A debris flow early warning method based on characteristics of short-time heavy rainfall

Abstract

The application provides a debris flow early warning method based on short-time heavy rainfall characteristics, and relates to the technical field of geological disaster monitoring and early warning, which comprises the following steps: obtaining real-time minute-level rainfall data; based on the real-time minute-level rainfall data, using a sliding time window algorithm, calculating short-duration cumulative rainfall; comparing each short-duration cumulative rainfall with the quantile regression threshold corresponding to each fixed time window set in advance to obtain initial early warning level data corresponding to each fixed time window; and determining the final debris flow early warning level data of the target area by using the highest early warning level output principle to determine the initial early warning level data corresponding to each fixed time window. The application solves the problem that the existing debris flow early warning model generalizes the rainfall process, is difficult to capture short-time heavy rainfall pulses, and leads to poor timeliness and low prediction accuracy in early warning in small catchments with rapid confluence response.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning technology, and in particular to a debris flow early warning method based on the characteristics of short-term heavy rainfall. Background Technology

[0002] Debris flows are a highly destructive geological hazard in mountainous areas, and accurate early warning is crucial for protecting people's lives and property. Current research and practical applications show that debris flow rainfall prediction models based on statistical empirical methods are widely used. These models typically determine the critical rainfall threshold for triggering debris flows by establishing statistical relationships between historical debris flow events and rainfall parameters. The rainfall parameters used in these models mainly include average rainfall intensity, antecedent rainfall, effective rainfall intensity, and rainfall duration.

[0003] However, existing prediction models have significant limitations. While the aforementioned conventional rainfall indicators can reflect the overall rainfall moisture conditions before debris flows to some extent, they often generalize and homogenize the entire rainfall process. This approach makes it difficult to fully reveal the profound impact of the temporal structure evolution within rainfall events on the debris flow initiation mechanism. In particular, in actual monitoring, there is often a lack of high temporal resolution rainfall data at the minute level, which makes it easy for early warning systems to overlook key triggering information such as the timing of short-term rainfall peaks, the concentration of rainfall in a very short period of time, the actual duration of heavy rainfall, and the drastic changes in rainfall intensity.

[0004] This technical limitation is particularly prominent in debris flow prediction in small watersheds in mountainous areas. Small watersheds typically have short confluence times and extremely rapid hydrological responses, meaning that debris flow initiation is often more controlled by intense rainfall pulses within sub-hourly scales (minutes to an hour), rather than solely by the cumulative total rainfall or the flattened average rainfall intensity of the entire macro-level rainfall event. In summary, existing traditional threshold models based on event rainfall, average rainfall intensity, or long-duration preceding rainfall are inadequate for accurately depicting the rapid changes within the rainfall process. They tend to ignore the influence of short-term rainfall patterns on debris flow threshold models and suffer from severe long-duration dilution effects, thus significantly limiting the model's prediction accuracy, short-term warning timeliness, and practical applicability. Summary of the Invention

[0005] To address the aforementioned shortcomings in existing technologies, this invention provides a debris flow early warning method based on short-duration heavy rainfall characteristics. This method solves the problems of existing debris flow early warning models generalizing the rainfall process and making it difficult to capture short-duration heavy rainfall pulses, resulting in poor early warning timeliness and low prediction accuracy in small watersheds with fast confluence responses.

[0006] To achieve the aforementioned objectives, the present invention employs the following technical solution: a debris flow early warning method based on short-duration heavy rainfall characteristics, comprising: S1: Acquire real-time minute-level rainfall data collected by rainfall monitoring equipment in the target area; S2: Based on real-time minute-level rainfall data, a sliding time window algorithm with a preset update step size is used to calculate the short-duration cumulative rainfall within multiple preset fixed time windows; S3: Compare the cumulative rainfall for each short duration with the quantile regression threshold corresponding to each pre-set fixed time window to obtain the initial warning level data corresponding to each fixed time window; S4: Based on the initial warning level data corresponding to each fixed time window, the final debris flow warning level data for the target area is determined by using the highest warning level output principle, thus completing the debris flow warning.

[0007] Further, S1 includes: Based on the data collected by rainfall monitoring equipment in the target area, raw minute-by-minute rainfall sequence data are obtained using data extraction methods; Based on the raw minute-level rainfall sequence data, event segmentation is performed using a preset rainfall event classification standard to obtain real-time minute-level rainfall data.

[0008] Further, S2 includes: Based on real-time minute-level rainfall data, a parallel operation mechanism is established using a sliding time window algorithm with a preset update step size to obtain multiple parallel sliding time window data. Based on multiple parallel sliding time window data, a real-time accumulation and comparison algorithm is used to continuously track candidate peak moments and obtain the short-duration cumulative rainfall corresponding to each window.

