A distributed flood forecasting method based on multi-layer soil moisture assimilation

Abstract

The application discloses a distributed flood forecasting method based on multi-layer soil humidity assimilation, and belongs to the field of flood forecasting. First, a distributed hydrological model is constructed, and a soil profile is divided into three layers with different characteristics, i.e., a surface layer, a lower layer and a deep layer; second, remote sensing observation data are collected and processed, including four parts of spatial scale unification, unit unification, time scale unification and bias correction; finally, a multi-layer soil humidity assimilation module is constructed based on the distributed hydrological model, and an ensemble Kalman filter algorithm and a Kalman gain transfer method are adopted to correct simulation errors of surface layer and lower layer soil humidity of the distributed hydrological model, and improve the flood forecasting precision. In the application, the Kalman gain of the surface layer is transferred to the lower layer, the soil humidity state of the surface layer and the lower layer before the occurrence of the field flood of the distributed hydrological model is corrected, the accuracy and reliability of the flood forecasting are improved, the simulation errors of the model soil humidity can be effectively reduced, and technical support can be provided for flood control and disaster reduction of large and medium-sized basins.

Description

Technical Field

[0001] This invention belongs to the field of flood forecasting and relates to a distributed flood forecasting method based on multi-layer soil moisture assimilation. Background Technology

[0002] Flood forecasting is a crucial component of flood control and disaster reduction systems, playing a vital role in ensuring watershed flood safety and the scientific management of water projects. Watershed hydrological models are the primary technical means of flood forecasting. Among them, distributed hydrological models, by dividing the watershed into grids, can meticulously account for the uneven spatial distribution of rainfall and underlying surface, and are widely used for flood simulation and forecasting in large and medium-sized watersheds. However, these models generally suffer from uncertainties in input data, parameters, and structure during application, leading to the gradual accumulation of errors in state variables. Particularly at the beginning and end of the flood season, simulated soil moisture values ​​deviate significantly from actual conditions, thus affecting flood forecast results. Soil moisture, as a crucial state variable determining runoff generation, directly impacts the accuracy of peak flood size, volume, and timing. Therefore, it is necessary to study how to correct errors in soil moisture state variables to improve the accuracy of flood forecasts using distributed hydrological models.

[0003] Current research primarily uses ground-based observations or satellite remote sensing data to correct soil moisture status in models. Ground-based observation data is limited by sparse stations, limited coverage, and insufficient spatial representativeness, and may contain systematic biases; directly substituting this data for model status can easily introduce errors. In recent years, satellite remote sensing soil moisture products (such as SMAP, SMOS, and ASCAT) have been used to assimilate and correct soil moisture status in models due to their ability to provide wide-coverage, high-resolution surface (0-5cm) soil moisture information. For example, Chinese invention patent CN119150531A proposes a method for calculating the flood resistance capacity of reservoirs considering soil moisture status. This method uses remote sensing observation data to assimilate the surface soil moisture status of the VIC model, improving the accuracy of reservoir inflow flood forecasts. Chinese invention patent CN114036127A proposes a method for improving runoff simulation in hydrological models. This method uses remote sensing observation data to assimilate the surface soil moisture status of the hydrological model, improving the runoff simulation performance of the hydrological model.

[0004] However, the above methods have obvious shortcomings: remote sensing products can usually only obtain accurate surface soil moisture information, which leads to the assimilation process being limited to surface correction. However, the lower soil layer is the main runoff-producing layer and is more critical to flood formation. Existing methods based on surface observation data cannot effectively correct the lower soil moisture, thus limiting the improvement of forecast accuracy. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes a distributed flood forecasting method based on multi-layer soil moisture assimilation. This distributed flood forecasting method is based on surface soil moisture data from remote sensing observations and an ensemble Kalman filter algorithm. By shifting the Kalman gain of the surface layer to the lower layers, it corrects the surface and lower layer soil moisture states before the occurrence of a flood in a distributed hydrological model. In other words, this invention can fully utilize remote sensing observation data and data assimilation methods to correct the multi-layer soil moisture states in a distributed hydrological model, improve the estimation of soil moisture content before a flood, and thus improve the accuracy and reliability of flood forecasting.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A distributed flood forecasting method based on multi-layer soil moisture assimilation, the distributed flood forecasting method comprising the following steps:

[0008] Step 1: Construct a distributed hydrological model. Specifically:

[0009] Based on the topography, soil type, vegetation cover, land use, and hydrological and meteorological data of the study watershed, the watershed was divided into grids, and a distributed hydrological model was established to provide a basic framework for process simulation and state updating of subsequent multi-layer soil moisture data assimilation. The distributed hydrological model divides the soil profile into three layers with different characteristics: surface, subsurface, and deep. The surface and subsurface layers are the main control layers affecting surface runoff processes; they directly respond to rainfall input and primarily influence infiltration and surface runoff generation. The deep layer mainly controls groundwater recharge and baseflow processes; its moisture changes are relatively slow, and its impact on flood forecasting results is relatively small. Therefore, subsequent assimilation work only needs to assimilate the soil moisture states of the surface and subsurface layers in the distributed hydrological model.

[0010] Step 2: Collect and process remote sensing observation data.

[0011] The primary source of remote sensing observation data is publicly available global soil moisture datasets. Firstly, soil moisture products differ from distributed models in terms of spatiotemporal scale and physical meaning (units). Secondly, soil moisture observation data suffers from significant systematic errors due to measurement and processing methods, while data assimilation presupposes zero mean and temporal incompatibility between observation and prediction errors. Therefore, this invention eliminates systematic errors between remote sensing observation data and distributed hydrological model simulation data through preprocessing. The specific steps are as follows:

[0012] Step 2.1: Collect publicly available remote sensing soil moisture products applicable to the study watershed to obtain remote sensing observation data, and collect the watershed vegetation cover coefficient (NDVI). The remote sensing observation data and the NDVI are determined based on the publicly available dataset of the target study watershed.

[0013] Step 2.2 involves preprocessing the remote sensing soil moisture data obtained in Step 2.1 to make it suitable for subsequent distributed hydrological model assimilation studies. The preprocessing includes four parts: spatial scale matching, unit unification, temporal scale matching, and bias correction.

[0014] (1) Spatial scale matching;

[0015] First, the remote sensing soil moisture data is cropped to the study watershed area using a mask. Then, the data quality is checked, outliers are corrected, and the spatial coverage within the study watershed is assessed. If the spatial coverage for a given time period is below 90%, data for that period is discarded; if the coverage is above 90%, missing values ​​within that time period are filled in using nearest-neighbor interpolation. Finally, bilinear interpolation is used to match the soil moisture product to the same spatial resolution as the distributed hydrological model.

[0016] (2) Units are standardized;

[0017] Converting the data units of remote sensing observation data to the soil moisture units (mm) used in distributed hydrological models facilitates subsequent assimilation studies.

[0018] (3) Time scale matching;

[0019] Unifying the time scale of remote sensing observation data with that of distributed hydrological models facilitates subsequent assimilation studies.

[0020] (4) Deviation correction;

[0021] The cumulative distribution function method is used to eliminate systematic errors between remote sensing observation data and soil moisture simulation by a distributed hydrological model. The formula for the cumulative distribution function method is as follows:

[0022] (1)

[0023] In the formula, This is data to be corrected; This is the correction result; This is the cumulative distribution function of the simulation results from the distributed hydrological model; It is the inverse function of the cumulative distribution function of remote sensing observation data.

[0024] In summary, the processed remote sensing observation dataset provides a data foundation for the assimilation of multi-layer soil moisture data.

[0025] Step 3: Construct a multi-layer soil moisture assimilation module.

[0026] The multi-layer soil moisture assimilation module is built upon a distributed hydrological model. It employs an ensemble Kalman filter algorithm and a Kalman gain shifting method to correct the simulation errors of surface and lower-layer soil moisture in the distributed hydrological model, thereby improving flood forecast accuracy. The module's construction approach is as follows: First, an assimilation time is set. When the distributed hydrological model reaches the specified assimilation time, the model operation is paused, and the soil moisture status file for that time is saved. Second, prediction and observation datasets are constructed separately to characterize the model's prediction and observation errors. Then, the surface Kalman gain matrix is ​​calculated using the prediction and observation datasets and shifted to the lower layer. Finally, based on the surface and lower-layer Kalman gain matrices, the updated soil moisture status value of the distributed hydrological model is calculated and written to the distributed hydrological model's soil moisture status file. After assimilation at one time, the distributed hydrological model reads the soil moisture status file as the initial state for the next time step, continues running, and cyclically executes the above assimilation process. Specific steps are as follows:

[0027] Step 3.1, set the assimilation time.

