Winter rainfall phase state prediction method based on thickness of cold and warm layers

Abstract

The invention relates to the field of weather prediction, in particular to a winter rainfall phase state prediction method based on the thickness of a cold and warm layer, and the method comprises the steps: defining a melting layer and a freezing layer; after a first absolute threshold value and a first relative threshold value are determined, the absolute threshold value and the relative threshold value after dynamic optimization are obtained by taking a threshold value adaptive matching algorithm as a basic framework; and based on the absolute threshold and the relative threshold after dynamic optimization, a multi-rule collaborative discrimination system is constructed in combination with a cold and warm layer boundary and parameter extraction result, and a final rainfall phase prediction result and phase transformation probability distribution are obtained. According to the method, the temperature vertical structure of the ice particles and the freezing rain can be more accurately and quantitatively described through the newly prepared absolute threshold value and the threshold value relative index, the method has clear physical significance, the threshold value intervals are mutually independent and do not coincide, and a good judgment effect can be achieved on the ice particles and the freezing rain.

Description

Technical Field

[0001] This invention relates to the field of meteorological forecasting, specifically to a method for predicting the phase of winter precipitation based on the thickness of the cold and warm layers. Background Technology

[0002] Extreme cold waves not only bring meteorological disasters such as low temperatures and strong winds, but also make the types and transitions of precipitation phases more complex. Winter precipitation phases mainly include rain, sleet, snow, hail, graupel, and freezing rain, which can be categorized into solid (snow, hail, graupel), liquid (rain), and mixed (sleet, freezing rain) phases. Clearly, in winter, the impact of solid and mixed precipitation is often much more severe than that of liquid precipitation. Firstly, due to density differences, the magnitude of solid and mixed precipitation is greater than that of liquid precipitation for the same amount. For example, 5 millimeters is light rain for rain but heavy snow for snow, highlighting the difference in impact. Secondly, the hazardous effects of solid and mixed precipitation are more diverse and widespread. Weather events like blizzards and hail can lead to low visibility, snow accumulation on roads, and icing, affecting multiple sectors such as transportation, agriculture, power, aviation, and infrastructure construction.

[0003] Current winter precipitation phase forecasting techniques are relatively mature in predicting and distinguishing between rain and snow, but there is still a lack of extensive research on other precipitation phase types. On the one hand, other precipitation phase types are indeed relatively rare and rarely occur independently, resulting in fewer case studies. On the other hand, traditional technical indicators are fundamentally insufficient in their understanding of atmospheric temperature stratification. While using specific isobaric surface temperature or thickness thresholds can reflect the basic situation of atmospheric temperature stratification, there are problems of jumps and discontinuities, which are insufficient to show the precise characteristics of atmospheric temperature stratification.

[0004] Therefore, the traditional precipitation phase threshold uses a combination of temperature and thickness thresholds of a fixed isobaric surface as a criterion to analyze the precipitation phase in conjunction with the horizontal circulation pattern. However, it is limited by the fact that past vertical observations were not refined enough and could not be used to further distinguish the solid precipitation phase types. Summary of the Invention

[0005] This invention addresses the technical problems existing in the prior art by providing a method for predicting the phase state of winter precipitation based on the thickness of the warm and cold layers.

[0006] The technical solution of this invention to solve the above-mentioned technical problems is as follows: a method for predicting the phase state of winter precipitation based on the thickness of the cold and warm layers, the method comprising: S101. Perform collaborative inversion using the acquired multi-source data to complete parameter extraction and define the melting layer and the freezing layer; S102. Based on the definition results and the extraction results of the cold and warm layer boundaries and parameters, after determining the first absolute threshold and the first relative threshold, a layer thickness evolution mechanism is introduced to capture the dynamic change trend of the thickness of the frozen layer and the thawing layer, based on the threshold adaptive matching algorithm as the basic framework, and a corrected initial threshold interval is generated. At the same time, a phase adaptation coefficient is generated through the phase-layer feature coupling mapping strategy. The initial threshold interval and the phase adaptation coefficient are embedded into the threshold adaptive matching algorithm to dynamically optimize the absolute threshold and the relative threshold, and obtain the dynamically optimized absolute threshold and the relative threshold. S103. Based on the dynamically optimized absolute and relative thresholds, combined with the cold and warm layer boundaries and parameter extraction results, a multi-rule collaborative discrimination system is constructed. The transition probability is quantified based on the transition state in the initial classification of precipitation phases to obtain the final precipitation phase prediction results and phase transition probability distribution.

[0007] In a preferred embodiment, the multi-source data acquired in S101 specifically includes: high-resolution radiosonde data at the second level, ERA5 reanalysis data, microwave radiometer atmospheric temperature profile data, and ground-based radar reflectivity data. A temperature field dataset is constructed by fusing multi-source data. After eliminating the observation error of the temperature field dataset based on the Bayesian iterative fusion algorithm, a continuous temperature profile with vertical sampling interval is generated. The observation error refers to the instantaneous fluctuation during the ascent of the radiosonde, the spatial smoothing effect of the reanalysis data, etc. After obtaining the continuous temperature profile, for the temperature plateau section within the 0℃ transition zone in the continuous temperature profile, the abrupt change point where the first derivative of temperature with respect to height changes from positive to negative and / or from negative to positive is identified by calculating the first derivative of temperature with respect to height, and the height of the 0℃ layer is determined. The 0℃ transition zone refers to the region within ±2℃ of the 0℃ critical temperature, and the temperature plateau section refers to the temperature gradient being less than or equal to 0.1℃ / m. When multiple 0°C layers exist, the 0°C boundaries of each layer are separated and the primary and secondary layers are marked by polynomial fitting with adaptive window length, and then the boundaries and parameters of the cold and warm layers are extracted.