[0009] Furthermore, the expression for the short-duration cumulative rainfall is: ; in, This indicates the cumulative rainfall over a short period of time. This represents the function that takes the maximum value. This indicates the length of a fixed time window determined by system presets. This indicates the end time of the current sliding window, obtained based on the real-time system clock. The first one extracted based on the rainfall event segmentation results At that moment, every minute This indicates that the data was acquired based on real-time, minute-level rainfall data. The minute-by-minute rainfall value at any given time.

[0010] Further, S3 includes: Obtain quantile regression threshold data containing multiple quantile gradients, constructed based on historical rainfall data and debris flow event data; The cumulative rainfall over short durations is compared and judged step by step across gradients with the quantile regression threshold data to obtain the corresponding over-limit status and obtain the initial warning level data corresponding to each fixed time window.

[0011] Furthermore, the expression for the quantile regression threshold data is as follows: ; in, This represents the critical short-duration rainfall intensity within a fixed time window corresponding to the quantile regression threshold data. This represents the rainfall duration from the start of rainfall to the moment when the peak short-duration rainfall intensity occurs, dynamically calculated based on the continuously updated time of the current maximum short-duration rainfall intensity. This represents the scaling coefficients obtained by using quantile regression fitting based on regional historical debris flow triggering event data and the corresponding quantile regression threshold. This represents the shape parameters obtained by quantile regression fitting based on regional historical debris flow triggering event data and the quantile regression threshold.

[0012] Further, S4 includes: Based on the initial warning level data corresponding to each fixed time window, the warning levels of all windows at the same time are extracted using a concurrent state evaluation mechanism to obtain the level distribution data. Based on the grade distribution data, the logic of prohibiting multi-window weighted averaging and directly selecting the highest exceedance level is used to determine the final debris flow warning level data of the target area, thus completing the debris flow warning.

[0013] The beneficial effects of this invention are as follows: This invention provides a debris flow early warning method based on the characteristics of short-duration heavy rainfall. By acquiring real-time minute-level rainfall data and employing a sliding time window algorithm with a preset update step size, it calculates the short-duration cumulative rainfall within multiple preset fixed time windows, breaking the limitation of traditional models that must rely on the total duration of the complete rainfall event. This feature can accurately depict the short-term changes and extreme rainfall concentration within the rainfall process in real time, effectively avoiding the dilution effect of traditional long-duration rainfall on short-duration heavy rain pulse data, and significantly improving the system's sensitivity to debris flow triggering processes in small watersheds with fast confluence responses.

[0014] The initial warning level is obtained by comparing the cumulative rainfall over each short duration with the corresponding quantile regression threshold, and the final warning level is determined by using the highest warning level as the output principle. This feature enables real-time and dynamic updates to the warning status during rainfall without waiting for the rainfall event to end. Simultaneously, the mechanism of parallel determination across multiple time scales and direct selection of the highest risk level prevents the strong rainfall anomalies in a single time window from being averaged or diluted by low-risk values ​​in other windows, thereby significantly improving the accuracy of debris flow short-term warnings and the timeliness of disaster prevention response. Attached Figure Description

[0015] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is an exemplary flowchart of a debris flow early warning method based on the characteristics of short-term heavy rainfall, as shown in some embodiments of this specification. Figure 2 These are exemplary schematic diagrams illustrating the method for dividing independent rainfall time and the definition of rainfall magnitude and duration in different scenarios, according to some embodiments of this specification; Figure 3 This is an exemplary schematic diagram showing the results of different quantile regression line ID models according to some embodiments of this specification; Figure 4 This is an exemplary schematic diagram illustrating the debris flow rainfall threshold and index evaluation under different short-term heavy rainfall indices according to some embodiments of this specification; Figure 5 This is an exemplary schematic diagram showing the predictive performance evaluation results of a conventional intensity-duration threshold model and a short-duration rainfall threshold model according to some embodiments of this specification. Detailed Implementation

[0016] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0017] Example Figure 1 This is an exemplary flowchart illustrating a debris flow early warning method based on short-duration heavy rainfall characteristics, according to some embodiments of this specification. Figure 1 As shown, the process includes the following steps. In some embodiments, the process may be executed by a processor.

[0018] S1: Acquire real-time minute-level rainfall data collected by rainfall monitoring equipment in the target area.

[0019] Rainfall monitoring equipment refers to hardware sensing facilities deployed in a target area for real-time observation and recording of precipitation. For example, rainfall monitoring equipment may include rain gauges, data transmission modules, and power supply systems. It can record the time of precipitation occurrence and the corresponding precipitation amount at a high frequency, and is the front-end data source of the early warning system.

