Method and system for retrieving degree-day factor based on snow water equivalent difference

Abstract

The application provides a method and system for inverting degree-day factors based on snow water equivalent difference, and relates to the technical field of hydrological remote sensing. The method of the application firstly acquires and unifies multi-source snow water equivalent data, air temperature, radiation, wind speed, relative humidity and other meteorological elements to form a standardized input data set, adopts data assimilation fusion to generate assimilated snow water equivalent and calculate uncertainty; combines meteorological conditions and uncertainty to determine the snow melting process, calculates the pixel-scale degree-day factor when the snow melting is determined, and constructs a time-space database; uses convolutional neural network and long short-term memory network to extract spatial and temporal characteristics, obtains historical sample degree-day factor distribution based on similarity matching for prediction, calculates the snow melting amount and estimates the uncertainty, so as to improve the snow melting simulation and prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of hydrological remote sensing technology, and in particular to a method and system for retrieving daily factors based on snow water equivalent difference. Background Technology

[0002] Snowmelt is a crucial component of the hydrological cycle in cold-region watersheds, directly impacting regional water resource allocation, flood control, disaster prevention and mitigation, and ecosystem stability. Accurate prediction of snowmelt processes is of great significance for reservoir management, agricultural production, and water resource management in the context of climate change. Currently, snowmelt simulation primarily employs two methods: one is based on the energy balance principle, calculating snowmelt through energy budgets such as radiation flux, sensible heat, and latent heat; the other is the day-degree factor method, which estimates snowmelt based on the empirical relationship between temperature and snowmelt amount. Due to its ease of calculation and low input data requirements, this method has been widely integrated into various hydrological models and watershed management tools. However, existing day-degree factor methods mostly use fixed values ​​or regional constants to represent the day-degree factor, failing to reflect the dynamic impacts of different underlying surface types, topographic conditions, and climate change on the snowmelt process, resulting in limited prediction accuracy.

[0003] On the other hand, snow water equivalent is an important indicator of snow moisture storage and a core data source for snowmelt simulation and forecasting. Currently, passive microwave remote sensing products are widely used to obtain large-scale snow water equivalent data. However, these products have low spatial resolution, typically between ten and twenty-five kilometers, which is insufficient to meet the research needs of complex terrain and detailed hydrological processes. Furthermore, in forested areas, wet snow areas, and regions with high surface roughness, the remote sensing inversion accuracy of snow water equivalent is low, with significant systematic errors. In addition, although deep learning technology has shown promising applications in image analysis and time series modeling, there is currently a lack of methods and systems to combine high spatial resolution snow water equivalent data with deep learning technology for dynamic inversion of degree-day factors. There is also a lack of a complete technical route for establishing a spatiotemporal database of degree-day factors based on multi-source data and using similarity matching for snowmelt forecasting. Existing research also rarely incorporates uncertainty information from snow water equivalent products into the degree-day factor estimation or snowmelt forecasting process, thus affecting the reliability of the forecast results.

[0004] Therefore, there is an urgent need to propose a new method and system that utilizes high-resolution snow water equivalent data and combines deep learning technology to dynamically reflect the spatiotemporal distribution of the degree-day factor, effectively reduce the impact of remote sensing errors, and improve the accuracy of snowmelt forecasting, so as to overcome the shortcomings of existing technologies. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, the purpose of this invention is to provide a method and system for inverting the degree-day factor based on the snow water equivalent difference, which can improve the accuracy of snowmelt simulation and forecast.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A method for retrieving daily factors based on snowmelt equivalent difference includes:

[0008] Acquire snow water equivalent data for the study area over several consecutive days, and unify the snow water equivalent data from multi-source remote sensing products, ground observation data, and numerical model simulation results with meteorological element data to the same spatial resolution and to a daily scale, merging them into a standardized input dataset; the meteorological element data includes air temperature, radiation, wind speed, and relative humidity;

[0009] The multi-source snow water equivalent data in the standardized input dataset are fused using data assimilation technology to obtain assimilated snow water equivalent data, and the corresponding uncertainty is calculated.

[0010] The change in snow water equivalent data after assimilation between two adjacent days is detected. When a decrease occurs, the meteorological conditions of temperature, radiation, wind speed and relative humidity are combined, and the reliability of the difference in snow water equivalent data is constrained according to the uncertainty, so as to determine whether it is a snow melting process.

[0011] When snowmelt is identified, the degree-day factor at the pixel scale is calculated based on the difference in snow water equivalent data after assimilation between two adjacent days and the positive cumulative temperature difference between the two days.

