^(m) is a vector containing the estimated values of all coefficients _bij at the mth iteration, b ^(m+1) is the updated coefficient vector containing the new estimated values of all coefficients _bij after the m+1th iteration, and the gradient of the objective function It is a vector of partial derivatives of the objective function with respect to each parameter, and the calculation formula is:

in, is the partial derivative of G(β ₀ ,b) with respect to the intercept term β ₀ , is the partial derivative vector with respect to the coefficient vector b, which contains the partial derivative of the objective function with respect to each element in the coefficient vector b; the Hessian matrix H is the second-order partial derivative matrix of the objective function, and the calculation formula is:

in, is the second-order partial derivative of the objective function with respect to the intercept term β ₀ , is the mixed second-order partial derivative vector of the objective function with respect to the intercept term _β0 and the coefficient vector b, is the transposed vector of the mixed second-order partial derivatives with respect to the coefficient vector b and the intercept term β ₀ , is the second-order partial derivative matrix with respect to the coefficient vector b, represents the partial derivative and T represents the transpose of the matrix.

In the embodiment of the present invention, step 1121 establishes a mathematical relationship between vegetation growth (NPP) and three environmental factors, namely, temperature (T), precipitation (P) and solar radiation (R), through a formula. Wherein, NPP is expressed as the sum of an intercept term, a linear combination of basis functions of the three environmental factors, and an error term, and each environmental factor is fitted by a basis function (B _ij ); the present invention provides a flexible model framework that can capture the nonlinear relationship that may exist between each environmental factor and NPP. Through the combination of basis functions, it can adapt to various complex data patterns and improve the expression ability of the model.

In the embodiment of the present invention, in step 1121, the calculation formula of B _1j (T) is:

B _1j (T) = T ^j ;

Where j is the order of the polynomial; B _2j (P) is calculated as:

B _2j (P) = max(0, P - θ _j );

Among them, θ _j is the segmentation point, B _2j (P) = max(0, P-θ _j ) is a piecewise function, also known as the ReLU (Rectified Linear Unit) function, P represents an input value, which can be any real number. In precipitation, P represents the precipitation at a certain point in time. The behavior of the function can be divided into two parts:

When P＜θ _j , the function value is 0, because P-θ _j is a negative number, and max(0, negative number)＝0;

When P≥θ _j , the function value is P-θ _j , because at this time P-θ _j is a non-negative number, and max(0, non-negative number)=non-negative number.

Solar radiation has significant diurnal and seasonal variations, similar to temperature, and these variations can be represented using polynomial basis functions. For example, a combination of sine functions can be used to represent the diurnal and seasonal variations of solar radiation:

Where t is the day of the year (from 0 to 364) and j is the frequency factor that controls the periodicity of the basis function to capture the annual cycle of solar radiation.

Step 1122 constructs a model matrix for each predictor variable. For the three predictor variables of temperature, precipitation and solar radiation, model matrices (B _i ) are constructed respectively. These matrices contain the value of each observation corresponding to the selected basis function. The relationship between the predictor variables and the basis function is matrixed to facilitate mathematical operations and parameter estimation. Constructing a model matrix for each predictor variable separately helps to maintain the modularity and scalability of the model.

In the embodiment of the present invention, step 1123 estimates the coefficient (b _ij ) and the intercept term β ₀ of the model by minimizing the objective function G(β ₀ , b). The objective function consists of two parts, a data fitting term and a smoothing penalty term. The data fitting term ensures that the model can fit the observed data well, while the smoothing penalty term is used to control the complexity of the model to prevent overfitting, balance the fitting ability and complexity of the model, and help improve the generalization performance of the model. By introducing the penalty term, the smoothness of the model can be directly controlled, thereby increasing the interpretability of the model.

In the embodiment of the present invention, in step 1123, the calculation process of the second-order derivative B″ _ij ( _xi ) of the basis function includes:

Suppose there is a cubic B-spline basis function _Bij (x), which is a cubic polynomial on a specific interval [ _tm , tm ₊₁ ], where _tm and _tm+1 are adjacent nodes. This cubic polynomial can be expressed as:

_Bij (x)= _am ( _xtm ) ³ + _bm ( _xtm ) ² + _cm ( _xtm )+ _dm ;

On this interval, where a _m , b _m , c _m , d _m are the coefficients of the polynomial, determined by the construction of the B-spline and the location of the nodes; calculate the second-order derivative of the above cubic polynomial, and take the derivative twice for B _ij (x), we get:

The above calculation formula is the second-order derivative on the interval [t _m , t _m+1 ]. For the entire basis function, this process needs to be repeated on each interval because each interval has its own polynomial representation and coefficients.

