Big data-based regional drought monitoring and evaluation method and system

Abstract

This invention discloses a method and system for regional drought monitoring and assessment based on big data. The method includes drought characteristic data collection, construction of a two-branch drought feature extraction network, construction of a regional drought boundary loss, toughness quasi-order constraint, construction of a drought assessment objective function, adaptive iterative optimization, establishment of a drought assessment model, and regional drought monitoring and assessment. This invention belongs to the field of data processing, specifically referring to a method and system for regional drought monitoring and assessment based on big data. This scheme constructs a two-branch neural network to adapt to spatial heterogeneity; designs a regional drought boundary loss, applying double-order penalties to samples near the threshold and direct-order penalties to samples experiencing severe drought, improving the stability of drought level classification; introduces toughness quasi-order constraints to maintain robust regional decision thresholds; and constructs a regional objective function with spatial masking, updating weights and biases through adaptive soft-critical iterative optimization, thereby improving the reliability of regional drought monitoring and assessment.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a regional drought monitoring and assessment method and system based on big data. Background Technology

[0002] Regional drought monitoring and assessment methods are conventional methods based on multi-source meteorological, vegetation, and soil data, employing a unified model and global loss to determine drought levels and invert spatial distribution. However, these methods generally suffer from several drawbacks: large fitting biases in wet-dry boundary regions, low accuracy in identifying extremely arid areas, and weak discrimination of boundary samples, leading to poor monitoring and assessment accuracy. Furthermore, they often suffer from insufficient adaptation to spatial heterogeneity, resulting in uneven assessment accuracy. Finally, they lack specificity for drought scenarios, being sensitive and unstable to samples near drought level thresholds, thus contributing to poor monitoring and assessment reliability. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a regional drought monitoring and assessment method and system based on big data. Addressing the problems of large fitting deviations in wet-dry boundary regions, low accuracy in identifying extreme drought areas, and weak boundary sample discrimination in general regional drought monitoring and assessment methods, leading to poor monitoring and assessment accuracy, this scheme constructs a dual-branch neural network, combined with spatial partitioning masks, to model the differentiated features of wet and arid regions, adapting to spatial heterogeneity; it designs a regional drought boundary loss, applying double-order penalties to samples near the threshold and direct-order penalties to severely drought samples, accurately distinguishing level boundary samples and reducing misclassification rates; it suppresses noise from extreme outliers, improving the stability of drought level classification; and it introduces a toughening quasi-order constraint, which... This approach highlights key drought factors such as precipitation and evapotranspiration, eliminates multi-source collinearity, reduces overfitting, and maintains robust regional decision thresholds, thereby improving the accuracy of regional drought monitoring and assessment. Addressing the shortcomings of general regional drought monitoring and assessment methods, such as insufficient spatial heterogeneity adaptation, uneven assessment accuracy, weak drought scenario specificity, and sensitivity and instability to samples near drought level thresholds, leading to poor monitoring and assessment reliability, this scheme constructs a regional objective function with spatial masking. It optimizes the separation of humid and arid regions, avoiding smoothed boundaries and significantly improving the accuracy of identifying severely drought-stricken areas. Furthermore, it utilizes adaptive soft-critical iterative optimization to update weights and biases, resulting in greater robustness to noise and outlier drought samples, thus further enhancing the reliability of regional drought monitoring and assessment.

[0004] The technical solution adopted by this invention is as follows: The regional drought monitoring and assessment method based on big data provided by this invention includes the following steps:

[0005] Step S1: Acquisition of drought characteristic data;

[0006] Step S2: Construction of a dual-branch drought feature extraction network;

[0007] Step S3: Constructing regional drought boundary losses;

[0008] Step S4: Toughness sequence constraint;

[0009] Step S5: Constructing the drought assessment objective function;

[0010] Step S6: Adaptive iterative optimization;

[0011] Step S7: Establishing a drought assessment model;

[0012] Step S8: Regional drought monitoring and assessment.