[0008] In a preferred embodiment, the definition of the melting layer and the freezing layer in S101 is specifically as follows: the melting layer is defined by the boundary of the 0°C layer, and the freezing layer is defined by the area above 0°C. S101 involves parameter extraction, specifically the following stratification parameter data: the vertical distance from the bottom of the thawed layer to the top of the frozen layer is used as the height of the frozen layer top; the vertical distance from the top of the frozen layer to the bottom of the frozen layer is used as the thickness of the frozen layer; the vertical distance from the top of the thawed layer to the bottom of the thawed layer is used as the thickness of the thawed layer; the presence indicator of the thawed layer; the average temperature of the thawed layer; the temperature difference between the frozen layer and the thawed layer; and the overall atmospheric temperature. The presence indicator of the thawed layer refers to the condition that if the thawed layer thickness is ≤50 meters, it is considered that there is no thawed layer.

[0009] In a preferred embodiment, S102 uses the height of the top of the frozen layer to characterize the position of the frozen layer in vertical space as a reference and determines it as a first absolute threshold, and uses the ratio of the thickness of the frozen layer to the thickness of the thawed layer to characterize the relative configuration relationship of the thicknesses of the cold and warm layers as a first relative threshold. The initial reference intervals of the first absolute threshold and the first relative threshold are determined statistically based on historical phase observation data. Based on the time series data of frozen layer thickness, thaw layer thickness, and frozen layer top height acquired within a periodic time, the time series data can fully reflect the dynamic evolution characteristics of the thickness and key height of the cold and warm layers over time. Input into the layer thickness evolution mechanism, the time coordinates corresponding to the parameter values ​​of frozen layer top height, frozen layer thickness, and thaw layer thickness are matched based on the timestamp information of the frozen layer thickness, thaw layer thickness, and frozen layer top height time series data. The time correlation between the layer parameter data and time nodes is established to form thickness-time data. The thickness-time data is then segmented into time series, dividing the interval range of adjacent time steps of the thickness-time data. The initial thickness value and the final thickness value in each interval are extracted, and the difference between the initial thickness value and the final thickness value is calculated to obtain the thickness change within adjacent time steps. The thickness change is normalized and then compared with the time step to eliminate the influence of different time step differences on the calculation results, thus obtaining the thickness gradient, a quantitative value that characterizes the degree of thickness change over time. The thickness gradient can also be understood as the degree of change of thickness at different times relative to the previous time.

[0010] In a preferred embodiment, after obtaining the thickness gradient, the thickness gradient time series within the periodic time is subjected to adjacent time difference calculation to obtain the instantaneous rate of change of the thickness gradient with time. The first derivative of the thickness gradient time series is subjected to adjacent time difference calculation to obtain the thickness change acceleration that represents the change of the first derivative with time. The result obtained by the adjacent time difference calculation of the thickness gradient is the first derivative of the thickness gradient, that is, the preliminary derivative of the thickness gradient. The adjacent time difference calculation of the first derivative of the thickness gradient time series is the further derivative of the first derivative of the thickness gradient. Finally, the dynamic trend of thickness evolution can be fully captured by the synergistic effect of the first and second derivatives. Based on historical phase state observation data, this study uses thickness gradient sample data corresponding to the thickening of the frozen layer and the shrinkage of the thawing layer. The samples cover different climatic backgrounds and different phase state transformation scenarios as the data foundation. Statistical analysis is performed on the extracted sample data to calculate the quantile interval of the gradient data. Combined with the actual state of phase state transformation corresponding to the sample data, the threshold range of special thickening of the frozen layer thickness gradient and special shrinkage of the thawing layer thickness gradient is determined. The corresponding actual state of phase state transformation refers to whether it triggers an actual phase state change. The special thickening and special shrinkage finally obtained in this step represent the two abrupt events of rapid thickening and sharp shrinkage of the frozen layer thickness gradient within a period of time, respectively. Based on the determined threshold ranges for special thickening and special shrinkage, all instantaneous rates of change and accelerations of thickness change in the thickness gradient time series are verified. Abnormal data in the frozen layer thickness time series, melt layer thickness time series, and frozen layer top height time series are removed to complete the initial cleaning of multi-source data. The multi-source data with abnormal data removed are then subjected to step S101 again to obtain a second absolute threshold and a second relative threshold. The threshold deviation is obtained by calculating the difference between the second absolute threshold and the second relative threshold and the first absolute threshold and the first relative threshold.

[0011] In a preferred embodiment, after obtaining the threshold deviation, the thickness gradient, instantaneous rate of change, and thickness change acceleration are fitted with the threshold deviation to generate a deviation mapping model between the thickness gradient and the threshold. The threshold ranges for the special thickening of the frozen layer thickness gradient and the special contraction of the thawing layer thickness gradient are introduced as constraints. The deviation mapping model outputs the corrected initial threshold interval, thus completing the introduction of the layer thickness evolution mechanism. Specifically, when a sudden change event in the frozen layer thickness is identified, the upward correction magnitude is adaptively determined based on the magnitude characteristics of the frozen layer thickness gradient, using the base interval of the first absolute threshold as a benchmark. When a sudden change event in the thawing layer thickness is identified, the downward correction magnitude is dynamically adjusted based on the intensity characteristics of the thawing layer thickness gradient, using the base interval of the relative threshold as a benchmark. After the correction is completed, the corrected initial threshold interval is output, which is the threshold of the second absolute threshold and the second relative threshold at this time.

