Hyperspectral remote sensing monitoring method for soil microbial community and storage medium

Abstract

The invention relates to the technical field of soil microbe monitoring, and discloses a hyperspectral remote sensing monitoring method for a soil microbe community and a storage medium, and the hyperspectral remote sensing monitoring method for the soil microbe community comprises the following steps: carrying out data enhancement and semi-supervised learning based on a generative adversarial network; constructing a generative adversarial network to generate pseudo-hyperspectral data conforming to physical constraints, and performing soil microbial community parameter prediction; constructing a physical information neural network, introducing a soil hydrothermal migration physical equation constraint, and inferring microorganism distribution and hydrothermal states of different depths of a soil profile; a nitrogen cycle kinetic equation is embedded in the physical information neural network, network output variables and physical constraint loss are expanded, and model parameters and process parameters are optimized; through the GAN data enhancement and semi-supervised learning technology, the prediction precision of the microbial community index can be maintained under the condition that the labeled sample size is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of soil microorganism monitoring, and more specifically, it relates to a hyperspectral remote sensing monitoring method and storage medium for soil microbial communities. Background Art

[0002] With the development of precision agriculture and soil ecosystem protection, the monitoring of soil microbial communities has increasingly become a key link, which is of great significance for crop yield, soil quality, and environmental protection. Soil microorganisms participate in the key processes of carbon and nitrogen cycles and play an irreplaceable role in nutrient transformation, environmental pollutant degradation, and soil structure improvement.

[0003] Currently, the monitoring of soil microbial communities mainly relies on laboratory analysis methods, such as DNA extraction and sequencing, cultivation counting, functional gene analysis, etc. Although these methods have high accuracy, they have disadvantages such as large sampling destructiveness, high analysis cost, long time consumption, and limited spatial coverage, making it difficult to meet the requirements of large-scale, rapid, and non-destructive monitoring. Hyperspectral remote sensing technology provides the possibility for rapid, non-destructive, and large-scale monitoring of soil properties through the acquisition of reflected spectral information of surface targets, and has made progress in the monitoring of physical and chemical properties such as soil organic matter and texture. However, applying hyperspectral technology to soil microbial community monitoring still faces multiple challenges.

[0004] The existing hyperspectral monitoring technologies for soil microbial communities mainly have the following problems: it is difficult to obtain labeled samples, and the high cost of laboratory analysis leads to sparse labeled training data, which limits the performance of deep learning models; it is difficult to obtain profile information, and existing hyperspectral technologies mainly target surface features and are difficult to directly obtain microbial information at different depths of the soil profile; there is a lack of effective combination of physical processes and data-driven approaches, either completely relying on statistical correlations and ignoring physical process constraints, or completely relying on mechanism models but with difficult parameterization; it is difficult to parameterize the key processes of the nitrogen cycle, and it is difficult to rapidly obtain microbial-driven nitrogen cycle process parameters over a large range; the lack of result uncertainty assessment reduces the reliability of decision support.

[0005] Therefore, there is an urgent need for a hyperspectral remote sensing monitoring method for soil microbial communities that can overcome the above technical problems to achieve precise prediction under sparse data conditions, information inference from the surface to the profile, the integration of physical and data-driven approaches, and the accurate inversion of process parameters. Summary of the Invention

[0006] The present invention provides a hyperspectral remote sensing monitoring method and storage medium for soil microbial communities, which solve the technical problems of difficult acquisition of labeled samples, difficult acquisition of soil profile information, and difficult parameterization of nitrogen cycle parameters in related technologies.

[0007] The present invention provides a hyperspectral remote sensing monitoring method for soil microbial communities, including: Based on data augmentation and semi-supervised learning of generative adversarial networks, hyperspectral remote sensing data and labeled hyperspectral microbial data pairs are obtained, a generative adversarial network is constructed to generate pseudo-hyperspectral microbial data pairs, a semi-supervised learning framework is constructed, and a prediction model for microbial community parameters is output; Based on the profile inference of physics-informed neural networks, according to the output of the prediction model, combined with hyperspectral features, meteorological data and spatial coordinates, a physics-informed neural network is constructed and the physical equations of soil water and heat transport are introduced for constraint, and the microbial distribution and water and heat status at different depths are inferred; Embed the nitrogen cycle kinetic equation into the physics-informed neural network, and use the profile microbial distribution as the driving force to output the key process parameters of the nitrogen cycle; Use the data assimilation method to combine the key process parameters of the nitrogen cycle and the profile microbial distribution with the observed data, input the hyperspectral data of the target area, generate the monitoring results of the spatial distribution of soil microbial communities, the profile microbial abundance distribution, and the nitrogen cycle parameter distribution, and conduct uncertainty assessment.

[0008] Further, the steps for the generative adversarial network to generate pseudo-hyperspectral data that conforms to physical constraints include: Construct a generative adversarial network including a generator and a discriminator. The generator receives a random noise vector and a microbial parameter conditional vector as inputs and generates hyperspectral data that conforms to the true distribution; Adopt an alternating optimization strategy to train the generator and the discriminator until the generator can generate pseudo-hyperspectral data that conforms to physical constraints; By inputting different microbial parameter conditional vectors, corresponding hyperspectral data is generated to form a large number of pseudo-labeled samples; Construct a semi-supervised learning framework and use the hyperspectral data of the true distribution, the pseudo-hyperspectral data that conforms to physical constraints, and the pseudo-labeled samples for model training.

[0009] Further, the constructed semi-supervised learning framework includes a student model and a teacher model. The student model updates its parameters through supervised training, and the parameters of the teacher model are the exponential moving averages of the parameters of the student model; the loss function of the semi-supervised learning framework includes supervised loss, consistency loss, and regularization loss.

[0010] Further, the construction of the physics-informed neural network includes: Construct a neural network including a forward propagation part and a physical constraint part; To ensure that the output of the neural network satisfies physical laws, introduce the physical equation constraints of the soil water and heat transport process; Construct a total loss function including data fitting loss and physical constraint loss; Using a generative adversarial network to generate pseudo-profile data that satisfies physical constraints and enhancing the training effect of a physics-informed neural network.

[0011] Further, the physical equation constraints include a water migration equation and a heat conduction equation.