[0020] In some embodiments, the processor can obtain rainfall monitoring equipment data by receiving and parsing electrical or digital signals transmitted in real time from on-site deployed hardware. Specifically, the processor receives raw telemetry messages from various monitoring stations through a network interface, and after protocol parsing and format conversion, extracts the station number, observation timestamp, and corresponding precipitation physical quantity values.

[0021] Real-time minute-level rainfall data refers to a set of precipitation observation values ​​that are continuously collected and updated with a minute-level time resolution. For example, real-time minute-level rainfall data may include a specific year, month, day, hour, and minute timestamp, the cumulative rainfall in millimeters within the corresponding minute, and status flags used to identify the continuity and validity of the data. It can finely depict the changes within short-term rainfall processes and strong rainfall pulses.

[0022] In some embodiments, the processor can obtain real-time minute-level rainfall data by performing time-series alignment and outlier filtering on the raw collected data returned by the rainfall monitoring equipment. Specifically, the processor interpolates or resamples the received irregular observation data according to a standard one-minute time interval. If multiple records exist within the same minute, they are accumulated and merged. If no data is returned for a certain minute, its precipitation is filled with zero by default, thereby forming a continuous minute-scale rainfall sequence on the time axis.

[0023] In some embodiments, the processor can obtain raw minute-level rainfall sequence data based on data collected by rainfall monitoring equipment in the target area using a data extraction method; based on the raw minute-level rainfall sequence data, it can perform event segmentation processing using a preset rainfall event segmentation standard to obtain real-time minute-level rainfall data.

[0024] The data collected by rainfall monitoring equipment refers to the initial meteorological information that is directly measured and packaged by the front-end sensing equipment and sent to the early warning system without in-depth processing. For example, the data collected by rainfall monitoring equipment may include the latitude and longitude coordinates of the station, the equipment operating status code, the battery voltage information, and discrete precipitation trigger pulse records that are not aligned with the time scale. This is the original basis for subsequent event segmentation and rainfall statistics.

[0025] In some embodiments, the processor can receive messages sent by front-end rain gauges via a wireless communication network through a polling or event-triggered mechanism to obtain data collected by the rainfall monitoring equipment. The processor establishes communication connections with each rainfall monitoring station, unpacks the received raw binary data stream or text data packet according to the set communication protocol, and extracts the original precipitation sensing records and the time stamp information provided by the equipment.

[0026] Raw minute-by-minute precipitation sequence data refers to continuous time series data formed after preliminary cleaning and time resolution unification of collected discrete precipitation data. For example, raw minute-by-minute precipitation sequence data can include minute nodes arranged strictly in chronological order, as well as the corresponding measured precipitation in millimeters at each minute node. It contains a complete and continuous record of all precipitation and non-precipitation periods during long-term observation.

[0027] In some embodiments, the processor can obtain raw minute-level rainfall sequence data by performing time-axis normalization mapping and data aggregation on the data collected by the rainfall monitoring equipment. The processor categorizes all valid collected data according to minute-level timestamps, sums up the rainfall corresponding to multiple tipping bucket rain gauge pulse signals within the same minute, and connects all the sorted minute nodes in chronological order into a long, unbroken sequence.

[0028] Rainfall event classification criteria refer to the rules and threshold conditions used to divide continuous long-term rainfall data into several independent precipitation events. For example, rainfall event classification criteria may include two core indicators: one is the threshold for the interval of consecutive periods without rainfall, such as greater than or equal to four hours; the other is the upper limit of cumulative trace precipitation over a longer duration, such as six hours, such as less than 0.5 mm, to define the physical boundary between two significant precipitation events.

[0029] In some embodiments, the processor can obtain rainfall event segmentation standard data by reading a preset configuration file or a system default parameter library. The processor retrieves rainless intervals and trace precipitation tolerance values ​​determined through hydrometeorological historical data analysis from local memory or a cloud database, and loads these specific values ​​into memory as a comparison benchmark for subsequent time series segmentation.

[0030] S2: Based on real-time minute-level rainfall data, a sliding time window algorithm with a preset update step size is used to calculate the short-duration cumulative rainfall within multiple preset fixed time windows.

[0031] The update step size refers to the time interval by which the window moves forward during sliding time window calculation. For example, the update step size can include a specific time value, such as ten minutes. This means that the early warning system will trigger a re-statistical analysis and peak comparison of rainfall data within the past time period every ten minutes, which determines the response frequency of the early warning system to rainfall change pulses.

[0032] In some embodiments, the processor can obtain update step size data by acquiring pre-set system operating parameters. Based on the real-time requirements of the early warning system and the allocation of computing resources, the processor reads the preset step size parameters from the system configuration file and converts them into the trigger cycle of the system's internal timer to control the execution frequency of subsequent sliding window calculation tasks.