[0013] Furthermore, in step S1, the drought characteristic data collection involves acquiring historical multi-source spatiotemporal data of meteorology, vegetation, and soil within the region, eliminating dimensional differences through unified normalization, and constructing drought level labels and spatial partitioning to build a drought characteristic set.

[0014] Furthermore, in step S2, the construction of the dual-branch drought feature extraction network is to adapt to the spatial heterogeneity of wet and dry areas. For the drought feature set, a dual-branch neural network structure is adopted: branch 1 focuses on learning the normal hydrological and ecological boundary of the wet area, and branch 2 focuses on learning the abnormal water shortage boundary of the arid area, so as to realize the spatially differentiated feature modeling; the bottom layer of the network shares the feature extraction layer, and the top layer makes independent decisions.

[0015] Furthermore, in step S3, the construction of the regional drought boundary loss involves applying a two-order penalty to the drought boundary samples and a direct-order penalty to the severe drought and extreme anomalous samples, thereby obtaining the regional drought boundary loss function.

[0016] Furthermore, in step S4, the toughness criterion constraint is to introduce a toughness criterion term to automatically screen key drought features, suppress overfitting and multicollinearity, and establish a stable regional decision threshold.

[0017] Furthermore, in step S5, the drought assessment objective function construction unifies the regional drought boundary loss, the toughness quasi-order term, and the bi-branch spatial fitting into a total objective, with each branch only optimizing the corresponding regional samples.

[0018] Furthermore, in step S6, the adaptive iterative optimization employs adaptive neighborhood update, constructs soft critical operation, and performs weight bias update.

[0019] Furthermore, in step S7, the drought assessment model is established by using the grading results as the supervision label to construct a regional drought assessment model. The model architecture follows the dual-branch neural network constructed in S2, with a shared feature extraction layer at the bottom layer and separate wet baseline and drought discrimination branches at the top layer. The model training uses the objective function constructed in S5 as the optimization objective, adopts the adaptive neighborhood update iterative training in S6, outputs the regional drought intensity index, and performs drought risk assessment based on the regional drought intensity index.

[0020] Furthermore, in step S8, the regional drought monitoring and assessment is based on the established drought assessment model, which acquires multi-source spatiotemporal data of meteorology, vegetation, and soil in the region in real time, and inputs them into the drought assessment model after preprocessing. Regional drought monitoring is achieved based on the assessment results output by the model.

[0021] The regional drought monitoring and assessment system based on big data provided by this invention includes a drought characteristic data acquisition module, a two-branch drought characteristic extraction network construction module, a regional drought boundary loss construction module, a toughness criterion constraint module, a drought assessment objective function construction module, an adaptive iterative optimization module, a drought assessment model establishment module, and a regional drought monitoring and assessment module.

[0022] The drought feature data acquisition module acquires historical multi-source spatiotemporal data within the region and constructs a drought feature set.

[0023] The dual-branch drought feature extraction network construction module is based on a drought feature set and constructs a dual-branch neural network. The bottom layer shares the feature extraction layer, and the top layer has separate independent decision branches that focus on the normal hydrological and ecological boundary of the humid area and the abnormal water shortage boundary of the arid area.

[0024] The regional drought boundary loss construction module constructs a regional drought boundary loss function, applying a two-order penalty to drought boundary samples and a direct-order penalty to severe drought samples;

[0025] The shrinkage quasi-sequence constraint module sets quasi-sequence coefficients and constructs shrinkage quasi-sequence terms;

[0026] The drought assessment objective function construction module integrates regional drought boundaries and resilience quasi-ordination terms to construct the drought assessment objective function;

[0027] The adaptive iterative optimization module uses adaptive neighborhood update to construct soft critical operation, and completes the iterative update of network weights and biases.

[0028] The drought assessment model building module establishes a drought assessment model based on a drought feature set, a two-branch drought feature extraction network, a drought assessment objective function, and adaptive iterative optimization.