[0012] Further, the steps of embedding the nitrogen cycle kinetics equation in the physics-informed neural network include: Adding a description equation for key processes of the nitrogen cycle in the physics-informed neural network framework; Expanding the output variables of the physics-informed neural network to add predictions of ammonium nitrogen and nitrate nitrogen concentrations, as well as outputs of the nitrification rate and denitrification rate varying with depth; Expanding the physical constraint loss of the physics-informed neural network to add a residual term of the nitrogen cycle equation; Adding parameter value range constraints to ensure that the inverted parameters are physically reasonable.

[0013] Further, the description equations for key processes of the nitrogen cycle include a nitrification process equation and a denitrification process equation, which describe the dynamic change processes of ammonium nitrogen and nitrate nitrogen, as well as the effects of microbial abundance, water content, temperature, and organic carbon on nitrification and denitrification processes.

[0014] Further, the steps of using the data assimilation method include adopting the ensemble Kalman filter method to update parameter estimates when observation data arrives by constructing multiple ensemble members of model parameters.

[0015] Further, the uncertainty assessment includes: Using the ensemble prediction method to calculate the prediction intervals of various monitoring results; Generating an uncertainty distribution map to identify areas with relatively low prediction reliability; Providing a confidence assessment of the prediction results to provide a reliability reference for decision-making applications; Outputting a quality control report of the monitoring results, including model performance indicators and applicability evaluations.

[0016] The present invention provides a storage medium, including a memory and one or more processors. Executable code is stored in the memory. When the one or more processors execute the executable code, it is used to implement the above-mentioned hyperspectral remote sensing monitoring method for soil microbial communities.

[0017] The beneficial effects of the present invention are as follows: Through GAN data augmentation and semi-supervised learning techniques, it is possible to maintain the prediction accuracy of microbial community indicators even when the number of labeled samples is reduced, and even slightly improve in the prediction of certain functional groups, reducing the dependence on on-site sampling and laboratory analysis, and reducing the monitoring cost and time; Based on the physics-informed neural network, it is able to infer the soil profile microbial distribution and hydrothermal state from surface hyperspectral remote sensing data, reduce the average relative error of profile microbial abundance prediction, reduce the average relative error of profile hydrothermal state prediction, achieve the ability of "seeing deep from the surface" from the surface to the profile, and reduce the workload of traditional drilling sampling; It realizes the inversion of the key process parameters of the nitrogen cycle driven by soil microorganisms, which are crucial for the nitrogen cycle model but difficult to directly measure by traditional methods, provides core parameter support for precision nitrogen fertilizer management and environmental impact assessment, and makes it possible to monitor large-scale nitrogen transformation; By integrating physical process constraints and data-driven technologies, it can not only provide accurate prediction results, but also quantify the uncertainty of the prediction, provide a reliability assessment for decision-making, and enhance the application value of the monitoring results in precision agriculture and environmental management. Description of the Drawings

[0018] Figure 1 is the flow chart of a hyperspectral remote sensing monitoring method for soil microbial communities in the present invention; Figure 2 is the flow chart of Step 1 in the present invention; Figure 3 is the flow chart of Step 2 in the present invention; Figure 4 is the flow chart of Step 3 in the present invention; Figure 5 is the flow chart of Step 4 in the present invention; Figure 6 is the comparative combination chart of the prediction accuracy and data efficiency of different methods in the microbial community in the present invention; Figure 7 [[ID=2)8]]is the heat map of the spatial distribution and uncertainty assessment of the key nitrogen cycle parameters in the present invention; Figure 8 is the line chart of the parameter prediction performance and verification at different depths of the soil profile in the present invention. Detailed Embodiments

[0019] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein, and the functions and arrangements of the elements discussed can be changed without departing from the scope of protection of the content of this specification. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0020] In at least one embodiment of the present invention, a hyperspectral remote sensing monitoring method for soil microbial communities is disclosed, as Figures 1 to 5 shown, including: Step 1: Based on data augmentation and semi-supervised learning of generative adversarial networks, obtain hyperspectral remote sensing data and labeled hyperspectral microbial data pairs, construct a generative adversarial network to generate pseudo-hyperspectral microbial data pairs, construct a semi-supervised learning framework, and output a microbial community parameter prediction model. In this step, the problem of sparse labeled samples in hyperspectral remote sensing monitoring is solved through a generative adversarial network (GAN), and the generalization ability of the deep learning model is improved. It should be understood that the specific implementation is as follows: Step 1.1: Data preparation Obtain a small number of labeled hyperspectral microbial data pairs (labeled data) and a large number of unlabeled hyperspectral data (unlabeled data). The hyperspectral features in the labeled data include surface reflectance curves (usually covering the wavelength range of 400 to 2500 nm), and the microbial labels include community composition (such as phylum-level taxonomic abundance), diversity indices (such as Shannon index), or specific functional group abundances.

[0021] Step 1.2: GAN model construction Construct a GAN model suitable for the characteristics of hyperspectral data, including a generator and a discriminator. The generator receives a random noise vector and a conditional vector (microbial parameters) as inputs, and generates hyperspectral data that conforms to the true distribution; the discriminator distinguishes between real hyperspectral data and generated pseudo-hyperspectral data.

[0022] According to the embodiments of the present application, the specific implementation of the GAN model includes the following structure: The generator adopts an encoder-decoder structure. The encoder encodes the random noise vector and the conditional vector into latent features, and the decoder reconstructs the latent features into hyperspectral data: The encoder contains 3 convolutional layers, each followed by batch normalization and a LeakyReLU activation function; the decoder contains 3 transposed convolutional layers for upsampling, and the last layer uses a Tanh activation function to ensure that the output value range is between [-1, 1], and then maps to the [0, 1] interval through a linear transformation as the reflectance value; The discriminator adopts a multi-layer convolutional neural network structure, containing 4 convolutional layers, each followed by batch normalization and a LeakyReLU activation function, and the last layer outputs a single scalar representing the probability that the input data is a real sample.

[0023] Optionally, in some embodiments, the generator may adopt an encoder-decoder structure based on the attention mechanism, where a self-attention layer is introduced between the encoder and the decoder to enhance the ability to capture long-range dependencies between features, which is particularly suitable for modeling the complex correlations between different bands in hyperspectral data.

