Multi-source data fusion diagnosis method for grassland water stress state

Abstract

The invention discloses a multi-source data fusion diagnosis method for a grassland water stress state, and relates to the technical field of grassland ecological hydrology monitoring. Sequentially performing system deployment, macroscopic wide-area scanning to generate a preliminary diagnostic graph, near-earth targeted observation to generate a verification data set and mechanism model data assimilation, and finally outputting a water stress level distribution graph. The problems that in the prior art, observation means are isolated, diagnosis results stay in representation, mechanism interpretation is insufficient, and system flexibility is poor are solved. According to the method, accurate and deep diagnosis of the vegetation water stress state is realized through multi-level cooperation and mechanism fusion, and the diagnosis accuracy and efficiency can be continuously improved through a self-learning optimization mechanism.

Description

Technical Field

[0001] This invention relates to the field of grassland eco-hydrological monitoring technology, specifically a multi-source data fusion diagnostic method for grassland water stress status. Background Technology

[0002] Currently, macroscopic monitoring technology based on satellite remote sensing has been widely used in large-scale vegetation condition assessment. This technology can perform preliminary inversion of regional vegetation moisture status by calculating remote sensing indicators such as vegetation indices and surface temperature. However, this type of technology has significant limitations: its spatial resolution is relatively low, and it is easily affected by weather conditions, leading to insufficient accuracy in identifying stressed areas. It struggles to capture subtle changes at the canopy scale and cannot distinguish between soil moisture stress and similar symptoms caused by other factors such as pests and diseases, thus often resulting in a high misjudgment rate.

[0003] To compensate for the limitations of satellite remote sensing, near-Earth observation technologies such as drones and ground sensor networks are employed to acquire detailed data with high spatiotemporal resolution. However, these technologies typically operate independently, lacking effective coordination with macroscopic observations. Their limited observation range makes it difficult to cover vast areas, resulting in high deployment and maintenance costs and a waste of observational resources. Furthermore, existing diagnostic methods largely rely on empirical interpretations of directly observed indicators. This approach struggles to effectively reveal the intrinsic physiological and ecological mechanisms of vegetation under water stress and cannot quantify key physiological parameters such as water use efficiency. Consequently, diagnostic results often remain superficial descriptions, lacking predictive and mechanistic explanatory power, making it difficult to achieve accurate, in-depth, and prospective diagnosis of grassland water stress. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-source data fusion diagnostic method for grassland water stress. By sequentially performing wide-area scanning to generate a preliminary diagnostic map, a drone and sensor are guided to conduct precise observations. Finally, a water-carbon coupling model is used to fuse multi-source data, and a sequential assimilation algorithm is used to optimize internal state variables, thereby achieving accurate estimation of root soil moisture content and vegetation water use efficiency, and generating a stress level distribution map. This method overcomes the shortcomings of traditional methods, such as superficial diagnosis and poor mechanistic interpretation, and achieves accurate quantification of the intrinsic physiological state of water stress, thus solving the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A multi-source data fusion diagnostic method for grassland water stress is proposed. This method is based on a collaborative system comprising a macroscopic observation layer, a near-ground sensing layer, and a mechanism diagnostic layer, and includes the following steps:

[0007] Initial system deployment, configuring data sources and model parameters for each layer;

[0008] The macroscopic observation layer performs wide-area scanning to acquire and process multi-source satellite remote sensing data and meteorological data, and generates a preliminary diagnostic map reflecting the probability of regional moisture status.

[0009] The near-ground perception layer initiates targeted observations based on the guidance of the preliminary diagnostic map, uses UAVs and ground sensor networks to acquire high spatiotemporal resolution data, and fuses them to generate a verification dataset for verification and calibration.

[0010] The mechanism diagnostic layer runs a water-carbon coupling process model and uses a sequential assimilation algorithm to fuse meteorological data from the macroscopic observation layer, remote sensing inversion products, and validation datasets from the near-ground sensing layer to constrain and optimize the model state variables and output core variables that reflect the vegetation water stress state.

[0011] Based on the optimized model state variables, a distribution map of grassland water stress levels is generated.

