Unsteady runoff response prediction method considering forest disturbance

Abstract

This invention discloses a method for predicting unsteady runoff response considering forest disturbance. The method includes data decomposition of the unsteady runoff sequence to obtain a basic runoff sequence and a disturbed runoff sequence; time-varying coefficient regression to obtain a basic runoff mapping function; correlation analysis and sensitivity analysis; using the correlation analysis results to correct the sensitivity analysis results to obtain optimized disturbance sensitivity; constructing a disturbed runoff prediction model; inputting the watershed characteristics and disturbance data of the watershed to be predicted into the basic runoff mapping function and the disturbed runoff prediction model respectively to obtain predicted basic runoff and predicted disturbed runoff; and superimposing these to obtain the predicted unsteady runoff. This method can quantify the disturbed runoff resulting from the interaction of wildfire disturbance, logging disturbance, vegetation adjustment disturbance, and conventional climate disturbance, and incorporates a competition-cooperative improved double cumulative curve (C-MDMC) to quantify the interaction effects between disturbances, outputting disturbed runoff prediction results. This provides a refined decision-making basis for watershed multi-scenario adaptive water resource management.

Description

Technical Field

[0001] This invention relates to the field of hydrological and water resources prediction technology, and in particular to a method for predicting unsteady runoff response considering forest disturbance. Background Technology

[0002] The relationship between forests and runoff exhibits significant unsteady characteristics in alpine and subalpine regions. As a key regulator of the watershed hydrological cycle, the structure, composition, and functional state of forests directly control runoff generation and confluence processes such as precipitation interception, evapotranspiration, soil infiltration, and groundwater recharge. However, in recent years, multiple disturbances, including wildfires, deforestation, and vegetation species adjustments, have been coupled and superimposed in time and space, posing a severe challenge to traditional runoff prediction methods based on steady-state assumptions. Accurately analyzing the unsteady response mechanisms of forest disturbances to runoff has significant scientific value and practical implications for water resource management, forest ecological restoration, and flood and drought disaster early warning.

[0003] Existing runoff prediction technologies suffer from the following shortcomings: First, traditional hydrological models typically use fixed parameter sets or time-segmented static parameter sets for runoff simulation, making it difficult to capture the continuous drift of watershed runoff coefficients over time under unsteady conditions, leading to systematic biases in runoff prediction during unsteady periods. Furthermore, conventional baseflow segmentation methods, based on globally fixed thresholds or steady-state assumptions, cannot adapt to the time-varying characteristics of baseflow / disturbance runoff attribution caused by differences in underlying surface conditions between wet and dry seasons, resulting in ambiguous physical meanings in the decomposition results. Simultaneously, existing forest disturbance runoff analyses often employ single sensitivity analysis or single correlation analysis, failing to effectively couple the cascade transmission effects between disturbance, forest condition, and runoff, and neglecting the nonlinear interactions of competition and cooperation among disturbances, leading to redundant coupling and spurious sensitivity problems in sensitivity estimation. Finally, existing machine learning prediction models are mostly black-box structures, lacking physical mechanism embedding, unable to explicitly distinguish between the direct hydrological effects of forest disturbance and the indirect effects mediated by forest condition, resulting in poor interpretability of prediction results and insufficient generalization ability to complex extreme climate-disturbance scenarios. Therefore, this invention proposes a nonsteady-state runoff response prediction method considering forest disturbance. Through ensemble empirical mode decomposition with climate-time adaptive constraints, time-varying coefficient regression, coupled correction of correlation analysis and sensitivity analysis, and a dual-path parallel prediction model, it achieves adaptive fusion prediction of physical transmission paths and baseflow-climate coupling paths. This not only significantly improves the accuracy and robustness of nonsteady-state runoff prediction, but also provides interpretable and implementable technical support for forest restoration and adaptive water resource management in subalpine watersheds under changing environments by explicitly quantifying the three-layer transmission chain of forest disturbance-forest state-disturbed runoff. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting unsteady runoff response that takes into account forest disturbance.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution: This invention includes the following steps: Obtain unsteady runoff sequences and corresponding watershed information and disturbance data; decompose the unsteady runoff sequences to obtain basic runoff sequences and disturbed runoff sequences. The basic runoff mapping function is obtained by performing time-varying coefficient regression on watershed characteristics, conventional disturbances, and basic runoff. Correlation analysis and sensitivity analysis were performed on forest disturbance, forest status and disturbance runoff respectively. The results of the correlation analysis were used to correct the results of the sensitivity analysis to obtain the optimized disturbance sensitivity. A disturbance runoff prediction model is constructed based on historical disturbance runoff and corresponding disturbance data, optimized disturbance sensitivity, and watershed information. The watershed characteristics and disturbance data of the watershed to be predicted are input into the basic runoff mapping function and the disturbance runoff prediction model, respectively, to obtain the predicted basic runoff and the predicted disturbance runoff. They are then superimposed to obtain the predicted unsteady runoff. The watershed information includes forest status, topographic features, and watershed characteristics; the forest status includes forest vegetation density, average vegetation height, and species richness; the topographic features include topographic gradient, average elevation, elevation difference, and soil geology; the watershed characteristics include upstream runoff, water supply area and corresponding population density, industrial GDP, and agricultural GDP. The disturbance data includes forest disturbances and conventional disturbances; the forest disturbances include wildfire disturbances, logging disturbances, and vegetation adjustment disturbances; the conventional disturbances are represented by climate data; The disturbed runoff includes the response duration and response intensity.

[0006] Furthermore, the method for performing data decomposition to obtain basic runoff and disturbed runoff includes: The precipitation index for each time period is calculated based on the climate data of the regular disturbance, and the time domain is divided into the wet season, the normal season and the dry season according to the precipitation index. The climate-adaptive noise amplitude is calculated based on the precipitation index. This amplitude is then superimposed onto the unsteady runoff sequence and subjected to empirical mode decomposition to obtain the intrinsic mode function components and trend residuals. The expression for the climate-adaptive noise amplitude is as follows: ; in for Noise amplitude at time, The reference noise amplitude, Climate sensitivity coefficient, for Precipitation index at any given time For ensemble trials Standard normally distributed random numbers generated at time 1; Hilbert transforms are performed on each intrinsic mode function component to calculate the instantaneous frequency of each intrinsic mode function. The energy density and average instantaneous frequency of each intrinsic mode function are also calculated at different water storage periods. Based on forest disturbance, a comprehensive disturbance intensity index is calculated, and the modal correlation coefficients between each intrinsic mode function and the disturbance intensity at different water storage periods are calculated. The expression is as follows: ; in for Water storage period The modal correlation coefficient between the intrinsic mode function and the disturbance intensity for The time range of the water storage period For the first eigenmode functions The value at time, For the first Each eigenmode function in The average value during the water storage period. for The overall disturbance intensity index at any given time. To normalize the comprehensive disturbance intensity index in Average value during the water storage period; The comprehensive discriminant index is calculated based on the modal correlation coefficient. The comprehensive discriminant indices are then arranged in ascending order. The adaptive discriminant threshold is determined using the maximum curvature method, expressed as follows: ; ; ; in for Water storage period The comprehensive discriminant index of the intrinsic mode functions, for Adaptive discrimination threshold for water storage period For threshold modality index, , , To determine the index weights, for Energy density during the water storage period , They are respectively The baseline instantaneous frequency and maximum energy density during the water storage period. This is an adjustment coefficient for wet and dry seasons. The length of the main channel of the river. The average slope This represents the number of intrinsic modes; Based on the comprehensive discrimination index and the adaptive discrimination threshold, the intrinsic mode functions during the water storage period are divided into basic runoff and disturbed runoff.

