Terraced field structure coupling model construction and simulation method based on multi-source data

Abstract

The invention relates to the technical field of water and soil engineering monitoring, and particularly discloses a terrace structure coupling model construction and simulation method based on multi-source data, and the method comprises the steps: synchronously obtaining a time sequence hyperspectral image, topographic survey and soil physical attribute data; correlation analysis is carried out on the spectral features and the soil hydraulic attributes, and a key feature parameter set is extracted; establishing a dynamic mapping relation from the spectral characteristics to the soil hydraulic parameters, and inverting a soil hydraulic parameter field with a vertical layered structure; taking the parameter field as core dynamic input, and combining terrace unit network topology defined by terrains to construct a parameterized terrace structure coupling model for simulating the coupling process of hydrology and erosion; during simulation, an inversion result of a subsequent time sequence is introduced, model parameters are dynamically updated through a data assimilation algorithm and fed back to the model, simulation of a next time sequence is driven, and closed-loop optimization is formed; according to the invention, the precision and the adaptive capability of the water and sediment process simulation of the terrace system are improved.

Description

Technical Field

[0001] This invention relates to the field of water and soil engineering monitoring technology, specifically to a method for constructing and simulating a terraced structure coupling model based on multi-source data. Background Technology

[0002] Terraced fields, as a key agricultural engineering measure for soil and water conservation, play a vital role in preventing soil erosion, regulating regional water cycles, and ensuring food security due to their structural stability and hydrological and ecological functions. Accurate simulation of the hydrological response and soil erosion processes of terraced field systems is fundamental to assessing their benefits, optimizing their design, and achieving sustainable management. Currently, this technical field mainly relies on distributed hydrological and soil erosion models based on physical processes or experience for simulation analysis.

[0003] The existing technology has the following shortcomings:

[0004] Existing terraced field hydrological erosion models, relying on static soil parameters, cannot accurately simulate the dynamic changes of soil properties (such as saturated hydraulic conductivity) under the influence of rainfall and erosion. By integrating time-series hyperspectral remote sensing inversion and data assimilation techniques, real-time dynamic updates and feedback corrections of the model's core parameter fields are achieved. This enables the model to accurately respond to the real evolution of soil structure over time, thereby significantly improving the accuracy and reliability of water and sediment process simulation in terraced field systems. Summary of the Invention

[0005] The purpose of this invention is to provide a method for constructing and simulating a coupled model of terraced field structures based on multi-source data, so as to solve the problems mentioned above.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A method for constructing and simulating a coupled model of terraced field structures based on multi-source data includes the following steps:

[0008] S1: Acquire multi-source data of the target terraced field area; the multi-source data includes: time-series hyperspectral image data, topographic measurement data, and soil physical property data collected simultaneously in the field;

[0009] S2: Based on multispectral image data and soil physical property data, conduct correlation analysis between spectral features and deep soil hydraulic properties, and extract and generate a set of key feature parameters that can characterize the dynamic changes in soil structure.

[0010] S3: Based on the key feature parameter set, establish a dynamic mapping relationship from spectral features to soil hydraulic parameters, and perform inversion processing on the target terrace area according to the dynamic mapping relationship to generate real-time inversion results of soil hydraulic parameter fields with vertical stratification structure for each spatial unit;

[0011] S4: The real-time inversion results are configured as the core dynamic parameter set in the basic framework of the terraced structure coupling model to form a parameterized terraced structure coupling model, which is used to simulate the coupling effect of hydrological processes and soil erosion processes.

[0012] S5: During the simulation, the terraced structure coupling model is run. During the simulation, the real-time inversion results of subsequent time series are introduced to dynamically update and optimize the parameter set in the terraced structure coupling model. The updated parameter set is then fed back to the terraced structure coupling model to drive the terraced structure coupling model to simulate the next time series.

[0013] As a further aspect of the present invention: S2 specifically includes:

[0014] Hyperspectral image data is preprocessed and feature-enhanced to extract a first set of spectral indices sensitive to soil physical structure;

[0015] Based on the hydraulic properties and mechanical composition data in the soil physical property data, a set of second attribute parameters describing the stability of soil structure is generated.

[0016] A multivariate coupling analysis was performed on the first set of spectral indices and the second set of attribute parameters. By analyzing the synergistic change patterns of the first set of spectral indices and the second set of attribute parameters, a set of key feature parameters was selected and fused to generate a set of key feature parameters.

[0017] As a further aspect of the present invention: the multivariate coupling analysis specifically includes:

[0018] Construct a dynamic mapping matrix between each index in the first set of spectral indices and each parameter in the second set of attribute parameters;

[0019] Calculate the co-evolution gradient of each mapping relationship in the dynamic mapping matrix;

[0020] Based on the co-evolution gradient, the candidate parameters in the first spectral index set and the second attribute parameter set are ranked by importance, and parameters whose gradient values ​​exceed a preset threshold are selected.

[0021] The selected first spectral index and the second attribute parameter are normalized and fused to generate a set of key feature parameters.