[0028] In the operation of the distributed hydrological model, the time settings for multi-layer soil moisture assimilation are first determined. Specifically, this includes the following:

[0029] All available remote sensing soil moisture data within one month prior to a flood event were selected as assimilation time points, which comprehensively reflect the soil moisture characteristics before the flood. When the model reaches an assimilation time point, the system pauses the distributed hydrological model's operation and saves the model's soil moisture state file at that moment. This file serves as the initial state file for the next time period's model operation, enabling continuous simulation after assimilation.

[0030] In summary, after completing the assimilation time setting, the distributed hydrological model pauses operation at the preset assimilation time and starts the multi-layer soil moisture assimilation module.

[0031] Step 3.2 Construct the prediction dataset.

[0032] First, by analyzing meteorological forcing data and infiltration curve parameters from the previous assimilation time to the current assimilation time... Apply perturbation to generate The distributed hydrological model parameter data is used as the initialization background field. The perturbation method is as follows: Gaussian white noise with a mean of 0 and a standard deviation of 0.4 is randomly added to the rainfall, temperature, and wind speed data in the meteorological input; the permeability curve parameters are used as the initialization background field. Randomly add Gaussian white noise with a mean of 0 and a standard deviation of 0.1 times its value.

[0033] Then, using the background meteorological data and model parameters of each set, the distributed hydrological model was driven to run, and the corresponding assimilation time was obtained. The predicted soil moisture values ​​are used to form a prediction data set.

[0034] In summary, the predicted dataset is obtained.

[0035] Step 3.3: Construct the observation dataset.

[0036] Based on the remote sensing observation data preprocessed in step 2, an observation dataset is constructed.

[0037] To characterize the uncertainty of observation errors, Gaussian perturbations are added to the observation data to generate... Set of observations:

[0038] (2)

[0039] In the formula, The observation representing the assimilation time; This represents the set of observation data after perturbation; This represents adding a mean of 0 and a standard deviation of 0 to the observed dataset. Gaussian perturbation.

[0040] Furthermore, the standard deviation Specifically, it refers to the standard deviation of the observation error, which is set as follows:

[0041] (3)

[0042] In the formula, 'a' represents the basic observation error; represents the observation error caused by vegetation cover; where b represents the weighting coefficient, ranging from [0,1], and is set empirically; NDVI is the watershed vegetation cover coefficient collected in step two.

[0043] In summary, the observation data set was obtained.

[0044] Step 3.4, Kalman gain shift and soil moisture status update value calculation.

[0045] First, at each assimilation time point, based on the predicted data set and observed data set obtained in steps 3.2 and 3.3, the surface Kalman gain matrix K is calculated using the ensemble Kalman filter algorithm, and then the Kalman gain matrix is ​​used... Calculate the updated value of surface soil moisture; then, scale the gain matrix accordingly. The process is then moved to the lower layer, where the updated soil moisture state value is calculated to achieve synchronous updates. Further, the moving method is as follows: a scaling factor F is added to the surface Kalman gain matrix K to create the lower layer Kalman gain matrix. The lower layer is updated by multiplying the surface soil moisture update amount by F, as shown in the following formula:

[0046] (4)

[0047] In the formula, This represents an update on the moisture status of the lower soil layer; This represents a dataset of predicted moisture conditions in the lower soil layers. Represents the surface Kalman gain matrix; The representative proportionality coefficient is determined based on the calibration of flood parameters for each flood event, and the parameter range is [0,1]. This represents an update of the surface soil moisture status, among which, This represents the set of observational data on surface soil moisture status. Represents the observation operator. Represents the model prediction matrix. This represents the observed noise disturbance term. Indicates the surface layer. Indicates the lower level. Indicates prediction, Indicates the member number of the set.

[0048] In summary, the updated values ​​of soil moisture status in the surface and lower layers were calculated based on the Kalman gain matrices of the surface and lower layers.

[0049] Step 3.5: Write the updated values ​​of surface and subsurface soil moisture status to the distributed hydrological model's state file. The distributed hydrological model reads the state file as the initial state for the next time step and continues to run until the next assimilation time step to continue performing the assimilation operation.

[0050] Thus, a multi-layered soil moisture assimilation module was constructed. Based on this module, the soil moisture content in the early stages of a flood event can be corrected, thereby improving the accuracy of watershed flood forecasting.

[0051] The beneficial effects of this invention are as follows:

[0052] (1) This invention combines remote sensing surface soil moisture data with data assimilation algorithms to achieve synchronous updates of surface and subsurface soil moisture state variables in a distributed hydrological model.