[0024] In some other embodiments, the discriminator may adopt an objective function based on the Wasserstein distance, and the training stability is ensured through a gradient penalty term. The loss function is modified as: ; where represents the loss function of the discriminator (DiscriminatorLoss); represents the expectation of the output of the discriminator sampled from the true data distribution for the true hyperspectral samples and is the probability distribution of the true hyperspectral data; represents the expectation of the output of the discriminator sampled from the noise distribution after the generator generates the pseudo-samples ; is a random noise vector is the noise distribution and is the generated hyperspectral data; represents the weight coefficient of the gradient penalty term, which controls the contribution of the gradient penalty term to the loss function; represents the expectation of the square of the difference between the L2 norm of the gradient of the discriminator sampled from the uniform interpolation distribution between the true samples and the generated samples and represents the gradient of and represents the L2 norm; represents the output of the discriminator for the input

[0025] ; where represents the loss function of the generator (GeneratorLoss); represents the expectation of the output of the generator sampled from the noise distribution Generate fake samples Post-discriminator The expected value of the output; represents a random noise vector, is the noise distribution; Representation Discriminator For the generator Output of the generated samples.

[0026] The loss functions of the generator G and the discriminator D are: ; in represents the generator loss; Represents the noise distribution Sampled random noise vector , is the noise distribution; Represents the distribution of microbial parameters Sampled conditional vector , is the probability distribution of microbial parameters; Representation Discriminator Generate data and conditions Output; Representation Generator by and Pseudo-hyperspectral data generated for input; Represents the expectation operator; ; in represents the discriminator loss; Represents the distribution of real hyperspectral data Sampling , is the probability distribution of real hyperspectral data; Represents the distribution of microbial parameters Sampled conditional vector ; Representation Discriminator For real samples and conditions The logarithmic loss of the output; Represents the discriminator's response to input data In the conditions Below is the probability estimate of the real sample; Represents the noise distribution Sampled random noise vector ; Representation Discriminator Generate samples and conditions log loss of the output; represents the expectation operator.

[0027] In the application scenario of farmland soil microbial monitoring, the GAN model can be used to generate hyperspectral data under different abundances of nitrogen cycle functional flora. For example, in the application of precision fertilization decision support, by inputting different abundances of nitrifying bacteria, the corresponding hyperspectral data can be generated to make up for the problem of sparse field sampling points and provide data support for subsequent fertilization zoning.

[0028] Step 1.3, GAN training and data generation; The generator is trained using an alternating optimization strategy and the discriminator until the generator can generate pseudo-hyperspectral data that meets the physical constraints. After training, by inputting different condition vectors of microbial parameters , the corresponding hyperspectral data is generated to form a large number of pseudo-labeled samples. The newly generated pseudo-samples need to meet the physical constraint conditions at the same time, such as the reflectance value range is between and the spectral curve is continuous and smooth, etc.

[0029] Step 1.4, semi-supervised learning model construction; Based on the real-labeled samples, the generated pseudo-labeled samples and a large number of unlabeled samples, a semi-supervised learning framework is constructed. According to an embodiment of the present application, the semi-supervised learning framework adopts the MeanTeacher model structure, which includes a student model and a teacher model. The student model updates the parameters through supervised training, and the teacher model parameters are the exponential moving averages of the student model parameters.

[0030] The specific implementation of this semi-supervised learning framework includes: The student model and the teacher model adopt the same network structure, which consists of four convolutional blocks and two fully connected layers. Each convolutional block contains a convolutional layer, a batch normalization layer, and a ReLU activation function; For the labeled data (including real-labeled samples and high-quality pseudo-labeled samples), the mean squared error (MeanSquaredError, MSE) between the predicted value and the real label is calculated as the supervised loss ; For the unlabeled data, different data augmentations are applied to the same input, and the predictions are made through the student model and the teacher model respectively. The mean squared error between the two prediction results is calculated as the consistency loss ; regularization loss L2 regularization is adopted to prevent the model from overfitting.

[0031] Optionally, in some embodiments, the semi-supervised learning framework may adopt the Pseudo-Labeling method. First, an initial model is trained using labeled data, then the model is used to predict unlabeled data to generate pseudo-labels, and finally, the labeled data and the data with pseudo-labels are jointly used for model training. To control the quality of pseudo-labels, a confidence threshold can be set, and only high-confidence pseudo-labels are used.

[0032] In some other embodiments, the semi-supervised learning framework may adopt the MixMatch method, which generates more diverse training samples by mixing augmented versions of labeled and unlabeled data. Specifically, multiple augmented versions are generated for each unlabeled sample, and the average of the prediction results of these augmented versions is used as the pseudo-label; then the labeled and unlabeled samples are mixed to generate mixed samples for calculating the loss function.

[0033] This framework includes three parts of the loss function: the supervised loss (calculated using real-labeled samples and pseudo-labeled samples), the consistency loss (calculated using unlabeled data), and the regularization loss . The total loss function is: ; where represents the total loss function; represents the supervised loss, which measures the prediction error of the model for labeled data; represents the consistency loss, which measures the consistency of the prediction results of the student model and the teacher model for unlabeled data; represents the regularization loss, usually L2 regularization, to prevent overfitting; represents the consistency loss 's weight coefficient, which controls its contribution to the total loss; represents the regularization loss 's weight coefficient, which controls its contribution to the total loss.

[0034] In the regional soil monitoring scenario, this semi-supervised learning framework can be used for predicting soil microbial functional groups at the regional scale. Application examples include: when monitoring large agricultural areas, only a small number of sampling points with laboratory measurement results are required, combined with a large number of locations with only hyperspectral data. Through this framework, a microbial function map of the entire region can be generated, providing decision support for regional soil health assessment and precision fertilization management.

[0035] Through the above steps, a semi-supervised learning model that can accurately predict soil microbial community parameters from hyperspectral data is obtained. Therefore, this model has good generalization ability.

[0036] Step 2, Profile inference based on physics-informed neural network. According to the output of the prediction model, combined with hyperspectral features, meteorological data, and spatial coordinates, construct a physics-informed neural network and introduce the soil water and heat transport equation constraints to infer the microbial distribution and water and heat status at different depths; In this step, the physics-informed neural network (PINN) method is used to solve the problem of inferring the soil profile microbial distribution and its coupling relationship with the water and heat processes from surface hyperspectral data. It should be noted that the specific implementation is as follows: Step 2.1, Construction of the PINN network architecture; Construct a physics-informed neural network, which includes a forward propagation part and a physical constraint part. The forward propagation part is a multi-layer neural network, and the input is the surface hyperspectral feature vector (extracted from the preprocessed hyperspectral data in Step 1), the meteorological data vector (including precipitation , temperature , etc.) and the spatial position coordinates and the depth coordinate , and the output is the microbial abundance , soil moisture content and soil temperature at this position and depth.