[0033] A fixed time window refers to a time segment with a defined and fixed length used to statistically analyze short-duration cumulative rainfall. For example, a fixed time window can include intervals with different time scales such as ten minutes, twenty minutes, thirty minutes, and sixty minutes. These windows of different lengths can respectively reflect the mobilization effect of extremely short-duration heavy rainfall pulses, rapid runoff processes, and continuous heavy rainfall on loose sediment sources.

[0034] In some embodiments, the processor can obtain fixed-time-window data by parsing multiple observation scale parameters set by the early warning model. Based on the short-term early warning model settings, the processor converts the system-preset time length parameters such as ten minutes, twenty minutes, thirty minutes, and sixty minutes into specific time difference variables, which are used to define the corresponding calculation range in a continuous time series.

[0035] In some embodiments, to evaluate the performance of the early warning model, the accuracy data is expressed as follows: ; in, This represents accuracy data, used to measure the overall classification accuracy of the model. This represents the number of true positive samples based on model validation results, i.e., the number of samples correctly predicted as debris flow triggering events. This represents the number of true negative samples based on model validation results, i.e., the number of samples correctly predicted as not triggering a rainfall event. This represents the number of false positive samples based on model validation results; that is, the number of samples that were incorrectly predicted as triggered but did not actually trigger the event. This represents the number of false negative samples based on model validation results, i.e., the number of samples that were actually triggered but were incorrectly predicted as not triggered.

[0036] In some embodiments, The expression for the evaluation indicator data is as follows: ; in, This indicates that the evaluation metrics data take into account both precision and recall. The weighting parameter represents the degree to which recall is important relative to precision. To represent precision data, its expression is: , Recall rate data is expressed as follows: .

[0037] Short-duration cumulative rainfall refers to the sum of all minute-level rainfall within a specified short fixed time window. For example, short-duration cumulative rainfall can include the total number of millimeters of precipitation generated in a target area over the past ten, twenty, or sixty minutes. It does not require normalization based on the total duration of the complete rainfall event and can directly and objectively reflect the concentration of rainfall within that short time window.

[0038] In some embodiments, the processor can obtain the short-duration cumulative rainfall by summing the minute-level precipitation sequences within a fixed time window. Based on the current time point and the length of the selected fixed time window, the processor backtracks to extract all real-time minute-level precipitation data within that time range, and algebraically sums the precipitation millimeters corresponding to these minute-level data to obtain the total rainfall for that time segment.

[0039] In some embodiments, for the first The expression for the total rainfall data of an independent rainfall event is as follows: ; in, Indicates the first Total rainfall data for each independent rainfall event. This indicates the first event determined based on the rainfall event classification criteria. The start time of an independent rainfall event. This indicates the first event determined based on the rainfall event classification criteria. The end time of an independent rainfall event. The first one extracted based on the rainfall event segmentation results At that moment, every minute This indicates that the rainfall data obtained is based on the real-time minute-level rainfall data. The minute-by-minute rainfall value at any given time.

[0040] In some embodiments, the processor can establish a parallel operation mechanism based on real-time minute-level rainfall data and a sliding time window algorithm with a preset update step size to obtain multiple parallel sliding time window data; based on the multiple parallel sliding time window data, a real-time accumulation and comparison algorithm is used to continuously track candidate peak moments to obtain the short-duration cumulative rainfall corresponding to each window.

[0041] Sliding time window data refers to a dynamic data set that slides forward over time according to a set update step size, containing the current time point and past time periods. For example, sliding time window data can include the timestamp of the current calculation moment, the start and end time range included in the window, and all real-time minute-level rainfall data sequences included in that time range, which is in a state of continuous dynamic updating and evolution.

[0042] In some embodiments, the processor can obtain sliding time window data by using the current system time as the termination time and extracting a time segment of a certain length in the historical direction. The processor periodically wakes up the computing process with a set update step size, reads the current real-time time of the system as the end of the window, subtracts the fixed length of the time window to obtain the beginning of the window, thereby extracting a subset of data within this dynamic interval from the complete real-time minute-level rainfall data queue.

[0043] A candidate peak moment refers to the current time point within a fixed time window during continuous monitoring of a rainfall event when the short-duration cumulative rainfall exceeds all previous statistical maximum values ​​for that rainfall event. For example, a candidate peak moment may include the specific date, time, and minute information of the record-breaking event, as well as the specific value of the record-breaking short-duration cumulative rainfall at that moment. It is a key benchmark point used to determine the maximum rainfall intensity and to calculate the duration of subsequent rainfall events.

[0044] In some embodiments, the processor can obtain candidate peak time data by comparing the short-duration cumulative rainfall calculated at the current moment with the stored historical maximum value. After each update step, the processor compares the currently calculated cumulative rainfall for a certain window with the historical maximum cumulative value for that window within the rainfall event recorded in memory. If the current value is greater than the historical maximum value, the maximum value in memory is updated to the current value, and the current system time is recorded as the candidate peak time for that window.