In an embodiment of the present invention, the cubic B-spline basis function allows the model to capture the nonlinear relationship between the predictor and the response variable. By using different polynomial coefficients on each interval, the basis function can adapt to the local changes of the data, thereby providing a more accurate fit. The calculation of the second-order derivative is the key to controlling the smoothness of the model. In the penalty term of GAM, the square sum of the second-order derivative is used to avoid overfitting, ensuring that the prediction curve of the model is smooth, rather than over-fluctuating, and this smoothness contributes to the generalization ability of the model on unknown data. The cubic B-spline basis function and its derivatives are continuous at the nodes, which ensures that the prediction results of the model are continuous throughout the domain of definition. This continuity is necessary for many practical applications because it ensures the stability and reliability of the prediction results. By checking the shape of the basis function and the value of the second-order derivative, an intuitive understanding of how the predictor affects the response variable can be obtained. For example, the sign and size of the second-order derivative can provide information about the rate of change of the response variable. The cubic B-spline basis function can adapt to various types of data distributions and patterns because they are implemented by fitting different polynomials on each interval.

Step 1124 iteratively updates the parameter estimates to obtain the optimal solution, uses an iterative optimization algorithm to minimize the objective function, and iteratively updates the parameter estimates until the convergence condition is met. Each iteration calculates the gradient of the objective function based on the current parameter values. and the inverse H ^-1 of the Hessian matrix, and then use this information to update the parameters. The iterative optimization algorithm can efficiently find the minimum value of the objective function and work effectively even when the parameter space is very complex. By gradually approaching the optimal solution, high-precision parameter estimates can be obtained with limited computing resources.

In a preferred embodiment of the present invention, the above step 12 may include:

Step 121, based on the known time point data, estimate the annual precipitation value of the missing time point by linear interpolation method to obtain aligned data;

Step 122, determining the geographical unit and time unit for matching according to the alignment data;

Step 123, extracting the annual precipitation value corresponding to each selected geographic unit and time unit, and extracting the multi-year average precipitation value corresponding to each geographic unit from the aligned data;

Step 124, matching the annual precipitation value, the multi-year average precipitation value and the global multi-year average vegetation net primary productivity value of the corresponding geographical unit and time unit to obtain a matching result, wherein the matching result includes an annual precipitation value, a multi-year average precipitation value and a global multi-year average vegetation net primary productivity value uniquely corresponding to each geographical location and time point.

In the embodiment of the present invention, step 121 can estimate the annual precipitation value of the missing time point through the linear interpolation method, thereby improving the integrity and continuity of the data set, and can reduce the deviation caused by missing data to a certain extent. Step 122, determining the geographical unit and time unit helps to clarify the spatial and temporal scale of the analysis, making the research more targeted and operational, and the clear division of geographical and temporal units helps to more accurately match the data from different data sources. Step 123, extracting the annual precipitation value and the multi-year average precipitation value of specific geographical and time units, facilitating comparative analysis of different regions and different time periods, these data provide the basis for subsequent matching with the global multi-year average vegetation net primary productivity value. Step 124, the matching results include geographical location, time point, annual precipitation, multi-year average precipitation and global multi-year average vegetation net primary productivity value, providing rich information for comprehensive analysis; by ensuring that each geographical location and time point uniquely corresponds to a set of data, the accuracy and reliability of the analysis are improved; the matching data structure supports the analysis of vegetation net primary productivity from different dimensions (such as space, time, precipitation, etc.).

In another preferred embodiment of the present invention, the above step 123 may include:

Step 1231, extracting the annual precipitation value by _Pij = α×Lat _i + β×Lon _i + γ×A _k + δ×Season(T _j )+∈ _ijk , wherein _Pij represents the predicted annual precipitation value under the geographical unit (latitude and longitude), time unit T _j (season) and altitude _Ak , Lat _i represents the latitude value of the geographical unit, Lon _i represents the longitude value of the geographical unit, _Ak represents the altitude value of the geographical unit, Season(T _j ) represents mapping the time unit T _j (such as month) to the corresponding seasonal category; α, β, γ and δ are the regression coefficients of latitude, longitude, altitude and season respectively, and ∈ _ijk represents the random error term;

Step 1232, pass Calculate the average precipitation value over many years, where: represents the multi-year average precipitation value of the geographical unit, _Pij represents the annual precipitation value under the geographical unit and time unit, _wj represents the weight of the jth time unit, _Qij represents the data quality control factor, N represents the total number of time units related to the geographical unit, and _Qij is a value between 0 and 1, where 0 indicates that the data is completely unreliable and 1 indicates that the data is completely reliable.

In the embodiment of the present invention, step 1231 takes into account multiple influencing factors such as geographical location (latitude and longitude), altitude and season, and can more comprehensively reflect the changes in precipitation under different geographical and time conditions; by estimating the regression coefficients of various influencing factors through regression analysis, a more accurate annual precipitation prediction model can be established, thereby improving the accuracy and reliability of the prediction; the random error term in the model can take into account unobserved influencing factors, so that the model has a certain adaptability to precipitation prediction in different geographical regions and time scales. Step 1232 introduces the weight of the time unit, which can reflect the importance difference of precipitation data in different time periods, such as data from more recent years may be more valuable for reference; by excluding outliers or other data that do not meet the quality standards through data quality control factors, it can ensure that reliable data is used when calculating the multi-year average precipitation, thereby improving the accuracy of the results, and the multi-year average precipitation value can reflect the precipitation characteristics of the geographical unit on a long time scale.