[0029] The regional drought monitoring and assessment module performs drought monitoring and assessment based on real-time multi-source spatiotemporal data within the region using a drought assessment model.

[0030] The beneficial effects achieved by the present invention using the above solution are as follows:

[0031] (1) To address the problems of large fitting deviations in wet and dry boundary areas, low discrimination accuracy in extreme drought areas, and weak discrimination of boundary samples in general regional drought monitoring and assessment methods, which lead to poor monitoring and assessment accuracy, this scheme constructs a dual-branch neural network and uses spatial partitioning masks to model the differentiated features of wet and arid areas to adapt to spatial heterogeneity; designs a regional drought boundary loss, applies double-order penalties to samples near the threshold and direct-order penalties to samples in severe drought, accurately distinguishes grade boundary samples, and reduces misclassification rate; suppresses noise from extreme outliers and improves the stability of drought grade classification; introduces toughness quasi-order constraints to automatically highlight key drought factors such as precipitation and evapotranspiration, eliminates multi-source feature collinearity, reduces overfitting, and maintains robust regional decision thresholds; thereby improving the accuracy of regional drought monitoring and assessment.

[0032] (2) In view of the problems that general regional drought monitoring and assessment methods have insufficient spatial heterogeneity adaptation, uneven assessment accuracy, weak drought scenario targeting, and sensitivity and instability to samples near the drought level threshold, which leads to poor monitoring and assessment reliability, this scheme constructs a regional objective function with spatial masking, optimizes the wet and dry areas separately, avoids smoothing the boundary, and significantly improves the discrimination accuracy of severely drought areas; through adaptive soft critical iterative optimization to update weights and biases, it is more robust to noise and outlier drought samples; thereby improving the reliability of regional drought monitoring and assessment. Attached Figure Description

[0033] Figure 1 A flowchart illustrating the regional drought monitoring and assessment method based on big data provided by this invention;

[0034] Figure 2 This is a schematic diagram of the regional drought monitoring and assessment system based on big data provided by the present invention.

[0035] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0036] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0037] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0038] Example 1, see Figure 1 The present invention provides a regional drought monitoring and assessment method based on big data, which includes the following steps:

[0039] Step S1: Acquire drought characteristic data, obtain historical multi-source spatiotemporal data of the region, and construct a drought characteristic set;

[0040] Step S2: Construction of a dual-branch drought feature extraction network. Based on the drought feature set, a dual-branch neural network is constructed. The bottom layer shares the feature extraction layer, and the top layer has separate independent decision branches focusing on the normal hydrological and ecological boundary of the humid area and the abnormal water shortage boundary of the arid area.

[0041] Step S3: Constructing the regional drought boundary loss function, applying a double-order penalty to drought boundary samples and a straight-order penalty to severe drought samples;

[0042] Step S4: Toughness sequence constraint, set sequence coefficients, and construct toughness sequence terms;

[0043] Step S5: Construction of drought assessment objective function, integrating regional drought boundaries and resilience quasi-ordination terms to construct drought assessment objective function;

[0044] Step S6: Adaptive iterative optimization, using adaptive neighborhood update to construct flexible critical operation, to complete the iterative update of network weights and biases;

[0045] Step S7: Drought assessment model establishment. Based on drought feature set, two-branch drought feature extraction network, drought assessment objective function and adaptive iterative optimization, a drought assessment model is established.

[0046] Step S8: Regional drought monitoring and assessment, based on the drought assessment model, to monitor and assess drought using multi-source spatiotemporal data within the real-time region.