[0022] As a further aspect of the present invention: S3 specifically includes:

[0023] Based on the soil profile layers, the key feature parameter set is vertically layered and reorganized to form feature parameter subsets corresponding to different soil layers;

[0024] Independent mapping units are trained based on a subset of feature parameters for each soil layer to establish a correspondence between spectral features and soil layer hydraulic parameters.

[0025] Based on the correspondence, synchronous layered inversion calculations are performed on each spatial unit to generate independent initial inversion values ​​of hydraulic parameters for each soil layer;

[0026] The initial inversion values ​​of different soil layers within the same spatial unit are vertically consistent and integrated to generate real-time inversion results of soil hydraulic parameter fields with vertical stratification.

[0027] As a further aspect of the present invention: the generation of independent initial inversion values ​​of hydraulic parameters for each soil layer specifically includes:

[0028] Based on the correspondence, dynamic inversion templates for different soil layers are generated. These templates define the matching rules between the spectral characteristic modes and hydraulic parameters of each soil layer.

[0029] Within each spatial cell, layered template matching calculations are performed in parallel by comparing and fitting the real-time spectral feature data of the spatial cell with the dynamic inversion templates of each soil layer layer by layer.

[0030] Based on the comparison and fitting results, the initial inversion values ​​of the independent hydraulic parameters of each soil layer in the spatial unit are calculated simultaneously.

[0031] As a further aspect of the present invention: the formation of the parameterized terraced structure coupling model specifically includes:

[0032] The real-time inversion results are combined with external driving data to form comprehensive input data. The real-time inversion results provide the soil hydraulic parameter field, and the external driving data include rainfall and evaporation sequences.

[0033] The integrated input data is loaded into the basic framework, and coupled simulation calculations are performed based on the terraced unit network topology and transport path defined in the basic framework.

[0034] Through coupled simulation calculations, the hydrological and erosion state variables of each terrace unit and the entire system are solved and output simultaneously. The erosion state variables include: surface runoff, soil moisture dynamics, and soil erosion and deposition in each terrace unit.

[0035] As a further aspect of the present invention: the execution of coupled simulation calculations specifically includes:

[0036] Based on the soil hydraulic parameter field and external driving data in the comprehensive input data, the calculation of water movement in the vertical layer of each terrace unit and the calculation of horizontal water transport between terrace units are executed sequentially within the terrace unit network topology to generate the water status output of each terrace unit.

[0037] Using the moisture status output of each terrace unit as input, and based on the transport path, within the network topology, the soil separation calculation of vertical stratification of each terrace unit and the sediment transport calculation between terrace units are executed simultaneously to generate the erosion status output of each terrace unit.

[0038] Integrate the water state output and erosion state output to form a complete result of hydrological and erosion state variables.

[0039] As a further aspect of the present invention: the dynamic updating and optimization specifically includes:

[0040] Based on the current simulated hydrological and erosion state variables of the terraced structure coupling model, the state differences are calculated by comparing them with the state variables measured in the same period or indirectly inferred from the real-time inversion results.

[0041] When the state difference exceeds the preset trigger threshold, the real-time inversion results of subsequent time series are used as the optimized observations, and the parameter set of the terraced structure coupling model currently in operation is used as the initial estimate.

[0042] The soil hydraulic parameters of each spatial unit in the parameter set are adjusted synchronously until the state difference is minimized, thereby generating an optimized parameter set.

[0043] As a further aspect of the present invention: the simulation of the next time-series driving terraced structure coupling model specifically includes:

[0044] The optimized parameter set completely replaces the original soil hydraulic parameter field in the terrace structure coupling model and serves as the initial core dynamic parameter set for the next simulation time series.

[0045] Based on the optimized parameter set, the internal state variables of each unit in the terraced structure coupling model are consistently corrected and reinitialized.

[0046] Starting with the corrected state variables and optimized parameter set, the system continues to receive new external driving data to drive the terraced structure coupling model to perform the next time-series coupling simulation calculation.

[0047] The beneficial effects of this invention are:

[0048] (1) By integrating time-series hyperspectral remote sensing inversion and ensemble data assimilation techniques, dynamic and high-precision updates of soil hydraulic parameters in terraced fields were achieved. This method overcomes the limitations of traditional terraced field structure coupling models that rely on static or empirical parameters, and can continuously correct and optimize the core parameter fields (such as the saturated hydraulic conductivity of different soil layers) in the terraced field structure coupling model based on the latest remote sensing observation data. This enables the terraced field structure coupling model to accurately capture transient changes in soil structure caused by cultivation, rainfall, or erosion (such as surface crusting), improves the spatiotemporal accuracy and reliability of hydrological and erosion process simulation, and provides a more accurate data foundation for the quantitative assessment of soil and water conservation benefits and risk warning of terraced fields.

