A method for quantitatively analyzing driving mechanism of lake water level evolution

Abstract

This application relates to the field of hydrological monitoring and water resources analysis technology, and discloses a quantitative analysis method for the driving mechanism of lake water level evolution. This method first acquires multi-source remote sensing data to construct a standardized dataset, and then uses laser altimetry and optical area co-assimilation technology to reconstruct a continuous monthly water level time series including the freezing period. The BFAST algorithm is used to detect structural abrupt changes and divide the evolution stages. A partial least squares structural equation model is constructed to solve the long-term driving weights of multiple hydrological and meteorological factors. Based on anomaly data of terrestrial water storage, ST-DBSCAN spatiotemporal clustering is performed to quantify the intensity of extreme hydrological events. Combined with precipitation lag response parameters, the contribution rate of each driving factor is comprehensively calculated. This invention solves the problems of missing data on the freezing period of high-latitude lakes and multivariate complex collinearity interference, achieving accurate quantitative analysis of the long-term evolution trend of lake water levels and the mechanism of extreme climate disturbances.

Description

Technical Field

[0001] This invention relates to the field of hydrological monitoring and water resources analysis technology, specifically a quantitative analysis method for the driving mechanism of lake water level evolution. Background Technology

[0002] Lakes, as a key component of the global water cycle, are highly sensitive to climate change and human activities, and their water level fluctuations directly affect regional water security and ecosystem stability. Accurately reconstructing long-term water level records and analyzing the underlying driving mechanisms is an important foundation for formulating watershed water resource management strategies.

[0003] Currently, monitoring lake water levels primarily relies on observations from ground-based hydrological stations and satellite remote sensing retrieval. While ground-based stations offer high accuracy, their sparse distribution and high maintenance costs make them difficult to cover remote or transboundary lakes. Satellite remote sensing technology, particularly water area extraction methods based on optical imagery, has become the mainstream approach for acquiring information on lake dynamics. However, existing technologies have limitations when dealing with lakes in high-latitude or high-altitude regions. These areas experience long winters, with lake surfaces remaining frozen for extended periods. Frequent cloud and fog interference further hinders the effective identification of land-water boundaries by optical sensors. Water level sequences obtained solely from optical data often suffer from significant seasonal gaps, making continuous monitoring of water level changes during the freezing period impossible and creating gaps in understanding the hydrological processes throughout the year.

[0004] In analyzing the driving mechanisms of water level changes, existing studies typically employ traditional statistical methods such as correlation analysis or multiple linear regression to directly link meteorological and hydrological factors like precipitation, evaporation, and runoff with water levels. However, in real natural environments, complex coupling relationships and multicollinearity exist among various water cycle variables. Traditional statistical models struggle to effectively eliminate these inter-variable interferences, making it impossible to accurately isolate and quantify the independent contribution rate of a single factor to water level changes, thus reducing the physical interpretability of attribution analysis.

[0005] Furthermore, existing methods primarily focus on long-term linear trends or periodic changes in water levels, tending to examine the evolution of average states on interannual or decadal scales, while paying insufficient attention to the response mechanisms of short-term extreme hydrological events. In reality, sudden climate anomalies such as torrential rains and floods or extreme droughts are often key factors triggering drastic fluctuations in water levels, exhibiting pulse-like and nonlinear characteristics with significant lag effects. Conventional time series analysis methods often treat these extreme fluctuations as noise or mask them during smoothing processes, lacking precise quantitative means to assess the intensity of extreme events and their lagged effects, making it difficult to reveal the dynamic response patterns of lake water levels under extreme climate forcing. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a quantitative analysis method for the driving mechanism of lake water level evolution. This method solves the problems of missing water level monitoring data during the freezing period of high-latitude lakes, the difficulty of decoupling the complex collinearity interference between multiple hydrological factors by traditional statistical models, and the lack of accurate quantitative analysis of extreme climate events and their lag response mechanisms.

[0007] To achieve the above objectives, the present invention provides a quantitative analysis method for the driving mechanism of lake water level evolution, comprising the following steps: Acquire multi-source heterogeneous remote sensing data and preprocess them to construct a standardized multidimensional dataset; Laser altimetry data and optical water area data are extracted from the multidimensional dataset, and a monthly continuous water level time series is reconstructed using the altimetry and area co-assimilation method. The BFAST algorithm is used to perform time series decomposition and detection on the continuous water level time series, identify structural abrupt change points in the water level evolution process, and divide the water level evolution stages accordingly. A partial least squares structural equation model is constructed using the continuous water level time series as the endogenous latent variable and the hydrological and meteorological variables in the multidimensional dataset as the exogenous latent variable, and the long-term driving weights of each hydrological and meteorological variable on the water level are calculated. ST-DBSCAN spatiotemporal density clustering is performed on the land water storage anomaly data in the multidimensional dataset to identify extreme hydrological events corresponding to the continuous water level time series fluctuations and to calculate the event intensity. Perform lag cross-correlation analysis on the precipitation data in the multidimensional dataset and the continuous water level time series to determine the lag response parameters of precipitation to water level; Based on the long-term driving weights, the event intensity, and the hysteresis response parameters, an attribution analysis is performed on the water level changes during the water level evolution stage, the contribution rate of each driving factor is calculated, and an analysis report on the driving mechanism of lake hydrological evolution is generated.