In the embodiment of the present invention, the specific mapping process of Season(T _j ) is:

Using the standard four-season division, spring, summer, autumn, and winter are divided into corresponding months. Then, the mapping process of Season(T _j ) can be expressed by the following steps:

Determine the specific value of the time unit T _j . This is a date, month, or other time identifier. In this example, assume that T _j is a month value ranging from 1 to 12.

According to the value of T _j , we can judge which season it belongs to by comparing T _j with the boundary value of seasonal division.

Map T _j to the corresponding seasonal category. This is done by assigning a season label (such as "spring", "summer", "autumn", "winter") or a season code (such as using the number 3 for spring, 6 for summer, 9 for autumn, and 12 for winter).

In a preferred embodiment of the present invention, the above step 13 may include:

Step 131, pass Calculate long-term average annual precipitation Where w _i is the weight of the i-th year, β is the decay factor, P _i is the annual precipitation in the i-th year, and N is the number of years used to calculate the long-term average;

Step 132, based on the long-term average annual precipitation, Calculate the standard deviation of long-term annual precipitation s _P , where γ _i is the robustness weight of the i-th year;

Step 133, based on the standard deviation of long-term annual precipitation, Calculate the precipitation anomaly rate Rate _i , where is the absolute deviation of the precipitation in the ith year from the weighted long-term average annual precipitation, and δ is the adjustment parameter.

In the embodiment of the present invention, step 131, by using the attenuation factor β, the recent annual precipitation has a greater weight when calculating the long-term average, which is more in line with the fact that recent data is more important in actual situations; by adjusting the weight w _i of each year, the importance of data in different years can be flexibly considered, for example, for years with higher data quality or more critical, a greater weight can be given; the long-term average annual precipitation can be used as an important benchmark value for evaluating the long-term climate conditions of a region, which helps to understand the climate characteristics of the region. Step 132, by calculating the standard deviation of the long-term annual precipitation, the interannual variation of precipitation in a region can be quantified, and by setting the robustness weight γ _i , it helps to reduce the impact of outliers on the standard deviation calculation and improve the robustness of the results. Step 133, the precipitation anomaly rate comprehensively considers the absolute deviation between the annual precipitation and the weighted long-term average annual precipitation and the proportion of the deviation to the weighted long-term average annual precipitation, providing a more comprehensive assessment of precipitation anomaly. By adjusting the parameter δ, the sensitivity to the deviation can be adjusted according to actual needs when calculating the precipitation anomaly rate, which increases the flexibility of the method. The precipitation anomaly rate can be used as an early warning indicator for extreme climate events such as droughts and floods, which helps to take timely response measures to reduce losses.

In a preferred embodiment of the present invention, the above step 14 may include:

Step 141, determining the study area and target vegetation;

Step 142, based on the actual situation of the study area, estimate the feasibility and potential risks of the field measurement to obtain an evaluation result; determine the field measurement plan based on the evaluation result, and determine a schedule based on the field measurement plan;

Step 143, measuring the aboveground vegetation and underground vegetation biomass of each sample plot according to the field measurement plan to obtain measurement data;

Step 144, converting the measured data into the global multi-year average vegetation net primary productivity as required.

In the embodiment of the present invention, step 141, by determining the research area, it is possible to focus on a specific geographical space, making the research more targeted and operational. Determining the target vegetation helps to select representative vegetation types, thereby improving the reliability and universality of the research results. Step 142, by estimating the feasibility and potential risks of field measurements, it is possible to identify and respond to possible problems in advance, thereby reducing the risks and uncertainties in the field measurement process. Determining the field measurement plan and schedule helps to reasonably allocate human, material and time resources to ensure the smooth progress of the measurement work. Step 143, by simultaneously measuring the above-ground and underground vegetation biomass, it is possible to more comprehensively evaluate the growth status of vegetation and ecosystem functions. Measuring each sample plot helps to obtain more accurate data, thereby improving the accuracy and reliability of subsequent analysis. Step 144, converting the measured data into the global multi-year average vegetation net primary productivity helps to unify data from different regions and different times under the same metric standard, facilitating global comparison and analysis. The converted data can be integrated with data from other global studies, thereby expanding its scope of application and providing strong support for research in the fields of global climate change and ecosystem services.

In another preferred embodiment of the present invention, the above step 141 may include:

According to the research purpose and actual conditions, select a suitable geographical area, such as a river basin or ecological zone; study the natural environmental conditions of the area, such as climate, topography, soil, etc., to ensure representativeness and diversity within the area; determine the boundaries of the study area, you can use natural geographical boundaries or customize boundaries according to research needs; mark the location and scope of the study area on the map, and calculate the area of the study area.

According to the vegetation type and research purpose of the study area, determine the vegetation type that needs to be measured and evaluated, such as forest, grassland, farmland, etc.; investigate and classify the vegetation types in the study area, and refer to the existing vegetation classification system or establish a classification system according to research needs; on the basis of vegetation type classification, further subdivide the target vegetation, such as coniferous forest, broad-leaved forest, shrub, etc., to ensure that the target vegetation is representative and typical; outline and mark the distribution area of the target vegetation, and calculate the area and proportion of each target vegetation.