[0049] (2) An adaptive terraced field structure coupling model with "simulation-assimilation" closed-loop feedback capability was constructed, realizing continuous simulation of the dynamic evolution of the system state. This terraced field structure coupling model can not only simulate the water and sediment process at a specific moment, but also automatically trigger parameter optimization and update the internal state by introducing subsequent time-series observation data, forming an evolutionary cycle of "simulation → evaluation → update → re-simulation". This mechanism enables the model to have the ability to learn and adapt to environmental changes, and can reflect the evolution trend of the terraced field system under the influence of climate change and human activities in a long-term and dynamic manner, thus providing a powerful dynamic decision support tool for the sustainable management of terraced fields, timely intervention in structural maintenance, and adaptive planning. Attached Figure Description

[0050] The invention will now be further described with reference to the accompanying drawings.

[0051] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0052] 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.

[0053] Please see Figure 1 As shown, this invention is a method for constructing and simulating a coupled model of terraced field structures based on multi-source data, comprising the following steps:

[0054] S1: Acquire multi-source data of the target terraced field area; the multi-source data includes: time-series hyperspectral image data, topographic measurement data, and soil physical property data collected simultaneously in the field;

[0055] S2: Based on multispectral image data and soil physical property data, conduct correlation analysis between spectral features and deep soil hydraulic properties, and extract and generate a set of key feature parameters that can characterize the dynamic changes in soil structure.

[0056] S3: Based on the key feature parameter set, establish a dynamic mapping relationship from spectral features to soil hydraulic parameters, and perform inversion processing on the target terrace area according to the dynamic mapping relationship to generate real-time inversion results of soil hydraulic parameter fields with vertical stratification structure for each spatial unit;

[0057] S4: The real-time inversion results are configured as the core dynamic parameter set in the basic framework of the terraced structure coupling model to form a parameterized terraced structure coupling model, which is used to simulate the coupling effect of hydrological processes and soil erosion processes.

[0058] S5: During the simulation, the terraced structure coupling model is run. During the simulation, the real-time inversion results of subsequent time series are introduced to dynamically update and optimize the parameter set in the terraced structure coupling model. The updated parameter set is then fed back to the terraced structure coupling model to drive the terraced structure coupling model to simulate the next time series.

[0059] In S1, the acquisition of the time-series hyperspectral image data specifically employs an unmanned aerial vehicle (UAV) platform equipped with a pushbroom or frame-type hyperspectral imaging sensor, conducting flight operations along a preset flight path over the target terraced field area. The flight altitude is set between 80 and 150 meters to obtain hyperspectral image data with a ground resolution better than 0.1 meters. The acquisition work needs to be carried out before and after key agricultural activities or weather events (such as after tilling or before and after rainfall) to form a time series. The acquired raw image data needs to undergo radiometric calibration, converting digital quantization values ​​into radiance values; subsequently, atmospheric correction is performed to eliminate the effects of atmospheric scattering and absorption, ultimately obtaining surface reflectance data of the ground features, forming a time-series hyperspectral image dataset.

[0060] The acquisition of the topographic survey data employs two complementary techniques. For the terraced micro-topographic structure, a lidar integrated with the aforementioned UAV platform or oblique photogrammetry is used to generate dense 3D point cloud data from overlapping images taken from different angles using a motion reconstruction algorithm. This point cloud data undergoes classification filtering to separate ground points, and then a digital elevation model is generated using spatial interpolation methods, with its spatial resolution matching that of the hyperspectral image data. For key topographic sections or concealed areas, real-time dynamic differential GPS receivers or total stations are simultaneously used to measure ground control points and supplement topographic feature points to verify and correct the accuracy of the digital elevation model generated by remote sensing.

[0061] The collection of soil physical property data through simultaneous field sampling was conducted concurrently with the acquisition of hyperspectral imagery data, following a pre-designed spatial sampling plan. Sampling points were strategically placed according to the type and spatial distribution of the terraced units. At each sampling point, soil samples were collected layer by layer using a ring cutter or undisturbed soil drill, with a typical sampling depth of 0 to 30 cm, and then packaged in 10-cm layers. The collected samples were sent to the laboratory for testing using standard soil physical analysis methods: saturated hydraulic conductivity was determined using the constant or variable head method; soil porosity and saturated water content were determined using the ring cutter immersion method; and the soil mechanical composition (sand, silt, and clay content) was determined using a laser particle size analyzer. The coordinates of all sample collection locations were recorded using high-precision global navigation satellite system equipment to ensure accurate spatial correspondence with remote sensing imagery and topographic data.

[0062] In S2, the acquired temporal hyperspectral imagery data is first preprocessed and feature-enhanced to extract a first set of spectral indices sensitive to soil physical structure. Preprocessing includes precise geometric registration of images at different time phases to ensure comparability of spectral information for the same geographic coordinates at different time points. Feature enhancement targets soil characteristics, selecting and calculating a series of known spectral indices related to soil physical structure. For example, the Normalized Difference Index (NDDI) is calculated, obtained by dividing the difference in reflectance of soil in two specific bands by their sum, to enhance the response to soil moisture and surface roughness; the Depth Absorption Index is calculated, quantified by analyzing the depth and area of ​​spectral absorption valleys in specific absorption bands (such as those associated with clay minerals or organic matter). By calculating these indices pixel-by-pixel on the full-time imagery, a first set of spectral indices that varies over time is formed, consisting of multiple indices.