[0008] Preferably, the step of acquiring multi-source heterogeneous remote sensing data and performing preprocessing includes: The geoid difference at the laser footpoint location is obtained by interpolating grid data provided by the Earth's gravity field model. The reference ellipsoid height of the laser altimetry data is subtracted from the geoid difference to convert it into normal high water level, and the monthly water level observation value is calculated. Radiometric calibration and atmospheric correction were performed on optical remote sensing images, and the normalized water index and improved normalized water index were calculated. The optimal segmentation threshold is selected by iterating through gray levels to maximize the inter-class variance between the background class and the target class. The lake area is then extracted using the optimal segmentation threshold to construct a time series of lake area.

[0009] Preferably, the step of reconstructing the continuous water level time series using the height and area co-assimilation method includes: A matching dataset is constructed by selecting laser height measurement water level data and optical water area data within the time matching window; Based on the matching dataset, a quadratic or cubic polynomial function relationship of water level with respect to area is established, and the optical water body area data is converted into an initial simulated water level using the polynomial function relationship; A linear regression assimilation model is constructed to describe the relationship between measured water level data of hydrological stations and the initial simulated water level. The initial simulated water level is then corrected using the linear regression assimilation model to generate a monthly continuous water level dataset.

[0010] Preferably, the step of identifying structural abrupt change points in the water level evolution process using the BFAST algorithm includes: A time series additive decomposition model is established to decompose the continuous water level time series into trend components, seasonal components, and residual components; Construct a piecewise linear fitting model for the trend component and a harmonic fitting model for the seasonal component; The number and location of breakpoints are determined using the least squares principle, and the significance of breakpoints is determined using the F-test. The continuous subinterval with a slope approaching zero before the break point is defined as the baseline period, and the subinterval with a slope not equal to zero after the break point is defined as the change period.

[0011] Preferably, the step of constructing a partial least squares structural equation model to solve for the long-term driving weights of multiple factors on water level includes: Lake water level is defined as an endogenous latent variable, and lake inflow, precipitation, snow water equivalent, lake surface evaporation and outflow are defined as exogenous latent variables. Construct an internal model describing the linear relationship between latent variables, and an external model describing the relationship between observed variables and latent variables; The partial least squares iterative estimation algorithm is used to solve the internal model and the external model, and the path coefficients and standardized weights are output.

[0012] Preferably, the step of performing ST-DBSCAN spatiotemporal density clustering based on the terrestrial water storage anomaly data in the multidimensional dataset includes: Pixels with standardized values ​​greater than a set positive threshold are selected as candidate points for extreme wetness, and pixels with standardized values ​​less than a set negative threshold are selected as candidate points for extreme drought. Calculate the spatial and temporal Euclidean distances between candidate points; If the spatial Euclidean distance is less than the set spatial search radius and the temporal Euclidean distance is less than the set time interval parameter, then the two points are determined to be density-reachable and clustered into the same extreme hydrological event.

[0013] Preferably, the step of identifying extreme hydrological events and calculating event intensity further includes: For all grid cells clustered into the same extreme hydrological event, calculate the product of the water storage anomaly value and the physical area of ​​each grid cell; The event intensity of the extreme hydrological event is obtained by summing the products of all grid cells over the duration of the event.

[0014] Preferably, the step of determining the hysteresis response relationship between precipitation and water level includes: The ratio of the covariance to the product of variance of the precipitation series and the water level series at different lag times is calculated to obtain the lag cross-correlation coefficient. The lag time corresponding to the lag cross-correlation coefficient with the largest absolute value is selected as the optimal lag time; Based on the optimal lag time, a linear regression model of precipitation increment events and water level rise events is established, and the least squares method is used to solve the regression coefficients to quantify the impact of precipitation increment on water level rise.

[0015] Preferably, the steps of the parsing driving mechanism include: The standardized weights output by the partial least squares structural equation model are integrated with the event intensity of the extreme hydrological event; The contribution rate of each driving factor and extreme hydrological event to water level change was quantitatively decomposed using effect values; By comparing the contribution rates of each driving factor, the type of dominant driving factor can be identified.