[0047] Example 2, see Figure 1This embodiment is based on the above embodiment. In step S1, drought characteristic data collection involves acquiring historical multi-source spatiotemporal data of meteorology, vegetation, and soil in the region to construct a drought characteristic set; eliminating dimensional differences through unified normalization; and constructing drought level labels and spatial partitioning. The meteorological dimension includes precipitation, average temperature, relative humidity, wind speed, and potential evapotranspiration; the vegetation dimension includes Normalized Difference Vegetation Index (NDVI) and Vegetation Status Index (VCI); the soil dimension includes soil volumetric water content at 0–10 cm and 10–40 cm depths; drought is divided into four levels: no drought, mild drought, moderate drought, and severe drought, and drought level labels are constructed (according to the "Meteorological Drought Level" GB / T20481-2017); spatial partitioning is based on the average dryness and wetness of the region, dividing it into humid and arid areas, and generating a spatial partitioning mask (setting a wetness index threshold of 1.0 to divide the entire region into humid and arid areas).

[0048] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the construction of the dual-branch drought feature extraction network is to adapt to the spatial heterogeneity of dry and wet regions and avoid the fitting deviation of a single network to the dry and wet boundaries. For the drought feature set, a dual-branch neural network structure is adopted: branch 1 focuses on learning the normal hydrological and ecological boundaries of wet areas, and branch 2 focuses on learning the abnormal water-scarce boundaries of arid areas, so as to realize spatially differentiated feature modeling. The bottom layer of the network shares the feature extraction layer to reduce the number of parameters and improve efficiency, and the top layer makes independent decisions to maintain the flexibility of regional discrimination. While ensuring computational efficiency, it significantly improves the evaluation accuracy of boundary areas and extreme arid areas.

[0049] Branch 1 The wetted reference plane is represented as: ;

[0050] Branch 2 The drought discrimination surface is represented as: ;in, and These are the weight matrices for branch 1 and branch 2, respectively; and These are the biases for branch 1 and branch 2, respectively; the L in the bottom right corner represents the network layer number. This is the data from the kth drought sample; It uses the ReLU activation function; number of network layers: 3–6; hidden layer dimensions: 64–256.

[0051] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the regional drought boundary loss construction applies a double-order penalty to drought boundary samples to improve boundary discrimination accuracy; and applies a straight-order penalty to severe drought / extreme anomaly samples to suppress outlier interference. The regional drought boundary loss function is expressed as follows: ; Where δ is the interval threshold, which controls the width of the transition zone, and its value ranges from 0.5 to 2.0; s is the drought interval term. It is the regional drought boundary loss function; It is a drought level label; It is a two-branch average output.

[0052] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the toughness quasi-order constraint is introduced because high-dimensional multi-source drought features are prone to overfitting and multicollinearity. This allows for the automatic screening of key drought features (precipitation, evapotranspiration), suppression of overfitting and multicollinearity, and stabilization of regional decision thresholds. Represented as: ;in, , and These are ordinal coefficients, each taking the value 1e. -5 ~1e -3 1e -4 ~1e -3 and 1e -5 ~1e -3 .

[0053] By performing the above operations, this scheme addresses the problems of large fitting biases in wet and dry boundary areas, low discrimination accuracy in extreme drought areas, and weak discrimination of boundary samples in general regional drought monitoring and assessment methods, which lead to poor monitoring and assessment accuracy. It constructs a dual-branch neural network, combined with spatial partitioning masks, to model the differentiated features of wet and arid areas, adapting to spatial heterogeneity. A regional drought boundary loss is designed, employing double-order penalties for samples near the threshold and direct-order penalties for severely drought samples to accurately distinguish level boundary samples and reduce misclassification rates. It suppresses noise from extreme outliers, improving the stability of drought level classification. A tough quasi-order constraint is introduced to automatically highlight key drought factors such as precipitation and evapotranspiration, eliminating multi-source feature collinearity, reducing overfitting, and maintaining robust regional decision thresholds. This ultimately improves the accuracy of regional drought monitoring and assessment.

[0054] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the drought assessment objective function is constructed by unifying the regional drought boundary loss, the resilience quasi-ordination term, and the bi-branch spatial fitting into a total objective. Each branch only optimizes the corresponding regional samples. The objective function is expressed as follows: ;in, and It is the penalty coefficient for dry and wet areas, with a value ranging from 1 to 100; and It is a spatial partition mask; , It is the item between branch 1 and branch 2.