[0012] In a preferred embodiment, step S102 generates phase adaptation coefficients through a phase-layer feature coupling mapping strategy, specifically as follows: Historical precipitation phase observation data and stratification parameter data are collected, and each historical precipitation phase observation data is matched with the stratification parameter data of the same period to form a pair of basic data samples; among them, the stratification parameter data are the core stratification parameters output by S101, which specifically include the height of the frozen layer top, the thickness of the frozen layer, the thickness of the thawed layer, the thickness ratio, the temperature difference between the frozen layer and the thawed layer, etc. The kernel partial least squares method is used to construct a parameter coupling matrix with two target phases, solid and liquid, as output variables and the height of the frozen layer top, the thickness ratio of the frozen layer thickness to the thawed layer thickness, and the temperature difference as input variables. This matrix is ​​used to quantify the intrinsic correlation between the layering parameters and the phases. In the parameter coupling matrix, the row vectors correspond to the height of the frozen layer top, the thickness ratio, and the temperature difference, and the column vectors correspond to the two target phases, solid and liquid. Based on the kernel function mapping of the kernel partial least squares method, the correlation strength between the layering parameters and the corresponding phases is obtained, and the elements in the coupling matrix represent the correlation strength between a layering parameter and a target phase. Using a single target phase as a unit, the correlation strength values ​​of the three types of layered parameters are summed, and then the proportion of the correlation strength of each layered parameter to the total is calculated to obtain the correlation weight of each layered parameter relative to the target phase. Based on the target phase state, the correlation weighted score of each single layer parameter is calculated, and then the weighted scores of all layer parameters are summed to obtain the initial comprehensive score. After mapping to a fixed interval, the phase adaptation coefficient is generated. The phase fit coefficients of all basic data samples are summed and averaged, and the interval system is preset. Specifically, basic data samples with phase fit coefficients exceeding the average value are marked as high-match intervals, and basic data samples with phase fit coefficients less than or equal to the average value are marked as low-match intervals. The layer parameter data in the basic data samples of low-match intervals are re-acquired to complete the mapping of phase-layer feature coupling.

[0013] In a preferred embodiment, in S102, after the modified initial threshold interval, phase adaptation coefficient, and correlation weight are input into the threshold adaptive matching algorithm, the current layer parameter data is adjusted using the modified initial threshold interval, including: The difference between the upper and lower limits of the initial threshold interval is calculated as half the width of the initial threshold interval. The absolute difference between the current layer parameter and the midpoint of the initial threshold interval is divided by the half width of the interval to obtain the proportion of the deviation between the current parameter and the midpoint of the interval. If the absolute difference between the half width of the initial threshold interval and the proportion of the deviation between the current parameter and the midpoint of the interval exceeds one-third, the adjustment is made by one-tenth of the half width of the initial threshold interval. This means that the current parameter is close to the boundary of the threshold interval and deviates significantly from the center. If the absolute difference between the half width of the initial threshold interval and the proportion of the deviation between the current parameter and the midpoint of the interval is less than or equal to one-third, the adjustment is made by one-twentieth of the half width of the initial threshold interval. This indicates that the current parameter is in the middle range of the threshold interval and deviates less. The interval adjustment results are verified as follows: The dynamic change trends of the frozen and thawing layers, captured by the layer thickness evolution mechanism, are examined. If, for example, the frozen layer is thickening and / or the thawing layer is shrinking, the interval adjustment direction aligns with these trends. For instance, if a rapid thickening trend is detected in the frozen layer, and the optimized absolute threshold interval is finely adjusted upwards to match this trend, then the condition is satisfied. If the threshold adjustment direction is opposite to the layer evolution trend, then the condition is not satisfied, and the interval adjustment direction needs to be readjusted and changed. The parameter coupling matrix is ​​called, and the parameter ranges corresponding to the dynamically optimized absolute and relative thresholds are matched with the correlation strength in the parameter coupling matrix. Based on the target phase, the correlation weighted score of each single layer parameter is calculated. Then, the weighted scores of all layer parameters are summed. If the comprehensive score exceeds the initial comprehensive score, the interval adjustment is determined to be correct; otherwise, it is incorrect, and the adjustment is readjusted. Finally, the dynamically optimized absolute and relative thresholds are obtained.

[0014] In a preferred embodiment, the multi-rule collaborative discrimination system constructed by S103, combining the cold and warm layer boundaries with the parameter extraction results, specifically includes: If the presence of a melt layer is marked as absent and the temperature of the entire layer of air is at or below 0°C, it is directly determined to be pure snow. If the temperature of the entire layer of air is at or below 0°C, but the presence of a melt layer is marked as present and the current thickness of the melt layer does not exceed the dynamically optimized relative threshold range of the pure snow phase, it is determined to be a transitional state from pure snow to sleet. If the existence of the melt layer is marked as present, and the current stratification parameters simultaneously satisfy the dynamically optimized absolute threshold range and the dynamically optimized relative threshold range corresponding to the ice particle phase, it is determined to be ice particles; if the relative threshold of the current stratification parameters exceeds the upper limit of the dynamically optimized relative threshold range corresponding to the ice particle phase, and the parameter corresponding to the melt layer thickness is at the end of the range of the dynamically optimized relative threshold range associated with the ice particle phase, it is determined to be a transition state from ice particles to pure snow, that is, sleet; When the existence of the melting layer is marked as present, and the current stratification parameters simultaneously satisfy the dynamically optimized absolute threshold range and the dynamically optimized relative threshold range corresponding to the freezing rain phase, it is determined to be freezing rain; if the relative threshold of the current stratification parameters is lower than the lower limit of the dynamically optimized relative threshold range corresponding to the freezing rain phase, and the parameter corresponding to the thickness of the frozen layer is at the edge of the adaptation range of the dynamically optimized absolute threshold range associated with the freezing rain phase, it is determined to be a transition state from freezing rain to rain. When the existence of the melting layer is marked as present, and the parameter corresponding to the thickness of the melting layer exceeds the upper limit of the dynamically optimized relative threshold range adaptation range associated with freezing rain and ice particle phases, and the parameter corresponding to the thickness of the frozen layer is lower than the lower limit of the dynamically optimized absolute threshold range adaptation range associated with freezing rain and ice particle phases, and the relative threshold of the current stratification parameter is lower than the lower limit of the dynamically optimized relative threshold range adaptation range for the rain phase, and the near-surface air temperature is higher than the critical temperature of 0℃, it is determined to be rain. When the near-surface air temperature of the current stratification parameters is at the critical temperature of 0℃, and the thickness ratio of the frozen layer to the thawed layer is within the transitional overlap range of the dynamically optimized relative threshold intervals corresponding to the freezing rain phase and the ice particle phase, it is determined to be sleet.