[0016] The step of acquiring and preprocessing multi-source heterogeneous remote sensing data further includes: Gravity satellite data was acquired, and anomaly signals of land water storage were extracted. The scale factor method was used to perform amplitude recovery correction on the anomaly signals of land water storage to eliminate signal leakage errors, and spatial resampling was performed. Meteorological reanalysis data is acquired, and raster data are aggregated into watershed-scale monthly series data using inverse distance weighted interpolation or kriging interpolation. This data, along with the land water storage anomaly signal, is then incorporated into the standardized multidimensional dataset.

[0017] This invention provides a quantitative analysis method for the driving mechanism of lake water level evolution. It has the following beneficial effects: 1. This invention constructs a synergistic assimilation model of laser altimetry and optical water surface area, leveraging the advantages of strong penetration and high vertical ranging accuracy of lidar to compensate for the inability of optical remote sensing to extract effective water surface during the winter ice-covered period or when obscured by clouds and fog. It can reconstruct a monthly water level time series with a long time span, good continuity, and including information on the freezing period, even in the absence of ground measurement stations. This ensures the integrity of hydrological data at the interannual and seasonal scales, providing a discontinuous data foundation for subsequent attribution analysis.

[0018] 2. This invention introduces a partial least squares structural equation model, using highly correlated variables such as precipitation, evaporation, and inflow as exogenous latent variables to construct a network model; it can solve the complex interactions between various natural factors, and quantitatively separate the independent contribution rate of each factor to water level changes by calculating path coefficients and standardized weights; it enables the analysis results to clearly identify the dominant driving factors, and improves the statistical reliability and physical interpretability of attribution analysis in a multivariate coupled environment.

[0019] 3. This invention realizes the automated extraction and parameterization of extreme wet and dry events within the watershed. By performing spatiotemporal matching and lag regression analysis on the identified event intensity, duration and water level fluctuation process, it can not only capture the direct conversion efficiency of precipitation increment on water level pulse rise, but also distinguish between normal seasonal fluctuations and sudden water level rises and falls caused by extreme climate anomalies, providing a quantitative analytical basis for responding to extreme hydrological disasters. Attached Figure Description

[0020] Figure 1 A flowchart illustrating the overall process of the quantitative analysis method for driving lake hydrological evolution provided in this embodiment of the invention; Figure 2 This is a schematic diagram illustrating the definition of water level change events in this invention; Figure 3 The following are the monthly water level change trend charts (a) and (bc) comparisons of water levels during the freezing and non-freezing periods of a certain lake reconstructed by this invention. Figure 4 This is a graph showing the weighted contribution of each factor to water level changes in the PLS-SEM model of this invention. Figure 5 This invention presents the monthly (a) and annual (b) precipitation changes in a certain lake basin from 2002 to 2023. Figure 6 The graph (ab) shows the relationship between the magnitude and duration of continuous precipitation increment events and water level rise events in this invention. Figure 7 This is a graph (ad) showing the intensity and duration of extreme wet and dry events and their correlation with water level fluctuations. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example: Please see the appendix Figure 1 This invention provides a method for quantitative analysis of the driving mechanism of lake water level evolution, comprising the following steps: First, multi-source heterogeneous remote sensing data acquisition and preprocessing are performed. This step acquires laser altimetry data, optical imagery data, terrestrial water storage anomaly data, and meteorological reanalysis data. For laser altimetry data, elevation datum conversion is performed using a geoid model to calculate monthly water level observations. For optical imagery data, radiometric calibration and atmospheric correction are performed, and the water body extent is extracted using the normalized water index and improved normalized water index, combined with an adaptive thresholding method, to construct a time series of lake area.

[0023] Next, water level time series reconstruction based on water level-area relationship curves is performed. This step involves spatiotemporally matching the acquired laser height measurement water level data with synchronized optical water area data. Based on the matched data pairs, a quadratic or cubic polynomial function relationship is established, and this function relationship is used to transform all optical water area sequences into water level change sequences. On this basis, a linear regression model is constructed in conjunction with ground-measured data to assimilate and correct the transformed water level sequences, generating a long-term monthly continuous water level dataset.

[0024] The next step involves using a breakpoint detection algorithm to identify key turning points in water level evolution. This step applies the BFAST algorithm to decompose the reconstructed water level time series, separating the series into trend, seasonal, and residual terms. Statistical tests are then used to identify structural change points in the trend term, determining the critical periods for sustained water level rises or falls. Based on this, the hydrological evolution process is divided into a baseline period and a change period.