[0063] Based on soil physical property data obtained from simultaneous field sampling, a set of second attribute parameters describing soil structural stability is generated. Specifically, this includes calculating a soil structural stability index using measured soil saturated hydraulic conductivity, porosity, and the percentage content of sand, silt, and clay particles. The index is calculated as follows: first, the texture category is determined by consulting a standard soil triangulation diagram based on mechanical composition data, and a preset basic structure coefficient is assigned; then, the logarithm of the saturated hydraulic conductivity and porosity are weighted and summed, and the result is multiplied by the basic structure coefficient to obtain the final soil structural stability index. This calculation is performed on different soil layers at each sampling point, forming a set of second attribute parameters describing the structural stability of the soil profile.

[0064] The first set of spectral indices and the second set of soil attribute parameters obtained above are subjected to multivariate coupling analysis to screen and fuse them to generate the final set of key feature parameters. The core of this analysis is to explore the synergistic change pattern between spectral indices and soil attribute parameters. In specific implementation, a dynamic mapping matrix is ​​first constructed. The rows of this matrix correspond to all indices in the first set of spectral indices, and the columns correspond to all parameters in the second set of soil attribute parameters. The value of each element in the matrix is ​​initialized by calculating the correlation coefficient between the index in that row and the parameter value in that column at the same sampling point location and at the same time. For points with time-series data, this correlation coefficient is calculated based on the time series.

[0065] The co-evolutionary gradient of each index-parameter pair represented by each element in the dynamic mapping matrix is ​​calculated. For points with time-series data, the co-evolutionary gradient is calculated as follows: select two consecutive observation time points, calculate the difference between the correlation coefficient of the index-parameter pair at the later time point and the earlier time point, and then divide this difference by the time interval between the two time points (in days). The quotient is the co-evolutionary gradient for that period. This gradient value quantifies the sensitivity and direction of the spectral index response to changes in soil properties.

[0066] Based on the calculated co-evolutionary gradients, all candidate parameters in the first set of spectral indices and the second set of attribute parameters are ranked by importance. The absolute values ​​of the co-evolutionary gradients of all mapping relationships involved by each index or parameter are averaged to obtain the comprehensive gradient score for that index or parameter. A preset importance threshold is established, determined based on the statistical distribution of the comprehensive gradient scores (e.g., the top 30% quantile). Indices and parameters with comprehensive gradient scores exceeding this preset threshold are considered key components with high sensitivity and strong indicative power for dynamic changes in soil structure.

[0067] The selected first spectral indices and second attribute parameters are normalized and fused to generate a key feature parameter set. Normalization employs a maximum-minimum normalization method, subtracting the minimum value from the original numerical sequence of each selected index or parameter, and then dividing by the difference between the maximum and minimum values ​​to transform it to the range of 0 to 1. All normalized hyperspectral index sequences and soil attribute parameter sequences are combined by sampling point or raster unit. Each spatial unit at each time point corresponds to a vector composed of multiple normalized values; this vector set constitutes the key feature parameter set used in subsequent steps. This parameter set integrates information from rapid remote sensing and precise ground measurement, jointly characterizing the soil structure state and its dynamic change potential.

[0068] In step S3, based on a pre-defined soil profile layering scheme, the key feature parameter set obtained in step S2 is vertically layered and reorganized. The profile is divided into surface, upper core, and lower core layers, with specific depth ranges determined according to typical soil profile characteristics of the region, for example, divided into three layers: 0-10 cm, 10-30 cm, and 30-60 cm. The reorganization process is as follows: for each spatial unit (corresponding to a sampling point or image pixel), the soil structure stability index component associated with that layer and the hyperspectral index component most sensitive to the spectral response of that layer are extracted and combined from its key feature parameter vector. Through this operation, the comprehensive feature parameter set originally representing the entire profile is decomposed into several feature parameter subsets, each subset specifically describing the structural and spectral characteristics of a specific layer in the soil profile.

[0069] Secondly, independent mapping relationships are established based on the feature parameter subsets of each soil layer. For each soil layer, the feature parameter subset of all training samples of that soil layer is used as the input variable, and the measured soil hydraulic parameters of the corresponding soil layer obtained through synchronous field sampling are used as the target variable. A supervised learning algorithm is used for training, the goal of which is to find a function that makes the predicted hydraulic parameter values ​​based on the input features as close as possible to the measured values. The optimization objective of this process is defined by a loss function. Assume that for the ... Layer, whose training sample set contains The nth sample, the nth The input feature vector of each sample is The measured target parameter value is (For example, saturated hydraulic conductivity), then the optimal function is found through an algorithm. Minimize the following mean squared error loss function: ;

[0070] This loss function calculates the average of the squared differences between the predicted and measured values ​​for all samples. The function is adjusted using an iterative optimization algorithm (such as gradient descent). The internal parameters of the soil layer are trained until the loss function converges to an acceptable minimum. At this point, a high-precision correspondence between the spectral characteristics and hydraulic parameters of the soil layer is considered to have been established. Each soil layer (LL) independently completes this training process, obtaining its own mapping function. .