[0012] The spatial distribution of the daily time factor and the corresponding meteorological elements, underlying surface type, elevation and slope aspect information are stored to construct a spatiotemporal database of the time factor.

[0013] A convolutional neural network is used to extract the spatial distribution features of the day-of-day factor, and a long short-term memory network is combined to analyze the time series changes of the spatial distribution features to generate a multidimensional feature vector of the day-of-day factor.

[0014] In the snowmelt forecasting process, based on the meteorological forecast values, underlying surface parameters and topographic information of the forecast period, a similarity matching algorithm is used to retrieve historical samples with the smallest Euclidean distance to the feature vector of the forecast period from the spatiotemporal database, obtain the corresponding degree-day factor distribution as the forecast input, calculate the snowmelt amount, and estimate the snowmelt uncertainty based on the variance of similar sample groups.

[0015] Preferably, a data assimilation technique is used to fuse the multi-source snowmelt equivalent data in the standardized input dataset to obtain assimilated snowmelt equivalent data, and the corresponding uncertainty is calculated, including:

[0016] The snow water equivalent obtained from the numerical model is set as the state variable, and the ground observation data and multi-source remote sensing data are set as the observation variables. An observation operator is established to map the state variables to the observation space, and the observation error covariance matrix and system process noise are set according to the accuracy experience of different data sources.

[0017] An ensemble model is used to generate a set of predicted states;

[0018] The predicted state set is updated based on the Kalman gain to obtain assimilated snow water equivalent data, and the process is carried out in parallel at each pixel point in the study area according to time steps.

[0019] The uncertainty at the cell level is calculated based on the deviation of observations from each data source relative to the multi-source average snow water equivalent data.

[0020] Preferably, the uncertainty The calculation formula is:

[0021]

[0022] Where x and y represent the horizontal and vertical coordinates of a spatial location, respectively; t represents a specific day or time. The snow water equivalent data value of the i-th data source; denoted as the average snow water equivalent of a pixel across multiple sources; N represents the number of data sources participating in the fusion.

[0023] Preferably, the detection is based on the change in assimilated snow water equivalent data over two consecutive days. When a decrease occurs, the meteorological conditions of air temperature, radiation, wind speed, and relative humidity are considered, and the uncertainty is used to apply a reliability constraint to the difference in snow water equivalent data to determine whether it is a snowmelt process. This includes:

[0024] a) Calculate the difference between the assimilated snow water equivalent data on day t and day t-1 at the pixel scale, and proceed to the subsequent discrimination when the difference is negative;

[0025] b) Determine that the air temperature on day t exceeds a preset temperature threshold;

[0026] c) Determine that the radiation on day t exceeds a preset radiation threshold;

[0027] d) Determine that the wind speed and relative humidity on day t satisfy their respective preset threshold conditions;

[0028] e) Compare the absolute value of the difference with the uncertainty, and determine that the absolute value is greater than twice the uncertainty;

[0029] When the conditions in steps b) to e) are met simultaneously, it is determined that the snow melting process occurs in the pixel on day t.

[0030] Preferably, when snowmelt is detected, a degree-day factor at the pixel scale is calculated based on the difference in assimilated snow water equivalent data between two adjacent days and the positive cumulative temperature difference between the two days, including:

[0031] Only for pixels identified as part of the snow melting process, the snow water equivalent difference between day t and day t-1 is calculated based on the assimilated snow water equivalent data.

[0032] Calculate the positive cumulative temperature difference between two adjacent days; the positive cumulative temperature difference is the cumulative value of the portion of the daily average temperature of the pixel on day t-1 and day t that exceeds the threshold temperature;

[0033] The degree-day factor of the pixel on day t is obtained by the ratio of the obtained snow water equivalent difference to the positive temperature difference.

[0034] Preferably, a convolutional neural network is used to extract the spatial distribution features of the day-of-day factor, and a long short-term memory network is combined to analyze the time series changes of the spatial distribution features to generate a multidimensional feature vector of the day-of-day factor, including:

[0035] The two-dimensional matrix of the daily life factor is used as input to the convolutional neural network;

[0036] Several convolutional and pooling layers are used to extract spatial feature maps, resulting in a spatial feature matrix that characterizes spatial texture, local extremum distribution, and edge features.

[0037] The spatial feature matrix arranged in chronological order, with the historical days length as a preset value, is input into the long short-term memory network to capture the weekly, monthly, and seasonal changes of the degree-day factor in the time series, thus obtaining a time feature vector.

[0038] The multidimensional feature vector of the degree-day factor is generated based on the spatial feature matrix and the temporal feature vector.