[0037] According to an embodiment of the present application, the specific structure of the PINN network includes: Input layer: Receive the surface hyperspectral feature vector (dimensionality reduced to 20 principal components), the meteorological data vector (5-dimensional, including precipitation, temperature, humidity, radiation intensity, and wind speed), the spatial coordinates (2-dimensional, longitude and latitude or XY coordinates), and the depth coordinate (1-dimensional); Hidden layer: Includes 5 fully connected layers, each layer contains 64 neurons, and the Swish activation function is used to improve the non-linear expression ability; Output layer: Output the microbial abundance (which can be a multi-dimensional vector representing the abundances of different taxa), the soil water content, and the soil temperature.

[0038] Optionally, in some embodiments, the PINN network can adopt a residual network structure, introduce skip connections to alleviate the gradient vanishing problem in the training of deep networks, and improve the model convergence speed and performance. Specifically, a skip connection is added between every two hidden layers, and the output of the previous layer is directly added to the output of the next layer.

[0039] In some other embodiments, the PINN network may adopt a multi-scale fusion structure, where multiple sub-networks in parallel process input features of different scales, and then the multi-scale features are fused for prediction. For example, three parallel sub-networks can be constructed to process hyperspectral features, meteorological data, and spatial-depth coordinates respectively, and then the outputs of the three sub-networks are concatenated or weighted and fused. This structure is particularly suitable for the fusion processing of multi-source heterogeneous data.

[0040] The network structure can be expressed as: ; where represents the microbial abundance at position , depth , and time ; represents the soil volumetric water content at position , depth , and time ; represents the soil temperature at position , depth , and time ; represents the neural network function, which represents the mapping from input features to output variables; , , , represent the surface hyperspectral feature vector, meteorological data vector, spatial coordinates, and depth coordinates respectively; represents the set of neural network parameters, including all trainable weights and biases.

[0041] Step 2.2, Definition of physical constraint conditions; To ensure that the neural network output satisfies physical laws, the following physical equation constraints are introduced for the soil water and heat transport process: Water transport equation (Richards equation): ; where represents the partial derivative of the soil volumetric water content with respect to time , indicating the change rate of water content over time; represents the gradient operator, indicating the spatial derivative; represents the unsaturated hydraulic conductivity, which is a function dependent on the water content ; represents the spatial gradient of the water potential; represents the source-sink term, such as root water uptake, indicating the loss or supply of water; Heat conduction equation: ; where represents the soil heat capacity, the heat capacity of the soil per unit volume; represents the soil temperature partial derivative with respect to time ; represents the soil thermal conductivity, indicating the heat transfer ability of the soil; represents the spatial gradient of the soil temperature; represents the density of water; represents the specific heat capacity of water; represents the water flow velocity vector, indicating the flow velocity of water in the soil; represents the gradient operator, indicating the spatial derivative; Step 2.3, Construction of the PINN loss function; The total loss function of PINN includes two parts: data fitting loss and physical constraint loss: ; where represents the total loss function of PINN; represents the weight coefficient of the data fitting loss, controlling its contribution to the total loss; represents the data fitting loss, measuring the error between the model output and the observed data; represents the weight coefficient of the physical constraint loss, controlling its contribution to the total loss; represents the physical constraint loss, measuring the degree to which the model output satisfies the physical equation; Data fitting loss is calculated based on the profile observation data of a small number of borehole sampling points: ; where represents the total number of observation points; represents the data fitting loss, measuring the error between the model output and the observed data; represents the measured value of microbial abundance at the th observation point; represents the predicted value of microbial abundance by the model at the th observation point; represents the measured value of soil volumetric water content at the th observation point; represents the predicted value of soil volumetric water content by the model at the th observation point; represents the measured value of soil temperature at the th observation point; represents the Predicted values of soil temperature models at individual observation points.

[0042] Physical constraint loss Based on the residual calculation of the above physical equations, the automatic differentiation technique is used to evaluate the difference between the left and right sides of the equations:

[0043] where represents the physical constraint loss, represents the number of sampling points (which can be virtual points) used to calculate the physical constraints; represents the residual of the Richards equation at point ; , , respectively represent the spatial coordinate, depth coordinate, and time coordinate of the th observation point; represents the residual of the heat conduction equation at point ; , , respectively represent the spatial coordinate, depth coordinate, and time coordinate of the th observation point.

[0044] Step 2.4, GAN-assisted generation of physically constrained samples; Using the GAN model trained in Step 1, pseudo-profile data that satisfy the physical constraints are generated to enhance the training effect of the PINN. These pseudo-samples not only conform to the statistical relationship between hyperspectral and microbial abundance but also satisfy the constraints of the hydrothermal transport physical process.

[0045] Through the above steps, a PINN model is trained to infer the microbial distribution and hydrothermal state at different depths of the soil profile from surface hyperspectral data. In addition, this model not only fits the actual observed data but also conforms to the physical laws of soil hydrothermal transport.

[0046] Step 3, embed the nitrogen cycle kinetic equation in the physics-informed neural network, using the profile microbial distribution as the driving force to output the key process parameters of the nitrogen cycle; In this step, the nitrogen cycle kinetic equation is embedded based on the PINN framework to realize the inversion of the key process parameters of the microbial-driven nitrogen cycle and solve the problem of difficult direct acquisition of parameters. It should be noted that the specific implementation is as follows: Step 3.1, embedding of the nitrogen cycle process kinetic equation; In the PINN framework of Step 2, equations describing the key processes of the nitrogen cycle are added. Taking nitrification and denitrification as examples, the following equations are introduced: Nitrification process equation: ; wherein represents the spatial gradient of the ammonium nitrogen concentration; represents the partial derivative of the ammonium nitrogen concentration with respect to time ; represents the gradient operator and represents the spatial derivative; represents the diffusion coefficient of ammonium ions, representing the diffusion ability in the soil; represents the nitrification rate constant, representing the rate parameter of the nitrification reaction; represents the influence function of environmental factors on nitrification, is the microbial abundance, is the soil volumetric water content, is the soil temperature; ; wherein represents the spatial gradient of the nitrate nitrogen concentration; represents the partial derivative of the nitrate nitrogen concentration with respect to time ; represents the diffusion coefficient of nitrate ions, representing the diffusion ability in the soil; represents the denitrification rate constant, representing the rate parameter of the denitrification reaction; represents the influence function of environmental factors on denitrification, is the microbial abundance, is the soil volumetric water content, is the soil temperature,<> is the organic carbon content; represents the gradient operator and represents the spatial derivative; represents the nitrification rate constant, representing the rate parameter of the nitrification reaction; represents the ammonium nitrogen concentration, representing the content in the soil ; represents the influence function of environmental factors on nitrification, is the microbial abundance, is the soil volumetric water content, is the soil temperature; ; wherein represents the influence function of environmental factors on nitrification, is the microbial abundance, is the soil volumetric water content, is the soil temperature; represents the influence function of water content on the process rate, see below; represents the influence function of temperature on the process rate; ; wherein represents the influence function of environmental factors on denitrification, is the microbial abundance, is the soil volumetric water content, is the soil temperature, is the organic carbon content; represents the influence function of water content on the process rate; represents the influence function of temperature on the process rate; represents the influence function of organic carbon on the process rate.