[0053] (2) This invention overcomes the problems of inconsistent resolution between remote sensing data and model grid and system bias by preprocessing remote sensing data, unifying spatiotemporal scale and bias correction; by adopting the Kalman gain shift method, it solves the limitation of traditional assimilation methods that can only update surface soil moisture and cannot correct the lower soil moisture in time, significantly improving the accuracy of flood forecasting and providing efficient and reliable technical support for flood control and disaster reduction in large and medium-sized watersheds. Attached Figure Description

[0054] Figure 1 This is a flowchart of the present invention.

[0055] Figure 2 This is a soil moisture distribution map before assimilation according to the present invention; Figure 2 (a) in the figure represents the spatial distribution of surface soil moisture before assimilation; Figure 2 (b) in the figure represents the spatial distribution of soil moisture in the lower layer before assimilation.

[0056] Figure 3 This is a soil moisture distribution diagram after assimilation according to the present invention; Figure 3 (a) in the figure represents the spatial distribution of surface soil moisture after assimilation; Figure 3 (b) in the figure represents the spatial distribution of soil moisture in the lower layer after assimilation.

[0057] Figure 4 This is a diagram showing the flood forecast results of this invention. Detailed Implementation

[0058] The present invention will be further described below with reference to specific implementation examples.

[0059] This invention proposes a distributed flood forecasting method based on multi-layer soil moisture assimilation, such as... Figure 1 As shown.

[0060] The upper reaches of the Jialing River were selected as the study area. Originating from the southern foothills of the Qinling Mountains, this basin flows through Gansu, Shaanxi, and Sichuan provinces, covering an area of ​​approximately 27,000 km², and is an important part of the Yangtze River basin. This region belongs to the semi-humid subtropical monsoon climate zone, characterized by hot and rainy summers and cold and dry winters. The average annual temperature is low, and the average annual precipitation is about 600 mm, mainly concentrated from May to October. The rain-flood process in the upper reaches of the Jialing River is complex, with significant spatial heterogeneity of the underlying surface, posing a considerable challenge to flood simulation. Especially at the beginning of the flood season, due to the uneven spatial and temporal distribution of rainfall and the difficulty in accurately depicting the initial soil moisture content, the accuracy of hydrological models in simulating soil moisture and runoff is often insufficient. Based on these findings, a distributed flood forecasting method based on multi-layer soil moisture assimilation is adopted to improve the accuracy of flood forecasting in the upper reaches of the Jialing River.

[0061] Step 1: Construct a distributed hydrological model for the watershed. Specifically:

[0062] Based on the topography, soil type, vegetation cover, land use, and hydrological and meteorological data of the study basin, the basin was divided into grids, and a distributed hydrological model was established to provide a basic framework for process simulation and state updates for subsequent multi-layer soil moisture data assimilation. The runoff generation module of the distributed hydrological model adopted the VIC model with a spatial resolution of 5 km and a temporal resolution of 6 h, dividing the basin into 1164 grids. The runoff collection module, based on collected hydrological station information within the basin, divided the study area into 63 sub-basins. Runoff generation and collection were calculated independently for each sub-basin, with an output node set at the sub-basin outlet, progressively evolving to the basin outlet.

[0063] After constructing a distributed hydrological model for the study area, and combining the measured flow data from four hydrological stations in the study area, the multi-objective genetic algorithm NSGA-II was used to calibrate the model's runoff generation and confluence parameters. Finally, a distributed hydrological model applicable to the upper reaches of the Jialing River was obtained, realizing a reasonable simulation of the runoff generation and confluence process in this region.

[0064] Step 2: Collect and process remote sensing observation data.

[0065] The remote sensing observation data comes from the SMAP soil moisture dataset. First, there are differences in spatiotemporal scale and physical meaning (units) between the soil moisture product and the distributed model. Second, due to measurement and processing methods, the soil moisture observation data has significant systematic errors, while data assimilation requires that the observation and prediction errors have zero mean and are temporally uncorrelated. Therefore, preprocessing is needed to eliminate the systematic errors between the remote sensing observation data and the distributed hydrological model simulation data. The specific steps are as follows:

[0066] Step 2.1: Collect the SMAP soil moisture dataset and the watershed vegetation cover coefficient NDVI. The data source is the National Snow and Ice Data Center (NSIDC).

[0067] Step 2.2 involves preprocessing the collected remote sensing soil moisture data to make it suitable for subsequent distributed hydrological model assimilation studies. Preprocessing includes four parts: spatial scale processing, unit unification, temporal scale processing, and bias correction.