[0045] In some embodiments, the expression for short-duration cumulative rainfall is: ; in, This indicates the cumulative rainfall over a short period of time. This represents the function that takes the maximum value. This indicates the length of a fixed time window determined by system presets. This indicates the end time of the current sliding window, obtained based on the real-time system clock. The first one extracted based on the rainfall event segmentation results At that moment, every minute This indicates that the data was acquired based on real-time, minute-level rainfall data. The minute-by-minute rainfall value at any given time.

[0046] S3: Compare the cumulative rainfall over each short duration with the quantile regression thresholds corresponding to each pre-set fixed time window to obtain the initial warning level data corresponding to each fixed time window.

[0047] Quantile regression thresholds refer to the critical rainfall intensity curve values ​​used to classify the probability of debris flow triggering at different probability levels based on the statistical relationship between historical debris flow events and rainfall data. For example, quantile regression thresholds can include critical rainfall intensity values ​​at different quantile levels such as 10%, 30%, and 50%, which correspond to the lower limit of the possible triggering of debris flows, the boundary of a significant increase in the probability of triggering, and the boundary of the typical triggering level, respectively.

[0048] In some embodiments, the processor can obtain quantile regression threshold data by calling a rainfall intensity and duration model pre-established in a double logarithmic coordinate system and substituting relevant parameters. The processor extracts the scale coefficients and shape parameters pre-fitted for different quantiles in the model, combines them with the rainfall duration data before the occurrence of peak short-duration rainfall intensity obtained in real time, and calculates the critical values ​​of rainfall intensity at each quantile level under the corresponding current duration conditions through exponential or logarithmic inverse operations.

[0049] In some embodiments, the expression for rainfall duration data is: ; in, Indicates the first Rainfall duration data for independent rainfall events, in hours. This indicates the first event determined based on the rainfall event classification criteria. The start time of an independent rainfall event, in minutes. This indicates the first event determined based on the rainfall event classification criteria. The end time of an independent rainfall event, in minutes.

[0050] In some embodiments, the expression for average rainfall intensity data is: ; in, Indicates the first Average rainfall intensity data for independent rainfall events, Indicates the first Total rainfall data for each independent rainfall event. Indicates the first Rainfall duration data for independent rainfall events.

[0051] Initial warning level data refers to a single-dimensional risk rating derived from a preliminary comparison of short-duration cumulative rainfall within a single fixed time window with the corresponding quantile regression threshold. For example, initial warning level data may include information such as a blue warning status for a ten-minute window and a non-compliance status for a sixty-minute window. It only reflects the debris flow triggering risk level corresponding to the concentration of rainfall at a single time scale.

[0052] In some embodiments, the processor can determine the initial warning level data by comparing the real-time calculated short-duration cumulative rainfall with multiple levels of quantile regression thresholds. For each sliding time window, the processor compares its current cumulative rainfall or converted rainfall intensity with the thresholds corresponding to the 10%, 30%, and 50th percentiles, respectively, and assigns an independent warning color label or risk number code to the time window based on the highest threshold limit exceeded.

[0053] In some embodiments, the processor can acquire quantile regression threshold data containing multiple quantile gradients, constructed based on historical rainfall data and debris flow event data; compare and distinguish each short-duration cumulative rainfall with the quantile regression threshold data across gradients to obtain the corresponding over-limit status, and obtain the initial warning level data corresponding to each fixed time window.

[0054] Historical rainfall data refers to a complete collection of records of all rainfall events accumulated in a target area over many years of long-term monitoring. For example, historical rainfall data may include detailed parameters such as the occurrence time, end time, minute-level rainfall sequence, maximum short-duration cumulative rainfall under different time windows, and total duration of rainfall events for each rainfall event in all years since the monitoring equipment was established. It is the basic database for building and validating early warning models.

[0055] In some embodiments, the processor can obtain historical rainfall data by accessing a central database of a meteorological or hydrological department to retrieve long-archived meteorological monitoring records. The processor sends a structured query request to the database server storing the historical ledger, retrieving and downloading all archived minute-level precipitation logs and event statistics forms according to the specified time range and target station number.

[0056] Debris flow event data refers to a detailed record of historical debris flow disasters that actually occurred within a target watershed. For example, debris flow event data can include the specific date and minute-accurate outbreak time of each debris flow disaster, the specific location coordinates of the disaster, the size of the debris flow, its duration, and the direct or indirect impacts caused. This data is used to perform timeline matching with rainfall data to extract triggering conditions.

[0057] In some embodiments, the processor can obtain debris flow event data by reading historical disaster investigation reports from disaster management departments or long-term observation records from field scientific observation stations. The processor imports and parses disaster catalog lists stored in text, spreadsheet, or structured database formats, extracts disaster occurrence timestamps and location coordinates, and converts them into a standardized data format that can be precisely matched with rainfall time series.