[0047] According to an embodiment of the present application, these influence functions can be specifically expressed as: When : ; Otherwise ; wherein represents the influence function of water content on the process rate; is the soil volumetric water content; is the lower limit of water content; is the optimal water content; is the upper limit of water content; ; wherein represents the influence function of temperature on the process rate; is the soil temperature; is the reference temperature; is the temperature sensitivity coefficient, representing the multiple by which the reaction rate increases for every 10°C increase in temperature; ; wherein represents the influence function of organic carbon on the process rate; is the organic carbon content; is the half-saturation constant, representing the saturation effect of organic carbon on the process rate; Step 3.2, expand the PINN output variables; Based on the PINN network output in Step 2, add the prediction of the concentrations of ammonium nitrogen and nitrate nitrogen , as well as the output of the key parameters and changing with depth.

[0048] According to an embodiment of the present application, based on step 2, the extended PINN network structure adds 4 additional output nodes in the output layer, corresponding to the ammonium nitrogen concentration, nitrate nitrogen concentration, nitrification rate constant, and denitrification rate constant respectively. The network structure can be expressed as: ; where represents the microbial abundance; represents the soil volumetric water content; represents the soil temperature; represents the ammonium nitrogen concentration; represents the nitrate nitrogen concentration; represents the nitrification rate constant; represents the denitrification rate constant; represents the extended neural network function; , , , , represent the surface hyperspectral feature vector, meteorological data vector, spatial coordinates, depth coordinates, and neural network parameters respectively.

[0049] Step 3.3, extending the PINN physical constraint loss; Based on the physical constraint loss in step 2, add the residual term of the nitrogen cycle equation: ; where represents the extended physical constraint loss; represents the physical constraint loss; represents the weight coefficient of the nitrogen cycle equation residual, controlling the contribution of the nitrogen cycle equation residual to the total loss; represents the number of sampling points used to calculate the physical constraint; represents the residual of the ammonium nitrogen kinetic equation at point ; represents the residual of the nitrate nitrogen kinetic equation at point ; , , respectively represent the spatial coordinates, depth coordinates, and time coordinates of the th observation point.

[0050] Step 3.4, increasing the parameter constraint conditions; To ensure that the inverted parameters are physically reasonable, add constraints on the parameter value range: ; ; where Represents the minimum value of the nitrification rate constant; Represents the maximum value of the nitrification rate constant; Represents the minimum value of the denitrification rate constant; Represents the maximum value of the denitrification rate constant; Represents the nitrification rate constant output by the model; Represents the denitrification rate constant output by the model.

[0051] These constraints can be achieved by adding penalty terms to the loss function or using parameter mapping methods (such as the sigmoid function) to ensure that the output parameters are within a reasonable range.

[0052] Step 3.5, data assimilation and parameter optimization; Adopt the data assimilation method to combine the surface hyperspectral data with a small amount of profile 、 concentration observation data to optimize the PINN model parameters and process parameters.

[0053] According to an embodiment of the present application, the data assimilation process is implemented by using the Ensemble Kalman Filter method, which updates the parameter estimation when the observation data arrives by constructing multiple ensemble members of the model parameters. The specific steps include: Initialization: Based on prior knowledge, generate multiple ensemble members (such as 100) for the PINN model parameters and process parameters; Prediction: Run the model using the parameters of each ensemble member to obtain the predicted values of the state variables; Update: When new observation data arrives, update the parameter values of each ensemble member according to the Kalman gain; Iteration: Repeat the prediction and update steps until the parameters converge.

[0054] Optionally, in some embodiments, the data assimilation process can adopt the Particle Filter method, which processes non-linear and non-Gaussian distribution situations through importance sampling and resampling steps. Specifically, assign a weight to each particle (i.e., the parameter ensemble member), indicating the degree of matching between the predicted value and the observed value of the particle; then resample according to the weights to generate a new particle set and add appropriate perturbations to maintain particle diversity. This method is particularly suitable for strong non-linear processes in the soil-microbial system.

[0055] In some other embodiments, the variational data assimilation method can be adopted to optimize the model parameters by minimizing the cost function. The cost function usually includes a background error term (the deviation from the prior parameter estimation) and an observation error term (the deviation from the observation data): ; where represents the cost function, the objective to be minimized; represents the parameter vector to be estimated; represents the prior parameter estimation vector; represents the background error covariance matrix, indicating the uncertainty of the prior parameters; represents the transpose operator; represents the observation operator that maps the parameter vector to the observation space; represents the observation vector containing the observation data; represents the observation error covariance matrix, indicating the uncertainty of the observation data; represents the inverse matrix operation.

[0056] The optimization objective is to minimize the total loss function of the extended PINN: ; where represents the total loss function of the extended PINN; represents the weight coefficient of the extended data loss term; represents the extended data loss term containing the fitting error of the nitrogen concentration observation data; represents the weight coefficient of the extended physical constraint loss; represents the extended physical constraint loss.

[0057] In the application scenario of environmental monitoring, this method can be used to monitor the nitrogen cycling process in agricultural and natural ecosystems, evaluate the nitrogen loss risk (such as nitrate leaching, nitrogen and nitrous oxide emissions), and provide a scientific basis for environmental protection and emission reduction measures.

[0058] Through the above steps, a model capable of retrieving the key parameters of the soil profile nitrogen cycle from surface hyperspectral data ( and ) is obtained. It can be seen that these parameters not only conform to the constraints of physical, chemical, and biological processes but also can explain the actually observed nitrogen concentration data.

[0059] Step 4: Use the data assimilation method to combine the key process parameters of the nitrogen cycle and the profile microbial distribution with the observation data, input the hyperspectral data of the target area, generate the monitoring results of the spatial distribution of soil microbial communities, the profile microbial abundance distribution, and the nitrogen cycle parameter distribution, and conduct uncertainty assessment; Based on the integrated model trained in the above three steps, realize the hyperspectral remote sensing monitoring of the soil microbial community in the target area.