[0068] 1) Uniform spatial scale;

[0069] First, the remote sensing soil moisture data is cropped to the study watershed area using a mask. Then, the data quality is checked, outliers are corrected, and the spatial coverage within the study watershed is assessed. If the spatial coverage for a given time period is below 90%, data for that period is discarded; if the coverage is above 90%, missing values ​​within that time period are filled in using nearest-neighbor interpolation. Finally, bilinear interpolation is used to match the soil moisture product to the same spatial resolution as the distributed hydrological model.

[0070] 2) Units of measurement are standardized;

[0071] The SMAP soil moisture dataset uses volumetric water content (m³ / m³). In the preprocessing step, the units of the remote sensing data need to be converted to the soil moisture units (mm) used by the VIC model to facilitate subsequent assimilation studies.

[0072] 3) Uniform time scale;

[0073] The collected remote sensing data was processed into a 6-hour scale to facilitate subsequent assimilation studies.

[0074] 4) Deviation correction;

[0075] The cumulative distribution function method is used to eliminate systematic errors between remote sensing observation data and soil moisture simulation by a distributed hydrological model. The resulting processed remote sensing observation dataset provides a data foundation for multi-level soil moisture data assimilation.

[0076] The formula for the cumulative distribution function method is as follows:

[0077] (1)

[0078] In the formula, This is data to be corrected; This is the correction result; This is the cumulative distribution function of the simulation results from the distributed hydrological model; It is the inverse function of the cumulative distribution function of remote sensing observation data.

[0079] Step 3: Construct a multi-layer soil moisture assimilation module.

[0080] The multi-layer soil moisture assimilation module is built upon a distributed hydrological model. It employs an ensemble Kalman filter algorithm and a Kalman gain shift method to correct simulation errors in surface and subsurface soil moisture, thereby improving flood forecast accuracy. The module's construction approach is as follows: First, an assimilation time is set. When the distributed hydrological model reaches the specified assimilation time, the model operation is paused, and the soil moisture status file for that time is saved. Second, prediction and observation datasets are constructed separately to characterize the model's prediction and observation errors. Then, the Kalman gain matrix of surface soil moisture is calculated using the prediction and observation datasets, and the surface Kalman gain is shifted to the subsurface soil. Finally, based on the surface and subsurface Kalman gain matrices, the updated values ​​of the distributed hydrological model's soil moisture status variables are calculated and written to the soil moisture status file. After assimilation at one time, the distributed hydrological model reads the soil moisture status file and continues running until the next assimilation time, repeating the above process cyclically. Specific steps are as follows:

[0081] Step 3.1, set up the assimilation window.

[0082] In the operation of the distributed hydrological model, the time settings for multi-layer soil moisture assimilation are first determined. This includes the following:

[0083] All periods within one month prior to a flood event with available remote sensing soil moisture data were selected as assimilation time points. This time period comprehensively reflects the soil moisture evolution characteristics before the flood, providing representative information for model state correction. When the model reaches an assimilation time point, the system pauses the distributed hydrological model's operation and saves the model's predicted soil moisture state file at that moment. This file serves as the initial state file for the next time period's model operation, enabling continuous simulation after assimilation. In summary, after completing the assimilation time settings, the distributed hydrological model pauses operation at the preset assimilation time and initiates the multi-layer soil moisture data assimilation module.

[0084] Step 3.2 Construct the model prediction dataset.

[0085] First, by analyzing meteorological forcing data and infiltration curve parameters from the previous assimilation time to the current assimilation time... Perturbation was applied to generate 50 sets of distributed hydrological model parameter data as the initialization background field. The perturbation method is as follows: Gaussian white noise with a mean of 0 and a standard deviation of 0.4 was randomly added to the rainfall, temperature, and wind speed data in the meteorological input; and permeability curve parameters were perturbed. Randomly add Gaussian white noise with a mean of 0 and a standard deviation of 0.1 times its value.

[0086] Then, using the background meteorological data and model parameters of each set, the distributed hydrological model was driven to run, resulting in 50 sets of predicted soil moisture values ​​corresponding to the assimilation time, forming a prediction dataset.

[0087] In summary, the predicted dataset is obtained.

[0088] Step 3.3: Construct the observation dataset.

[0089] Based on the remote sensing observation data preprocessed in step 2, an observation dataset is constructed.