[0025] After completing the time series decomposition, a partial least squares structural equation model is constructed to quantify the contributions of multiple factors. This step defines lake level as an endogenous latent variable and lake inflow, precipitation, snow water equivalent, lake surface evaporation, and outflow as exogenous latent variables. Observational index data corresponding to each latent variable are collected, and the scores and path coefficients of the latent variables are estimated through iterative algorithms. The standardized weights of each exogenous latent variable on the changes of the endogenous latent variable are calculated.

[0026] Simultaneously or subsequently, an algorithm based on spatiotemporal density clustering is executed to identify extreme wet and dry events in the watershed. This step uses the ST-DBSCAN algorithm to perform spatiotemporal clustering processing on the terrestrial water storage anomaly data, sets a standard deviation threshold to define candidate pixels for extreme events, and sets spatial search radius and time interval parameters; the algorithm aggregates candidate pixels based on density accessibility and extracts the intensity, duration, and spatial extent features of extreme wet and dry events.

[0027] Further, cross-correlation analysis is performed to determine the lag response relationship between precipitation and water level. This step calculates the lag cross-correlation coefficient between the precipitation series and the water level series, and identifies the optimal lag time corresponding to the maximum correlation coefficient by traversing different lag durations. Based on this optimal lag time, a linear regression model is established for precipitation increment events and water level rise events to assess the magnitude of the impact of precipitation on water level fluctuations.

[0028] Finally, the comprehensive extreme event characteristics and weighted results analysis driving mechanism is executed. This step combines the calculated factor weights with the extracted extreme event intensities to quantitatively decompose the contribution rate of each driving factor to water level changes, identify the dominant driving factor types, and generate the analysis results of the lake hydrological evolution driving mechanism.

[0029] In one embodiment of the present invention, the quantitative analysis method for the driving mechanism of lake hydrological evolution first performs the steps of acquiring multi-source heterogeneous remote sensing data and preprocessing it to construct a standardized multidimensional dataset. This step ensures the uniformity of all data sources in terms of spatial and temporal references in subsequent analyses. The specific implementation process includes the following sub-steps: Processing of laser altimetry data: Laser altimetry data from ICESat / ICESat-2 satellites was acquired, and laser footpoint data within the lake boundary were extracted. The original recorded reference ellipsoidal height was converted to normal high water level using an Earth gravity field model. For each valid laser footpoint, the geoid difference at that point was obtained through grid data interpolation provided by the EGM2008 model, and the normal high water level was calculated using the following formula: ; In the formula, This represents the normal high after conversion, i.e., the lake water level value; The ellipsoidal height of the laser foot point (based on satellite observations of the WGS84 ellipsoid); This represents the geoid difference at the footpoint location; after single-point transformation, it is expressed as 3 times the standard deviation. The criteria for outlier removal include eliminating noise points that deviate from the mean by more than three standard deviations, and calculating the monthly water level observation value by taking the arithmetic mean of the normal heights of all valid lake surface laser footprints within a single month. : ; In the formula, This refers to the number of valid lake surface footholds within a single month. For the first The normal height of each foot point.

[0030] Preprocessing of optical remote sensing images: Optical images from Landsat series satellites are acquired, and image data with cloud cover below a preset threshold are filtered out; radiometric calibration is performed to convert the raw digital quantization values ​​of image pixels into apparent radiance, using the following formula: ; In the formula, Apparent radiance; The original grayscale value of the image pixel; This refers to the band gain coefficient. The above coefficients are band offset coefficients, all of which are read from the image metadata file. For atmospheric correction processing, those skilled in the art can use the FLAASH model or other equivalent atmospheric correction algorithms to convert apparent radiance into surface reflectance.

[0031] Water body index calculation and water body extent extraction. Based on atmospherically corrected surface reflectance data, a normalized water body index and an improved normalized water body index are calculated to enhance water body characteristics and suppress vegetation and soil background. The calculation formulas are as follows: ; ; In the formula, Reflectivity in the green band; Reflectivity in the near-infrared band; The reflectivity is in the mid-infrared band; in actual operation, the following is adopted. and Logical intersection or preferred based on terrain features As the primary criterion for judgment.

[0032] Lake area extraction based on adaptive threshold. The optimal segmentation threshold for water index images is automatically determined using the maximum inter-class variance method; assuming the grayscale range of the water index images is [missing information]. The total number of pixels is Iterate through all possible gray levels The image is segmented into background categories. (Non-water bodies) and target categories (Water bodies), calculate inter-class variance Selecting Reaching the maximum value As the optimal segmentation threshold: ; use The image was binarized, and the total number of water body pixels was counted. Combined with the spatial resolution of the image (i.e., the actual area represented by a single pixel) The lake's water area was calculated. : ; By performing the above operations on each image in the time series, a time series of lake area can be constructed.