[0060] Step 4.1: Input data of the target area; Input the hyperspectral remote sensing data of the area to be monitored into the trained model system, including: Surface hyperspectral data: Hyperspectral images of the target area obtained by satellites, drones or ground equipment; Auxiliary data: Meteorological data (temperature, humidity, precipitation, etc.) and geographical location information of the target area; Spatial coordinates: Geographical coordinate information of each pixel in the target area.

[0061] Step 4.2, Model prediction calculation; Calculate the input data through the integrated prediction model system: First, through the semi-supervised learning model trained in Step 1, predict the surface microbial community parameters from the hyperspectral data; Then, through the physics-informed neural network trained in Step 2, infer the microbial distribution and hydrothermal state at different depths of the soil profile; Finally, through the nitrogen cycle parameter inversion model trained in Step 3, obtain the key process parameters of the microbial-driven nitrogen cycle.

[0062] Step 4.3, Output of remote sensing monitoring results; After the model prediction calculation, obtain the complete remote sensing monitoring results of the target area, including: Spatial distribution map of soil microbial communities: showing the spatial distribution characteristics of different microbial functional groups in the target area; Profile microbial abundance distribution: providing information on the microbial abundance distribution at different depth levels of the soil profile; Nitrogen cycle process parameter distribution: including the spatial distribution of key parameters such as nitrification rate and denitrification rate; Soil environmental status information: Profile distribution of environmental factors such as soil water content and temperature.

[0063] Step 4.4, Uncertainty assessment and reliability quantification; Conduct uncertainty assessment on all prediction results: Using the ensemble prediction method, calculate the prediction intervals of various monitoring results; Generate an uncertainty distribution map to identify areas with low prediction reliability; Provide a confidence assessment of the prediction results to provide a reliability reference for decision-making applications; Output a quality control report for the monitoring results, including model performance indicators and applicability evaluation.

[0064] Through the above steps, a complete hyperspectral remote sensing monitoring result of the soil microbial community is formed, providing a scientific basis for precision agriculture management, environmental monitoring and ecological assessment.

[0065] A storage medium, comprising a memory and one or more processors, wherein an executable code is stored in the memory, and when the one or more processors execute the executable code, it is used to implement the above-mentioned hyperspectral remote sensing monitoring method for soil microbial communities.

[0066] Herein, the present invention provides an implementation example: A real application example of the present invention in the monitoring of soil microbial communities in farmland is provided: This application example is for the hyperspectral remote sensing monitoring of soil microbial communities in a certain farmland ecosystem. The farmland covers an area of about 100 hectares, mainly growing wheat and corn in rotation. In the past, a unified fertilization method was adopted, resulting in low nitrogen fertilizer use efficiency in some areas and a risk of nitrate leaching. The purpose of this example is to achieve high-precision monitoring of the distribution of soil microbial communities and key parameters of the nitrogen cycle through the method of this application, providing decision-making support for precision fertilization.

[0067] In this farmland, a drone is used to carry a hyperspectral imager (wavelength range 400 - 2500 nm, spectral resolution 10 nm) to obtain surface hyperspectral data. At the same time, 25 sampling points are set in the farmland for soil sample collection. Among them, only 10 points are used for the determination of soil microbial community composition (using 16S rRNA sequencing) and the determination of the abundance of nitrogen cycle functional genes. These 10 points are used as labeled data; the other 15 points only measure the basic physical and chemical properties and are used as unlabeled data. In addition, 5 points are selected from the 10 labeled points for 0 - 100 cm profile stratified sampling, and a sample is taken every 20 cm to determine the microbial composition and nitrogen forms in each layer, which are used for PINN model training and verification.

[0068] The preprocessing of hyperspectral data includes atmospheric correction, geometric correction, denoising, and band selection. Finally, 50 characteristic bands are selected for subsequent analysis. The soil microbial data is processed through a bioinformatics analysis process to obtain the information of microbial community composition, and the functional microbial groups related to the nitrogen cycle are focused on, including nitrifying bacteria (AOB, AOA) and denitrifying bacteria.

[0069] To solve the problem of insufficient labeled samples, a conditional GAN model is constructed. The generator adopts an encoder-decoder structure. The input layer receives a 128-dimensional random noise vector and a 10-dimensional microbial parameter vector (including the relative abundances of key microbial groups and diversity indices), and is processed through 3 encoding convolutional layers and 3 decoding transposed convolutional layers to output a 50-band hyperspectral feature vector. The discriminator contains 4 convolutional layers and 1 fully connected layer, and outputs the true / false discrimination result.

[0070] The model training adopts the mini-batch gradient descent method with a batch size of 32, the learning rate is initially set to 0.0002, and the Adam optimizer is used. During the training process, the generator is updated once every 5 times of training the discriminator to prevent the generator from crashing prematurely. To ensure the quality of the generated samples, spectral smoothness constraints and inter-band correlation constraints are introduced. After the training is completed, 100 new pseudo-hyperspectral microbial data pairs are generated by inputting different microbial parameter conditions.

[0071] When constructing the semi-supervised learning framework, the MeanTeacher structure is adopted. Both the student model and the teacher model adopt a four-layer convolutional neural network structure. For 10 real labeled samples and 100 generated pseudo-labeled samples, the label prediction error is calculated as the supervised loss; for 15 unlabeled samples, different data augmentations (such as random band masking, adding random noise, etc.) are applied, and through the predictions of the student model and the teacher model, the mean square error of the two prediction results is calculated as the consistency loss. During the model training, the consistency loss weight coefficient is set to 0.5 and linearly increases to 1.0 as the training progresses; the regularization loss weight coefficient is set to 0.001.

[0072] The constructed PINN network contains 5 hidden layers, with 64 neurons in each layer, and the Swish activation function is used. The input features include: a 20-dimensional hyperspectral feature vector extracted from the discriminator trained by GAN, a 5-dimensional meteorological data vector (obtained from the nearest weather station), a 2-dimensional spatial coordinate (farmland grid coordinate), and a 1-dimensional depth coordinate (within the range of 0 - 100 cm).

[0073] In terms of physical constraints, the Richards water movement equation and the heat conduction equation are introduced. The parameter settings include: the unsaturated hydraulic conductivity of the soil is parameterized using the van Genuchten model, the soil heat capacity and the heat conduction coefficient are estimated based on the soil texture and organic matter content. In the loss function, the data fitting loss weight is set to 1.0, and the physical constraint loss weight is initially set to 0.1 and gradually increases to 0.5 during the training process.