[0058] An "over-limit status" refers to a logical judgment result where the rainfall index value obtained from real-time monitoring or calculation exceeds a pre-set safety threshold or risk level threshold. For example, an over-limit status may include a Boolean value or classification flag to indicate whether the current ten-minute short-duration cumulative rainfall has exceeded the 30% quantile regression threshold, which indicates that the system needs to switch from normal monitoring to triggering the corresponding level of warning.

[0059] In some embodiments, the processor can obtain out-of-limit status data by performing a logical judgment operation to compare the current calculated rainfall value with a preset threshold. The processor extracts the current short-duration cumulative rainfall value of a certain time window, uses it as the comparison number, extracts the corresponding first-order quantile regression threshold as the comparison benchmark number, performs a greater than or equal to logical operation, and generates a status code indicating that the limit has been exceeded if the condition is met; otherwise, it generates a safe status code indicating that the limit has not been exceeded.

[0060] In some embodiments, the expression for the quantile regression threshold data is: ; in, This represents the critical short-duration rainfall intensity within a fixed time window corresponding to the quantile regression threshold data. This represents the rainfall duration from the start of rainfall to the moment when the peak short-duration rainfall intensity occurs, dynamically calculated based on the continuously updated time of the current maximum short-duration rainfall intensity. This represents the scaling coefficients obtained by using quantile regression fitting based on regional historical debris flow triggering event data and the corresponding quantile regression threshold. This represents the shape parameters obtained by quantile regression fitting based on regional historical debris flow triggering event data and the quantile regression threshold.

[0061] In some embodiments, the expression for short-duration rainfall intensity data is: ; in, This indicates that the window length is Time, with For short-duration rainfall intensity data at the termination time, This indicates the length of a fixed time window determined by system presets. This indicates that the data is obtained by accumulating the real-time minute-level rainfall data within a window of length [missing information]. Time, with The short-duration cumulative rainfall at the time of termination.

[0062] In some embodiments, to extract the indicators corresponding to debris flow triggering events, the expression for the triggering short-term heavy rainfall intensity data corresponding to the debris flow event is: ; in, This indicates the intensity of the short-term heavy rainfall that triggered the debris flow event. This indicates the time of occurrence of a debris flow, obtained from debris flow event data. This indicates the end time of the current sliding window, obtained based on the real-time system clock. This indicates that the window length is Time, with This refers to the short-duration rainfall intensity data at the termination time. The formula indicates that the maximum short-duration rainfall intensity of each window is extracted as a trigger indicator before or at the time of the debris flow, avoiding the use of rainfall information after the disaster.

[0063] S4: Based on the initial warning level data corresponding to each fixed time window, the final debris flow warning level data for the target area is determined by using the highest warning level output principle, thus completing the debris flow warning.

[0064] The final debris flow warning level data refers to the highest risk level instruction issued by the warning system after comprehensively considering the initial warning status of all concurrently operating time windows. For example, the final debris flow warning level data may include the final determined blue, yellow, or red warning instruction code, as well as the corresponding defense response recommendations. It directly determines the level of actual response actions that disaster prevention and mitigation departments will take.

[0065] In some embodiments, the processor can obtain the final debris flow warning level data by performing the highest-level extraction operation on the initial warning level data corresponding to all fixed time windows acquired at the same time. The processor summarizes the warning levels of the four windows at the current time (ten minutes, twenty minutes, thirty minutes, and sixty minutes), and without performing a weighted average calculation, directly selects the one with the highest risk level among these four states according to a preset risk level sequence, and uses it as the current final warning level issued for the area.

[0066] In some embodiments, the processor can extract the warning levels of all windows at the same time based on the initial warning level data corresponding to each fixed time window using a concurrent state evaluation mechanism to obtain the level distribution data; based on the level distribution data, the processor can determine the final debris flow warning level data of the target area by using the logic of prohibiting multi-window weighted averaging and directly selecting the highest exceedance level, thereby completing the debris flow warning.

[0067] Warning levels refer to a tiered risk scaling system that classifies the potential for debris flow disasters based on rainfall conditions. For example, warning levels can include three distinct tiers: blue, yellow, and red. Blue indicates that rainfall conditions are approaching the lower limit of the debris flow triggering limit; yellow indicates a significantly increased triggering probability; and red indicates that rainfall conditions have reached the typical triggering level, requiring immediate emergency response. This system intuitively transforms complex numerical indicators into easily understandable disaster prevention warning signals. Specific debris flow warning levels based on short-duration heavy rainfall characteristics are shown in Table 1.