[0012] Preferably, the initial deployment steps of the system include:

[0013] Clearly define the boundaries, grassland types, and terrain features of the target monitoring area;

[0014] Configure data source access channels connecting multiple types of satellites and regional meteorological data services for the macroscopic observation layer;

[0015] The near-ground sensing layer is equipped with a drone platform carrying a multispectral camera and a thermal infrared imager, and IoT sensor nodes containing multi-layer soil temperature and humidity sensors and canopy micro-weather stations are deployed in the monitoring area.

[0016] The physiological and ecological parameters of the water-carbon coupling process model are initialized for the mechanism diagnostic layer.

[0017] Preferably, the step of the macroscopic observation layer performing a wide-area scan includes:

[0018] Satellite images are acquired at a preset frequency and then preprocessed with radiometric calibration, atmospheric correction, geometric fine correction and cloud masking to generate standard products.

[0019] Parallel computation of multiple remote sensing indices and meteorological drought indices;

[0020] The remote sensing index and the meteorological drought index are registered and overlaid on a unified spatial grid. The preliminary diagnostic map is generated using multivariate logistic regression or multi-index weighting methods, and a diagnostic uncertainty assessment is attached.

[0021] Preferably, the steps for the near-Earth sensing layer to initiate targeted observations include:

[0022] Receive the preliminary diagnostic map and its uncertainty assessment information, and combine it with the distribution of fixed observation plots and equipment constraints to generate the optimal UAV flight path and data acquisition strategy;

[0023] Control the drone to perform the mission and simultaneously acquire canopy spectral and thermal infrared images;

[0024] Control the ground-based IoT sensor network to continuously record soil volumetric water content, soil temperature, and canopy air temperature data.

[0025] Preferably, the step of generating the validation dataset further includes:

[0026] The images collected by the drone were stitched together, orthorectified, and processed for radiometric consistency to extract high-resolution vegetation index and canopy temperature distribution maps.

[0027] Information extracted by UAVs and measurement data from ground sensors are fused using spatial interpolation to generate a refined validation dataset that matches the pixel scale of the macroscopic observation layer. The validation dataset includes soil moisture distribution maps, average canopy temperature, and coefficient of variation.

[0028] Preferably, the step of fusing data using a sequential assimilation algorithm in the mechanism diagnosis layer includes:

[0029] Run a set of model state variables, where each set member represents a possible state trajectory, and predict the values ​​of each observable state variable;

[0030] When the actual observation data arrives, compare the observed values ​​with the predicted values ​​of all ensemble members;

[0031] Based on the differences, all model state variables are adjusted in reverse according to Bayes' theorem to achieve the best match between the model simulation trajectory and the observed facts.

[0032] Preferably, the core variables of the output include root zone soil moisture content and vegetation water use efficiency;

[0033] The generated grassland water stress level distribution map is achieved by classifying the core variables based on a pre-determined classification threshold.

[0034] Preferably, the method further includes a self-learning optimization step:

[0035] Archive the input data, intermediate results, assimilated model state, final diagnostic conclusions, and subsequent field validation data for each diagnostic task to build a diagnostic case knowledge base;

[0036] Regularly utilize the diagnostic case knowledge base to perform optimization tasks.

[0037] Preferably, the optimization tasks in the self-learning optimization step include:

[0038] The first optimization task is to re-estimate and optimize the key physiological and ecological parameters related to specific grassland types and soil characteristics in the water-carbon coupling process model by assimilating multi-source observation data from historical periods.

[0039] The second optimization task is to analyze historical cases using machine learning algorithms, dynamically adjust the weights and rules of the remote sensing index combinations used in the preliminary diagnosis of the macroscopic observation layer, and optimize the path planning algorithm for UAV inspections in the near-ground perception layer.

[0040] Preferably, the macroscopic observation layer, the near-ground sensing layer, and the mechanism diagnosis layer are connected via a standardized data interface and an automated task scheduling mechanism.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] 1. This invention constructs a collaborative diagnostic framework that deeply integrates macroscopic observation, near-ground sensing, and mechanism diagnosis. Through standardized data interfaces and task scheduling mechanisms, it organically integrates large-scale satellite scanning, high-precision UAV and sensor verification, and a water-carbon coupling model based on physical processes. This realizes the transformation from observation of superficial data to interpretation of deep mechanisms, significantly improving the systematicness, accuracy, and adaptability of diagnosis.