The spatial distribution information of the study area and the target vegetation is superimposed and integrated to generate a distribution map of the target vegetation in the study area. The distribution map is analyzed and evaluated to ensure that the distribution of the target vegetation in the study area is representative and balanced. Based on the distribution of the study area and the target vegetation, the location and number of sample points for field measurements are preliminarily determined.

In another preferred embodiment of the present invention, the above step 142 may include:

Analyze the natural conditions of the study area, such as topography, transportation, and climate, and evaluate the difficulty and accessibility of field measurements. Based on the above factors, an assessment result of the feasibility of field measurements is obtained, such as high, medium, or low levels.

Identify the natural risks that may be encountered during the field measurement, such as extreme weather, wild animals, geological disasters, etc., and evaluate the possibility of their occurrence and the extent of their impact; identify the technical risks that may be encountered during the field measurement, such as instrument failure, improper measurement methods, data quality issues, etc., and evaluate their possibility of occurrence and the extent of their impact; based on the above factors, draw an assessment result of the potential risks of the field measurement, such as high risk, medium risk, and low risk levels.

Based on the feasibility and risk assessment results, determine the overall plan for field measurement, including the measurement area, measurement object, measurement method, sample point setting, personnel division of labor, etc.; based on the measurement plan, estimate the workload and time required for field measurement, consider factors such as traffic and weather, and preliminarily formulate the start and end time and milestone nodes of the measurement.

Match the measurement schedule with the vegetation growth cycle, phenological characteristics, etc., select the appropriate measurement time window, and avoid conducting measurements during periods when vegetation growth is inactive or difficult to identify.

Dynamically adjust and optimize the schedule, and promptly revise and improve the measurement plan based on the actual measurement progress and situation to ensure that the measurement task is completed on time.

In another preferred embodiment of the present invention, the above step 143 may include:

Step 1431, classifying the ground vegetation by type to obtain a classification result;

Step 1432, measuring the coverage area and average height of each vegetation, selecting representative soil samples in each sample plot for root sampling, cleaning and processing the root samples, separating the roots and measuring their total weight;

Step 1433, estimating the root biomass per unit volume according to the soil bulk density and the sampling volume;

Step 1434, for the aboveground vegetation, by Calculate the biomass of each vegetation pass The total aboveground biomass is calculated as follows: represents the coverage area of the i-th type of aboveground vegetation, represents the density of the i-th aboveground vegetation (plants/unit area), represents the height correction coefficient of the i-th aboveground vegetation, represents the unit biomass of the i-th aboveground vegetation dry matter (weight after removing water), represents the growth correction coefficient of the i-th aboveground vegetation, represents the moisture content correction coefficient of the i-th aboveground vegetation, G represents the set of vegetation functional groups, _wgi represents the weight of the i-th vegetation in the g-th functional group, _TFg represents the terrain correction factor of the g-th functional group, and _Ktopo represents the terrain correction coefficient of the entire sample plot;

Step 1435, for underground vegetation, estimate the root biomass of the entire sample plot based on the sampling results; The total aboveground and underground biomass was calculated, where z represents the stratified set of soil depths and Bi _be,z represents the underground biomass at the zth soil depth.

In an embodiment of the present invention, by classifying the aboveground vegetation by type, measuring the coverage area and average height of each type of vegetation, and selecting representative soil samples for root sampling, the biomass of each type of vegetation can be estimated more accurately. At the same time, sampling and measuring the root system of underground vegetation also provide direct underground biomass data, further improving the accuracy of the estimation. In the process of calculating biomass, this step introduces multiple correction factors, such as height correction factor, growth correction factor and moisture content correction factor, to consider the effects of different heights, growth stages and moisture content on biomass. In addition, the terrain correction factor and the terrain correction factor of the entire sample plot are also considered to adjust the biomass estimation value to make it more in line with the actual situation. This step not only estimates the biomass of aboveground vegetation, but also estimates the root biomass of underground vegetation by sampling and measurement, and adds the two to obtain the total biomass, so that the biomass distribution and total amount of the ecosystem can be more fully understood. The whole step follows a scientific method and a systematic process, from vegetation classification, measurement to calculation and analysis, each step has a clear purpose and operating specifications. This helps to ensure the reliability and repeatability of biomass estimation. The biomass estimation results obtained through this step can provide basic data for studies on energy flow, material cycle and carbon storage in ecosystems, and help to gain a deeper understanding of the structure and function of ecosystems.

In another preferred embodiment of the present invention, the above step 144 may include:

Step 1441, selecting representative sample plots in the study area to perform simultaneous measurements of biomass and net primary productivity; measuring aboveground and belowground biomass and measuring net primary productivity;

Step 1442, perform quality control on the measured data, remove outliers and mismeasured values; calculate the aboveground and underground biomass and net primary productivity of each sample plot; classify the biomass and net primary productivity data according to factors such as vegetation type and growth stage.

Step 1443, using biomass as an independent variable and net primary productivity as a dependent variable, a linear regression model is established, and regression models for the aboveground and underground parts are established separately, such as:

_NPPs = a × AGB + b;

NPP _r = c × BGB + d;

Wherein, NPP _s and NPP _r are aboveground and belowground net primary productivity, AGB and BGB are aboveground and belowground biomass, respectively, and a, b, c, and d are regression coefficients.