[0055] Example 7, see Figure 1 This embodiment is based on the above embodiment. In step S6, the adaptive iterative optimization uses adaptive neighborhood update to construct a soft critical operation. , is represented as: Perform a weight bias update, represented as: ; ; ; Where t is the network weight value to be updated, specifically the network weight value; It is a symbolic function; , and These are the network weights updated after the j-th, j-1-th, and j-2-th iterations, respectively, representing the weights of all layers in the dual-branch architecture, where j is the iteration number. It is the adaptive Lipschitz constant; It is the weight update amount; It is the gradient of the loss term with respect to the weights in the objective function, i.e. Gradient with respect to the weights; , and These are the bias terms updated after the j-th, j-1-th, and j-2-th iterations, respectively; It is the offset update amount; It is the gradient of the loss term with respect to the bias in the objective function; γ is the momentum coefficient, with a value of 0.9~0.99.

[0056] Example 8, see Figure 1 This embodiment is based on the above embodiment. In step S7, the drought assessment model is established using the grading results as supervision labels to construct a regional drought assessment model that is adapted to the spatial heterogeneity of dry and wet regions, is noise-resistant, and has high accuracy. The model architecture follows the dual-branch neural network constructed in S2, with a shared feature extraction layer at the bottom layer and separate wet baseline and drought discrimination branches at the top layer, corresponding to branch 1 and branch 2 in S2, respectively. The model training uses the objective function constructed in S5 as the optimization objective and adopts the adaptive neighborhood update iterative training in S6. The accuracy rate is used as the model performance evaluation standard until the model loss tends to stabilize and the generalization performance is optimal. The model input is the drought feature set preprocessed in S1, and the output is the regional drought intensity index. Drought risk assessment is performed based on the regional drought intensity index. The outputs of the two branches are fused into the regional drought intensity index, which is expressed as: Based on the drought index threshold, the drought level is classified, and the spatial distribution, area proportion, and center of gravity shift of the drought in the region are output to complete the assessment, which is represented as follows: ;in, It is the drought intensity index at spatial location x; the larger the value, the more severe the drought. It is the dual-branch output fusion weight, used to balance the contributions of the wet baseline branch and the drought discrimination branch; Sigmoid(·) is the Sigmoid function; DroughtClass is the final drought level; , and These are drought classification thresholds, with values ​​of 0.2~0.35, 0.35~0.55, and 0.55~0.75, respectively.

[0057] By performing the above operations, this scheme addresses the problems of insufficient spatial heterogeneity adaptation and uneven assessment accuracy in general regional drought monitoring and assessment methods; weak drought scenario targeting; sensitivity and instability to samples near drought level thresholds; and consequently, poor monitoring and assessment reliability. It constructs a regional objective function with spatial masking, optimizing the humid and arid regions separately to avoid smoothing the boundaries and significantly improving the accuracy of identifying severely drought-stricken areas. Furthermore, by using adaptive soft-critical iterative optimization to update weights and biases, it exhibits greater robustness to noise and outlier drought samples, thereby improving the reliability of regional drought monitoring and assessment.

[0058] Example 9, see Figure 1 This embodiment is based on the above embodiment. In step S8, the regional drought monitoring and assessment is based on the established drought assessment model. Multi-source spatiotemporal data of meteorology, vegetation and soil in the region are acquired in real time. After preprocessing, the data is input into the drought assessment model. Regional drought monitoring is realized based on the assessment results output by the model. If the assessment result is moderate drought or severe drought, an early warning is issued to the management personnel.

[0059] Example 10, see Figure 2 Based on the above embodiments, the regional drought monitoring and assessment system based on big data provided by the present invention includes a drought characteristic data acquisition module, a two-branch drought characteristic extraction network construction module, a regional drought boundary loss construction module, a toughness criterion constraint module, a drought assessment objective function construction module, an adaptive iterative optimization module, a drought assessment model establishment module, and a regional drought monitoring and assessment module.