[0039] Preferably, during the snowmelt forecasting process, based on the meteorological forecast values, underlying surface parameters, and topographic information for the forecast period, a similarity matching algorithm is used to retrieve historical samples with the smallest Euclidean distance to the feature vectors of the forecast period from the spatiotemporal database. The corresponding degree-day factor distribution is then obtained as the forecast input, the snowmelt amount is calculated, and the snowmelt uncertainty is estimated based on the variance of the similar sample groups, including:

[0040] Based on the meteorological forecasts, underlying surface parameters, and topographic information for the forecast period, a feature vector for the forecast period is constructed.

[0041] Using Euclidean distance as a similarity metric, the feature vectors of historical samples in the spatiotemporal database are sorted from smallest to largest distance, and similar sample groups are selected.

[0042] The degree-day factor distribution corresponding to the historical sample with the smallest Euclidean distance in the similar sample group is used as the degree-day factor input value for the prediction period;

[0043] The snow melt amount for the forecast period is calculated per pixel based on the positive accumulated temperature for the forecast period and the input value of the degree-day factor.

[0044] The uncertainty of the predicted snowmelt amount is estimated based on the variance of the similar sample groups, thus obtaining the snowmelt uncertainty.

[0045] Preferably, the prediction period feature vector consists of the statistical characteristics of the daily average temperature, accumulated temperature, radiation, wind speed, and relative humidity of the current day and several past days, as well as the land surface type, elevation, slope, and aspect, the spatial feature values ​​extracted by the convolutional neural network, and the mean and variance of the degree-day factor over the past n days.

[0046] A system for retrieving daily factors based on snowmelt equivalent difference includes:

[0047] The data acquisition and preprocessing unit is used to acquire snow water equivalent data for multiple consecutive days in the study area, unify the snow water equivalent data from multi-source remote sensing products, ground observation data and numerical model simulation results with meteorological element data to the same spatial resolution, unify them to a daily scale, and merge them into a standardized input dataset; the meteorological element data includes air temperature, radiation, wind speed and relative humidity;

[0048] The data assimilation and uncertainty calculation unit is used to fuse the multi-source snow water equivalent data in the standardized input dataset using data assimilation technology to obtain assimilated snow water equivalent data and calculate the corresponding uncertainty.

[0049] The snow melting process discrimination unit is used to detect changes in the assimilated snow water equivalent data between two adjacent days. When a decrease occurs, it combines the meteorological conditions of the air temperature, radiation, wind speed and relative humidity, and applies a reliability constraint on the difference in snow water equivalent data based on the uncertainty, in order to determine whether it is a snow melting process.

[0050] The day factor calculation unit is used to calculate the day factor at the pixel scale when it is determined to be snow melting, based on the difference between the assimilated snow water equivalent data of two adjacent days and the positive cumulative temperature difference between the two days.

[0051] The daily life factor database construction unit is used to store the spatial distribution of the daily life factors and the corresponding meteorological elements, underlying surface type, elevation and slope aspect information, and to construct the spatiotemporal database of the daily life factors.

[0052] The feature extraction unit is used to extract the spatial distribution features of the day-of-day factor using a convolutional neural network, and to analyze the time series changes of the spatial distribution features using a long short-term memory network to generate a multidimensional feature vector of the day-of-day factor.

[0053] The similarity matching and snowmelt amount prediction unit is used in the snowmelt forecasting process to retrieve historical samples with the smallest Euclidean distance to the feature vector of the forecast period from the spatiotemporal database based on the meteorological forecast value, underlying surface parameters and topographic information of the forecast period, and obtain the corresponding degree-day factor distribution as the prediction input, calculate the snowmelt amount, and estimate the snowmelt uncertainty based on the variance of similar sample groups.

[0054] The present invention discloses the following technical effects:

[0055] This invention overcomes the limitations of traditional methods, such as fixed daily temperature factors and lack of spatial adaptability, by dynamically retrieving daily temperature factors at the pixel scale using the difference in snow water equivalent between two adjacent days and the positive cumulative temperature difference. It can reflect the impact of different underlying surfaces, topography, and climatic conditions on the snowmelt process. Furthermore, this invention constructs a spatiotemporal database of daily temperature factors that integrates high-resolution snow water equivalent data, meteorological factors, and topographic information, providing rich and reusable reference samples for snowmelt simulation and regional hydrological forecasting.