[0043] 2. This invention achieves accurate quantitative diagnosis of vegetation water stress status through sequential assimilation of multi-source heterogeneous data and model constraints. In the process of integrating macroscopic remote sensing information and near-ground fine observation data, the model not only optimizes directly observable variables, but also inversely estimates key intrinsic physiological state variables such as root zone soil moisture content and water use efficiency. This makes the diagnostic results no longer limited to surface indices, but can reveal the actual impact of water stress on plant physiological functions. It effectively overcomes the shortcomings of traditional methods that rely solely on empirical thresholds, are susceptible to interference, and have unclear mechanisms, thus improving the accuracy and reliability of the diagnosis.

[0044] 3. The present invention sets up a self-learning optimization mechanism, which enables the system to continuously evolve. By continuously archiving diagnostic cases and using machine learning to analyze historical data, the system can dynamically optimize the localized parameters of the mechanistic model, the diagnostic rules at the macro level, and the observation strategies at the near-surface level. This allows the entire system to learn from actual operating experience, gradually reduce its dependence on initial prior knowledge, and automatically adapt to different grassland types, seasonal changes, and environmental dynamics. Ultimately, this leads to an increase in diagnostic efficiency and accuracy, as well as the optimized allocation of observation resources. Attached Figure Description

[0045] Figure 1 This is a flowchart of the multi-source data fusion diagnostic method for grassland water stress state according to the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] To address the problems of isolated observation methods, superficial diagnostic results, insufficient mechanistic interpretation, and poor system flexibility in existing technologies, please refer to [link / reference]. Figure 1 This embodiment provides the following technical solution:

[0048] A multi-source data fusion diagnostic method for grassland water stress is proposed. This method is based on a multi-source data fusion diagnostic system, which mainly includes a macroscopic observation layer, a near-ground sensing layer, and a mechanism diagnostic layer.

[0049] The macroscopic observation layer is used to obtain preliminary surface information over a large area and periodically, and to identify potential stress areas through satellite remote sensing, and to conduct preliminary detection and screening.

[0050] The near-ground sensing layer is used, under the guidance of the macroscopic observation layer, to acquire detailed surface information with high spatiotemporal resolution for key areas or fixed observation plots using UAVs and ground sensor networks, and to conduct on-site verification and precision measurement.

[0051] The mechanism diagnosis layer is used to fuse and interpret multi-source heterogeneous data input from the macroscopic observation layer and the near-ground sensing layer. By running a mechanism model that simulates the physiological and energy exchange processes of vegetation water, it outputs variables that can reflect the state of vegetation water stress.

[0052] The three layers are connected via standardized data interfaces and automated task scheduling mechanisms.

[0053] I. Initial System Deployment

[0054] First, the boundaries, main grassland types, and terrain features of the grassland target monitoring area are clearly defined.

[0055] The macroscopic observation layer is equipped with data source access channels that connect to optical satellites, thermal infrared satellites, and microwave satellites to acquire multispectral information, surface temperature information, and surface soil moisture information. It also accesses regional meteorological data services to acquire data on driving factors such as temperature, precipitation, and solar radiation within the target monitoring area.

[0056] The near-ground perception layer is equipped with UAVs and control platforms capable of autonomous flight path planning and execution, and is equipped with high-precision multispectral cameras and thermal infrared imagers. In addition, several fixed observation plots are selected in the target monitoring area, and IoT sensor nodes are deployed. Each IoT sensor node includes multi-layer soil temperature and humidity sensors and canopy micro-weather stations.

[0057] The mechanistic diagnostic layer selects a water-carbon coupling process model as the basis, such as a model based on the energy balance and photosynthesis-transpiration coupling relationship, and initializes the physiological and ecological parameters of the model according to the vegetation type of the target monitoring area. The physiological and ecological parameters include the maximum carboxylation rate, stomatal conductance model parameters, root distribution function and soil hydraulic properties parameters.

[0058] II. Wide-area scanning of the macroscopic observation layer

[0059] The wide-area scanning process is automatically triggered at a preset time frequency.

[0060] The macroscopic observation layer first acquires the latest images of the target monitoring area from the satellite platform. After preprocessing processes such as radiometric calibration, atmospheric correction, geometric fine correction, and cloud masking, standard surface reflectance and surface temperature products are generated.

[0061] A series of remote sensing indices are calculated in parallel based on multispectral information, surface temperature information, and surface soil moisture information, including but not limited to vegetation index, moisture index, and moisture stress index based on the spatial relationship between temperature and vegetation index.