[0015] In a preferred embodiment, after the initial classification of precipitation phases by the constructed multi-rule collaborative discrimination system, the instantaneous rate of change and acceleration of thickness change obtained based on the stratification thickness evolution mechanism, as well as the near-surface temperature change trend and water vapor transport intensity obtained based on third-party software, are used as input variables. Through an integrated model combining logistic regression and random forest, and using historical phase transformation cases as training samples, the probability of the initial classification result transforming into other phases is output.

[0016] The beneficial effects of this invention are: the newly proposed absolute threshold and threshold relative index can more accurately and quantitatively describe the temperature vertical structure of ice particles and freezing rain, which has clear physical meaning. Moreover, the threshold intervals are independent of each other and do not overlap, which can play a better role in distinguishing between the two. This solves the problem that traditional fixed isobaric surface thresholds are difficult to distinguish between similar phases such as ice particles and freezing rain, and makes phase discrimination more consistent with the real characteristics of atmospheric temperature stratification. Attached Figure Description

[0017] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0020] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0021] As attached Figure 1 As shown in the figure, this embodiment provides: a method for predicting the phase state of winter precipitation based on the thickness of the cold and warm layers, the method including: S101. Perform collaborative inversion using the acquired multi-source data to complete parameter extraction and define the melting layer and the freezing layer; The multi-source data acquired by S101 specifically includes: high-resolution radiosonde data at the second level, ERA5 reanalysis data, microwave radiometer atmospheric temperature profile data, and ground-based radar reflectivity data. A temperature field dataset is constructed by fusing multi-source data. After eliminating the observation error of the temperature field dataset based on the Bayesian iterative fusion algorithm, a continuous temperature profile with vertical sampling interval is generated. The observation error refers to the instantaneous fluctuation during the ascent of the radiosonde, the spatial smoothing effect of the reanalysis data, etc. After obtaining the continuous temperature profile, for the temperature plateau section within the 0℃ transition zone in the continuous temperature profile, the abrupt change point where the first derivative of temperature with respect to height changes from positive to negative and / or from negative to positive is identified by calculating the first derivative of temperature with respect to height, and the height of the 0℃ layer is determined. The 0℃ transition zone refers to the region within ±2℃ of the 0℃ critical temperature, and the temperature plateau section refers to the temperature gradient being less than or equal to 0.1℃ / m. When multiple 0°C layers exist, the 0°C boundaries of each layer are separated and the primary and secondary layers are marked by polynomial fitting with adaptive window length, and then the boundaries and parameters of the cold and warm layers are extracted.

[0022] S101 defines the melting layer and the freezing layer as follows: the boundary of the 0°C layer is used as the dividing line, the melting layer is defined by the area above 0°C, and the freezing layer is defined by the area below 0°C. S101 extracts the following stratification parameters: the vertical distance from the bottom of the thawed layer to the top of the frozen layer is used as the height of the frozen layer top; the vertical distance from the top of the frozen layer to the bottom of the frozen layer is used as the thickness of the frozen layer; the vertical distance from the top of the thawed layer to the bottom of the thawed layer is used as the thickness of the thawed layer; the presence indicator of the thawed layer; the average temperature of the thawed layer; the temperature difference between the frozen layer and the thawed layer; and the overall atmospheric temperature. The presence indicator of the thawed layer refers to the condition that if the thawed layer thickness is ≤50 meters, it is considered that there is no thawed layer.

[0023] S102. Based on the definition results and the extraction results of the cold and warm layer boundaries and parameters, after determining the first absolute threshold and the first relative threshold, a layer thickness evolution mechanism is introduced to capture the dynamic change trend of the thickness of the frozen layer and the thawing layer, based on the threshold adaptive matching algorithm as the basic framework, and a corrected initial threshold interval is generated. At the same time, a phase adaptation coefficient is generated through the phase-layer feature coupling mapping strategy. The initial threshold interval and the phase adaptation coefficient are embedded into the threshold adaptive matching algorithm to dynamically optimize the absolute threshold and the relative threshold, and obtain the dynamically optimized absolute threshold and the relative threshold. S102 uses the height of the top of the frozen layer to characterize the position of the frozen layer in vertical space as the first absolute threshold, and uses the ratio of the thickness of the frozen layer to the thickness of the thawed layer to characterize the relative configuration relationship of the thicknesses of the cold and warm layers as the first relative threshold. The initial reference intervals of the first absolute threshold and the first relative threshold are determined statistically based on historical phase observation data. Based on the time series data of frozen layer thickness, thaw layer thickness, and frozen layer top height acquired within a periodic time, the time series data can fully reflect the dynamic evolution characteristics of the thickness and key height of the cold and warm layers over time. Input into the layer thickness evolution mechanism, the time coordinates corresponding to the parameter values ​​of frozen layer top height, frozen layer thickness, and thaw layer thickness are matched based on the timestamp information of the frozen layer thickness, thaw layer thickness, and frozen layer top height time series data. The time correlation between the layer parameter data and time nodes is established to form thickness-time data. The thickness-time data is then segmented into time series, dividing the interval range of adjacent time steps of the thickness-time data. The initial thickness value and the final thickness value in each interval are extracted, and the difference between the initial thickness value and the final thickness value is calculated to obtain the thickness change within adjacent time steps. The thickness change is normalized and then compared with the time step to eliminate the influence of different time step differences on the calculation results, thus obtaining the thickness gradient, a quantitative value that characterizes the degree of thickness change over time. The thickness gradient can also be understood as the degree of change of thickness at different times relative to the previous time.