Step 1444, for the aboveground part, the conversion factor is the slope coefficient a of the regression model, which represents the net primary productivity corresponding to a unit of aboveground biomass; for the underground part, the conversion factor is the slope coefficient c of the regression model, which represents the net primary productivity corresponding to a unit of underground biomass.

Step 1445, using the established conversion factors, convert the aboveground and underground biomass of each plot into the corresponding net primary productivity; calculate the average value of the net primary productivity of all plots in the study area as the representative value of the net primary productivity of the area; multiply the average value by the area of the study area to obtain the total net primary productivity of the study area.

Step 1446, obtaining a global vegetation distribution map corresponding to the selected vegetation classification from the global land cover database; determining other areas of the same vegetation category as the study area based on the global vegetation distribution map; assuming that other areas of the same vegetation category as the study area have similar net primary productivity levels; assigning the net primary productivity representative value of the study area to all areas of the same vegetation category worldwide; for each vegetation category, multiplying its area in the world to obtain the global total net primary productivity of the category;

Step 1447, add up the global total net primary productivity of all vegetation categories to obtain the global total net primary productivity; collect global vegetation distribution data within a certain time range (such as 2000-2010), repeat the operation, and calculate the global total net primary productivity for each year; take the average of the global total net primary productivity over many years to obtain the global multi-year average net primary productivity; divide the multi-year average by the global total land area to obtain the global multi-year average net primary productivity density.

In the embodiment of the present invention, by selecting representative sample plots to measure biomass and net primary productivity simultaneously, the accuracy and representativeness of the data are ensured. The quality control of the measured data further improves the reliability of the data. The relationship between the above-ground and underground biomass and net primary productivity is established using a linear regression model; by introducing a conversion factor (the slope coefficient of the regression model), the biomass data can be easily converted into net primary productivity, and this method has flexibility and wide applicability. The representative value of the net primary productivity of the study area is extended to all areas of the same vegetation category worldwide, realizing the estimation from the local to the global scale, and providing an important basis for the global scale ecosystem assessment. By collecting global vegetation distribution data within a certain time range and repeating the above operation, the global total net primary productivity of each year can be calculated, and then the temporal variation trend of the net primary productivity can be analyzed.

In a preferred embodiment of the present invention, the above step 15 may include:

Step 151, obtaining the net primary productivity of aboveground vegetation and the net primary productivity of underground vegetation data sets and their latitude and longitude coordinates;

Step 152, obtaining an observation data set of net primary productivity of aboveground vegetation and net primary productivity of underground vegetation;

Step 153, formatting the observation data set to obtain formatted data;

Step 154, extracting and collating the latitude and longitude coordinates of the observation point in each formatted data;

Step 155, obtain the global data set, and perform spatial matching with the global data set based on the latitude and longitude coordinates of the observation point. For each observation point, extract the multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate values corresponding to the observation point.

In an embodiment of the present invention, step 151, obtaining the above-ground and underground vegetation net primary productivity data set provides the necessary data basis for subsequent analysis, and at the same time obtains the longitude and latitude coordinates to ensure the accuracy of the data's geographic spatial positioning, which is convenient for subsequent spatial matching and analysis. Step 152, the observation data set is used to verify the model output or remote sensing data, and provides an important reference for the ground truth. Combining model output, remote sensing data and ground observation data can improve the comprehensiveness and accuracy of the analysis. Step 153, formatting ensures the consistency and comparability of the data, which is convenient for subsequent data integration and analysis. The unified data format can reduce errors and redundancy in the data processing process and improve the efficiency of data processing. Step 154, extracting and organizing the longitude and latitude coordinates ensures the precise geographic positioning of each observation point, which provides an accurate basis for subsequent spatial matching. Organizing and verifying the longitude and latitude coordinates helps to improve data quality and reduce data deviations caused by geographic positioning errors. Step 155, through spatial matching, the data of the observation point can be associated with the global background data to provide more comprehensive information for analysis, and multiple indicators such as the multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate can be extracted, which is helpful for comprehensive analysis of the relationship between vegetation growth and climate factors from multiple dimensions.

The spatial matching in step 155 is to match the latitude and longitude coordinates of the observation point with the grid cells in the global data set to extract the multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate values at the location of the observation point. The following is a specific method for performing spatial matching:

Step 1551, determine the spatial resolution of the global dataset, the global dataset is provided in the form of a grid, such as 0.5°×0.5°, 1km×1km, etc.

Step 1552, determine the grid size and boundary coordinates of the data set, such as the longitude range of the data set is -180° to 180°, the latitude range is -90° to 90°, and the grid size is 0.5°×0.5°.

Step 1553, convert the observation point coordinates into the coordinate system of the global dataset. The latitude and longitude coordinates of the observation point use the geographic coordinate system (such as WGS84), while the global dataset may use a different projection coordinate system (such as equal area projection); use GIS software or programming tools to convert the observation point coordinates from the geographic coordinate system to the same projection coordinate system as the global dataset.