[0062] Process concurrent gridded meteorological data to calculate a meteorological drought index, including a standardized precipitation evapotranspiration index.

[0063] Remote sensing indices and meteorological drought indices are registered and overlaid on a unified spatial grid within the target monitoring area. Using multivariate logistic regression or multi-index weighting methods, a preliminary diagnostic map reflecting the probability of water status in the target monitoring area is generated. The preliminary diagnostic map not only identifies the spatial range and possible level of suspected water stress, but also includes an assessment of diagnostic uncertainty, providing a basis for decision-making in subsequent refined observation resource allocation.

[0064] III. The near-Earth sensing layer initiates targeted observations based on the output results of the macroscopic observation layer.

[0065] The near-ground perception layer receives the preliminary diagnostic map and its uncertainty assessment information, and combines the distribution location of fixed observation plots, UAV endurance, and airspace control constraints to generate the optimal UAV flight path and ground sensor data acquisition strategy.

[0066] The drone automatically executes flight missions according to instructions, and its onboard multispectral camera and thermal infrared imager simultaneously acquire canopy spectral and thermal infrared images with spatial resolution ranging from centimeters to meters.

[0067] During the drone's flight, the ground-based IoT sensor network continuously records data from various nodes at a high temporal frequency, including soil volumetric water content, soil temperature, and canopy air temperature.

[0068] After the observation mission is completed, the canopy spectral images and thermal infrared images collected by the UAV are stitched together, orthorectified and radiometrically consistent, and high-resolution vegetation index and canopy temperature distribution map are extracted.

[0069] Information collected by drones and measurement data from IoT sensors are fused using spatial interpolation methods to generate a validation dataset that matches the pixel scale of the macroscopic observation layer but with higher accuracy. This validation dataset includes, but is not limited to, refined soil moisture distribution maps, average canopy temperature, coefficient of variation, and multi-source observation data sequences. This data can directly verify and calibrate the initial diagnostic results of the macroscopic observation layer, and simultaneously serve as highly reliable observations input into the mechanistic diagnostic layer, driving the deep data fusion and model assimilation process.

[0070] IV. Mechanism Diagnostic Layer Operational Diagnostic Model

[0071] Run a dynamic water-carbon coupling process model with a time step typically in the form of days or hours.

[0072] At the beginning of each diagnostic cycle, the model uses the assimilated state of the previous cycle as the initial condition, and the meteorological data provided by the macro observation layer as the external forced input of the model. The regional vegetation index and surface temperature inverted by the macro observation layer, as well as the fine soil moisture and canopy temperature data provided by the near-ground sensing layer, are all defined as observable state variables of the model.

[0073] The mechanism diagnostic layer employs a sequential assimilation algorithm, which runs a set of model state variables, each set member representing a possible state trajectory. During the forward simulation process, the model predicts the values ​​of each observable state variable for each set member at future observation times.

[0074] When actual observation data from the macroscopic observation layer and the near-ground sensing layer arrive, the sequential assimilation algorithm compares the observed values ​​with the predicted values ​​of all set members. Based on the differences, it adjusts all model state variables in reverse according to Bayes' theorem, including both directly observable and indirectly observable variables.

[0075] Through continuous assimilation of multi-time-period, multi-source data, the model's state is progressively constrained and optimized, and its simulated trajectory achieves optimal agreement with all independent observations. At this point, the model's internal state variables are considered the current optimal estimates.

[0076] Among them, the root zone soil moisture content can be directly used to assess the water storage status of the soil reservoir and serve as a direct indicator of the degree of water stress; vegetation water use efficiency, i.e. the amount of carbon fixed per unit of water consumption, is a more sensitive indicator of the impact of water stress on plant physiological functions, and its abnormal reduction usually indicates severe water stress.

[0077] The system classifies core state variables based on pre-determined grading thresholds obtained through experiments or literature, and finally generates a spatiotemporally continuous grassland water stress level distribution map, while simultaneously outputting derived eco-hydrological variables such as evapotranspiration and gross primary productivity (GPP).

[0078] V. Self-learning optimization

[0079] The system also integrates a self-learning optimization program to systematically archive all data and results from each complete diagnostic task, including input data at each layer, intermediate results, assimilated model state and final diagnostic conclusions, as well as subsequently collected field validation data, forming a continuously enriched diagnostic case knowledge base.