[0033] Standardization processing of other auxiliary data. Gravity satellite data (GRACE / GRACE-FO) was acquired, and terrestrial water storage anomaly signals were extracted and spatially resampled to meet the needs of watershed-scale analysis. Simultaneously, meteorological reanalysis data (including precipitation, evaporation, runoff, and snow water equivalent) were acquired, and raster data were aggregated into watershed-scale monthly series data using inverse distance weighted interpolation or Kriging interpolation methods. These processed data collectively constitute a standardized multidimensional dataset.

[0034] Please see the appendix Figure 5 , Figure 5 This presentation showcases the processed precipitation series data from a lake basin, a core component of the dataset. Figure 5 (a) The vertical axis represents precipitation, and the pink line represents monthly precipitation, clearly showing the seasonal pulse characteristics of precipitation on a monthly scale (distinct rainy and dry seasons). The blue bars identify extreme wet events superimposed on seasonal fluctuations. Figure 5 (b) With year on the horizontal axis and annual precipitation on the vertical axis, the data is presented through the display of annual cumulative precipitation and linear regression fitting analysis (showing the trend equation as y = 4.42x + 1228.9, with a coefficient of determination R). 2 =0.03), reflecting the interannual climate oscillation pattern of the basin between 2002 and 2023; the high-precision meteorological forcing data provides a reliable input benchmark for the subsequent decoupling of the natural driving mechanism of water level evolution.

[0035] Subsequently, this embodiment employs a method based on height measurement and area co-assimilation for water level reconstruction. This process specifically includes the following steps: Construct synchronous observation data pairs. Select ICESat / ICESat-2 laser elevation water level data and Landsat optical water area data acquired during the multi-source heterogeneous remote sensing data acquisition and preprocessing process; set the time matching window to... Each day, water level measurements that are synchronized or nearly synchronized in time are selected. With water area Construct a water level area matching dataset.

[0036] Construct and solve a polynomial fitting model of water level and area. Establish a quadratic or cubic polynomial function relationship between water level and area; taking a quadratic polynomial model as an example, its functional relationship is: ; The least squares method is used to estimate the parameters of the above model, and the objective function is to minimize the sum of squared residuals. The fitting coefficients are then determined by solving this function. The optimal solution is obtained by using this functional relationship to convert the entire optical water area sequence into the initial simulated water level.

[0037] A linear regression assimilation correction model was constructed using collected measured water level data from hydrological stations. Initial simulated water level for polynomial model inversion Corrections were made; the linear regression assimilation model was constructed as follows: ; In the formula, The slope of the linear regression; The intercept is used for linear regression; the model parameters are solved using the least squares method. and .

[0038] Generate a continuous monthly water level dataset. Utilize long-term, high-frequency Landsat water area sequences obtained through acquisition and preprocessing of multi-source heterogeneous remote sensing data. The initial simulated water level is calculated by substituting it into the optimal polynomial model determined by constructing and solving the polynomial fitting model of water level area. Then, the initial simulated water level is corrected by using the linear assimilation parameters determined by constructing the linear regression assimilation correction model, and the final monthly continuous water level is obtained.

[0039] Based on the application results of a certain lake as an example, please refer to the appendix. Figure 3 ,like Figure 3 As shown in (a), with Year on the horizontal axis and Water Level on the vertical axis, the reconstructed water level time series maintains a high degree of continuity across the time span, clearly reflecting the interannual fluctuation trend of water level. Addressing the issue of high-latitude lakes freezing in winter, which prevents the extraction of water area from optical images, this method effectively recovers water level information during the freezing period—that is, thanks to direct observation of ice / water surface height using laser altimetry data—by means of directly observing the ice / water surface height. Figure 3 (b) Icecover is marked with a dashed box; such as Figure 3 (b) Comparison of the monthly water level process lines for the two periods of 2002-2012 and 2012-2023, and Figure 3(c) shows the interannual evolution of the ice-free season and the ice season. The reconstructed data fully preserves the differences in water level statistical characteristics between the ice-free and non-ice-free seasons, verifying the robustness of the co-assimilation method in all-season water level monitoring.

[0040] Furthermore, this embodiment employs the BFAST algorithm to process the reconstructed continuous water level time series based on the water level-area relationship curve, identifying structural abrupt changes in water level evolution; the specific implementation steps are as follows: Establish a time series additive decomposition model. Reconstruct the monthly water level time series. Decomposed into trend, seasonal fluctuations, and random noise: ; Construct a piecewise linear fitting model for the trend component. Assume the trend term... Given a piecewise linear function, its expression within the interval between the breakpoints is: .