[0056] Furthermore, this invention introduces deep learning technology, utilizing convolutional neural networks to extract the spatial distribution characteristics of the degree-day factor and combining it with a long short-term memory network to capture its temporal series variation patterns. This achieves accurate modeling of the spatiotemporal evolution of the degree-day factor, improving the accuracy of database construction and the adaptability of predictions. Through multi-source data assimilation, the uncertainty of snow water equivalent data is further estimated and recorded, effectively reducing the impact of remote sensing observation errors on degree-day factor inversion and snowmelt forecasting, and enhancing the reliability of the results.

[0057] This invention also proposes a similarity matching method based on the feature vector of the degree-day factor. In the snowmelt forecasting process, this method can quickly retrieve historical samples with the smallest Euclidean distance to the feature vector of the forecast period, significantly improving the rationality and accuracy of the forecast results. This method has relatively simple data input requirements, strong versatility and flexibility, and can be coupled with various hydrological models. It is applicable to different types of watersheds, such as high-altitude mountainous areas and water conservation areas, and is particularly suitable for hydrological forecasting and snowmelt simulation in snow-dominated watersheds. Therefore, this invention significantly improves the accuracy and reliability of snowmelt forecasting, overcoming the problems of static parameter setting, large errors, and poor applicability of traditional methods, and has broad engineering application value and scientific research significance. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 A flowchart of the method provided in an embodiment of the present invention;

[0060] Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0063] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, this invention provides a method for inverting the degree-day factor based on the difference in snow water equivalent, comprising:

[0064] Step 100: Obtain snow water equivalent data for the study area over several consecutive days, unify the snow water equivalent data from multi-source remote sensing products, ground observation data, and numerical model simulation results with meteorological element data to the same spatial resolution and to a daily scale, and merge them into a standardized input dataset; meteorological element data include air temperature, radiation, wind speed, and relative humidity;

[0065] Step 200: Use data assimilation technology to fuse the multi-source snow water equivalent data in the standardized input dataset to obtain assimilated snow water equivalent data, and calculate the corresponding uncertainty;

[0066] Step 300: Detect the change in snow water equivalent data after assimilation between two adjacent days. When a decrease occurs, combine the meteorological conditions of temperature, radiation, wind speed and relative humidity, and apply reliability constraints to the difference in snow water equivalent data based on uncertainty to determine whether it is a snow melting process.

[0067] Step 400: When it is determined to be snow melting, calculate the degree-day factor at the pixel scale based on the difference in snow water equivalent data after assimilation between two adjacent days and the positive cumulative temperature difference between the two days;

[0068] Step 500: Store the spatial distribution of daily day length factors and the corresponding meteorological elements, underlying surface type, elevation and aspect information to construct a spatiotemporal database of day length factors;

[0069] Step 600: Use a convolutional neural network to extract the spatial distribution features of the degree-day factor, and combine it with a long short-term memory network to analyze the time series changes of the spatial distribution features, and generate a multi-dimensional feature vector of the degree-day factor;

[0070] Step 700: During the snowmelt forecasting process, based on the meteorological forecast values, underlying surface parameters and topographic information of the forecast period, the similarity matching algorithm is used to retrieve the historical samples with the smallest Euclidean distance to the feature vector of the forecast period in the spatiotemporal database, obtain the corresponding degree-day factor distribution as the forecast input, calculate the snowmelt amount, and estimate the snowmelt uncertainty based on the variance of similar sample groups.

[0071] Specifically, this embodiment first obtains a high spatial resolution snow water equivalent dataset for multiple consecutive days within the study area, integrates snow water equivalent information from multiple data sources, including multi-source remote sensing products, ground observation data, and numerical model simulation results, and introduces relevant meteorological element data, such as temperature, radiation, wind speed, and relative humidity.

[0072] This invention obtains multi-day snow water equivalent products and corresponding meteorological driving factor data covering the study area by accessing authoritative data platforms such as the National Tibetan Plateau Scientific Data Center, the National Snow and Ice Data Center, the European Centre for Medium-Range Weather Forecasts, the National Glacier, Permafrost and Desert Scientific Data Center, the Chinese Academy of Sciences Scientific Data Platform, and the China Meteorological Administration. The data sources encompass remote sensing inversion products, ground station measured data, and land surface model simulation results, exhibiting high spatial coverage and temporal continuity.

[0073] To achieve consistent processing of multi-source data, this embodiment first resamples or interpolates all snow water equivalent data and meteorological data to a unified spatial resolution; then, the temporal resolution of each data is unified to a daily scale, finally obtaining a standardized input dataset for subsequent assimilation analysis, which has spatiotemporal consistency and strong comparability.

[0074] Furthermore, this embodiment uses data assimilation technology to fuse multi-source high-resolution snow water equivalent data to obtain assimilated snow water equivalent data, and simultaneously calculates the uncertainty information of snow water equivalent in the database.