[0007] Furthermore, the method for obtaining the basic runoff mapping function includes: Watershed characteristics at each water storage period are extracted. Using watershed characteristics and conventional disturbances as inputs and basic runoff as output, time-varying coefficient regression is performed to obtain the basic runoff mapping function. The time-varying coefficients of the basic runoff mapping function are updated by Kalman filtering.

[0008] Furthermore, the method for performing correlation analysis and sensitivity analysis includes: Wildfire disturbance, logging disturbance, and vegetation adjustment disturbance are extracted at each time point. The wildfire disturbance includes wildfire type, wildfire form, wildfire area, and duration. The logging disturbance includes logging frequency, logging quantity, and logging type. The vegetation adjustment disturbance includes adjusting vegetation type, adjusting quantity, and adjusting ecological purpose. Grey relational analysis was used to calculate the correlation strength among various forest disturbances; Sobol sensitivity analysis was performed on forest disturbance and forest state to obtain the first sensitivity, on forest state and disturbed runoff to obtain the second sensitivity, and on forest disturbance and disturbed runoff to obtain the third sensitivity.

[0009] Furthermore, the method for obtaining optimized perturbation sensitivity includes: The sensitivity analysis results are corrected using the correlation analysis results to obtain the optimized perturbation sensitivity, expressed as: ; ; ; in , , These are the first optimized perturbation sensitivity, the second optimized perturbation sensitivity, and the third optimized perturbation sensitivity, respectively. The strength of the correlation between forest disturbance and forest state. The strength of the correlation between forest disturbance and disturbed runoff. The strength of the association between forest condition and disturbed runoff. , , These are respectively the first sensitivity, the second sensitivity, and the third sensitivity. This is the negative feedback adjustment coefficient. This is the forward propagation coefficient.

[0010] Furthermore, the method for constructing a disturbance runoff prediction model includes: Historical disturbance runoff and corresponding disturbance data, optimized disturbance sensitivity, and watershed information are combined into a comprehensive set. The comprehensive set is randomly divided into a training set and a test set in a 7:3 ratio. The disturbance runoff prediction model is trained using the training set and the model performance is verified using the test set. The disturbance runoff prediction model includes an input layer, a physical conduction prediction layer, a climate coupling prediction layer, a time prediction layer, an adaptive fusion layer, and an output layer. The physical transmission prediction layer uses forest disturbance, optimized disturbance sensitivity, forest state, and water storage period as pre-inputs to calculate the forest state change, indirect path disturbance runoff, and direct path disturbance runoff caused by the disturbance, respectively. A water storage period gating function is then introduced to predict the disturbance runoff, expressed as follows: in for The output of the physical conduction prediction layer at any given time, i.e., the predicted value of disturbance runoff. , , To optimize perturbation sensitivity, This represents the weight vector of the direct impact of forest disturbance on disturbed runoff. This represents the weight vector of the indirect impact of forest condition on disturbed runoff. This is the perturbation-state transition matrix. for Forest perturbation vector at time step. for The forest state vector at time step 1. This is the gate function for predicting the water storage period of the physical conduction layer. for The category of water storage period corresponding to the current moment; The climate-coupled prediction layer embeds a basic runoff mapping function to calculate basic runoff based on watershed characteristics. It uses forest disturbance, conventional disturbance, basic runoff, topographic features, and water storage period as pre-inputs. The baseflow saturation is calculated using the predicted basic runoff value against the hydrological background. Enhanced disturbance vectors are obtained by quantifying conventional disturbances through effective climate disturbances. A water storage period gating function is then introduced to predict disturbed runoff. The expression is as follows: ; ; ; ; in for The output of the climate-coupled prediction layer at any given time, i.e., the predicted value of perturbation runoff. For base current saturation, To enhance the forest perturbation vector, For effective climate disturbance, To increase the weighting of forest disturbance impacts, For the effective weighting of climate disturbances, This is the gating function for climate-coupled prediction of the water storage period in the layer. for Basic runoff at any given time This is the historical basic runoff reference average. The saturation rate coefficient, This is the climate amplification factor. , , This is the climate effect coefficient. To standardize precipitation anomalies, To standardize temperature anomalies; The time prediction layer uses a BP neural network to capture disturbed runoff and corresponding disturbance data, optimizes the nonlinear relationship between disturbance sensitivity, basic runoff, and forest state structure, and predicts the duration of disturbance runoff. The adaptive fusion layer dynamically determines the fusion weight by calculating the residual entropy and physical dominance of the physical conduction prediction layer and the climate coupling prediction layer, and then weights and fuses the disturbance runoff predicted by the intermediate layer to obtain the final predicted disturbance runoff.

[0011] Furthermore, the disturbance runoff prediction model employs a four-objective hybrid loss function to improve the model's prediction accuracy, expressed as follows: ; ; ; ; ; in It is a four-objective mixed loss function. For fitting loss, For physical loss weights, For physical loss, To balance the loss weights, To balance the losses, To maintain the total amount of loss weight, For the total amount of loss due to conservation, For the sample size, for The predicted perturbation runoff output by the perturbation runoff prediction model at any given time. for The historical observed disturbed runoff at any given time was obtained by data decomposition of the non-steady-state runoff series. for The perturbation runoff predicted by the physical conduction prediction layer at any given time. for Predicted runoff values ​​for reconstructed physical transport chains at any given time. , They are respectively The fusion weights of the time-space physical transmission prediction layer and the climate coupling prediction layer for The predicted basic runoff output by the climate coupling prediction layer at any given time. for The predicted perturbation runoff output by the perturbation runoff prediction model at any given time. for The original total runoff at that time.