Step 1554, determine the grid cell where the observation point is located, and calculate its row number and column number in the global dataset grid according to the projection coordinates of the observation point, where:

Row number = (Y _max -Y _obs )/size _y ;

Column number = (X _obs - X _min )/size _x ;

Among them, Y _max and X _min are the northernmost and westernmost coordinates of the dataset, Y _obs and X _obs are the projection coordinates of the observation point, and size _y and size _x are the latitudinal and longitudinal sizes of the grid.

Step 1555, after rounding, obtain the row and column number of the grid cell where the observation point is located, and extract the corresponding multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate values from the global data set according to the row and column number of the grid cell where the observation point is located; if the observation point is located on the grid boundary, the proximity principle can be selected to extract the data value of the grid cell closest to the observation point.

Step 1556, processing missing data or abnormal values. If the data value of the grid cell where the observation point is located is missing or abnormal, it can be estimated by interpolation or averaging of neighboring cells;

Step 1557, combine the longitude and latitude coordinates of each observation point with the extracted multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate values into a new data table, and save the matching results.

In a preferred embodiment of the present invention, after the above step 17, the following steps may also be included:

Step 18, comparing the predicted global datasets of aboveground vegetation net primary productivity and belowground vegetation net primary productivity with the datasets of aboveground vegetation net primary productivity and belowground vegetation net primary productivity measured on the ground, respectively, to obtain comparison results;

Step 19: evaluate the prediction performance of the machine learning model based on the comparison result to obtain an evaluation result.

In the embodiment of the present invention, step 18 can directly verify the prediction accuracy of the machine learning model by comparing the predicted data with the ground measured data, thereby ensuring the reliability of the prediction results. Step 19 can quantitatively evaluate the prediction performance of the machine learning model by comparing the results. The evaluation results can provide guidance for further optimization of the model and provide stronger support for decision-making such as ecosystem management and climate change adaptation.

As shown in FIG2 , an embodiment of the present invention further provides a remote sensing prediction device 20 for net primary productivity of vegetation, comprising:

The acquisition module 21 is used to obtain the global multi-year average vegetation net primary productivity product generated based on the parameter model; obtain environmental covariates based on the global multi-year average vegetation net primary productivity product, and the environmental covariates include global annual precipitation data and multi-year average precipitation data; calculate the annual precipitation anomaly rate based on the global annual precipitation data; obtain the ground-measured net primary productivity of vegetation above ground and the net primary productivity of vegetation below ground data sets;

Processing module 22 is used to obtain the longitude and latitude coordinates corresponding to the net primary productivity of aboveground vegetation and the net primary productivity of underground vegetation datasets, and extract the global multi-year average net primary productivity of vegetation, the multi-year average precipitation and the annual precipitation anomaly rate at the observation point in the global dataset; construct and train a machine learning model based on the global multi-year average net primary productivity of vegetation, the multi-year average precipitation and the annual precipitation anomaly rate; according to the machine learning model, input the global multi-year average net primary productivity of vegetation, the multi-year average precipitation and the annual precipitation anomaly rate datasets on a global scale to generate a predicted global dataset of net primary productivity of aboveground vegetation and net primary productivity of underground vegetation.

Optionally, obtain the global multi-year average vegetation net primary productivity products generated based on the parameter model, including:

Determine the data source for the global multi-year average vegetation net primary productivity product;

According to the data source, determine the relationship between vegetation growth and environmental factors, and build a parameter model;

Obtaining input data, and inputting the input data into the parameter model, so that the parameter model calculates the net primary productivity of vegetation year by year;

The net primary productivity of vegetation over many years was screened to obtain screening data;

The screening data were aligned and matched in time and space to calculate multi-year averages;

Based on the multi-year average, the multi-year average vegetation net primary productivity product is generated.

Optionally, obtain environmental covariates based on the global multi-year average vegetation net primary productivity product. Environmental covariates include global annual precipitation data and multi-year average precipitation data, including:

According to the known time point data, the annual precipitation values of the missing time points are estimated by linear interpolation method to obtain aligned data;

Determine the geographical unit and time unit for matching according to the alignment data;

From the aligned data, annual precipitation values corresponding to each selected geographic unit and time unit are extracted, as well as multi-year average precipitation values corresponding to each geographic unit are extracted;

The annual precipitation value and the multi-year average precipitation value are matched with the global multi-year average vegetation net primary productivity value of the corresponding geographical unit and time unit to obtain the matching results, wherein the matching results include an annual precipitation value, a multi-year average precipitation value and a global multi-year average vegetation net primary productivity value uniquely corresponding to each geographical location and time point.

Optionally, calculate the annual precipitation anomaly rate based on global annual precipitation data, including:

pass Calculate long-term average annual precipitation Where w _i is the weight of the i-th year, β is the decay factor, P _i is the annual precipitation in the i-th year, and N is the number of years used to calculate the long-term average;

According to the long-term average annual precipitation, Calculate the standard deviation of long-term annual precipitation s _P , where γ _i is the robustness weight of the i-th year;

According to the standard deviation of long-term annual precipitation, Calculate the precipitation anomaly rate Rate _i , where is the absolute deviation of the precipitation in the ith year from the weighted long-term average annual precipitation, and δ is the adjustment parameter.