[0074] The training process uses the data of 5 profile sampling points, with a total of 25 depth-level observation data. At the same time, the pseudo-samples generated by GAN are used to assist in training to enhance the model's inference ability for profile information under different soil conditions. After the training is completed, the model can infer the microbial abundance, soil water content, and temperature distribution at different depths (0 - 100 cm) from the surface hyperspectral data at any location.

[0075] Based on the PINN framework, the nitrogen cycle dynamics equation is embedded. In the nitrification process equation, the conversion of ammonium nitrogen to nitrate nitrogen is affected by microbial abundance, soil moisture and temperature. In implementation, the moisture effect function The parameter settings are: , , ; Temperature influence function middle The value is set to 2.0, the reference temperature is 20℃.

[0076] The expanded PINN network adds four output nodes: ammonium nitrogen concentration, nitrate nitrogen concentration, nitrification rate constant, and denitrification rate constant. To ensure that the parameters are physically reasonable, the nitrification rate constant range is set to 0.01 to 0.5d , the denitrification rate constant ranges from 0.001 to 0.1d The network output is mapped to a reasonable parameter range through the sigmoid function.

[0077] The data assimilation process employed an ensemble Kalman filter to construct 100 ensemble members, each containing different initial PINN and soil parameters. Based on observations of ammonium and nitrate concentrations at five profile sampling points (a total of 25 depth levels), the ensemble member parameters were updated through multiple iterations. Ultimately, the spatial distribution of nitrification and denitrification rates at different locations and depths across the entire farmland was inverted.

[0078] Based on the trained integrated model, a hyperspectral remote sensing monitoring system was conducted on an entire 100-hectare farmland area. The trained model system was fed with hyperspectral image data (2m spatial resolution, 2500 pixels) from the entire farmland acquired by a drone.

[0079] Input data preparation: The hyperspectral imagery of the entire farmland was preprocessed, including radiometric correction, geometric registration, and spectral feature extraction, ultimately obtaining a 50-dimensional hyperspectral feature vector for each pixel. Meteorological data for the corresponding period (average daily temperature of 15.2°C, relative humidity of 65%, and precipitation of 12 mm) and the spatial coordinates of each pixel were also collected.

[0080] Model prediction calculation: The preprocessed hyperspectral data is predicted through three sub-models in turn: First, the surface microbial community parameters were predicted using a semi-supervised learning model to obtain the abundance of nitrifying bacteria, denitrifying bacteria, and microbial diversity index for each pixel. Then, the soil profile information is inferred through the PINN model to obtain the microbial abundance distribution, soil water content, and temperature distribution at each 20-cm layer within the depth range of 0-100 cm; Finally, the nitrification rate constant and denitrification rate constant at each depth layer are obtained through the nitrogen cycle parameter inversion model.

[0081] Monitoring result output: Through model prediction and calculation, the complete remote sensing monitoring results within the farmland are obtained, including: Spatial distribution map of soil microbial communities: Spatial distribution maps of nitrifying bacteria, denitrifying bacteria, and total microbial diversity are generated, and 3 high-nitrification activity areas, 2 high-denitrification activity areas, and 4 microbial diversity hotspots are identified; Profile microbial abundance distribution: The three-dimensional distribution of microbial abundance at 5 depth layers (0-20 cm, 20-40 cm, 40-60 cm, 60-80 cm, 80-100 cm) is provided, showing that the microbial abundance decreases with depth, but there is a local enrichment phenomenon in the 40-60 cm layer; Distribution of nitrogen cycle process parameters: Spatial distribution maps of nitrification rate and denitrification rate are obtained. The nitrification rate varies within the range of 0.05-0.35 d-1-1, and the denitrification rate varies within the range of 0.002-0.08 d-1-1; Soil environmental status information: The soil water content varies between 22% and 38%, and the soil temperature varies between 13.5 and 16.8 °C, showing obvious spatial heterogeneity.

[0082] Uncertainty assessment: Using the ensemble prediction method, the 95% confidence intervals of each monitoring result are calculated. The results show that the average uncertainty of microbial abundance prediction is ±12.5%, and the average uncertainty of nitrogen cycle parameter prediction is ±18.3%. An uncertainty distribution map is generated, identifying that the prediction reliability of an area of approximately 15 hectares in the southeast corner of the farmland is relatively low (uncertainty > 25%), and it is recommended to increase verification sampling in this area.

[0083] Based on the inversion results, the farmland is divided into 5 fertilization management areas. For areas with high nitrifying bacteria abundance and fast nitrification rate, reduce the amount of nitrogen fertilizer applied at one time and adopt a strategy of applying it in several times; for areas with strong denitrification, increase drainage measures to reduce the anaerobic environment, and at the same time use nitrification inhibitors to delay the conversion of ammonium nitrogen to nitrate nitrogen, improve nitrogen fertilizer utilization rate, and reduce nitrogen loss.

[0084] As Figures 6 to 8 shown, this application example focuses on verifying two main technical effects: the improvement of data utilization efficiency and the breakthrough in nitrogen cycle parameterization.

[0085] In this application scenario, compared with the traditional method that requires a large amount of on-site sampling and laboratory analysis, the present invention only uses 10 labeled samples and 15 unlabeled samples, reducing the data acquisition cost and time. Through GAN data augmentation and semi-supervised learning, the prediction accuracy has reached the level of a traditional model trained with 30 labeled samples.

[0086] Specifically, the root mean square error (RMSE) for predicting the abundance of nitrifying bacteria is compared as follows: traditional supervised learning method (30 labeled samples): 0.42; traditional supervised learning method (10 labeled samples): 0.68; the present invention (10 labeled samples + GAN augmentation + semi-supervised learning): 0.39. This shows that in the case of reducing the data volume by about 67%, the present invention not only maintains the prediction accuracy but also slightly improves it (the RMSE is reduced by about 7%).

[0087] At the same time, the present invention significantly reduces the sampling and analysis costs. The traditional method requires microbial community sequencing and functional gene analysis at 30 points, with a total cost of about 45,000 yuan and a time consumption of about 15 days; while the present invention only requires microbial analysis at 10 points, with a cost of about 15,000 yuan and a time consumption of about 7 days (including model training time).