[0012] The beneficial effects of this invention are: This invention is a method for predicting unsteady runoff response considering forest disturbance. Compared with existing technologies, this invention has the following technical advantages: This invention introduces ensemble empirical mode decomposition based on climate-time adaptive constraints, dynamically adjusts the decomposition noise amplitude according to the standardized precipitation index, and constructs a hydrophysical discrimination index that differentiates between wet and dry seasons, thereby realizing the time-varying determination of whether the same intrinsic mode function belongs to basic runoff or disturbed runoff at different times. This invention upgrades traditional multiple linear regression to time-varying coefficient regression and uses Kalman filter state transition equations to achieve continuous time-varying updates of coefficients, thus solving the coefficient drift problem of basic runoff mapping under unsteady conditions. This invention optimizes disturbance sensitivity by performing correlation and sensitivity analysis between forest disturbance, forest state, and disturbance runoff, and by using the correlation analysis results to correct the sensitivity analysis results. It quantifies the three-layer transmission chain of disturbance → state, state → runoff, and disturbance → runoff, thereby eliminating spurious sensitivity and redundant coupling in disturbance sensitivity analysis. This invention constructs a dual-path parallel prediction structure with a physical conduction layer and a climate coupling layer, and sets an adaptive fusion layer based on sliding window residual entropy and physical dominance to dynamically determine the fusion weight. Through a hybrid loss function, it ensures the logical consistency and water balance of the entire chain from data decomposition, baseflow mapping, disturbance prediction to result integration, and realizes explicit dual-path adaptive fusion prediction with physical mechanisms. This invention integrates wildfire disturbance, logging disturbance, vegetation adjustment disturbance and conventional climate disturbance into a unified prediction framework, and embeds a competition-cooperative improvement dual cumulative curve to quantify the interaction effects between disturbances. It can output multi-dimensional disturbance runoff prediction results including response duration and response intensity, providing a refined decision-making basis for watershed multi-scenario water resource adaptive management. Attached Figure Description

[0013] Figure 1 The flowchart illustrates the steps of the non-steady-state runoff response prediction method considering forest disturbance in this invention. Detailed Implementation

[0014] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0015] The present invention provides a method for predicting the unsteady runoff response considering forest disturbance, comprising the following steps: like Figure 1 As shown, this embodiment includes the following steps: Obtain unsteady runoff sequences and corresponding watershed information and disturbance data; decompose the unsteady runoff sequences to obtain basic runoff sequences and disturbed runoff sequences. The basic runoff mapping function is obtained by performing time-varying coefficient regression on watershed characteristics, conventional disturbances, and basic runoff. Correlation analysis and sensitivity analysis were performed on forest disturbance, forest status and disturbance runoff respectively. The results of the correlation analysis were used to correct the results of the sensitivity analysis to obtain the optimized disturbance sensitivity. A disturbance runoff prediction model is constructed based on historical disturbance runoff and corresponding disturbance data, optimized disturbance sensitivity, and watershed information. The watershed characteristics and disturbance data of the watershed to be predicted are input into the basic runoff mapping function and the disturbance runoff prediction model, respectively, to obtain the predicted basic runoff and the predicted disturbance runoff. They are then superimposed to obtain the predicted unsteady runoff. The watershed information includes forest status, topographic features, and watershed characteristics; the forest status includes forest vegetation density, average vegetation height, and species richness; the topographic features include topographic gradient, average elevation, elevation difference, and soil geology; the watershed characteristics include upstream runoff, water supply area and corresponding population density, industrial GDP, and agricultural GDP. The disturbance data includes forest disturbances and conventional disturbances; the forest disturbances include wildfire disturbances, logging disturbances, and vegetation adjustment disturbances; the conventional disturbances are represented by climate data; The disturbed runoff includes the response duration and response intensity.