[0090] To characterize the uncertainty of observation errors, a Gaussian perturbation was added to the observation data, generating 50 sets of observations, as shown below:

[0091] (2)

[0092] In the formula, The observation representing the assimilation time; This represents the set of observation data after perturbation; This represents adding a mean of 0 and a standard deviation of 0 to the observed dataset. The Gaussian perturbation. The standard deviation. Specifically, it refers to the standard deviation of the observation error, which is set as follows:

[0093] (3)

[0094] In the formula, NDVI represents the observation error caused by vegetation cover.

[0095] In summary, the observation data set was obtained.

[0096] Step 3.4, Kalman gain shift and soil moisture status update value calculation.

[0097] First, at each assimilation time point, based on the predicted data set and observed data set obtained in steps 3.2 and 3.3, the surface Kalman gain matrix K is calculated using the ensemble Kalman filter algorithm, and then the Kalman gain matrix is ​​used... Calculate the updated value of surface soil moisture; then, scale the gain matrix accordingly. The process is then moved to the lower layer, where the updated soil moisture state value is calculated to achieve synchronous updates. Further, the moving method is as follows: a scaling factor F is added to the surface Kalman gain matrix K to create the lower layer Kalman gain matrix. The lower layer is updated by multiplying the surface soil moisture update amount by F, as shown in the following formula:

[0098] (4)

[0099] In the formula, This represents an update on the moisture status of the lower soil layer; This represents a dataset of predicted moisture conditions in the lower soil layers. Represents the surface Kalman gain matrix; The representative proportionality coefficient is determined based on the calibration of flood parameters for each flood event, and the parameter range is [0,1]. This represents an update of the surface soil moisture status, among which, This represents the set of observational data on surface soil moisture status. Represents the observation operator. Represents the model prediction matrix. This represents the observed noise disturbance term. Indicates the surface layer. Indicates the lower level. Indicates prediction, Indicates the member number of the set.

[0100] The method for determining the proportionality coefficient F is as follows: the proportionality coefficient F of each sub-basin is calibrated using the NSGA-II method to take into account the spatial heterogeneity of the basin. In order to maximize the assimilation benefit, the calibration range of the proportionality coefficient F is set to [0.8, 1] in this embodiment.

[0101] Table 1, Proportionality coefficient F-value table

[0102]

[0103] In summary, the updated values ​​of soil moisture status in the surface and lower layers were calculated based on the Kalman gain matrices of the surface and lower layers.

[0104] Step 3.5: Write the updated values ​​of surface and subsurface soil moisture status to the distributed hydrological model's state file. The distributed hydrological model reads the state file as the initial state for the next time step and continues to run until the next assimilation time step to continue performing the assimilation operation.

[0105] Thus, a multi-layered soil moisture assimilation module was constructed. Based on this module, the soil moisture content in the early stages of a flood event can be corrected, thereby improving the accuracy of watershed flood forecasting.

[0106] To verify the application effect of the multi-layer soil moisture assimilation module in flood forecasting, this study selected five representative flood events within the watershed for assimilation experiments. By introducing the multi-layer soil moisture data assimilation module into the VIC model framework, the synchronous updating of surface and subsurface soil moisture states was achieved. The results show that the simulation accuracy of each flood event was improved to varying degrees after assimilation. The simulation effects before and after assimilation for each flood event are compared in Table 2.

[0107] Table 2, Assimilation Results

[0108]

[0109] Overall results show that the multi-layer soil moisture assimilation method outperformed the unassimilated distributed hydrological model in the analysis of the five flood events. In terms of flood volume and peak flow simulation, assimilation significantly reduced the relative error and improved the NSE value. Specifically, the average relative error of flood volume decreased from 0.27 to 0.13, the average relative error of peak flow decreased from 0.30 to 0.12, and the average NSE increased from 0.55 to 0.77, demonstrating that the multi-layer assimilation method can effectively enhance the flood forecasting capability of the distributed hydrological model.

[0110] Among them, the flood event with the flood number "20190510" showed the most significant improvement in simulation accuracy after assimilation, with the runoff error reduced by about 50% compared to before assimilation. Figure 2 (a) and (b) in the figure show the spatial distribution of soil moisture before the assimilation of the flood event. Figure 3 In the figures (a) and (b), the spatial distribution of soil moisture after assimilation is shown. Figure 4 This study compares the flood process before and after assimilation. The results show that the overall soil moisture in both the surface and subsurface layers increased after assimilation, particularly in the upper left corner of the watershed. Simultaneously, assimilation significantly improved flood forecast accuracy, indicating that the distributed hydrological model underestimated soil moisture levels in this region, especially in the second layer. This may be related to the fact that the flood event occurred at the beginning of the flood season, and the distributed hydrological model's ability to simulate soil moisture conditions was insufficient. After assimilation, the observed information was effectively integrated into the distributed hydrological model, thereby correcting the soil moisture simulation error and improving the accuracy of flood simulation.