[0024] After obtaining the thickness gradient, the thickness gradient time series within the periodic time is subjected to adjacent time difference calculation to obtain the instantaneous rate of change of the thickness gradient with time. The first derivative of the thickness gradient time series is subjected to adjacent time difference calculation to obtain the thickness change acceleration, which represents the change of the first derivative with time. The result of the thickness gradient adjacent time difference calculation is the first derivative of the thickness gradient, that is, the preliminary derivative of the thickness gradient. The adjacent time difference calculation of the first derivative of the thickness gradient time series is the further derivative of the first derivative of the thickness gradient. Finally, the synergistic effect of the first and second derivatives can completely capture the dynamic trend of thickness evolution. Based on historical phase state observation data, this study uses thickness gradient sample data corresponding to the thickening of the frozen layer and the shrinkage of the thawing layer. The samples cover different climatic backgrounds and different phase state transformation scenarios as the data foundation. Statistical analysis is performed on the extracted sample data to calculate the quantile interval of the gradient data. Combined with the actual state of phase state transformation corresponding to the sample data, the threshold range of special thickening of the frozen layer thickness gradient and special shrinkage of the thawing layer thickness gradient is determined. The corresponding actual state of phase state transformation refers to whether it triggers an actual phase state change. The special thickening and special shrinkage finally obtained in this step represent the two abrupt events of rapid thickening and sharp shrinkage of the frozen layer thickness gradient within a period of time, respectively. Based on the determined threshold ranges for special thickening and special shrinkage, all instantaneous rates of change and accelerations of thickness change in the thickness gradient time series are verified. Abnormal data in the frozen layer thickness time series, melt layer thickness time series, and frozen layer top height time series are removed to complete the initial cleaning of multi-source data. The multi-source data with abnormal data removed is then subjected to step S101 again to obtain the second absolute threshold and the second relative threshold. The threshold deviation is obtained by calculating the difference between the second absolute threshold and the second relative threshold and the first absolute threshold and the first relative threshold.

[0025] After obtaining the threshold deviation, the thickness gradient, instantaneous rate of change, and acceleration of thickness change are fitted with the threshold deviation to generate a deviation mapping model between the thickness gradient and the threshold. The threshold ranges of special thickening of the frozen layer thickness gradient and special contraction of the thawed layer thickness gradient are introduced as conditional constraints. The deviation mapping model outputs the corrected initial threshold range, thus completing the introduction of the layer thickness evolution mechanism.

[0026] Specifically, when a sudden change in the thickness of the frozen layer is detected, the upward correction magnitude is adaptively determined based on the magnitude characteristics of the frozen layer thickness gradient, using the base interval of the first absolute threshold as a benchmark; when a sudden change in the thickness of the melted layer is detected, the downward correction magnitude is dynamically adjusted based on the intensity characteristics of the melted layer thickness gradient, using the base interval of the relative threshold as a benchmark. After the correction is completed, the initial threshold interval after correction is output, which is the threshold of the second absolute threshold and the second relative threshold at this time.

[0027] S102 generates phase adaptation coefficients through a phase-layer feature coupling mapping strategy, specifically: Historical precipitation phase observation data and stratification parameter data are collected, and each historical precipitation phase observation data is matched with the stratification parameter data of the same period to form a pair of basic data samples; among them, the stratification parameter data are the core stratification parameters output by S101, which specifically include the height of the frozen layer top, the thickness of the frozen layer, the thickness of the thawed layer, the thickness ratio, the temperature difference between the frozen layer and the thawed layer, etc. The kernel partial least squares method is used to construct a parameter coupling matrix with two target phases, solid and liquid, as output variables and the height of the frozen layer top, the thickness ratio of the frozen layer thickness to the thawed layer thickness, and the temperature difference as input variables. This matrix is ​​used to quantify the intrinsic correlation between the layering parameters and the phases. In the parameter coupling matrix, the row vectors correspond to the height of the frozen layer top, the thickness ratio, and the temperature difference, and the column vectors correspond to the two target phases, solid and liquid. Based on the kernel function mapping of the kernel partial least squares method, the correlation strength between the layering parameters and the corresponding phases is obtained, and the elements in the coupling matrix represent the correlation strength between a layering parameter and a target phase. Using a single target phase as a unit, the correlation strength values ​​of the three types of layered parameters are summed, and then the proportion of the correlation strength of each layered parameter to the total is calculated to obtain the correlation weight of each layered parameter relative to the target phase. Based on the target phase state, the correlation weighted score of each single layer parameter is calculated, and then the weighted scores of all layer parameters are summed to obtain the initial comprehensive score. After mapping to a fixed interval, the phase adaptation coefficient is generated. The phase fit coefficients of all basic data samples are summed and averaged, and the interval system is preset. Specifically, basic data samples with phase fit coefficients exceeding the average value are marked as high-match intervals, and basic data samples with phase fit coefficients less than or equal to the average value are marked as low-match intervals. The layer parameter data in the basic data samples of low-match intervals are re-acquired to complete the mapping of phase-layer feature coupling.