[0016] In this embodiment, the method for obtaining basic runoff and disturbed runoff by data decomposition includes: The precipitation index for each time period is calculated based on the climate data of the regular disturbance, and the time domain is divided into the wet season, the normal season and the dry season according to the precipitation index. The climate-adaptive noise amplitude is calculated based on the precipitation index. This amplitude is then superimposed onto the unsteady runoff sequence and subjected to empirical mode decomposition to obtain the intrinsic mode function components and trend residuals. The expression for the climate-adaptive noise amplitude is as follows: ; in for Noise amplitude at time, The reference noise amplitude, Climate sensitivity coefficient, for Precipitation index at any given time For ensemble trials Standard normally distributed random numbers generated at time 1; Hilbert transforms are performed on each intrinsic mode function component to calculate the instantaneous frequency of each intrinsic mode function. The energy density and average instantaneous frequency of each intrinsic mode function are also calculated at different water storage periods. Based on forest disturbance, a comprehensive disturbance intensity index is calculated, and the modal correlation coefficients between each intrinsic mode function and the disturbance intensity at different water storage periods are calculated. The expression is as follows: ; in for Water storage period The modal correlation coefficient between the intrinsic mode function and the disturbance intensity for The time range of the water storage period For the first eigenmode functions The value at time, For the first Each eigenmode function in The average value during the water storage period. for The overall disturbance intensity index at any given time. To normalize the comprehensive disturbance intensity index in Average value during the water storage period; The comprehensive discriminant index is calculated based on the modal correlation coefficient. The comprehensive discriminant indices are then arranged in ascending order. The adaptive discriminant threshold is determined using the maximum curvature method, expressed as follows: ; ; ; in for Water storage period The comprehensive discriminant index of the intrinsic mode functions, for Adaptive discrimination threshold for water storage period For threshold modality index, , , To determine the index weights, for Energy density during the water storage period , They are respectively The baseline instantaneous frequency and maximum energy density during the water storage period. This is an adjustment coefficient for wet and dry seasons. The length of the main channel of the river. The average slope This represents the number of intrinsic modes; Based on the comprehensive discrimination index and the adaptive discrimination threshold, the intrinsic mode functions during the water storage period are divided into basic runoff and disturbed runoff. In actual assessments, the precipitation index is based on... Calculate, where, for Cumulative precipitation in the 12 months prior to the date , The mean and standard deviation of historical cumulative precipitation are used, and the time domain is divided into high-water period, normal-water period and low-water period according to [1, +∞), (-1, 1), (-∞, -1]. When calculating the amplitude of climate-adaptive noise, the total number of ensemble experiments is set according to the number of ensemble experiments of EEMD (100-200). The mean of the random numbers of all ensemble experiments is taken as the standard normal distribution random number, and the amplitude of climate-adaptive noise is calculated. Then, after being superimposed on the unsteady runoff sequence, empirical mode decomposition is performed to obtain the intrinsic mode function components and trend residuals. The Hilbert transform of each intrinsic mode function component is used to calculate the instantaneous frequency of each intrinsic mode function at the corresponding time, and the energy density and average instantaneous frequency of each intrinsic mode function are calculated within different water storage periods. The weights of the discrimination index are set to 0.3 / 0.2 / 0.5, 0.4 / 0.25 / 0.35, and 0.55 / 0.25 / 0.2 for the wet season, the normal season, and the dry season, respectively. Take respectively The weighted average of the standardized intensity of wildfire / logging / vegetation adjustment disturbances and the corresponding historical maximum values ​​of disturbances is used as the comprehensive disturbance intensity index. The wet / dry season adjustment coefficients are 1.5 / 1 / 0.8 for wet / dry seasons, respectively. The comprehensive discrimination index is calculated based on the modal correlation coefficient, and the comprehensive discrimination indexes are arranged in ascending order. The maximum curvature method is used to determine the adaptive discrimination threshold. When the comprehensive discrimination index is less than the adaptive discrimination threshold, the first... Each eigenmode function in During periods of water storage, the flow is classified as basic runoff; otherwise, it is classified as disturbed runoff (the same intrinsic mode function may be classified as disturbed runoff / reflecting rapid surface runoff during the wet season and as basic runoff / reflecting slow release of groundwater during the dry season). Taking the forest disturbance that occurred in July 2025 in a subalpine watershed (a tributary of the Shangzhagun River) in Southwest China as an example, the main channel length of 25km and the average slope of 0.25 (m / m) were extracted. Based on the climate data of the regular disturbance, the precipitation index at a certain time node in July was calculated to be 1.45. The precipitation index at multiple time nodes in July was calculated to determine that this period was the high water season. The reference noise amplitude is taken as 0.15m. 3 Given a climate sensitivity coefficient of 0.6, calculate the adaptive noise amplitude for a specific time point in July. ,in, The mean of the standard normally distributed random numbers generated for 150 ensemble trials; The adaptive noise amplitudes of all time-series nodes in July were superimposed onto the unsteady runoff sequence, and empirical mode decomposition was performed to obtain the intrinsic mode function components (the number of intrinsic modes is 6) and trend residuals. The high-water season adjustment coefficient was 1.5, the discrimination index weights were 0.30 / 0.20 / 0.50, and the base instantaneous frequency was calculated to be 0.0850 cycles / day. Taking the aforementioned time series nodes as an example, the average instantaneous frequency, energy density, and modal correlation coefficient with the comprehensive disturbance intensity index of the six intrinsic mode functions of the time series nodes during the high water period are extracted. Based on this, the comprehensive discrimination index is calculated to be 2.3745, 1.3673, 0.6846, 0.4065, 0.1989, and 0.1082. The composite discriminant indices are calculated and sorted in ascending order. The maximum curvature method is used to determine the threshold mode index as 4 (corresponding to the original mode as 3), and the corresponding adaptive discriminant threshold is 0.6846. The intrinsic mode functions with a composite discriminant index less than or equal to 0.6846 are assigned to the basic runoff, and the intrinsic mode functions with a composite discriminant index greater than 0.6846 are assigned to the disturbed runoff. The components of the intrinsic mode functions are superimposed according to the runoff classification results to obtain the basic runoff and the disturbed runoff. Among them, the trend residuals, which reflect the long-term trend (extremely low frequency), are all assigned to the basic runoff.

[0017] In this embodiment, the method for obtaining the basic runoff mapping function includes: Watershed characteristics at each water storage period are extracted. Using watershed characteristics and conventional disturbances as inputs and basic runoff as output, time-varying coefficient regression is performed to obtain the basic runoff mapping function. The time-varying coefficients of the basic runoff mapping function are updated by Kalman filtering. In practical assessments, the basic runoff mapping function and the time-varying coefficient update formulas are expressed as follows: ; ; in for Basic runoff at time, for The regression intercept at time t represents the baseline basic runoff value when all watershed characteristics are zero. , They are respectively time Time-varying regression coefficients and eigenvalues ​​of watershed characteristics or conventional disturbances. for The regression residuals (error terms) at time step 1 reflect the random perturbations not explained by the model. for time The state transition noise of the time-varying regression coefficients corresponding to watershed characteristics or conventional disturbances follows a Gaussian distribution. A basic runoff mapping function was established using time-varying coefficient regression (TVCM). The time-varying coefficients were updated using the Kalman filter state transition equation, yielding monthly coefficient values ​​of 3.5 / 0.25 / 0.018 / -0.005 / 0.12 / 0.2 (corresponding to intercept, upstream runoff, water supply area, population density, industrial GDP, agricultural GDP, projected rainfall, and projected temperature). Using the watershed characteristics for July 2025 / high-water season (upstream runoff 15.2 m³ / s, water supply area 120 km², population density 45 people / km², industrial GDP 230 million yuan, agricultural GDP 180 million yuan, projected rainfall 269 mm, projected rainfall 23℃) as input, the basic runoff for that month was predicted to be 9.871 m³ / s based on the basic runoff mapping function. 3 / s.