[0080] The self-learning optimization program periodically performs two optimization tasks using a diagnostic case knowledge base:

[0081] First, optimize the localization parameters of the mechanism model. By assimilating a large amount of multi-source observation data from historical periods, re-estimate and optimize the key parameters in the model that are related to specific grassland types and soil properties, so that the model's eco-hydrological process simulation is more in line with the actual local conditions.

[0082] Second, we optimize the diagnostic rules of the macro-observation layer and the task planning strategy of the near-ground perception layer. By analyzing historical cases through machine learning algorithms, we can find out which combinations of remote sensing indices have the strongest predictive ability for the water stress level output by the assimilation model under specific seasons and grassland types. This allows us to dynamically adjust the weights and rules of the preliminary diagnosis of the macro-observation layer and improve its screening efficiency. By analyzing which areas have had the most effective reduction in diagnostic uncertainty in historical tasks, we can optimize the path planning algorithm for UAV inspections and achieve the optimal allocation of observation resources.

[0083] The self-learning optimization mechanism enables the entire system to continuously learn from historical experience, reduce reliance on prior knowledge, and gradually adapt to the dynamic changes in environmental conditions and vegetation status.

[0084] Working Principle: This invention is based on a multi-source data fusion diagnostic system, which consists of a macroscopic observation layer, a near-ground sensing layer, and a mechanism diagnostic layer. Through layered collaboration, it achieves accurate diagnosis of grassland water stress status. The system first undergoes initial deployment, defining the monitoring area boundaries and grassland types, configuring satellite data sources, UAV platforms, and ground sensor networks, and initializing the parameters of the water-carbon coupling mechanism model.

[0085] The macroscopic observation layer periodically triggers wide-area scans, acquiring multispectral, surface temperature, and soil moisture information through satellite remote sensing. Combined with meteorological data, it calculates remote sensing and drought indices, generating preliminary diagnostic maps that identify suspected water stress areas and their uncertainties, providing guidance for subsequent observations. The near-ground sensing layer, based on the output of the macroscopic layer, initiates targeted observations. UAVs collect high-resolution canopy images along planned routes, while ground sensors continuously record soil and canopy data. This data is processed to generate a validation dataset, used to calibrate the macroscopic observation results and input into the mechanistic diagnostic layer.

[0086] The mechanistic diagnostic layer operates a dynamic water-carbon coupling model, driven by meteorological data. It employs a sequential assimilation algorithm to fuse multi-source observational data, comparing predicted and actual observations to inversely adjust model state variables. This outputs core indicators such as root zone soil moisture content and vegetation water use efficiency, ultimately generating a water stress level distribution map and derived variables. The system also integrates a self-learning optimization mechanism, archives historical diagnostic cases, and optimizes model parameters and observation strategies through machine learning, continuously improving diagnostic accuracy and efficiency to adapt to dynamic environmental changes.

[0087] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0088] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A multi-source data fusion diagnostic method for grassland water stress status, characterized in that, Includes the following steps: Step 1: Initial Deployment: Define the boundaries, grassland types, and terrain features of the grassland target monitoring area, and configure the data sources, equipment, and model parameters for each layer; Step 2, Wide-area scanning and initial diagnosis: Wide-area scanning is triggered periodically to acquire multispectral, thermal infrared and microwave data of the target monitoring area from multiple types of satellite platforms, and regional meteorological data is also connected. After preprocessing, multiple remote sensing indices and meteorological drought indices are calculated in parallel. Then, registration and overlay analysis are performed on a unified spatial grid through multivariate logistic regression or multi-index weighting methods to generate an initial diagnostic map that identifies the spatial range, probability level and diagnostic uncertainty of suspected water stress. Step 3: Targeted observation and data fusion: Based on the guidance of the preliminary diagnostic map, plan and execute UAV flight missions to acquire high-resolution canopy images, while activating the ground sensor network for continuous data recording, and process and spatially fuse the acquired data to generate a refined validation dataset. Step 4, State Diagnosis: Run a water-carbon coupled process model and use a sequential assimilation algorithm to update and optimize the set of model state variables by using meteorological driving data provided by the macroscopic observation layer, remote sensing inversion products, and the validation dataset provided by the near-ground sensing layer as observation constraints, and output the root layer soil moisture content and vegetation water use efficiency as core diagnostic variables.