[0041] Construct a harmonic fitting model for the seasonal components. For the seasonal components... Fourier series function fitting is used: ; Perform breakpoint significance testing and optimal breakpoint identification. Determine the number and location of breakpoints using the least squares principle, and then employ... The significance level is determined using the test method. The formula for calculating the statistic is: ; Define the baseline and variation periods of water level evolution. Ensure the slope before the breakpoint is met. The longest continuous subinterval is defined as the base period; after the breakpoint The sub-interval is defined as the period of change.

[0042] Based on this, this embodiment adopts a dual-model strategy. First, it constructs a partial least squares structural equation model to quantify the contributions of multiple factors.

[0043] Define the model variables. Define the lake water level as an endogenous latent variable, and define the lake inflow, precipitation, snow water equivalent, lake surface evaporation, and outflow as exogenous latent variables.

[0044] Model building and solution.

[0045] The internal model expression is: ; The external model expression is: ; The model is solved using a partial least squares iterative estimation algorithm to calculate path coefficients and standardized weights; please refer to the appendix. Figure 4 Taking the analysis results of a certain lake as an example, the figure compares RFimportance (Random Forest Importance) and PLS-SEMweigts (Partial Least Squares Structural Equation Model Weights). The blue bars visually represent the control effect of various hydrological and meteorological factors—including Rin (inflow), P (precipitation), SWE (snow water equivalent), E (evaporation), and Rout (outflow)—on water level changes. Positive weights indicate that the factor has a positive effect on water level (e.g., P (precipitation), Rin (inflow)), while negative weights indicate a negative effect (e.g., E (evaporation), Rout (outflow)). Through comparison... Figure 4 The absolute value of the standardized weights (PLS-SEMweights) of each factor can be used to quantitatively determine the relative importance ranking of each factor in the long-term water level evolution, thereby identifying the primary driving force affecting the water level fluctuation of the lake.

[0046] Subsequently, this embodiment implements the identification of extreme dry and wet events in the watershed based on the spatiotemporal density clustering algorithm, and the identification of extreme hydrological events based on the ST-DBSCAN algorithm.

[0047] Candidate point selection. Using terrestrial water storage anomaly data retrieved from GRACE / GRACE-FO, standardized values ​​greater than one standard deviation are defined. Pixels with a positive threshold (i.e., less than -1 standard deviation) are considered extreme wetness candidate points. Pixels that meet the negative threshold (i.e., those with a negative threshold) are considered candidates for extreme drought.

[0048] Spatiotemporal density clustering. A spatiotemporal density clustering algorithm is used to aggregate candidate pixels, with a set spatial search radius. Calculate the time interval and minimum number of neighboring pixels; calculate the spatial Euclidean distance between data points. Euclidean distance of time Only when and Only when both conditions are met is it determined that the two points are density-reachable, and thus they are clustered into the same event.

[0049] Calculate the intensity characteristics of extreme events. Event intensity is defined as the sum of the products of water storage anomalies and their areas in all relevant grid cells during the event's duration, calculated using the following formula: ; In the formula, Indicates the intensity of the event; Indicates the first Time of the first The actual physical area of ​​each grid cell; Indicates the first Time of the first Seasonal TWSA outliers in individual grid cells.

[0050] Please see the appendix Figure 7 As shown, the horizontal axis represents the year, illustrating the spatiotemporal characteristics of extreme wet and dry events in a certain lake basin identified using the algorithm described above, and their correspondence with water level evolution; among which, Figure 7 (a) Extreme wetevents and Figure 7 (c) Extreme drought events are represented by bubble color intensity (TWSAmean, i.e., event intensity) and bubble size (Duration), quantifying the characteristics of each independent extreme event and revealing that high-intensity events are often accompanied by long durations. Figure 7 (b) and Figure 7 (d) Map the time windows of these events onto the Waterlevel process line, where Figure 7 (b) Using + (m) (water level rise) indicates a positive fluctuation. Figure 7 (d) Using - (m) (the magnitude of water level drop) represents negative fluctuations, which intuitively confirms the high synchronicity between extreme wet events at the watershed scale and the sharp rise in lake water levels, while extreme drought events correspond to a continuous decline in water levels. Through this correlation analysis, the periods of sudden changes in water levels caused by extreme climate anomalies can be clearly distinguished.

[0051] Finally, this embodiment performs lag cross-correlation analysis to determine the lag response relationship of precipitation level, and conducts lag cross-correlation analysis and comprehensive attribution.