[0018] In this embodiment, the method for performing correlation analysis and sensitivity analysis includes: Wildfire disturbance, logging disturbance, and vegetation adjustment disturbance are extracted at each time point. The wildfire disturbance includes wildfire type, wildfire form, wildfire area, and duration. The logging disturbance includes logging frequency, logging quantity, and logging type. The vegetation adjustment disturbance includes adjusting vegetation type, adjusting quantity, and adjusting ecological purpose. Grey relational analysis was used to calculate the correlation strength among various forest disturbances; Sobol sensitivity analysis was performed on forest disturbance and forest state to obtain the first sensitivity, Sobol sensitivity analysis was performed on forest state and disturbed runoff to obtain the second sensitivity, and Sobol sensitivity analysis was performed on forest disturbance and disturbed runoff to obtain the third sensitivity. In actual assessments, in the aforementioned calculation of the intensity of wildfire / logging / vegetation adjustment disturbances, the product of wildfire area, duration, and wildfire type is taken as the wildfire disturbance intensity; the product of logging frequency, logging quantity, and logging type per unit time is taken as the logging disturbance intensity; and the product of adjusted vegetation type (different vegetation types have different weights) and adjusted quantity is taken as the vegetation adjustment disturbance intensity. Taking the forest disturbance that occurred in July 2025 in a subalpine watershed (a tributary of the Upper Zhagun River) in southwestern China as an example: The wildfire season lasted from July 1st to 10th, during which wildfire disturbances occurred (the wildfire type was climate-driven / high-temperature and drought-induced, the wildfire form was surface fire, the wildfire area was 150 hectares, and the duration was 10 days). Localized logging disturbances were also observed (logging occurred twice, approximately 120 dead standing trees were felled, the logging type was emergency selective logging / only clearing high-risk fallen trees and trees in the firebreak) and vegetation adjustment disturbances (no tree species were changed, only water spraying and temporary mulching were applied to the fire edge). During this phase, wildfire was the dominant disturbance. The intensity of each disturbance was calculated, and after standardization, the forest disturbance vector was obtained. ; From July 11th to 20th, the clearing of charred timber took place. Wildfire disturbance weakened (the wildfire type was residual fire, the form was sporadic surface embers, the wildfire area was 150 hectares, and the duration was 5 days). The main focus was on clearing charred timber by felling (felling occurred 5 times, with approximately 800 burned and charred trees felled; clear-cutting was used, and the burned area was thoroughly cleared in preparation for subsequent land preparation). Vegetation adjustment and disturbance also began (500 native herbs and shrubs were planted to initially cover the wildfire-disturbed area; the ecological purpose of this adjustment was soil and water conservation). During this stage, the wildfire gradually subsided, and the focus shifted to high-intensity felling and clearing. The intensity of each disturbance was calculated and standardized to obtain the forest disturbance vector. ; From July 21st to 31st, the planting period was for new trees. There was no wildfire disturbance. The period also included clearing remaining logging disturbances (one logging operation, approximately 30 trees felled, selectively clearing local obstructions) and vegetation adjustment disturbances (planting native fire-resistant species such as Alpine pine, Minjiang fir, and redwood, totaling 2000 trees; the ecological purpose of the adjustment was forest restoration). During this phase, the wildfires were completely quelled, and the focus shifted to large-scale vegetation adjustment (planting). The intensity of each disturbance was calculated and standardized to obtain the forest disturbance vector. ; Sobol sensitivity of forest disturbance, forest condition, and disturbed runoff during historical high-water seasons was calculated, yielding first sensitivity / second sensitivity / third sensitivity values ​​of 0.35 / 0.42 / 0.28. Grey relational analysis was then used to calculate the correlation strength of forest disturbance, forest condition, and disturbed runoff for the current month according to the high-water season. / / The values ​​are 0.72, 0.58, and 0.65 respectively.

[0019] In this embodiment, the method for obtaining optimized perturbation sensitivity includes: The sensitivity analysis results are corrected using the correlation analysis results to obtain the optimized perturbation sensitivity, expressed as: ; ; ; in , , These are the first optimized perturbation sensitivity, the second optimized perturbation sensitivity, and the third optimized perturbation sensitivity, respectively. The strength of the correlation between forest disturbance and forest state. The strength of the correlation between forest disturbance and disturbed runoff. The strength of the association between forest condition and disturbed runoff. , , These are respectively the first sensitivity, the second sensitivity, and the third sensitivity. This is the negative feedback adjustment coefficient. This is the forward propagation coefficient; In the actual assessment, the optimized disturbance sensitivity is calculated according to the water storage period. The first sensitivity reflects the impact of forest disturbance on forest state. The first optimized disturbance sensitivity represents the net disturbance-state sensitivity after weighting by correlation strength and removing redundant indirect coupling. The second sensitivity reflects the impact of forest state on disturbed runoff. The second optimized disturbance sensitivity represents the net sensitivity of forest state to the impact of disturbed runoff. The third sensitivity reflects the impact of forest disturbance on disturbed runoff. Taking the forest disturbance generated in July 2025 in a subalpine watershed (a tributary of the Shangzhagun River) in Southwest China as an example, with a negative feedback adjustment coefficient of 0.15 and a positive propagation coefficient of 0.2, the sensitivity analysis results were corrected using correlation analysis results to obtain optimized disturbance sensitivities of 0.2453 / 0.2367 / 0.1943.