Optionally, obtain ground-truth net primary productivity of aboveground vegetation and net primary productivity of belowground vegetation datasets, including:

Identify study area and target vegetation;

According to the actual situation of the study area, estimate the feasibility and potential risks of field measurement to obtain the evaluation results; determine the field measurement plan based on the evaluation results, and determine the schedule based on the field measurement plan;

According to the field measurement plan, the above-ground vegetation and underground vegetation biomass of each sample plot are measured to obtain measurement data;

The measured data are converted into global multi-year average vegetation net primary productivity as needed.

Optionally, obtain the latitude and longitude coordinates corresponding to the aboveground net primary productivity and underground net primary productivity datasets, and extract the global multi-year average vegetation net primary productivity, multi-year average precipitation, and annual precipitation anomaly rate at the observation point in the global dataset, including:

Get the aboveground vegetation net primary productivity and belowground vegetation net primary productivity datasets and their latitude and longitude coordinates

Obtain observational datasets of net primary productivity of above-ground vegetation and net primary productivity of below-ground vegetation;

Formatting the observation data set to obtain formatted data;

Extract and organize the latitude and longitude coordinates of each observation point in the formatted data;

The global dataset is obtained and spatially matched with the global dataset according to the latitude and longitude coordinates of the observation point. For each observation point, the multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate values corresponding to the observation point are extracted.

Optionally, after inputting the global multi-year average vegetation net primary productivity, multi-year average precipitation, and annual precipitation anomaly rate datasets at the global scale according to the machine learning model to generate the predicted global datasets of aboveground vegetation net primary productivity and belowground vegetation net primary productivity, it also includes:

The predicted global datasets of aboveground net primary productivity and belowground net primary productivity were compared with the ground-measured datasets of aboveground net primary productivity and belowground net primary productivity to obtain comparative results.

Based on the comparison results, the prediction performance of the machine learning model is evaluated to obtain an evaluation result.

It should be noted that the device is a device corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

The embodiment of the present invention further provides a computing device, comprising: a processor, a memory storing a computer program, wherein when the computer program is executed by the processor, the method described above is executed. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

The embodiment of the present invention also provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the method described above. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, ROM, RAM, magnetic disks, or optical disks.

In addition, it should be noted that in the apparatus and method of the present invention, it is obvious that each component or each step can be decomposed and/or recombined. These decompositions and/or recombinations should be regarded as equivalent schemes of the present invention. Moreover, the steps of performing the above-mentioned series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order, and some steps can be performed in parallel or independently of each other. For those of ordinary skill in the art, it is understood that all or any steps or components of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or a network of computing devices in hardware, firmware, software or a combination thereof, which can be achieved by those of ordinary skill in the art using their basic programming skills after reading the description of the present invention.

Therefore, the purpose of the present invention can also be achieved by running a program or a group of programs on any computing device. The computing device can be a well-known general device. Therefore, the purpose of the present invention can also be achieved by simply providing a program product containing a program code that implements the method or device. That is to say, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any well-known storage medium or any storage medium developed in the future. It should also be pointed out that in the device and method of the present invention, it is obvious that each component or each step can be decomposed and/or recombined. These decompositions and/or recombinations should be regarded as equivalent schemes of the present invention. In addition, the steps of performing the above-mentioned series of processing can naturally be performed in chronological order according to the order of description, but it is not necessary to perform them in chronological order. Some steps can be performed in parallel or independently of each other.

The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims (10)