[0111] The above-described embodiments are merely illustrative of the implementation methods of the present invention, but should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A distributed flood forecasting method based on multi-layer soil moisture assimilation, characterized in that, The distributed flood forecasting method includes the following steps: Step 1: Construct a distributed hydrological model; specifically: Based on the topography, soil type, vegetation cover, land use, and hydrological and meteorological data of the study watershed, the watershed is divided into grids, and a distributed hydrological model is established to provide a basic framework for process simulation and state updating for subsequent multi-layer soil moisture data assimilation. The distributed hydrological model divides the soil profile into three layers with different characteristics: surface layer, subsurface layer, and deep layer. Step 2: Collect and process remote sensing observation data; The remote sensing observation data comes from a soil moisture dataset; preprocessing is used to eliminate systematic errors between the remote sensing observation data and the distributed hydrological model simulation data. Step 3: Construct a multi-layered soil moisture assimilation module; A multi-layer soil moisture assimilation module is constructed based on a distributed hydrological model. An ensemble Kalman filter algorithm and a Kalman gain shifting method are used to correct the simulation error of surface and lower soil moisture in the distributed hydrological model, thereby improving the accuracy of flood forecasting. First, the assimilation time is set. Second, prediction and observation datasets are constructed to characterize the model's prediction and observation errors. Then, the surface Kalman gain matrix is ​​calculated using the prediction and observation datasets, and this gain matrix is ​​proportionally shifted to the lower layer to calculate the updated soil moisture state value, achieving synchronous updates. Finally, based on the surface and lower layer Kalman gain matrices, the updated soil moisture state value of the distributed hydrological model is calculated and written to the distributed hydrological model's soil moisture state file. After assimilation at one time step, the distributed hydrological model reads the soil moisture state file as the initial state for the next time step, continues running, and cyclically executes the above assimilation process. Specifically: Step 3.1, set the assimilation time; During the operation of the distributed hydrological model, the time setting for multi-layer soil moisture assimilation is first determined. After the assimilation time setting is completed, when the distributed hydrological model reaches the specified assimilation time, the model operation is paused and the soil moisture status file at that time is saved. Then, the multi-layer soil moisture data assimilation module is started. Step 3.2 Construct the prediction dataset; First, by perturbing the meteorological forcing data and infiltration curve parameters from the previous assimilation time to the current assimilation time, a process is generated. A set of distributed hydrological model parameter data is used as the initial background field. Then, the distributed hydrological model is driven by the meteorological data of each set of background fields and the model parameters to obtain the corresponding assimilation time. The predicted soil moisture values ​​are used to form a prediction data set; in summary, the prediction data set is obtained. Step 3.3: Construct the observation dataset; Based on the preprocessed remote sensing data from step 2, an observation dataset is constructed; Gaussian perturbation is added to the observation data to generate... Set of observations: （2）； In the formula, The observation representing the assimilation time; This represents the set of observation data after perturbation; This represents adding a mean of 0 and a standard deviation of 0 to the observed dataset. Gaussian perturbation ; Step 3.4, Kalman gain shift and soil moisture status update calculation; First, at each assimilation time point, based on the predicted data set and observed data set obtained in steps 3.2 and 3.3, the surface Kalman gain matrix K is calculated using the ensemble Kalman filter algorithm, and then the Kalman gain matrix is ​​used... Calculate the updated value of surface soil moisture; then, scale the gain matrix accordingly. Move to the lower layer and calculate the updated value of the soil moisture status in the lower layer to achieve synchronous updating; In summary, the updated values ​​of soil moisture status in the surface and lower layers were calculated based on the Kalman gain matrices of the surface and lower layers. Step 3.5: Write the updated values ​​of surface and subsurface soil moisture status to the distributed hydrological model state file. The distributed hydrological model reads the state file as the initial state for the next time step and continues to run until the next assimilation time to continue the assimilation operation. Thus, a multi-layered soil moisture assimilation module was constructed, which can correct the soil moisture content in the early stages of a flood event.