[0020] In this embodiment, the method for constructing a disturbance runoff prediction model includes: Historical disturbance runoff and corresponding disturbance data, optimized disturbance sensitivity, and watershed information are combined into a comprehensive set. The comprehensive set is randomly divided into a training set and a test set in a 7:3 ratio. The disturbance runoff prediction model is trained using the training set and the model performance is verified using the test set. The disturbance runoff prediction model includes an input layer, a physical conduction prediction layer, a climate coupling prediction layer, a time prediction layer, an adaptive fusion layer, and an output layer. The physical transmission prediction layer uses forest disturbance, optimized disturbance sensitivity, forest state, and water storage period as pre-inputs to calculate the forest state change, indirect path disturbance runoff, and direct path disturbance runoff caused by the disturbance, respectively. A water storage period gating function is then introduced to predict the disturbance runoff, expressed as follows: in for The output of the physical conduction prediction layer at any given time, i.e., the predicted value of disturbance runoff. , , To optimize perturbation sensitivity, This represents the weight vector of the direct impact of forest disturbance on disturbed runoff. This represents the weight vector of the indirect impact of forest condition on disturbed runoff. This is the perturbation-state transition matrix. for Forest perturbation vector at time step. for The forest state vector at time step 1. This is the gate function for predicting the water storage period of the physical conduction layer. for The category of water storage period corresponding to the current moment; The climate-coupled prediction layer embeds a basic runoff mapping function to calculate basic runoff based on watershed characteristics. It uses forest disturbance, conventional disturbance, basic runoff, topographic features, and water storage period as pre-inputs. The baseflow saturation is calculated using the predicted basic runoff value against the hydrological background. Enhanced disturbance vectors are obtained by quantifying conventional disturbances through effective climate disturbances. A water storage period gating function is then introduced to predict disturbed runoff. The expression is as follows: ; ; ; ; in for The output of the climate-coupled prediction layer at any given time, i.e., the predicted value of perturbation runoff. For base current saturation, To enhance the forest perturbation vector, For effective climate disturbance, To increase the weighting of forest disturbance impacts, For the effective weighting of climate disturbances, This is the gating function for climate-coupled prediction of the water storage period in the layer. for Basic runoff at any given time This is the historical basic runoff reference average. The saturation rate coefficient, This is the climate amplification factor. , , This is the climate effect coefficient. To standardize precipitation anomalies, To standardize temperature anomalies; The time prediction layer uses a BP neural network to capture disturbed runoff and corresponding disturbance data, optimizes the nonlinear relationship between disturbance sensitivity, basic runoff, and forest state structure, and predicts the duration of disturbance runoff. The adaptive fusion layer dynamically determines the fusion weight by calculating the residual entropy and physical dominance of the physical conduction prediction layer and the climate coupling prediction layer, and weights and fuses the disturbance runoff predicted by the intermediate layer to obtain the final predicted disturbance runoff. In practical assessments, within the physical transmission prediction layer, the transmission chain of forest disturbance to runoff includes both the indirect path of "forest disturbance → forest condition → disturbed runoff" and the direct path of "forest disturbance → disturbed runoff," through forest disturbance and the first optimized disturbance sensitivity. The current effective forest state is obtained by calculating the changes in forest state. The effective forest state is then evaluated using the second optimization perturbation sensitivity. Driving disturbed runoff, and through indirect path disturbances, forest disturbances bypass forest state mediators, optimizing disturbance sensitivity through a third mechanism. The calculation expressions for each path directly affect the disturbed runoff are as follows: ; ; ; in For indirect path disturbance runoff, For direct path disturbance runoff, For an effective forest state, The changes in forest state directly affect the weight vector. The direct impact of forest disturbance on disturbed runoff, and the indirect impact on the weight vector. The disturbance-state transition matrix represents the indirect effects of forest state on disturbed runoff (including the influence of vegetation density on runoff generation through interception capacity and the influence of species richness on soil infiltration through root structure). The elements represent the physical impact coefficients of a certain type of forest disturbance on a certain forest state (such as the negative impact coefficient of logging on vegetation density, the damage coefficient of wildfire on species richness), which are calibrated through historical disturbance-state observation data. Gating function during water storage period This reflects the modulation of the underlying surface conditions of the watershed on the disturbance runoff generation efficiency at different times (soil saturation during the wet season results in a significantly higher disturbance runoff generation coefficient for the same amount of water than during the dry season). Reflecting the specific response of the baseflow-climate pathway at different times; In the climate-coupled prediction layer, the baseflow saturation, watershed soil moisture content, and groundwater recharge status directly determine the runoff generation efficiency of disturbance events. The saturation rate coefficient is greater than 0. When the basic runoff is high (soil moisture saturation), the baseflow saturation approaches 1, and the forest disturbance runoff generation efficiency is maximized. When the basic runoff is low (dry season), the baseflow saturation approaches 0, the soil absorption capacity is strong, and the disturbance runoff generation is significantly suppressed. Enhance forest perturbation vector Runoff response used to significantly amplify wildfire disturbances in extreme climates (such as drought and high temperatures) (impaired vegetation recovery, enhanced surface runoff pathway connectivity, and climate effect coefficient). Characterizing the nonlinear amplification of runoff by the extreme effects of precipitation-temperature interaction, the standardized precipitation anomaly refers to the deviation between the observed precipitation value at a certain moment and the historical average precipitation value for the same period. The standardized temperature anomaly refers to the deviation between the observed temperature value at a certain moment and the historical average temperature value for the same period. These are expressed as: ; ; in for Actual precipitation observation value at any time for Actual temperature observation at any given time , These are the historical average precipitation and temperature for the same period, respectively. , These are the standard deviations of historical precipitation and temperature for the same period, respectively. In the adaptive fusion layer, the prediction residuals of the data windows of the physical transmission prediction layer and the climate coupling prediction layer are calculated separately, and the residual entropy is calculated based on the prediction residuals (the smaller the entropy, the higher the prediction stability). , The physical dominance of the physical conduction prediction layer and the climate coupling prediction layer are calculated based on the optimized perturbation sensitivity, baseflow saturation, and effective climate perturbation, respectively. The fusion weight is calculated by combining the prediction residuals and the physical dominance. After normalization, the perturbation runoff predicted by the physical conduction prediction layer and the climate coupling prediction layer is weighted and fused to obtain the predicted perturbation runoff. The expressions for the physical dominance and fusion weight are as follows: ; ; ; ; in For the physical dominance of the physical transmission prediction layer, For the physical dominance of the climate coupling prediction layer, For the fusion weights of the physical transmission prediction layer, For the fusion weights of the climate coupling prediction layer; The watershed characteristics of the watershed to be predicted are input into the basic runoff mapping function to obtain the predicted basic runoff, and the disturbance data of the watershed to be predicted are input into the disturbance runoff prediction model to obtain the predicted disturbance runoff. The two are then superimposed to obtain the predicted unsteady runoff.