[0060] The drought feature data acquisition module acquires historical multi-source spatiotemporal data within the region and constructs a drought feature set.

[0061] The dual-branch drought feature extraction network construction module is based on a drought feature set and constructs a dual-branch neural network. The bottom layer shares the feature extraction layer, and the top layer has separate independent decision branches that focus on the normal hydrological and ecological boundary of the humid area and the abnormal water shortage boundary of the arid area.

[0062] The regional drought boundary loss construction module constructs a regional drought boundary loss function, applying a two-order penalty to drought boundary samples and a direct-order penalty to severe drought samples;

[0063] The shrinkage quasi-sequence constraint module sets quasi-sequence coefficients and constructs shrinkage quasi-sequence terms;

[0064] The drought assessment objective function construction module integrates regional drought boundaries and resilience quasi-ordination terms to construct the drought assessment objective function;

[0065] The adaptive iterative optimization module uses adaptive neighborhood update to construct soft critical operation, and completes the iterative update of network weights and biases.

[0066] The drought assessment model building module establishes a drought assessment model based on a drought feature set, a two-branch drought feature extraction network, a drought assessment objective function, and adaptive iterative optimization.

[0067] The regional drought monitoring and assessment module performs drought monitoring and assessment based on real-time multi-source spatiotemporal data within the region using a drought assessment model.

[0068] It should be noted that, in this document, 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.

[0069] 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.

[0070] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for regional drought monitoring and evaluation based on big data, characterized in that: The method includes the following steps: Step S1: Acquire drought characteristic data, obtain historical multi-source spatiotemporal data of the region, and construct a drought characteristic set; Step S2: Construction of a dual-branch drought feature extraction network. Based on the drought feature set, a dual-branch neural network is constructed. The bottom layer shares the feature extraction layer, and the top layer has separate independent decision branches focusing on the normal hydrological and ecological boundary of the humid area and the abnormal water shortage boundary of the arid area. Step S3: Constructing the regional drought boundary loss function, applying a double-order penalty to drought boundary samples and a straight-order penalty to severe drought samples; Step S4: Toughness sequence constraint, set sequence coefficients, and construct toughness sequence terms; Step S5: Construction of drought assessment objective function, integrating regional drought boundaries and resilience quasi-ordination terms to construct drought assessment objective function; Step S6: Adaptive iterative optimization, using adaptive neighborhood update to construct flexible critical operation, to complete the iterative update of network weights and biases; Step S7: Drought assessment model establishment. Based on drought feature set, two-branch drought feature extraction network, drought assessment objective function and adaptive iterative optimization, a drought assessment model is established. Step S8: Regional drought monitoring and assessment, based on the drought assessment model, to monitor and assess drought using multi-source spatiotemporal data within the real-time region; The drought characteristic data collection involves acquiring historical multi-source spatiotemporal data of meteorology, vegetation, and soil in the region. The meteorological dimension includes precipitation, average temperature, relative humidity, wind speed, and potential evapotranspiration; the vegetation dimension includes Normalized Difference Vegetation Index (NDVI) and Vegetation Status Index (VCI); and the soil dimension includes soil volumetric water content at 0–10 cm and 10–40 cm depths. In step S3, the construction of the regional drought boundary loss involves applying a double-order penalty to drought boundary samples and a straight-order penalty to severely drought samples to suppress outlier interference. The regional drought boundary loss function is expressed as follows: ; Where δ is the interval threshold, which controls the width of the transition zone; s is the drought interval term; It is the regional drought boundary loss function; It is a drought level label; It is a two-branch average output; This is the data from the kth drought sample; In step S4, the toughness quasi-order constraint introduces a toughness quasi-order term to automatically screen key drought characteristics, suppress overfitting and multicollinearity, and establish a stable regional decision threshold; the toughness quasi-order term... Represented as: ;in, , and These are ordinal coefficients; In step S5, the drought assessment objective function is constructed by unifying the regional drought boundary loss, the resilience quasi-ordination term, and the bi-branch spatial fitting into a total objective. Each branch only optimizes the corresponding regional samples. The objective function is expressed as follows: ;in, and It is the penalty coefficient for dry and wet areas, with a value ranging from 1 to 100; and It is a spatial partition mask; , These are the drought interval terms for branches 1 and 2; In step S6, the adaptive iterative optimization employs adaptive neighborhood update to construct a flexible critical operation. , is represented as: Perform a weight bias update, represented as: ; ; ; Where t is the network weight value to be updated, specifically the network weight value; It is a symbolic function; , and These are the network weights updated after the j-th, j-1-th, and j-2-th iterations, respectively. It is the adaptive Lipschitz constant; It is the weight update amount; It is the gradient of the loss term in the objective function with respect to the weights; It is the offset update amount; γ is the gradient of the loss term with respect to the bias in the objective function; γ is the momentum coefficient.