[0068] Table 1. Debris Flow Warning Levels Based on Short-Term Heavy Rainfall Characteristics

[0069] In some embodiments, the processor can obtain warning level data by mapping a preset dictionary of rainfall exceeding threshold relationships. Based on the system's built-in warning level configuration table, the processor maps events with rainfall exceeding the decile threshold to a low-risk level, events exceeding the thirtieth percentile threshold to a medium-risk level, and events exceeding the fiftieth percentile threshold to a high-risk level, thereby assigning specific alarm level labels to different numerical exceedance states.

[0070] Level distribution data refers to a snapshot of the state set of warning levels triggered by multiple fixed time windows running in parallel at a certain moment. For example, level distribution data can include detailed information recorded simultaneously in an array or list: a 10-minute window is in a yellow warning state, a 20-minute window is in a blue warning state, and 30-minute and 60-minute windows are in a no-warning state. It fully demonstrates the multi-dimensional risk situation brought about by the concentrated distribution of current short-term rainfall at different time scales.

[0071] In some embodiments, the processor can obtain the level distribution data by concurrently reading and summarizing the initial warning states generated at the end of the current calculation cycle for all sliding time windows. At the end of the calculation cycle of each update step, the processor traverses the calculation threads or memory regions that manage each time window, extracts the independent warning level codes derived from the comparison of their respective short-duration cumulative rainfall, and packages and integrates these independent status codes into a structure or array under the same timestamp.

[0072] In some embodiments, the range of values ​​for the constant parameters involved in the above formula is as follows: the length of the fixed time window. This parameter is typically a sub-hour-scale parameter reflecting the hydrological response process of a small watershed. Its value range can be set according to the specific characteristics of the watershed. Preferably, The value can be 10, 20, 30, or 60 minutes. The parameter used to adjust the recall weight... Considering the high sensitivity to the risk of missed reports in debris flow early warning systems, its value is usually greater than 1. In a preferred embodiment of the present invention, The value is 2, that is, it adopts... The indicators are designed to further emphasize the model's ability to identify actual debris flow triggering events. The fitted quantile levels of the quantile regression thresholds are typically set between 0.1 and 0.5 to construct different levels of early warning defenses. In a preferred embodiment of the invention, 0.1, 0.3, and 0.5 are used as quantile benchmarks for classifying blue, yellow, and red warning levels. The formula... or in logarithmic coordinates The scaling factor involved (or intercept) and shape parameters (or slope) The value is not a fixed constant, but a regional empirical parameter obtained by those skilled in the art based on long-sequence historical rainfall data and debris flow event data of the target area, using statistical software to perform regression analysis fitting at specified quantiles (such as 0.1, 0.3, 0.5). Its specific value naturally adjusts with the change of historical data samples in the watershed.

[0073] In some embodiments, such as Figure 2The diagram illustrates the method for dividing rainfall into independent events and the definitions of rainfall magnitude and duration under different scenarios. The blue bars on the left represent 10-minute rainfall, and the red smooth curve represents cumulative rainfall. By setting interval standards for rainfall events, such as a continuous dry period D ≥ 4 hours with cumulative rainfall Ra = 0 mm, or D ≥ 6 hours with Ra ≤ 0.5 mm, continuous rainfall sequences are divided into independent rainfall events. The diagram visually indicates the interval between the preceding rainfall event and subsequent events using dashed boxes and double-headed arrows. The time of debris flow occurrence is marked with a red star, and the diagram also schematically indicates... I 10. I 20. I 30. I The specific positions of different short-duration heavy rainfall indicators (e.g., 60) on the time axis of the rainfall process.

[0074] In some embodiments, such as Figure 3 As shown, the regression lines at different quantiles are... ID A schematic diagram of the results of the (rainfall intensity-duration) model. Subplot (a) shows the scatter distribution of historical rainfall events in the target area from 2000 to 2025 in a double logarithmic coordinate system, with 0.1, 0.3, and 0.5 quantile regression lines plotted in pink, purple, and green respectively, visually demonstrating the power-law relationship that the rainfall intensity threshold decreases with increasing rainfall duration. Subplot (b) shows the results calculated using different models at different quantile levels (0.1, 0.3, 0.5). F 2 values ​​and Ac A bar chart comparing two evaluation metrics is used to assess the model's goodness of fit and overall performance.

[0075] In some embodiments, such as Figure 4 The figure shown is a detailed schematic diagram illustrating the debris flow rainfall threshold and index evaluation results under different short-term heavy rainfall indices. Figures (a) to (h) are divided into four sub-figures, each corresponding to a short-term heavy rainfall index. I 10. I 20. I 30. I The calculation and evaluation results of 60. The left-hand plot of each subplot (e.g., a, c, e, g) shows the scatter distribution of rainfall for different rainfall durations, and the corresponding 0.1, 0.3, and 0.5 quantile regression threshold lines; the right-hand plot of each subplot (e.g., b, d, f, h) shows the corresponding evaluation index ( F 2 values ​​and Ac A bar chart comparing values ​​at different quantile levels.