[0021] In this embodiment, the disturbance runoff prediction model employs a four-objective hybrid loss function to improve the model's prediction accuracy, expressed as follows: ; ; ; ; ; in It is a four-objective mixed loss function. For fitting loss, For physical loss weights, For physical loss, To balance the loss weights, To balance the losses, To maintain the total amount of loss weight, For the total amount of loss due to conservation, For the sample size, for The predicted perturbation runoff output by the perturbation runoff prediction model at any given time. for The historical observed disturbed runoff at any given time was obtained by data decomposition of the non-steady-state runoff series. for The perturbation runoff predicted by the physical conduction prediction layer at any given time. for Predicted runoff values ​​for reconstructed physical transport chains at any given time. , They are respectively The fusion weights of the time-space physical transmission prediction layer and the climate coupling prediction layer for The predicted basic runoff output by the climate coupling prediction layer at any given time. for The predicted perturbation runoff output by the perturbation runoff prediction model at any given time. for The original total runoff observed at that time; In actual assessments, the weights for physical loss, balanced loss, and total conservation loss are set at 0.3, 0.05, and 0.2, respectively. Taking a forest disturbance that occurred in July 2025 in a subalpine watershed (a tributary of the Upper Zhagun River) in Southwest China as an example (altitude range 2800-3800m, slope range 15°-40°, predominantly northwest slope (NW) with some southeast slope (SE), average altitude 3200m, elevation difference 1000m, soil composition 30cm alpine meadow soil + 50cm dark brown soil, rock strata mainly metamorphic sandstone and slate, weathered layer thickness 0.5-2.0m), the watershed experienced a three-stage continuous disturbance: "wildfire → clearing charred timber → planting new trees." The initial vector of the forest state was... (Corresponding to standardized vegetation density, height index, and species richness index); Based on actual rainfall of 280 mm, historical average of 220 mm, and standard deviation of 45 mm, the standardized precipitation anomaly is calculated to be 1.333. Based on actual temperature of 22.5 ℃, historical average of 18.0 ℃, and standard deviation of 3.0 ℃, the standardized temperature anomaly is calculated to be 1.5. Taking climate effect coefficients of 0.5 / 0.3 / 0.2, the effective climate disturbance is calculated to be 1.5167. Taking a saturation rate coefficient of 2 and a historical basic runoff reference average of 8.5 m, the calculation is as follows: 3 / s, the calculated base current saturation is 0.9020; the gate function value during the high-water season. , Take 1.35 / 1.2 respectively; In early July (wildfire season), the forest disturbance vector (High-intensity wildfires, early stages of deforestation and clearing, weak vegetation adjustment) Input the disturbance runoff prediction model, and take the direct influence weight vector as... The indirect influence weight vector is The forest state change was calculated sequentially in the physical transmission prediction layer. Effective forest status is (Normalized values ​​of forest vegetation density, average vegetation height, and species richness) are used to calculate the predicted disturbed runoff based on physical transmission predictions, which is 0.7871 m. 3 / s (indirect path disturbance runoff 0.3285m) 3 / s, direct path disturbance runoff 0.2545m 3 / s); In the climate coupled prediction layer, based on the effective climate perturbation of 1.5167 and the climate amplification factor of 0.3, the enhanced forest perturbation vector is calculated. Combined with enhanced perturbation weights With a climate disturbance weight of 0.35, the predicted runoff is calculated to be 1.1179 m. 3 / s; In the adaptive fusion layer, the observed perturbation runoff is 0.95m. 3 / s, the prediction residuals for the data windows of the physical conduction prediction layer and the climate coupled prediction layer are calculated to be 0.1629 / 0.1679 respectively. The residual entropy is calculated to be 0.45 / 0.52 based on the prediction residuals. The physical dominance is calculated to be 0.2524 and 2.27 based on the optimized perturbation sensitivity, baseflow saturation, and effective climate perturbation, respectively. The normalized fusion weights are calculated to be 0.1065 / 0.8935. The predicted perturbation runoff for the physical conduction prediction layer and the climate coupled prediction layer is 0.7871m. 3 / s, 1.1179m 3 The per-s weighted fusion, through the perturbation runoff prediction model, outputs a predicted perturbation runoff of 1.0826 m. 3 / s (duration 5 days, predicted by BP neural network), superimposed with the predicted basic runoff, yields a predicted unsteady runoff of 10.954m. 3 / s (assuming the basic runoff remains constant, in actual assessments the basic runoff follows the changes in watershed characteristics and regular disturbances). In mid-July (the period for clearing scorched wood), the forest perturbation vector will be... (Wildfire remnants, intensive clearing of charred timber, preliminary adjustments) Input the disturbance runoff prediction model, using the effective forest state at the end of Stage 1 as the initial state, and perform predictions. The output predicted disturbance runoff is 0.96m. 3 The predicted unsteady runoff was 10.83 m³ / s (lasting 10 days, predicted by a BP neural network) and superimposed with the predicted basic runoff to obtain the predicted unsteady runoff. 3 / s, In late July (the period for planting new trees), the forest disturbance vector will be... (Wildfire quelled, minimal clearing, large-scale vegetation adjustment) Input the disturbance runoff prediction model, using the effective forest state at the end of Stage 2 as the initial state, and perform predictions. The output predicted disturbance runoff is 0.96m. 3 / s, superimposed with the predicted basic runoff, yields a predicted unsteady runoff of 10.83m. 3 / s (duration is 10 days, predicted by BP neural network); As the disturbance shifted from "high-intensity wildfire" to "ecological restoration (planting)," the unsteady runoff disturbance runoff changed from 1.08 → 0.96 → 0.75 m. 3 / s decreases.

[0022] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting unsteady runoff response considering forest disturbance, characterized in that, Includes the following steps: S1. Obtain the unsteady runoff sequence and corresponding watershed information and disturbance data, and perform data decomposition on the unsteady runoff sequence to obtain the basic runoff sequence and the disturbance runoff sequence; S2. Time-varying coefficient regression of watershed characteristics, conventional disturbances and basic runoff is performed to obtain the basic runoff mapping function; S3. Conduct correlation analysis and sensitivity analysis on forest disturbance, forest status and disturbance runoff respectively, and use the correlation analysis results to correct the sensitivity analysis results to obtain optimized disturbance sensitivity; S4. Construct a disturbance runoff prediction model based on historical disturbance runoff and corresponding disturbance data, optimized disturbance sensitivity, and watershed information; S5. Input the watershed characteristics and disturbance data of the watershed to be predicted into the basic runoff mapping function and the disturbance runoff prediction model respectively to obtain the predicted basic runoff and the predicted disturbance runoff, and superimpose them to obtain the predicted unsteady runoff; The watershed information includes forest status, topographic features, and watershed characteristics; the forest status includes forest vegetation density, average vegetation height, and species richness; the topographic features include topographic gradient, average elevation, elevation difference, and soil geology; the watershed characteristics include upstream runoff, water supply area and corresponding population density, industrial GDP, and agricultural GDP. The disturbance data includes forest disturbances and conventional disturbances; the forest disturbances include wildfire disturbances, logging disturbances, and vegetation adjustment disturbances; the conventional disturbances are represented by climate data; The disturbed runoff includes the response duration and response intensity.

2. The nonsteady-state runoff response prediction method considering forest disturbance according to claim 1, characterized in that, The method for obtaining basic runoff and disturbed runoff by data decomposition includes: The precipitation index for each time period is calculated based on the climate data of the regular disturbance, and the time domain is divided into the wet season, the normal season and the dry season according to the precipitation index. The climate-adaptive noise amplitude is calculated based on the precipitation index. This amplitude is then superimposed onto the unsteady runoff sequence and subjected to empirical mode decomposition to obtain the intrinsic mode function components and trend residuals. The expression for the climate-adaptive noise amplitude is as follows: ； in for Noise amplitude at time, The reference noise amplitude, Climate sensitivity coefficient, for Precipitation index at any given time For ensemble trials Standard normally distributed random numbers generated at time 1; Hilbert transforms are performed on each intrinsic mode function component to calculate the instantaneous frequency of each intrinsic mode function. The energy density and average instantaneous frequency of each intrinsic mode function are also calculated at different water storage periods. Based on forest disturbance, a comprehensive disturbance intensity index is calculated, and the modal correlation coefficients between each intrinsic mode function and the disturbance intensity at different water storage periods are calculated. The expression is as follows: ； in for Water storage period The modal correlation coefficient between the intrinsic mode function and the disturbance intensity for The time range of the water storage period For the first eigenmode functions The value at time, For the first Each eigenmode function in The average value during the water storage period. for The overall disturbance intensity index at any given time. To normalize the comprehensive disturbance intensity index in Average value during the water storage period; The comprehensive discriminant index is calculated based on the modal correlation coefficient. The comprehensive discriminant indices are then arranged in ascending order. The adaptive discriminant threshold is determined using the maximum curvature method, expressed as follows: ； ； ； in for Water storage period The comprehensive discriminant index of the intrinsic mode functions, for Adaptive discrimination threshold for water storage period For threshold modality index, , , To determine the index weights, for Energy density during the water storage period , They are respectively The baseline instantaneous frequency and maximum energy density during the water storage period. This is an adjustment coefficient for wet and dry seasons. The length of the main channel of the river. The average slope This represents the number of intrinsic modes; Based on the comprehensive discrimination index and the adaptive discrimination threshold, the intrinsic mode functions during the water storage period are divided into basic runoff and disturbed runoff.