2. The multi-source data fusion diagnostic method for grassland water stress state according to claim 1, characterized in that, For step one, the initial deployment specifically includes: Configure data source access channels for the macroscopic observation layer to connect optical satellites, thermal infrared satellites, and microwave satellites; The near-ground perception layer is equipped with a drone platform with autonomous flight path planning capabilities and its onboard high-precision multispectral camera and thermal infrared imager. In addition, IoT sensor nodes containing multi-layer soil temperature and humidity sensors and canopy micro-weather stations are deployed in the target monitoring area. A water-carbon coupling process model was selected as the basis for the mechanism diagnostic layer, and the physiological and ecological parameters of the model were initialized according to the vegetation type of the target monitoring area, including the maximum carboxylation rate, stomatal conductance model parameters, root distribution function and soil hydraulic properties parameters.

3. The multi-source data fusion diagnostic method for grassland water stress state according to claim 1, characterized in that, For step two, the calculated remote sensing indices include vegetation index, moisture index, and moisture stress index based on the spatial relationship between temperature and vegetation index. The calculated meteorological drought index includes standardized precipitation evapotranspiration index.

4. The multi-source data fusion diagnostic method for grassland water stress state according to claim 1, characterized in that, For step three, the specific process of generating the validation dataset includes: The canopy spectral images and thermal infrared images collected by UAVs were stitched together, orthorectified, and processed for radiometric consistency to extract high-resolution vegetation indices and canopy temperature distribution maps. By integrating vegetation index, canopy temperature distribution map, and soil volumetric water content, soil temperature, and canopy air temperature data recorded by ground-based IoT sensor networks using spatial interpolation methods, a refined soil moisture distribution map, average canopy temperature, and coefficient of variation that match the pixel scale of the macroscopic observation layer are generated.

5. The multi-source data fusion diagnostic method for grassland water stress state according to claim 1, characterized in that, For step four, the specific process of the sequential assimilation algorithm includes: Run a set of model state variables, where each set member represents a possible state trajectory, and predict the values ​​of each observable state variable at future observation times; When actual observation data from the macroscopic observation layer and the near-Earth sensing layer arrive, compare the observed values ​​with the predicted values ​​of all ensemble members; Based on the differences, all model state variables are adjusted in reverse according to Bayes' theorem to make the model's simulated trajectory best match all independent observation facts.

6. The multi-source data fusion diagnostic method for grassland water stress state according to claim 1, characterized in that, The method also includes a result generation and self-learning optimization step, which classifies the core diagnostic variables according to a predetermined grading threshold, generates a spatiotemporally continuous grassland water stress level distribution map, and archives the data of the entire diagnostic process to update the diagnostic case knowledge base for periodic optimization of model parameters and decision-making strategies at each level.

7. The multi-source data fusion diagnostic method for grassland water stress state according to claim 6, characterized in that, The periodic optimization of decision-making strategies at each level includes: By assimilating multi-source observation data from historical periods, key parameters related to specific grassland types and soil properties in the water-carbon coupling process model are re-estimated and optimized. By analyzing historical diagnostic cases using machine learning algorithms, the weights and rules of the remote sensing index combinations used in the preliminary diagnosis of the macroscopic observation layer are dynamically adjusted, and the path planning algorithm for UAVs in the near-ground perception layer is optimized.

8. The multi-source data fusion diagnostic method for grassland water stress state according to claim 1, characterized in that, When planning the flight path of the UAV in the near-ground perception layer, the suspected stress areas marked in the preliminary diagnosis map, the diagnostic uncertainty assessment information, the distribution of fixed observation plots, the UAV's endurance and airspace control constraints are comprehensively considered.

9. The multi-source data fusion diagnostic method for grassland water stress state according to claim 1, characterized in that, At the beginning of each diagnostic cycle, the mechanism diagnostic layer uses the model state assimilated from the previous cycle as the initial condition. While outputting the distribution map of water stress levels, the mechanism diagnostic layer simultaneously outputs evapotranspiration and eco-hydrological variables derived from total primary productivity (GPP).

10. The multi-source data fusion diagnostic method for grassland water stress state according to claim 1, characterized in that, The time step of the water-carbon coupling process model is either a day or an hour.

Images