[0052] Calculate the lag cross-correlation coefficient of the precipitation water level series.

[0053] ; In the formula, This refers to the lag time.

[0054] Identify the optimal lag time. Select the lag time corresponding to the largest absolute value. As the optimal lag time, and requiring significance testing .

[0055] Construct a linear regression model for water level rise due to increased precipitation. (Combined with the attached...) Figure 2As shown, the horizontal axis represents Time, and the vertical axis represents Event (in this case, water level). The specific definition method for the water level rise event is as follows: In the reconstructed continuous water level time series, all local minima (i.e., in the figure) are identified. Corresponding time ) and the subsequent local maxima (i.e., in the figure) Corresponding time ); Define from the trough moment To the next peak moment For water level rise events; define precipitation increment events. and water level rise events (corresponding to the pink area marked in the image) The water level rise event The calculation is the difference between the peak water level and the trough water level (i.e., the formula shown in the figure). The precipitation increment event; Calculated as the duration of the rising event ( to ), combined with optimal lag time The corrected cumulative precipitation is used to build a model: ; The regression coefficients are solved using the least squares method. and And calculate the coefficient of determination. Evaluate the model's explanatory power.

[0056] Please see the appendix Figure 6 As shown, the horizontal axis represents the year, illustrating the correspondence between continuous precipitation increment events and water level rise events extracted from a certain lake between 2002 and 2023; among them, Figure 6 (a) The Precipitation section uses the depth of bubble color to represent + (mm) (precipitation increment) reveals a linear positive correlation between the cumulative magnitude of precipitation events and the magnitude of water level rise. The high coefficient of determination indicates that precipitation is the decisive factor in triggering short-term pulse-like rises in water level. Figure 6 (b) The Waterlevel section uses bubble color to represent + (m) (water level rise), combined with the bubble size representing Duration in the figure, shows the synchronicity of the event duration, indicating that long-duration heavy precipitation processes usually correspond to longer-duration water level rise processes. Based on the regression slope 'a' fitted in the figure, the specific conversion efficiency of unit precipitation increment into water level rise is accurately quantified.

[0057] This study integrates extreme event characteristics and weighted results to analyze the driving mechanism. It combines standardized path coefficients obtained from PLS-SEM with extreme event intensity characteristics; and utilizes effect values... Quantitatively analyze the contribution rate of each driving factor and extreme event to water level change; if the contribution rate of a certain factor is higher than that of other factors, then that factor is determined to be the dominant driving factor.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A quantitative analysis method for the driving mechanism of lake water level evolution, characterized in that, Includes the following steps: Acquire multi-source heterogeneous remote sensing data and preprocess them to construct a standardized multidimensional dataset; Laser altimetry data and optical water area data are extracted from the multidimensional dataset, and a monthly continuous water level time series is reconstructed using the altimetry and area co-assimilation method. The BFAST algorithm is used to perform time series decomposition and detection on the continuous water level time series, identify structural abrupt change points in the water level evolution process, and divide the water level evolution stages accordingly. A partial least squares structural equation model is constructed using the continuous water level time series as the endogenous latent variable and the hydrological and meteorological variables in the multidimensional dataset as the exogenous latent variable, and the long-term driving weights of each hydrological and meteorological variable on the water level are calculated. ST-DBSCAN spatiotemporal density clustering is performed on the land water storage anomaly data in the multidimensional dataset to identify extreme hydrological events corresponding to the continuous water level time series fluctuations and to calculate the event intensity. Perform lag cross-correlation analysis on the precipitation data in the multidimensional dataset and the continuous water level time series to determine the lag response parameters of precipitation to water level; Based on the long-term driving weights, the event intensity, and the hysteresis response parameters, an attribution analysis is performed on the water level changes during the water level evolution stage, the contribution rate of each driving factor is calculated, and an analysis report on the driving mechanism of lake hydrological evolution is generated.

2. The quantitative analysis method for the driving mechanism of lake water level evolution according to claim 1, characterized in that, The steps of acquiring multi-source heterogeneous remote sensing data and performing preprocessing include: The geoid difference at the laser footpoint location is obtained by interpolating grid data provided by the Earth's gravity field model. The reference ellipsoid height of the laser altimetry data is subtracted from the geoid difference to convert it into normal high water level, and the monthly water level observation value is calculated. Radiometric calibration and atmospheric correction were performed on optical remote sensing images, and the normalized water index and improved normalized water index were calculated. The optimal segmentation threshold is selected by iterating through gray levels to maximize the inter-class variance between the background class and the target class. The lake area is then extracted using the optimal segmentation threshold to construct a time series of lake area.