2. The distributed flood forecasting method based on multi-layer soil moisture assimilation according to claim 1, characterized in that, Step 2 is described in detail below: Step 2.1: Obtain remote sensing observation data based on the remote sensing soil moisture products of the study watershed, and collect the watershed vegetation cover coefficient (NDVI). Step 2.2 involves preprocessing the remote sensing soil moisture data obtained in Step 2.1 to obtain a processed remote sensing observation dataset, making it suitable for subsequent distributed hydrological model assimilation studies. The preprocessing includes four parts: spatial scale matching, unit unification, temporal scale matching, and bias correction.

3. The distributed flood forecasting method based on multi-layer soil moisture assimilation according to claim 2, characterized in that, In step 2.1, the remote sensing observation data and the vegetation cover coefficient NDVI are determined based on the publicly available dataset of the target study watershed.

4. The distributed flood forecasting method based on multi-layer soil moisture assimilation according to claim 3, characterized in that, Step 2.2 specifically includes: (1) Spatial scale matching; First, the remote sensing soil moisture data is cropped to the study watershed area using a mask. Then, the quality of the remote sensing soil moisture data is checked, outliers are corrected, and its spatial coverage in the study watershed is evaluated. If the spatial coverage of a certain period is less than 90%, the data for that period is removed. If the coverage is greater than 90%, the missing values ​​in the data for that period are filled in using the nearest neighbor interpolation method. Finally, the soil moisture product is matched to the same spatial resolution as the distributed hydrological model using bilinear interpolation. (2) Units are standardized; Convert the data units of remote sensing observation data to the soil moisture units used by the distributed hydrological model; (3) Time scale matching; Unify the time scale of remote sensing observation data with that of distributed hydrological models; (4) Deviation correction; The cumulative distribution function method is used to eliminate the systematic error between remote sensing observation data and soil moisture simulation by distributed hydrological model.

5. A distributed flood forecasting method based on multi-layer soil moisture assimilation according to claim 4, characterized in that, The formula for the cumulative distribution function method is as follows: （1）； In the formula, This is data to be corrected; This is the correction result; This is the cumulative distribution function of the simulation results from the distributed hydrological model; It is the inverse function of the cumulative distribution function of remote sensing observation data.

6. A distributed flood forecasting method based on multi-layer soil moisture assimilation according to claim 1, characterized in that, Step 3.1 specifically involves: All time periods with available remote sensing soil moisture data within one month prior to a flood event are selected as assimilation time points. These time periods can comprehensively reflect the soil moisture characteristics before the flood. When the model reaches an assimilation time point, the system pauses the distributed hydrological model operation and saves the model soil moisture state file at that moment. This file serves as the initial state file for the next time period of model operation, enabling continuous simulation after assimilation.

7. A distributed flood forecasting method based on multi-layer soil moisture assimilation according to claim 1, characterized in that, In step 3.2, the perturbation method is as follows: Gaussian white noise with a mean of 0 and a standard deviation of 0.4 is randomly added to the rainfall, temperature, and wind speed data in the meteorological input; the permeation curve parameters are... Randomly add Gaussian white noise with a mean of 0 and a standard deviation of 0.1 times its value.

8. A distributed flood forecasting method based on multi-layer soil moisture assimilation according to claim 1, characterized in that, In step 3.3, the standard deviation The standard deviation of the observation error is set as follows: （3）； In the formula, 'a' represents the basic observation error; represents the observation error caused by vegetation cover; where b represents the weighting coefficient, ranging from [0,1], and is set empirically; NDVI is the watershed vegetation cover coefficient collected in step two.

9. A distributed flood forecasting method based on multi-layer soil moisture assimilation according to claim 1, characterized in that, In step 3.4, the shift method is as follows: A scaling factor F is added to the surface Kalman gain matrix K to create the lower Kalman gain matrix. The lower layer is updated by multiplying the surface soil moisture update by F times, as shown in the following formula: （4）； In the formula, This represents an update on the moisture status of the lower soil layer; This represents a dataset of predicted moisture conditions in the lower soil layers. Represents the surface Kalman gain matrix; The representative proportionality coefficient is determined based on the calibration of flood parameters for each flood event, and its range is [0,1]. This represents an update of the surface soil moisture status, among which, This represents the set of observational data on surface soil moisture status. Represents the observation operator. Represents the model prediction matrix. This represents the observed noise disturbance term. Indicates the surface layer. Indicates the lower level. Indicates prediction, Indicates the member number of the set.

Images