[0028] In S102, after the modified initial threshold interval, phase adaptation coefficient, and associated weights are input into the threshold adaptive matching algorithm, the modified initial threshold interval is used to adjust the current layer parameter data, including: The difference between the upper and lower limits of the initial threshold interval is calculated as half the width of the initial threshold interval. The absolute difference between the current layer parameter and the midpoint of the initial threshold interval is divided by the half width of the interval to obtain the proportion of the deviation between the current parameter and the midpoint of the interval. If the absolute difference between the half width of the initial threshold interval and the proportion of the deviation between the current parameter and the midpoint of the interval exceeds one-third, the adjustment is made by one-tenth of the half width of the initial threshold interval. This means that the current parameter is close to the boundary of the threshold interval and deviates significantly from the center. If the absolute difference between the half width of the initial threshold interval and the proportion of the deviation between the current parameter and the midpoint of the interval is less than or equal to one-third, the adjustment is made by one-twentieth of the half width of the initial threshold interval. This indicates that the current parameter is in the middle range of the threshold interval and deviates less. The interval adjustment results are verified as follows: The dynamic change trends of the frozen and thawing layers, captured by the layer thickness evolution mechanism, are examined. If, for example, the frozen layer is thickening and / or the thawing layer is shrinking, the interval adjustment direction aligns with these trends. For instance, if a rapid thickening trend is detected in the frozen layer, and the optimized absolute threshold interval is finely adjusted upwards to match this trend, then the condition is satisfied. If the threshold adjustment direction is opposite to the layer evolution trend, then the condition is not satisfied, and the interval adjustment direction needs to be readjusted and changed. The parameter coupling matrix is ​​called, and the parameter ranges corresponding to the dynamically optimized absolute and relative thresholds are matched with the correlation strength in the parameter coupling matrix. Based on the target phase, the correlation weighted score of each single layer parameter is calculated. Then, the weighted scores of all layer parameters are summed. If the comprehensive score exceeds the initial comprehensive score, the interval adjustment is determined to be correct; otherwise, it is incorrect, and the adjustment is readjusted. Finally, the dynamically optimized absolute and relative thresholds are obtained.

[0029] S103. Based on the dynamically optimized absolute and relative thresholds, combined with the cold and warm layer boundaries and parameter extraction results, a multi-rule collaborative discrimination system is constructed. The transition probability is quantified based on the transition state in the initial classification of precipitation phases to obtain the final precipitation phase prediction results and phase transition probability distribution.

[0030] The multi-rule collaborative discrimination system constructed by S103, combining the cold and warm layer boundaries and parameter extraction results, is as follows: If the presence of a melt layer is marked as absent and the temperature of the entire layer of air is at or below 0°C, it is directly determined to be pure snow. If the temperature of the entire layer of air is at or below 0°C, but the presence of a melt layer is marked as present and the current thickness of the melt layer does not exceed the dynamically optimized relative threshold range of the pure snow phase, it is determined to be a transitional state from pure snow to sleet. If the existence of the melt layer is marked as present, and the current stratification parameters simultaneously satisfy the dynamically optimized absolute threshold range and the dynamically optimized relative threshold range corresponding to the ice particle phase, it is determined to be ice particles; if the relative threshold of the current stratification parameters exceeds the upper limit of the dynamically optimized relative threshold range corresponding to the ice particle phase, and the parameter corresponding to the melt layer thickness is at the end of the range of the dynamically optimized relative threshold range associated with the ice particle phase, it is determined to be a transition state from ice particles to pure snow, that is, sleet; When the existence of the melting layer is marked as present, and the current stratification parameters simultaneously satisfy the dynamically optimized absolute threshold range and the dynamically optimized relative threshold range corresponding to the freezing rain phase, it is determined to be freezing rain; if the relative threshold of the current stratification parameters is lower than the lower limit of the dynamically optimized relative threshold range corresponding to the freezing rain phase, and the parameter corresponding to the thickness of the frozen layer is at the edge of the adaptation range of the dynamically optimized absolute threshold range associated with the freezing rain phase, it is determined to be a transition state from freezing rain to rain. When the existence of the melting layer is marked as present, and the parameter corresponding to the thickness of the melting layer exceeds the upper limit of the dynamically optimized relative threshold range adaptation range associated with freezing rain and ice particle phases, and the parameter corresponding to the thickness of the frozen layer is lower than the lower limit of the dynamically optimized absolute threshold range adaptation range associated with freezing rain and ice particle phases, and the relative threshold of the current stratification parameter is lower than the lower limit of the dynamically optimized relative threshold range adaptation range for the rain phase, and the near-surface air temperature is higher than the critical temperature of 0℃, it is determined to be rain. When the near-surface air temperature of the current stratification parameters is at the critical temperature of 0℃, and the thickness ratio of the frozen layer to the thawed layer is within the transitional overlap range of the dynamically optimized relative threshold intervals corresponding to the freezing rain phase and the ice particle phase, it is determined to be sleet.

[0031] After the initial classification of precipitation phases using the constructed multi-rule collaborative discrimination system, the instantaneous rate of change and acceleration of thickness change obtained from the stratification thickness evolution mechanism, as well as the near-surface temperature change trend and water vapor transport intensity obtained from third-party software, are used as input variables. Through an integrated model combining logistic regression and random forest, and using historical phase transformation cases as training samples, the probability of the initial classification result transforming into other phases is output.

[0032] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0033] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0034] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0035] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0036] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0037] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0038] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A winter precipitation phase prediction method based on the thickness of cold and warm layers, characterized by, The method comprises: S101, collaborative inversion is performed on the obtained multi-source data, parameter extraction is completed, and the thawing layer and the frozen layer are defined; S102, after the first absolute threshold and the first relative threshold are determined according to the definition result and the extraction result of the cold and warm layer boundary and the parameter, a threshold self-adaptive matching algorithm is taken as a basic framework, a layer interface thickness evolution mechanism is introduced to capture the dynamic change trend of the thickness of the frozen layer and the thawing layer, a corrected threshold initial interval is generated, and a phase state adaptation coefficient is generated through a phase state-layer interface feature coupling mapping strategy; the threshold initial interval and the phase state adaptation coefficient are embedded into the threshold self-adaptive matching algorithm, the absolute threshold and the relative threshold are dynamically optimized, and the dynamically optimized absolute threshold and the relative threshold are obtained; S103, based on the dynamically optimized absolute threshold and the relative threshold, a multi-rule collaborative discrimination system is constructed in combination with the cold and warm layer boundary and the parameter extraction result, and a transition state in the initial classification of the precipitation phase state is quantified to obtain a final precipitation phase state prediction result and a phase state conversion probability distribution.