3. The nonsteady-state runoff response prediction method considering forest disturbance according to claim 1, characterized in that, The method for obtaining the basic runoff mapping function includes: Watershed characteristics at each water storage period are extracted. Using watershed characteristics and conventional disturbances as inputs and basic runoff as output, time-varying coefficient regression is performed to obtain the basic runoff mapping function. The time-varying coefficients of the basic runoff mapping function are updated by Kalman filtering.

4. The nonsteady-state runoff response prediction method considering forest disturbance according to claim 1, characterized in that, The methods for performing correlation analysis and sensitivity analysis include: Wildfire disturbance, logging disturbance, and vegetation adjustment disturbance are extracted at each time point. The wildfire disturbance includes wildfire type, wildfire form, wildfire area, and duration. The logging disturbance includes logging frequency, logging quantity, and logging type. The vegetation adjustment disturbance includes adjusting vegetation type, adjusting quantity, and adjusting ecological purpose. Grey relational analysis was used to calculate the correlation strength among various forest disturbances; Sobol sensitivity analysis was performed on forest disturbance and forest state to obtain the first sensitivity, on forest state and disturbed runoff to obtain the second sensitivity, and on forest disturbance and disturbed runoff to obtain the third sensitivity.

5. The nonsteady-state runoff response prediction method considering forest disturbance according to claim 1, characterized in that, The method for obtaining optimized perturbation sensitivity includes: The sensitivity analysis results are corrected using the correlation analysis results to obtain the optimized perturbation sensitivity, expressed as: ； ； ； in , , These are the first optimized perturbation sensitivity, the second optimized perturbation sensitivity, and the third optimized perturbation sensitivity, respectively. The strength of the correlation between forest disturbance and forest state. The strength of the correlation between forest disturbance and disturbed runoff. The strength of the association between forest condition and disturbed runoff. , , These are respectively the first sensitivity, the second sensitivity, and the third sensitivity. This is the negative feedback adjustment coefficient. This is the forward propagation coefficient.

6. The nonsteady-state runoff response prediction method considering forest disturbance according to claim 1, characterized in that, The method for constructing a disturbance runoff prediction model includes: Historical disturbance runoff and corresponding disturbance data, optimized disturbance sensitivity, and watershed information are combined into a comprehensive set. The comprehensive set is randomly divided into a training set and a test set in a 7:3 ratio. The disturbance runoff prediction model is trained using the training set and the model performance is verified using the test set. The disturbance runoff prediction model includes an input layer, a physical conduction prediction layer, a climate coupling prediction layer, a time prediction layer, an adaptive fusion layer, and an output layer. The physical transmission prediction layer uses forest disturbance, optimized disturbance sensitivity, forest state, and water storage period as pre-inputs to calculate the forest state change, indirect path disturbance runoff, and direct path disturbance runoff caused by the disturbance, respectively. A water storage period gating function is then introduced to predict the disturbance runoff, expressed as follows: in for The output of the physical conduction prediction layer at any given time, i.e., the predicted value of disturbance runoff. , , To optimize perturbation sensitivity, This represents the weight vector of the direct impact of forest disturbance on disturbed runoff. This represents the weight vector of the indirect impact of forest condition on disturbed runoff. This is the perturbation-state transition matrix. for Forest perturbation vector at time step. for The forest state vector at time step 1. This is the gate function for predicting the water storage period of the physical conduction layer. for The category of water storage period corresponding to the current moment; The climate-coupled prediction layer embeds a basic runoff mapping function to calculate basic runoff based on watershed characteristics. It uses forest disturbance, conventional disturbance, basic runoff, topographic features, and water storage period as pre-inputs. The baseflow saturation is calculated using the predicted basic runoff value against the hydrological background. Enhanced disturbance vectors are obtained by quantifying conventional disturbances through effective climate disturbances. A water storage period gating function is then introduced to predict disturbed runoff. The expression is as follows: ； ； ； ； in for The output of the climate-coupled prediction layer at any given time, i.e., the predicted value of perturbation runoff. For base current saturation, To enhance the forest perturbation vector, For effective climate disturbance, To increase the weighting of forest disturbance impacts, For the effective weighting of climate disturbances, This is the gating function for climate-coupled prediction of the water storage period in the layer. for Basic runoff at any given time This is the historical basic runoff reference average. The saturation rate coefficient, This is the climate amplification factor. , , This is the climate effect coefficient. To standardize precipitation anomalies, To standardize temperature anomalies; The time prediction layer uses a BP neural network to capture disturbed runoff and corresponding disturbance data, optimizes the nonlinear relationship between disturbance sensitivity, basic runoff, and forest state structure, and predicts the duration of disturbance runoff. The adaptive fusion layer dynamically determines the fusion weight by calculating the residual entropy and physical dominance of the physical conduction prediction layer and the climate coupling prediction layer, and then weights and fuses the disturbance runoff predicted by the intermediate layer to obtain the final predicted disturbance runoff.

7. The nonsteady-state runoff response prediction method considering forest disturbance according to claim 6, characterized in that, The disturbance runoff prediction model employs a four-objective hybrid loss function to improve prediction accuracy, expressed as follows: ； ； ； ； ； in It is a four-objective mixed loss function. For fitting loss, For physical loss weights, For physical loss, To balance the loss weights, To balance the losses, To maintain the total amount of loss weight, For the total amount of loss due to conservation, For the sample size, for The predicted perturbation runoff output by the perturbation runoff prediction model at any given time. for The historical observed disturbed runoff at any given time was obtained by data decomposition of the non-steady-state runoff series. for The perturbation runoff predicted by the physical conduction prediction layer at any given time. for Predicted runoff values ​​for reconstructed physical transport chains at any given time. , They are respectively The fusion weights of the time-space physical transmission prediction layer and the climate coupling prediction layer for The predicted basic runoff output by the climate coupling prediction layer at any given time. for The predicted perturbation runoff output by the perturbation runoff prediction model at any given time. for The original total runoff at that time.

Images