[0088] The present invention has successfully realized the inversion of key parameters of the soil profile nitrogen cycle from surface hyperspectral data, providing key support for precision nitrogen fertilizer management. The traditional method needs to obtain these parameters through indoor incubation experiments or in-situ tracer experiments, with high determination costs, long cycles, and limited spatial representativeness at each point.

[0089] The accuracy of the inverted parameters is verified by comparing with laboratory measured values. At 25 depth levels of 5 profile verification points, the correlation coefficient between the inverted nitrification rate constant and the indoor incubation measured value reaches 0.82, and the average relative error is 16.8%; the R of the inverted denitrification rate constant and the measured value 2 is 0.78, and the average relative error is 19.3%. This accuracy has good practical value for large-scale soil nitrogen cycle parameter monitoring.

[0090] The adjustment of fertilization management based on the inverted parameters improves the nitrogen fertilizer use efficiency. In the comparative experiment in the farmland test area, the nitrogen fertilizer use efficiency in the traditional uniform fertilization area is 35.2%, while in the area with precision fertilization according to the present invention, the nitrogen fertilizer use efficiency has increased to 46.8%, with an increase of 33%. At the same time, the nitrate leaching monitoring results show that the nitrogen leaching amount in the precision fertilization area is reduced by 28.5% compared with the traditional fertilization area, reducing the environmental pollution risk.

[0091] In addition, the uncertainty assessment function of the present invention provides reliability information for decision-making. By quantifying the parameter uncertainty obtained from ensemble forecasting, a conservative fertilization strategy is recommended for low-reliability regions (where uncertainty is higher than 30%), further reducing the risk.

[0092] The embodiments of the present invention have been described above, but the embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative rather than restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. A hyperspectral remote sensing monitoring method for soil microbial communities, characterized in that, Including: Based on data augmentation and semi-supervised learning of generative adversarial networks, obtain hyperspectral remote sensing data and labeled hyperspectral microbial data pairs, construct a generative adversarial network to generate pseudo-hyperspectral microbial data pairs, construct a semi-supervised learning framework, and output a prediction model for microbial community parameters; Based on the profile inference of physics-informed neural networks, according to the output of the prediction model, combined with hyperspectral features, meteorological data and spatial coordinates, construct a physics-informed neural network and introduce the constraints of soil water and heat transport equations to infer the microbial distribution and water and heat status at different depths; Embed the nitrogen cycle kinetic equation into the physics-informed neural network, and use the profile microbial distribution as the driving force to output the key process parameters of the nitrogen cycle; Use the data assimilation method to combine the key process parameters of the nitrogen cycle and the profile microbial distribution with the observed data, input the hyperspectral data of the target area, generate the monitoring results of the spatial distribution of the soil microbial community, the profile microbial abundance distribution, and the nitrogen cycle parameter distribution, and conduct uncertainty assessment.

2. The hyperspectral remote sensing monitoring method for soil microbial communities according to claim 1, wherein The steps of constructing a generative adversarial network to generate pseudo-hyperspectral microbial data pairs include: Construct a generative adversarial network containing a generator and a discriminator. The generator receives a random noise vector and a microbial parameter conditional vector as inputs and generates hyperspectral data that conforms to the true distribution; Adopt an alternating optimization strategy to train the generator and the discriminator until the generator can generate pseudo-hyperspectral data that conforms to physical constraints; By inputting different microbial parameter conditional vectors, generate corresponding hyperspectral data to form pseudo-labeled samples; Construct a semi-supervised learning framework and use the hyperspectral data of the true distribution, the pseudo-hyperspectral data that conforms to physical constraints, and the pseudo-labeled samples for model training.

3. The hyperspectral remote sensing monitoring method for soil microbial communities according to claim 2, characterized in that The constructed semi-supervised learning framework includes a student model and a teacher model. The student model updates its parameters through supervised training, and the parameters of the teacher model are the exponential moving averages of the parameters of the student model; the loss function of the semi-supervised learning framework includes supervised loss, consistency loss, and regularization loss.

4. A hyperspectral remote sensing monitoring method for soil microbial communities according to claim 1, characterized in that The construction of the physics-informed neural network includes: Construct a neural network containing a forward propagation part and a physical constraint part; To ensure that the output of the neural network satisfies physical laws, introduce the physical equation constraints of the soil water and heat transport process; Construct a total loss function including data fitting loss and physical constraint loss; Use a generative adversarial network to generate pseudo-profile data that satisfies physical constraints.

5. A hyperspectral remote sensing monitoring method for soil microbial communities according to claim 4, characterized in that, The physical equation constraints include the water transport equation and the heat conduction equation.

6. The hyperspectral remote sensing monitoring method for soil microbial communities according to claim 1, wherein The steps of embedding the nitrogen cycle kinetic equation into the physics-informed neural network include: Add a description equation for the key process of the nitrogen cycle in the physics-informed neural network framework; Expand the output variables of the physics-informed neural network, increase the prediction of the ammonium nitrogen and nitrate nitrogen concentrations, and the output of the nitrification rate and denitrification rate changes with depth; Expand the physical constraint loss of the physics-informed neural network and add the residual term of the nitrogen cycle equation; Add parameter value range constraints to ensure that the inverted parameters are physically reasonable.

7. A hyperspectral remote sensing monitoring method for soil microbial communities according to claim 6, characterized in that, The description equations for the key processes of the nitrogen cycle include the nitrification process equation and the denitrification process equation, which describe the dynamic change processes of ammonium nitrogen and nitrate nitrogen, as well as the effects of microbial abundance, water content, temperature, and organic carbon on the nitrification and denitrification processes.

8. The hyperspectral remote sensing monitoring method for soil microbial communities according to claim 1, wherein The steps of using the data assimilation method include adopting the ensemble Kalman filter method, and updating the parameter estimation when the observation data arrives by constructing multiple ensemble members of the model parameters.

9. The hyperspectral remote sensing monitoring method for soil microbial communities according to claim 1, characterized in that, The uncertainty assessment includes: Using the ensemble prediction method to calculate the prediction intervals of the monitoring results; Generating an uncertainty distribution map to identify the areas with relatively low prediction reliability; Providing a confidence assessment of the prediction results to provide a reliability reference for decision-making applications; Outputting a quality control report of the monitoring results, including model performance indicators and applicability evaluations.

10. A storage medium, characterized in that, It includes a memory and one or more processors. Executable code is stored in the memory. When the one or more processors execute the executable code, it is used to implement a hyperspectral remote sensing monitoring method for soil microbial communities described in any one of claims 1-9.

Images