2. The winter precipitation phase prediction method based on the thickness of cold and warm layers according to claim 1, characterized in that, The multi-source data obtained in S101 is specifically: high-resolution sounding second-level data, ERA5 reanalysis data, microwave radiometer atmospheric temperature profile data, and ground-based radar reflectivity data; A temperature field data set is constructed through multi-source data fusion, and after the observation error of the temperature field data set is eliminated based on a Bayesian iterative fusion algorithm, a continuous temperature profile with a vertical sampling interval is generated; After the continuous temperature profile is obtained, the temperature flat section in the 0℃ transition zone in the continuous temperature profile is calculated by calculating the first derivative of the temperature with respect to the height, the mutation point of the first derivative from positive to negative and / or from negative to positive is identified, and the height of the 0℃ layer is determined; When there are multiple 0℃ layers, after the 0℃ boundaries of each layer are separated and the primary and secondary layers are marked through polynomial fitting with adaptive window length adjustment, the cold and warm layer boundaries and parameters are extracted.

3. The winter precipitation phase prediction method based on the thickness of cold and warm layers of claim 2, wherein, In S101, the thawing layer and the frozen layer are defined, specifically: taking the 0℃ layer boundary as the division basis, taking the region above 0℃ as the thawing layer, and taking the region below 0℃ as the frozen layer; In S101, the parameters are extracted, specifically: the vertical distance from the bottom of the thawing layer to the top of the frozen layer is taken as the height of the top of the frozen layer; the vertical distance from the top of the frozen layer to the bottom of the frozen layer is taken as the thickness of the frozen layer; the vertical distance from the top of the thawing layer to the bottom of the thawing layer is taken as the thickness of the thawing layer; the existence of the thawing layer is marked; the average temperature of the thawing layer; the temperature difference between the frozen layer and the thawing layer; and the temperature of the whole layer of atmosphere.

4. The winter precipitation phase prediction method based on the thickness of cold and warm layers of claim 1, wherein, In S102, the height of the top of the frozen layer is taken as the position reference of the frozen layer in the vertical space to determine the first absolute threshold, and the ratio of the thickness of the frozen layer to the thickness of the thawing layer is taken as the relative configuration relationship of the cold and warm layer thickness to determine the first relative threshold; the initial reference interval of the first absolute threshold and the first relative threshold is determined based on historical phase state observation data statistics. Based on the frozen layer thickness time series data, the thawing layer thickness time series data and the frozen layer top height time series data obtained within the period of time, input into the layer thickness evolution mechanism, match the time coordinates of the parameter values of the frozen layer top height, the frozen layer thickness and the thawing layer thickness based on the time stamp information of the frozen layer thickness time series data, the thawing layer thickness time series data and the frozen layer top height time series data, establish the time association of the layer parameter data and the time node, form the thickness-time data, and perform time sequence segmentation on the thickness-time data, divide the interval range of the adjacent time steps of the thickness-time data, and extract the initial thickness value and the terminal thickness value in each interval, calculate the difference between the initial thickness value and the terminal thickness value to obtain the thickness change within the adjacent time steps; After the thickness change is normalized, the thickness gradient representing the degree of thickness change with time is obtained by ratio calculation with the time step.

5. The winter precipitation phase prediction method based on the thickness of cold and warm layers according to claim 4, characterized in that, After obtaining the thickness gradient, the thickness gradient time series within the period of time is subjected to adjacent time difference operation to obtain the instantaneous change rate representing the change of the thickness gradient with time, and the first derivative of the thickness gradient time series is subjected to adjacent time difference operation to obtain the thickness change acceleration representing the change of the first derivative with time; Based on the historical phase state observation data, the thickness gradient sample data corresponding to the frozen layer thickening and the thawing layer shrinkage, and the sample data covering different climate backgrounds and different phase state conversion scenarios are taken as the data basis, the extracted sample data is subjected to statistical analysis, the quantile interval of the gradient data is calculated, and the threshold range of the special thickening of the frozen layer thickness gradient and the special shrinkage of the thawing layer thickness gradient is determined in combination with the real state of the phase state conversion corresponding to the sample data; Based on the determined threshold range of the special thickening and the special shrinkage, verify all the instantaneous change rates and the thickness change accelerations of the thickness gradient time series, eliminate the abnormal data of the frozen layer thickness time series data, the thawing layer thickness time series data and the frozen layer top height time series data, and re-perform the S101 step on the multi-source data after eliminating the abnormal data to obtain the second absolute threshold and the second relative threshold. The threshold deviation is obtained by calculating the difference between the second absolute threshold and the second relative threshold and the first absolute threshold and the first relative threshold.

6. The winter precipitation phase prediction method based on the thickness of cold and warm layers of claim 5, wherein, After obtaining the threshold deviation, the thickness gradient, the instantaneous change rate and the thickness change acceleration are fitted with the threshold deviation to generate a deviation mapping model of the thickness gradient and the threshold, and the threshold range of the special thickening of the frozen layer thickness gradient and the special shrinkage of the thawing layer thickness gradient is introduced as a conditional constraint. The deviation mapping model outputs the corrected threshold initial interval, and the introduction of the layer thickness evolution mechanism is completed.