[0071] Using the established correspondence between soil layers, synchronous layered inversion calculations are performed on all spatial units of the entire target terraced area. Specifically, dynamic inversion templates for different soil layers are generated. This template is not a fixed table, but rather a mapping function for each soil layer. The operational logic, defined by the preprocessing rules of the spatial unit and its corresponding feature parameter subsets, specifies how to convert the real-time spectral feature data of any spatial unit into matching rules for the hydraulic parameters of each soil layer. During inversion, for each spatial unit, the following operations are performed in parallel: extract the latest key feature parameter vector of that unit and, according to the aforementioned hierarchical recombination rules, decompose it into feature parameter subsets for each soil layer; then, the... Subset of feature parameters of the layer Input to the pre-trained mapping function of this layer In this process, the predicted hydraulic parameters of the element in this soil layer are directly calculated: The calculation process is performed in parallel or iteratively on all spatial units and all soil layers, efficiently generating independent initial inversion values ​​of hydraulic parameters for each spatial unit and each soil layer.

[0072] Finally, since the above-mentioned stratified independent inversion may neglect the vertical continuity of soil profile physical properties, it is necessary to perform vertical consistency correction and integration on the initial inversion values ​​of different soil layers within the same spatial unit. The correction process is based on soil hydraulic parameters (such as saturated hydraulic conductivity). Prior knowledge that the soil depth decreases exponentially or exhibits a specific trend with depth. First, for a spatial unit, the initial inversion values ​​of each soil layer are obtained. According to the corresponding depth The sequence is plotted. Then, a weighted smoothing technique is used to correct the sequence so that the corrected values ​​are close to the initial inversion values ​​and conform to the vertical trend. Specifically, this can be achieved by defining a quadratic smoothing objective function. Let... For the first After layer correction, a set of hydraulic parameter values ​​is then sought. Minimize the following objective function : ;

[0073] In this formula, This represents the total number of soil layers. The first term on the right side of the equation is the data fidelity term, and we require the corrected value. Do not deviate excessively from the initial inversion value. , The first term is a weighting coefficient set based on the reliability of the initial inversion values ​​or the importance of the corresponding strata, with a higher weight for the surface layer. The second term is a smoothing constraint term, which penalizes sharp fluctuations in the vertical sequence by calculating the second difference between the values ​​of three adjacent soil layers, thus promoting a smoother final profile curve. This is the smoothing coefficient, used to control the smoothing intensity. Its value needs to be determined experimentally based on the drastic changes in the actual soil profile. By solving for the minimum value of this objective function, a set of hydraulic parameter values ​​after vertical consistency correction can be obtained. This correction calculation is performed on each spatial cell, and the final output is a soil hydraulic parameter field with a vertically layered structure and physical continuity across all spatial cells, which is the real-time inversion result required in this step. This result accurately reflects the spatial distribution of hydraulic properties within the soil profile at the time of inversion.

[0074] In step S4, the primary operation in forming the parameterized terraced structure coupling model is constructing the model's integrated input data. This integrated input data consists of two parts: the first part is the real-time inversion result of the soil hydraulic parameter field, the final output of step S3. This provides the model with key hydraulic properties of each spatial unit (corresponding to a single field surface unit of the terrace) at different soil depths, such as the saturated hydraulic conductivity and field capacity of each soil layer. The second part is the external driving data, mainly including time-series rainfall and evaporation data. The rainfall series consists of rainfall data recorded at fixed time intervals (e.g., hourly or daily); the evaporation series consists of potential evaporation or reference crop evapotranspiration data recorded at the same time intervals. These two parts of data must be strictly aligned in terms of spatiotemporal reference, meaning that the spatial extent of the soil parameter field, the time start point and time step of the external driving data must completely match the model's preset simulation period. Through data registration and format conversion, the above two parts of data are combined into a structured integrated input dataset that can be directly read by the model.

[0075] Subsequently, the aforementioned comprehensive input data is loaded into the basic framework of the pre-constructed terraced structure coupling model. This basic framework is essentially a computational graph based on physical rules, defined by two core elements: the terraced unit network topology and the material transport paths. The terraced unit network topology is obtained through hydrological analysis of the high-precision topographic survey data acquired in step S1. Specifically, firstly, each independent terraced surface unit is identified based on the digital elevation model and defined as the basic computational unit of the model; then, based on topographic relief and water flow direction, the upstream and downstream connections between these units are determined, forming a directed network. The material transport paths define the movement of water and sediment within this network, including vertical movement within units (such as infiltration and soil water redistribution) and horizontal movement between units (such as surface runoff flowing from upstream to downstream units and interflow exchange). The loading process involves assigning soil hydraulic parameters from the comprehensive input data to the corresponding terraced units according to their spatial location and linking rainfall and evaporation sequences as temporal drivers to the entire network. Based on this, the model performs coupled simulation calculations step by step according to its embedded physical equations.