2. The regional drought monitoring and assessment method based on big data according to claim 1, characterized in that: In step S2, the construction of the dual-branch drought feature extraction network is to adapt to the spatial heterogeneity of wet and dry regions. For the drought feature set, a dual-branch neural network structure is adopted: branch 1 focuses on learning the normal hydrological and ecological boundary of the wet region, and branch 2 focuses on learning the abnormal water shortage boundary of the arid region, so as to realize the modeling of spatially differentiated features; the bottom layer of the network shares the feature extraction layer, and the top layer makes independent decisions.

3. The regional drought monitoring and assessment method based on big data according to claim 1, characterized in that: In step S7, the drought assessment model is established by using the grading results as supervision labels to construct a regional drought assessment model. The model architecture follows the dual-branch neural network constructed in S2, with the bottom layer sharing the feature extraction layer and the top layer having separate wet baseline branches and drought discrimination branches. The model training uses the objective function constructed by S5 as the optimization objective, adopts the adaptive neighborhood update iterative training of S6, outputs the regional drought intensity index, and performs drought risk assessment based on the regional drought intensity index.

4. The regional drought monitoring and assessment method based on big data according to claim 1, characterized in that: In step S1, the drought feature data collection also eliminates dimensional differences through unified normalization; and drought level labels are constructed and spatial partitions are performed to build a drought feature set.

5. A regional drought monitoring and assessment system based on big data, used to implement the regional drought monitoring and assessment method based on big data as described in any one of claims 1-4, characterized in that: It includes a drought characteristic data acquisition module, a two-branch drought characteristic extraction network construction module, a regional drought boundary loss construction module, a toughness criterion-order constraint module, a drought assessment objective function construction module, an adaptive iterative optimization module, a drought assessment model establishment module, and a regional drought monitoring and assessment module; The drought feature data acquisition module acquires historical multi-source spatiotemporal data within the region and constructs a drought feature set. The dual-branch drought feature extraction network construction module is based on a drought feature set and constructs a dual-branch neural network. The bottom layer shares the feature extraction layer, and the top layer has separate independent decision branches that focus on the normal hydrological and ecological boundary of the humid area and the abnormal water shortage boundary of the arid area. The regional drought boundary loss construction module constructs a regional drought boundary loss function, applying a two-order penalty to drought boundary samples and a direct-order penalty to severe drought samples; The shrinkage quasi-sequence constraint module sets quasi-sequence coefficients and constructs shrinkage quasi-sequence terms; The drought assessment objective function construction module integrates regional drought boundaries and resilience quasi-ordination terms to construct the drought assessment objective function; The adaptive iterative optimization module uses adaptive neighborhood update to construct soft critical operation, and completes the iterative update of network weights and biases. The drought assessment model building module establishes a drought assessment model based on a drought feature set, a two-branch drought feature extraction network, a drought assessment objective function, and adaptive iterative optimization. The regional drought monitoring and assessment module performs drought monitoring and assessment based on real-time multi-source spatiotemporal data within the region using a drought assessment model.

Images