7. The winter precipitation phase prediction method based on the thickness of cold and warm layers of claim 1, wherein, The S102 generates a phase state adaptation coefficient through a phase state-layer junction feature coupling mapping strategy, specifically: Collect historical precipitation phase state observation data and layer parameter data, and each piece of historical precipitation phase state observation data matches the layer parameter data of the same period to form paired basic data samples; The kernel partial least squares method is used to take the solid phase and the liquid phase as output variables, to take the height of the frozen layer, the thickness ratio of the frozen layer and the thawing layer, and the temperature difference as input variables, to construct a parameter coupling matrix, and to take the row vector of the matrix in the parameter coupling matrix to correspond to the height of the frozen layer, the thickness ratio, and the temperature difference, to take the column vector to correspond to the solid phase and the liquid phase, and to obtain the correlation strength of the layer junction parameters and the corresponding phase state based on the kernel function mapping of the kernel partial least squares method. And taking a single target phase state as a unit, the corresponding layer junction parameter correlation strength value is summed, and the proportion of each layer junction parameter correlation strength in the total is calculated, to obtain the correlation weight of each layer junction parameter relative to the target phase state. Based on the target phase state as a unit, the correlation weight weighted score of a single layer junction parameter is calculated, and the weighted scores of all layer junction parameters are summed to obtain an initial comprehensive score, which is mapped to a fixed interval to generate a phase state adaptation coefficient. The phase state adaptation coefficients of all basic data samples are summed and averaged, and an interval system is preset, specifically: the basic data samples with a phase state adaptation coefficient greater than the average value are marked as a high matching interval, and the basic data samples with a phase state adaptation coefficient less than or equal to the average value are marked as a low matching interval, and the layer junction parameter data in the low matching interval basic data samples is reacquired to complete the mapping of the phase state-layer junction feature coupling.

8. The winter precipitation phase prediction method based on the thickness of cold and warm layers of claim 7, wherein, In S102, after the corrected threshold initial interval, the phase state adaptation coefficient, and the correlation weight are input into the threshold self-adaptive matching algorithm, the corrected threshold initial interval is used to adjust the interval of the current layer junction parameter data, including: The half of the difference between the upper limit and the lower limit of the threshold initial interval is taken as the threshold initial interval half-width, the absolute difference between the current layer junction parameter and the midpoint of the threshold initial interval is divided by the interval half-width to obtain the proportion of the deviation of the current parameter from the interval midpoint, and the absolute difference between the threshold initial interval half-width and the deviation proportion of the current parameter from the interval midpoint is more than one-third, which is adjusted by one-tenth of the threshold initial interval half-width, and the absolute difference between the threshold initial interval half-width and the deviation proportion of the current parameter from the interval midpoint is less than or equal to one-third, which is adjusted by one-twentieth of the threshold initial interval half-width; The interval adjustment result is verified, specifically: through the dynamic change trend of the thickness of the frozen layer and the thawing layer captured by the layer junction thickness evolution mechanism, when the frozen layer is thickening and / or the thawing layer is shrinking, the interval adjustment direction is consistent with the trend of the frozen layer thickening and / or the thawing layer shrinking; The parameter coupling matrix is called, the parameter range corresponding to the dynamically optimized absolute threshold and relative threshold is matched with the correlation strength in the parameter coupling matrix, and based on the target phase state as a unit, the correlation weight weighted score of a single layer junction parameter is calculated, and the weighted scores of all layer junction parameters are summed to obtain a comprehensive score greater than the initial comprehensive score, which is determined as correct interval adjustment, and the dynamically optimized absolute threshold and relative threshold are finally obtained.

9. The winter precipitation phase prediction method based on the thickness of cold and warm layers of claim 1, wherein, The S103 combines the cold and warm layer boundary and the parameter extraction result, and constructs a multi-rule collaborative discrimination system, specifically: When the melting layer existence identifier is non-existent, and the whole layer atmospheric temperature is in the interval of 0℃ and below, it is directly determined as pure snow; when the whole layer atmospheric temperature is in the interval of 0℃ and below, but the melting layer existence identifier is existent, and the current melting layer thickness does not exceed the dynamically optimized relative threshold interval adaptive range related to the pure snow phase state, it is determined as a pure snow to sleet transitional state; When the melting layer existence identifier is existent, and the current layer parameter simultaneously satisfies the dynamically optimized absolute threshold interval corresponding to the ice grain phase state and the dynamically optimized relative threshold interval corresponding to the ice grain phase state, it is determined as ice grain; if the relative threshold of the current layer parameter exceeds the upper limit of the dynamically optimized relative threshold interval corresponding to the ice grain phase state, and the parameter corresponding to the melting layer thickness is at the end of the dynamically optimized relative threshold interval adaptive range related to the ice grain phase state, it is determined as an ice grain to pure snow transitional state; When the melting layer existence identifier is existent, and the current layer parameter simultaneously satisfies the dynamically optimized absolute threshold interval corresponding to the frozen rain phase state and the dynamically optimized relative threshold interval corresponding to the frozen rain phase state, it is determined as frozen rain; if the relative threshold of the current layer parameter is lower than the lower limit of the dynamically optimized relative threshold interval corresponding to the frozen rain phase state, and the parameter corresponding to the freezing layer thickness is at the edge of the dynamically optimized absolute threshold interval adaptive range related to the frozen rain phase state, it is determined as a frozen rain to rain transitional state; When the melting layer existence identifier is existent, and the parameter corresponding to the melting layer thickness exceeds the upper limit of the dynamically optimized relative threshold interval adaptive range related to the frozen rain and ice grain phase states, the parameter corresponding to the freezing layer thickness is lower than the lower limit of the dynamically optimized absolute threshold interval adaptive range related to the frozen rain and ice grain phase states, the relative threshold of the current layer parameter is lower than the lower limit of the dynamically optimized relative threshold interval adaptive range related to the rain phase state, and the near-surface air temperature is higher than the 0℃ critical temperature, it is determined as rain; When the near-surface air temperature of the current layer parameter is at the 0℃ critical temperature, and the thickness ratio of the freezing layer and the melting layer is in the transitional overlapping range of the dynamically optimized relative threshold intervals corresponding to the frozen rain and ice grain phase states, it is determined as sleet.

10. The winter precipitation phase prediction method based on the thickness of cold and warm layers of claim 9, wherein, After the initial classification of the precipitation phase state in the constructed multi-rule collaborative discrimination system, the instantaneous change rate and thickness change acceleration obtained based on the layer thickness evolution mechanism, and the near-surface air temperature change trend and water vapor transport intensity obtained based on the third-party software are used as input variables, an integrated model combining logistic regression and random forest is used, historical phase state transformation cases are used as training samples, and the probability of the initial classification result to other phase state transformations is output.

Images