[0076] The coupled simulation calculation is performed sequentially in two main stages based on the causal relationship of the physical process and computational efficiency. The first stage is the calculation of water movement. Within each time step, for each terrace unit, the vertical water flux within the unit is calculated first using its configured vertical stratified soil hydraulic parameters and current soil moisture state, combined with the rainfall and evaporation data received in this step. This includes calculating the infiltration of surface rainfall using a variant based on the Green-Ampt infiltration model; calculating the water redistribution and interflow between soil layers using a simplified form of the Richards equation or the storage-discharge relationship; and finally updating the soil moisture content of each soil layer within the unit and calculating the surface runoff generated by the failure to infiltrate. Then, according to the unit network topology, the surface runoff and designated interflow generated by each unit are merged into its downstream adjacent units according to their migration paths, completing the horizontal water movement for this time step. The updated soil moisture content and the total runoff generated by all units constitute the water state output for this step.

[0077] The second stage uses the results of water movement calculations as input to perform soil erosion and sediment transport calculations. This stage is closely linked to the first stage within the same time step framework. For each terrace unit, soil separation calculations mainly occur at the surface. Using the surface runoff (or runoff depth) and runoff shear force obtained in the previous stage, combined with the unit slope and soil erodibility parameters (which can be derived from soil mechanical composition data), the potential rate of soil stripped by runoff within the unit is calculated using the water erosion force component in the modified general soil loss equation. Simultaneously, the separation rate is corrected to account for the weakening effect of increased soil moisture content on soil shear strength. For sediment transport, based on the transport path, the amount of sediment transported from upstream units to this unit is calculated, and it is determined whether deposition occurs within this unit or it continues to be carried by runoff. Deposition determination is based on a comparison between runoff sediment transport capacity and the current sediment load, using a simplified form of Einstein's deposition probability calculation. Finally, the net soil erosion (separation minus deposition) and the amount of sediment transported downstream for each unit are updated, generating the erosion status output.

[0078] Finally, at the end of each simulation time step, the model synchronously integrates the moisture and erosion state outputs of all terrace units within that step. The moisture state output includes at least: the volumetric soil water content of each soil layer within each terrace unit, the surface runoff at the unit outlet, and the inter-soil flow at the unit interface. The erosion state output includes at least: the net surface soil erosion thickness (or mass) of each terrace unit, and the sediment transport rate at the unit outlet. These variables together constitute the complete hydrological and erosion state variables output by the model, comprehensively reflecting the spatiotemporal dynamics of the terrace system. By continuously executing the above calculation steps, the model can simulate the complete coupled process of water and sediment generation, transport, and accumulation in the terrace structure network under specific soil conditions and driven by external climate.

[0079] It should be noted that the terraced structure coupling model of this invention is a numerical simulation structure based on physical mechanisms. Its core lies in the mathematical abstraction and procedural integration of the physical spatial structure of the terraced system, soil medium properties, and hydrological erosion processes. The model uses the terraced unit network topology interpreted by the digital elevation model as its static spatial framework, defining the paths of water and sediment transport. It uses the soil hydraulic parameter field with a vertically layered structure, dynamically obtained through hyperspectral inversion, as its core dynamic attribute, describing the spatiotemporal variability of the soil medium. Through a set of computational logic based on physical laws, it combines external meteorological driving data with the aforementioned spatial framework and dynamic attributes. Within discrete time steps, it sequentially solves the processes of water movement, soil separation, and sediment transport within and between each terraced unit, ultimately synchronously outputting a series of state variables including runoff, soil moisture content, and erosion deposition, thereby achieving a dynamic, distributed, and mechanistic simulation of the coupling effects of water and soil processes in the terraced system.

[0080] In step S5, the dynamic update and optimization begins with assessing the reliability of the terraced structure coupling model simulation, i.e., calculating state differences. Specifically, after the terraced structure coupling model completes a simulation period (e.g., one day or a rainfall event), the key state variables output by the model during that period are acquired. Variables with observational comparison are selected, such as the total surface runoff at the watershed outlet or the time series of soil surface water content for key terraced units. Simultaneously, measured state variable data for the same period are acquired, or the corresponding state variables are indirectly deduced based on newly acquired hyperspectral data during that period using the method in step S3. The calculation of state differences employs the root mean square error method, which involves squaring the difference between the simulated value and the observed value at each corresponding time point, then calculating the average of these squared values, and finally taking the square root of this average to obtain a scalar value representing the overall deviation.

[0081] When the calculated state difference value exceeds a preset trigger threshold, the system automatically initiates the parameter optimization process. This trigger threshold is dynamically determined based on the statistical distribution of historical simulation errors; for example, it may be set to the 75th percentile of historical state difference sequences. This means that when the current error exceeds the historical 75% error level, the terraced structure coupling model is considered to have significant biases, requiring parameter optimization. The optimization uses the newly generated "real-time inversion results of subsequent time series" as the optimization objective and the soil hydraulic parameter field currently used in the terraced structure coupling model as the initial estimate to be adjusted.