1. A remote sensing prediction method for vegetation net primary productivity, characterized in that the method comprises: Obtain the global multi-year average vegetation net primary productivity product generated based on the parameter model; According to the global multi-year average vegetation net primary productivity product, environmental covariates are obtained, including global annual precipitation data and multi-year average precipitation data; Based on the global annual precipitation data, the annual precipitation anomaly rate is calculated; Obtain ground-truthed net primary productivity of above-ground vegetation and net primary productivity of below-ground vegetation; Obtain the longitude and latitude coordinates corresponding to the aboveground net primary productivity and underground net primary productivity datasets, and extract the global multi-year average vegetation net primary productivity, multi-year average precipitation, and annual precipitation anomaly rate at the observation point in the global dataset; Build and train a machine learning model based on the global multi-year average vegetation net primary productivity, multi-year average precipitation, and annual precipitation anomaly rate; According to the machine learning model, the global multi-year average vegetation net primary productivity, multi-year average precipitation, and annual precipitation anomaly rate datasets at the global scale are input to generate the predicted global datasets of aboveground vegetation net primary productivity and belowground vegetation net primary productivity. 2. The remote sensing prediction method of vegetation net primary productivity according to claim 1 is characterized in that obtaining the global multi-year average vegetation net primary productivity product generated based on the parameter model comprises: Determine the data source for the global multi-year average vegetation net primary productivity product; According to the data source, determine the relationship between vegetation growth and environmental factors, and build a parameter model; Obtaining input data, and inputting the input data into the parameter model, so that the parameter model calculates the net primary productivity of vegetation year by year; The net primary productivity of vegetation over many years was screened to obtain screening data; The screening data were aligned and matched in time and space to calculate multi-year averages; Based on the multi-year average, the multi-year average vegetation net primary productivity product is generated. 3. The remote sensing prediction method of vegetation net primary productivity according to claim 2 is characterized in that environmental covariates are obtained based on the global multi-year average vegetation net primary productivity product, and the environmental covariates include global annual precipitation data and multi-year average precipitation data, including: According to the known time point data, the annual precipitation values of the missing time points are estimated by linear interpolation method to obtain aligned data; Determine the geographical unit and time unit for matching according to the alignment data; From the aligned data, annual precipitation values corresponding to each selected geographic unit and time unit are extracted, as well as multi-year average precipitation values corresponding to each geographic unit are extracted; The annual precipitation value and the multi-year average precipitation value are matched with the global multi-year average vegetation net primary productivity value of the corresponding geographical unit and time unit to obtain the matching results, wherein the matching results include an annual precipitation value, a multi-year average precipitation value and a global multi-year average vegetation net primary productivity value uniquely corresponding to each geographical location and time point. 4. The remote sensing prediction method of vegetation net primary productivity according to claim 3 is characterized in that the annual precipitation anomaly rate is calculated based on global annual precipitation data, comprising: pass Calculate long-term average annual precipitation Where w _i is the weight of the i-th year, β is the decay factor, P _i is the annual precipitation in the i-th year, and N is the number of years used to calculate the long-term average; According to the long-term average annual precipitation, Calculate the standard deviation of long-term annual precipitation s _P , where γ _i is the robustness weight of the i-th year; According to the standard deviation of long-term annual precipitation, Calculate the precipitation anomaly rate Rate _i , where is the absolute deviation of the precipitation in the ith year from the weighted long-term average annual precipitation, and δ is the adjustment parameter. 5. The remote sensing prediction method of vegetation net primary productivity according to claim 4, characterized in that obtaining the ground-measured net primary productivity of vegetation above ground and the ground-measured net primary productivity of vegetation below ground datasets comprises: Identify study area and target vegetation; According to the actual situation of the study area, estimate the feasibility and potential risks of field measurement to obtain the evaluation results; determine the field measurement plan based on the evaluation results, and determine the schedule based on the field measurement plan; According to the field measurement plan, the above-ground vegetation and underground vegetation biomass of each sample plot are measured to obtain measurement data; The measured data are converted into global multi-year average vegetation net primary productivity as needed. 6. The remote sensing prediction method of vegetation net primary productivity according to claim 5 is characterized in that the latitude and longitude coordinates corresponding to the aboveground vegetation net primary productivity and underground vegetation net primary productivity data sets are obtained, and the global multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate at the observation point are extracted from the global data set, including: Get the aboveground vegetation net primary productivity and belowground vegetation net primary productivity datasets and their latitude and longitude coordinates Obtain observational datasets of net primary productivity of above-ground vegetation and net primary productivity of below-ground vegetation; Formatting the observation data set to obtain formatted data; Extract and organize the latitude and longitude coordinates of each observation point in the formatted data; The global dataset is obtained and spatially matched with the global dataset according to the latitude and longitude coordinates of the observation point. For each observation point, the multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate values corresponding to the observation point are extracted. 7. The remote sensing prediction method of vegetation net primary productivity according to claim 6 is characterized in that after inputting the global multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate data sets on a global scale according to the machine learning model to generate the predicted global data sets of above-ground vegetation net primary productivity and underground vegetation net primary productivity, it also includes: The predicted global datasets of aboveground net primary productivity and belowground net primary productivity were compared with the ground-measured datasets of aboveground net primary productivity and belowground net primary productivity to obtain comparative results. Based on the comparison results, the prediction performance of the machine learning model is evaluated to obtain an evaluation result. 8. A remote sensing prediction device for vegetation net primary productivity, characterized by comprising: The acquisition module is used to obtain the global multi-year average vegetation net primary productivity product generated based on the parameter model; obtain environmental covariates based on the global multi-year average vegetation net primary productivity product, which includes global annual precipitation data and multi-year average precipitation data; calculate the annual precipitation anomaly rate based on the global annual precipitation data; and obtain the ground-measured net primary productivity of vegetation above ground and net primary productivity of vegetation below ground data sets; The processing module is used to obtain the longitude and latitude coordinates corresponding to the aboveground vegetation net primary productivity and underground vegetation net primary productivity datasets, and extract the global multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate at the observation point in the global dataset; build and train a machine learning model based on the global multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate; according to the machine learning model, input the global multi-year average vegetation net primary productivity, multi-year average precipitation and annual precipitation anomaly rate datasets on a global scale to generate predicted global datasets of aboveground vegetation net primary productivity and underground vegetation net primary productivity. 9. A computing device, comprising: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in any one of claims 1 to 7. 10. A computer-readable storage medium, characterized in that a program is stored in the computer-readable storage medium, and when the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.