[0076] In some embodiments, such as Figure 5 As shown, this is the traditional intensity-duration threshold ( ID This diagram illustrates a comparison of the predictive performance evaluation results of the traditional model and the short-duration rainfall threshold model. Columns (a)-(c) on the left side of the diagram show the performance of the traditional model at regression levels of 0.1, 0.3, and 0.5 quantiles, respectively. ID The model and the present invention I 10. I 20. I 30. I 60-level short-term heavy rainfall index Ac Value comparison; the right columns (d)-(f) show the values ​​of each model at the corresponding quantile levels. F 2. Value Comparison. The specific values ​​on the bar chart directly demonstrate that the short-duration rainfall threshold model proposed in this invention outperforms traditional models in all quantile gradients. ID The model demonstrates the significant advantages of this method in terms of prediction accuracy and effectiveness.

Claims

1. A debris flow early warning method based on short-term heavy rainfall characteristics, characterized in that, include: S1: Acquire real-time minute-level rainfall data collected by rainfall monitoring equipment in the target area; S2: Based on real-time minute-level rainfall data, a sliding time window algorithm with a preset update step size is used to calculate the short-duration cumulative rainfall within multiple preset fixed time windows; S3: Compare the cumulative rainfall for each short duration with the quantile regression threshold corresponding to each pre-set fixed time window to obtain the initial warning level data corresponding to each fixed time window; S4: Based on the initial warning level data corresponding to each fixed time window, the final debris flow warning level data for the target area is determined by using the highest warning level output principle, thus completing the debris flow warning.

2. The debris flow early warning method based on short-term heavy rainfall characteristics according to claim 1, characterized in that, S1 includes: Based on the data collected by rainfall monitoring equipment in the target area, raw minute-by-minute rainfall sequence data are obtained using data extraction methods; Based on the raw minute-level rainfall sequence data, event segmentation is performed using a preset rainfall event classification standard to obtain real-time minute-level rainfall data.

3. The debris flow early warning method based on short-term heavy rainfall characteristics according to claim 1, characterized in that, S2 includes: Based on real-time minute-level rainfall data, a parallel operation mechanism is established using a sliding time window algorithm with a preset update step size to obtain multiple parallel sliding time window data. Based on multiple parallel sliding time window data, a real-time accumulation and comparison algorithm is used to continuously track candidate peak moments and obtain the short-duration cumulative rainfall corresponding to each window.

4. The debris flow early warning method based on short-term heavy rainfall characteristics according to claim 3, characterized in that, The expression for the short-duration cumulative rainfall is: ； in, This indicates the cumulative rainfall over a short period of time. This represents the function that takes the maximum value. This indicates the length of a fixed time window determined by system presets. This indicates the end time of the current sliding window, obtained based on the real-time system clock. The first one extracted based on the rainfall event segmentation results At that moment, every minute This indicates that the data was acquired based on real-time, minute-level rainfall data. The minute-by-minute rainfall value at any given time.

5. The debris flow early warning method based on short-term heavy rainfall characteristics according to claim 1, characterized in that, S3 includes: Obtain quantile regression threshold data containing multiple quantile gradients, constructed based on historical rainfall data and debris flow event data; The cumulative rainfall over short durations is compared and judged step by step across gradients with the quantile regression threshold data to obtain the corresponding over-limit status and obtain the initial warning level data corresponding to each fixed time window.

6. The debris flow early warning method based on short-term heavy rainfall characteristics according to claim 5, characterized in that, The expression for the quantile regression threshold data is: ； in, This represents the critical short-duration rainfall intensity within a fixed time window corresponding to the quantile regression threshold data. This represents the rainfall duration from the start of rainfall to the moment when the peak short-duration rainfall intensity occurs, dynamically calculated based on the continuously updated time of the current maximum short-duration rainfall intensity. This represents the scaling coefficients obtained by using quantile regression fitting based on regional historical debris flow triggering event data and the corresponding quantile regression threshold. This represents the shape parameters obtained by quantile regression fitting based on regional historical debris flow triggering event data and the quantile regression threshold.

7. The debris flow early warning method based on short-term heavy rainfall characteristics according to claim 1, characterized in that, S4 includes: Based on the initial warning level data corresponding to each fixed time window, the warning levels of all windows at the same time are extracted using a concurrent state evaluation mechanism to obtain the level distribution data. Based on the grade distribution data, the logic of prohibiting multi-window weighted averaging and directly selecting the highest exceedance level is used to determine the final debris flow warning level data of the target area, thus completing the debris flow warning.

Images