[0082] The core of the parameter optimization process is the use of a set-based data assimilation algorithm. The specific implementation steps are as follows: First, taking the current soil hydraulic parameter field of the terraced structure coupling model as the center, a set of 80 slightly different parameter fields is generated through random perturbation; each parameter field is called a set member. Then, each set member is used to drive the terraced structure coupling model, rerunning the simulation period to be optimized, generating a sequence of simulated state variables corresponding to each member. Next, the mean of the simulated state variables of all 80 set members, and the covariance between the simulated state variables and the parameter fields are calculated. Subsequently, the real-time inversion results of subsequent time series are compared with the simulated states of each set member, and the parameter fields of each set member are quantitatively adjusted based on the calculated covariance. The adjustment principle is: the closer the simulation results are to the observations, the greater the weight of the parameter fields of the member being retained; simultaneously, the difference between the observed state and the simulated state is reasonably allocated and corrected to the soil hydraulic parameters of each spatial unit and each soil layer using covariance information. By iterating the above steps once or multiple times, a new, optimized set of parameters is finally obtained, and the mean of all members of this set is taken as the final optimized set of parameters.

[0083] The process of driving the terraced structure coupling model to the next time-series simulation involves integrating new knowledge into the continuously running terraced structure coupling model after parameter optimization. First, parameter replacement is performed: the optimized parameter set generated in the previous step completely covers the soil hydraulic parameter field originally stored in the terraced structure coupling model.

[0084] Subsequently, consistency correction and reinitialization of the state variables are performed. Due to changes in soil hydraulic parameters, the equilibrium relationship between state variables such as soil moisture content stored in each terrace unit of the terraced structure coupling model and the soil matrix has changed. Therefore, it is necessary to correct these state variables based on the new parameter set. Specifically, while keeping the total water volume in each unit constant, the potential energy state and water content values ​​of this water in each soil layer are recalculated and reassigned according to the new soil moisture characteristic curve parameters, so that the state variables are physically consistent with the new parameter field.

[0085] After completing the above corrections, the terraced structure coupled model is ready to restart. Using the corrected state variables of each unit as initial conditions and the optimized parameter set as core attributes, it continues to read new external driving data. Starting from this point, the terraced structure coupled model executes the simulation of the next time series according to the coupled simulation calculation process described in step S4. Thus, the terraced structure coupled model enters a cycle of "simulation-evaluation-assimilation update-resimulation," and its simulation capabilities continuously optimize themselves as more time-series observation data is incorporated.

[0086] The working principle of this invention is as follows: First, multi-source data, including temporal hyperspectral imagery, high-precision topographic surveys, and soil physical properties obtained from field sampling, are acquired simultaneously for the target area. Next, by analyzing the correlation between spectral characteristics and deep soil hydraulic properties, a set of key feature parameters characterizing the dynamic changes in soil structure is constructed and selected. Then, based on this parameter set, a dynamic mapping relationship from spectral data to soil hydraulic parameters is established, and a layered inversion is performed across the entire region to generate real-time inversion results of a soil hydraulic parameter field with a vertically layered structure. Finally, using these inversion results as core dynamic parameters, combined with the terraced unit network topology defined by topographic data, a parameterized terraced structure coupling model is constructed. This model achieves simultaneous simulation of hydrological and erosion states by sequentially solving the coupling process of water movement and soil erosion. Finally, the inversion results of subsequent time series are introduced into the simulation process. The soil hydraulic parameter field inside the model is dynamically updated and optimized through a set-based data assimilation algorithm. The optimized parameters are then fed back to the model to drive the simulation of the next time series, thus forming a closed loop of "simulation-evaluation-assimilation update-resimulation". This enables the model to adapt to the dynamic changes in soil conditions and achieve continuous and high-precision simulation of the coupled evolution of water and soil processes in the terraced field system.

[0087] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for constructing and simulating a coupled model of terraced field structures based on multi-source data, characterized in that, Includes the following steps: S1: Acquire multi-source data of the target terraced field area; the multi-source data includes: time-series hyperspectral image data, topographic measurement data, and soil physical property data collected simultaneously in the field; S2: Based on multispectral image data and soil physical property data, conduct correlation analysis between spectral features and deep soil hydraulic properties, and extract and generate a set of key feature parameters that can characterize the dynamic changes in soil structure. S3: Based on the key feature parameter set, establish a dynamic mapping relationship from spectral features to soil hydraulic parameters, and perform inversion processing on the target terrace area according to the dynamic mapping relationship to generate real-time inversion results of soil hydraulic parameter fields with vertical stratification structure for each spatial unit; S4: The real-time inversion results are configured as the core dynamic parameter set in the basic framework of the terraced structure coupling model to form a parameterized terraced structure coupling model, which is used to simulate the coupling effect of hydrological processes and soil erosion processes. S5: During the simulation, the terraced structure coupling model is run. During the simulation, the real-time inversion results of subsequent time series are introduced to dynamically update and optimize the parameter set in the terraced structure coupling model. The updated parameter set is then fed back to the terraced structure coupling model to drive the terraced structure coupling model to simulate the next time series.