3. The quantitative analysis method for the driving mechanism of lake water level evolution according to claim 1, characterized in that, The steps for reconstructing continuous water level time series using the height and area co-assimilation method include: A matching dataset is constructed by selecting laser height measurement water level data and optical water area data within the time matching window; Based on the matching dataset, a quadratic or cubic polynomial function relationship of water level with respect to area is established, and the optical water body area data is converted into an initial simulated water level using the polynomial function relationship; A linear regression assimilation model is constructed to describe the relationship between measured water level data of hydrological stations and the initial simulated water level. The initial simulated water level is then corrected using the linear regression assimilation model to generate a monthly continuous water level dataset.

4. The quantitative analysis method for the driving mechanism of lake water level evolution according to claim 1, characterized in that, The steps for identifying structural abrupt changes in the water level evolution process using the BFAST algorithm include: A time series additive decomposition model is established to decompose the continuous water level time series into trend components, seasonal components, and residual components; Construct a piecewise linear fitting model for the trend component and a harmonic fitting model for the seasonal component; The number and location of breakpoints are determined using the least squares principle, and the significance of breakpoints is determined using the F-test. The continuous subinterval with a slope approaching zero before the break point is defined as the baseline period, and the subinterval with a slope not equal to zero after the break point is defined as the change period.

5. The quantitative analysis method for the driving mechanism of lake water level evolution according to claim 1, characterized in that, The steps for constructing a partial least squares structural equation model to solve for the long-term driving weights of multiple factors on water level include: Lake water level is defined as an endogenous latent variable, and lake inflow, precipitation, snow water equivalent, lake surface evaporation and outflow are defined as exogenous latent variables. Construct an internal model describing the linear relationship between latent variables, and an external model describing the relationship between observed variables and latent variables; The partial least squares iterative estimation algorithm is used to solve the internal model and the external model, and the path coefficients and standardized weights are output.

6. The quantitative analysis method for the driving mechanism of lake water level evolution according to claim 1, characterized in that, The step of performing ST-DBSCAN spatiotemporal density clustering based on the terrestrial water storage anomaly data in the multidimensional dataset includes: Pixels with standardized values ​​greater than a set positive threshold are selected as candidate points for extreme wetness, and pixels with standardized values ​​less than a set negative threshold are selected as candidate points for extreme drought. Calculate the spatial and temporal Euclidean distances between candidate points; If the spatial Euclidean distance is less than the set spatial search radius and the temporal Euclidean distance is less than the set time interval parameter, then the two points are determined to be density-reachable and clustered into the same extreme hydrological event.

7. The quantitative analysis method for the driving mechanism of lake water level evolution according to claim 6, characterized in that, The steps of identifying extreme hydrological events and calculating event intensity also include: For all grid cells clustered into the same extreme hydrological event, calculate the product of the water storage anomaly value and the physical area of ​​each grid cell; The event intensity of the extreme hydrological event is obtained by summing the products of all grid cells over the duration of the event.

8. The quantitative analysis method for the driving mechanism of lake water level evolution according to claim 1, characterized in that, The steps for determining the hysteresis response relationship between precipitation and water level include: The ratio of the covariance to the product of variance of the precipitation series and the water level series at different lag times is calculated to obtain the lag cross-correlation coefficient. The lag time corresponding to the lag cross-correlation coefficient with the largest absolute value is selected as the optimal lag time; Based on the optimal lag time, a linear regression model of precipitation increment events and water level rise events is established, and the least squares method is used to solve the regression coefficients to quantify the impact of precipitation increment on water level rise.

9. The quantitative analysis method for the driving mechanism of lake water level evolution according to claim 1, characterized in that, The steps involved in resolving the driver mechanism include: The standardized weights output by the partial least squares structural equation model are integrated with the event intensity of the extreme hydrological event; The contribution rate of each driving factor and extreme hydrological event to water level change was quantitatively decomposed using effect values; By comparing the contribution rates of each driving factor, the type of dominant driving factor can be identified.

10. The method for quantitative analysis of the driving mechanism of lake water level evolution according to claim 1, characterized in that, The step of acquiring and preprocessing multi-source heterogeneous remote sensing data further includes: Gravity satellite data was acquired, and anomaly signals of land water storage were extracted. The scale factor method was used to perform amplitude recovery correction on the anomaly signals of land water storage to eliminate signal leakage errors, and spatial resampling was performed. Meteorological reanalysis data is acquired, and raster data are aggregated into watershed-scale monthly series data using inverse distance weighted interpolation or kriging interpolation. This data, along with the land water storage anomaly signal, is then incorporated into the standardized multidimensional dataset.

Images