2. The method for constructing and simulating a terraced structure coupling model based on multi-source data according to claim 1, characterized in that, S2 specifically includes: Hyperspectral image data is preprocessed and feature-enhanced to extract a first set of spectral indices sensitive to soil physical structure; Based on the hydraulic properties and mechanical composition data in the soil physical property data, a set of second attribute parameters describing the stability of soil structure is generated. A multivariate coupling analysis was performed on the first set of spectral indices and the second set of attribute parameters. By analyzing the synergistic change patterns of the first set of spectral indices and the second set of attribute parameters, a set of key feature parameters was selected and fused to generate a set of key feature parameters.

3. The method for constructing and simulating a terraced structure coupling model based on multi-source data according to claim 2, characterized in that, The multivariate coupling analysis specifically includes: Construct a dynamic mapping matrix between each index in the first set of spectral indices and each parameter in the second set of attribute parameters; Calculate the co-evolution gradient of each mapping relationship in the dynamic mapping matrix; Based on the co-evolution gradient, the candidate parameters in the first spectral index set and the second attribute parameter set are ranked by importance, and parameters whose gradient values ​​exceed a preset threshold are selected. The selected first spectral index and the second attribute parameter are normalized and fused to generate a set of key feature parameters.

4. The method for constructing and simulating a terraced structure coupling model based on multi-source data according to claim 1, characterized in that, S3 specifically includes: Based on the soil profile layers, the key feature parameter set is vertically layered and reorganized to form feature parameter subsets corresponding to different soil layers; Independent mapping units are trained based on a subset of feature parameters for each soil layer to establish a correspondence between spectral features and soil layer hydraulic parameters. Based on the correspondence, synchronous layered inversion calculations are performed on each spatial unit to generate independent initial inversion values ​​of hydraulic parameters for each soil layer; The initial inversion values ​​of different soil layers within the same spatial unit are vertically consistent and integrated to generate real-time inversion results of soil hydraulic parameter fields with vertical stratification.

5. The method for constructing and simulating a terraced structure coupling model based on multi-source data according to claim 4, characterized in that, The generation of independent initial inversion values ​​of hydraulic parameters for each soil layer specifically includes: Based on the correspondence, dynamic inversion templates for different soil layers are generated. These templates define the matching rules between the spectral characteristic modes and hydraulic parameters of each soil layer. Within each spatial cell, layered template matching calculations are performed in parallel by comparing and fitting the real-time spectral feature data of the spatial cell with the dynamic inversion templates of each soil layer layer by layer. Based on the comparison and fitting results, the initial inversion values ​​of the independent hydraulic parameters of each soil layer in the spatial unit are calculated simultaneously.

6. The method for constructing and simulating a terraced structure coupling model based on multi-source data according to claim 1, characterized in that, The formation of the parameterized terraced structure coupling model specifically includes: The real-time inversion results are combined with external driving data to form comprehensive input data. The real-time inversion results provide the soil hydraulic parameter field, and the external driving data include rainfall and evaporation sequences. The integrated input data is loaded into the basic framework, and coupled simulation calculations are performed based on the terraced unit network topology and transport path defined in the basic framework. Through coupled simulation calculations, the hydrological and erosion state variables of each terrace unit and the entire system are solved and output simultaneously. The erosion state variables include: surface runoff, soil moisture dynamics, and soil erosion and deposition in each terrace unit.

7. The method for constructing and simulating a terraced structure coupling model based on multi-source data according to claim 6, characterized in that, The execution of the coupled simulation calculation specifically includes: Based on the soil hydraulic parameter field and external driving data in the comprehensive input data, the calculation of water movement in the vertical layer of each terrace unit and the calculation of horizontal water transport between terrace units are executed sequentially within the terrace unit network topology to generate the water status output of each terrace unit. Using the moisture status output of each terrace unit as input, and based on the transport path, within the network topology, the soil separation calculation of vertical stratification of each terrace unit and the sediment transport calculation between terrace units are executed simultaneously to generate the erosion status output of each terrace unit. Integrate the water state output and erosion state output to form a complete result of hydrological and erosion state variables.

8. The method for constructing and simulating a terraced structure coupling model based on multi-source data according to claim 1, characterized in that, The aforementioned dynamic updating and optimization specifically includes: Based on the current simulated hydrological and erosion state variables of the terraced structure coupling model, the state differences are calculated by comparing them with the state variables measured in the same period or indirectly inferred from the real-time inversion results. When the state difference exceeds the preset trigger threshold, the real-time inversion results of subsequent time series are used as the optimized observations, and the parameter set of the terraced structure coupling model currently in operation is used as the initial estimate. The soil hydraulic parameters of each spatial unit in the parameter set are adjusted synchronously until the state difference is minimized, thereby generating an optimized parameter set.

9. The method for constructing and simulating a terraced structure coupling model based on multi-source data according to claim 1, characterized in that, The simulation of the next time-series driving terraced structure coupling model specifically includes: The optimized parameter set completely replaces the original soil hydraulic parameter field in the terrace structure coupling model and serves as the initial core dynamic parameter set for the next simulation time series. Based on the optimized parameter set, the internal state variables of each unit in the terraced structure coupling model are consistently corrected and reinitialized. Starting with the corrected state variables and optimized parameter set, the system continues to receive new external driving data to drive the terraced structure coupling model to perform the next time-series coupling simulation calculation.

Images