[0075] To achieve effective fusion of multi-source high spatial resolution snow water equivalent data, an ensemble Kalman filter method is used for data assimilation to obtain more accurate and consistent snow water equivalent products. The uncertainty of the snow water equivalent at each time point is calculated for result evaluation and reliability control.

[0076] Furthermore, the data assimilation process in this embodiment includes the following operations:

[0077] The construction of the assimilation system includes using the snow water equivalent obtained from numerical simulation as the state variable. Using ground observation data and multi-source remote sensing data as observation variables Observation operators used to map state variables to the observation space Set the observation error covariance matrix based on the accuracy experience of different data sources. System process noise used to represent nondeterministic disturbances in state evolution. ;

[0078] Using ensemble models for prior estimation:

[0079]

[0080] in, For the set of predicted states, This represents the state transition model.

[0081] The updated state vector is calculated using the following formula:

[0082]

[0083] in, For the updated set of states, These are actual observed values. Map the predicted values ​​to the observation space. The Kalman gain matrix is ​​expressed as:

[0084]

[0085] in, To predict the covariance matrix, the process runs in parallel at each pixel in the study area at each time step, achieving a high-resolution snow water equivalent assimilation estimate for the entire region.

[0086] Furthermore, this embodiment detects the change in snow water equivalent between two consecutive days. When a decrease occurs, it comprehensively judges whether it is a snow melting process by combining meteorological conditions such as temperature, radiation, wind speed, and humidity.

[0087] The snow melting process is determined based on the following conditions:

[0088] Determine the daily average temperature on day t. satisfy:

[0089]

[0090] in Threshold temperature;

[0091] Determine the solar radiation on day t. satisfy:

[0092]

[0093] in The preset radiation threshold;

[0094] Determine the wind speed on day t. With humidity satisfy:

[0095]

[0096]

[0097] in and This is an empirical threshold;

[0098] Snow water equivalent of two consecutive days ( Uncertainty if the difference is greater than twice the snow water equivalent:

[0099]

[0100] If all of the above conditions are met, it is determined to be a snow melting process.

[0101] This embodiment performs parallel processing on a pixel-by-pixel basis throughout the entire study area and updates it daily. The thresholds used can be set based on historical meteorological data, literature research, or regional experimental statistics. It can be used to construct a "daily snowmelt area map" or a "snow retreat process sequence" to provide support for snowmelt runoff modeling, water resource assessment, and other tasks.

[0102] Furthermore, when it is determined to be snow melting, the difference in snow water equivalent between two adjacent days and the positive cumulative temperature difference between the two days are used to calculate the degree-day factor at the pixel scale;

[0103] In determining a certain pixel After the snowmelt process occurs, in order to further quantify the snow water equivalent melting rate driven by unit temperature during this process, this invention calculates the degree-day factor at the pixel scale. ). It reflects the reduction in snow water equivalent per unit of positive accumulated temperature and is often used in snowmelt volume modeling and hydrological simulation.

[0104] Only for the first The pixels that were identified as being in the snowmelt process were then processed. The calculation requires input data including: snow water equivalent after assimilation over two consecutive days. , and the corresponding temperature data Daily life factor ( The calculation formula is:

[0105]

[0106] Among them, the positive accumulated temperature between two days The calculation formula is:

[0107]

[0108] in, For the first The average temperature of the celestial phase, This is the threshold temperature.

[0109] Optionally, this embodiment stores the spatial distribution of daily degree factors and corresponding meteorological elements, underlying surface type, elevation, slope aspect and other information to construct a spatiotemporal database of degree factors;

[0110] After completing the calculation of the daily life factor, the two-dimensional spatial distribution of the daily life factor is... As core variables, meteorological elements (such as daily average temperature, radiation, wind speed, and humidity) and underlying surface parameters (such as land cover type, elevation, slope, and aspect) at the corresponding time are stored together in a structured spatiotemporal database of degree-day factors. The database adopts a structure design that supports spatial indexing and temporal retrieval. Through this database, rapid historical backtracking, sample matching, and model training data preparation can be achieved.

[0111] Specifically, this embodiment uses a convolutional neural network to extract the spatial distribution features of the degree-day factor, and combines it with a long short-term memory network to analyze its time series changes, generating a multi-dimensional feature vector of the degree-day factor;

[0112] Convolutional neural networks are used to extract spatial features of the degree-day factor distribution, including spatial texture, local extremum distribution, and edge features, including:

[0113] Input the two-dimensional matrix of daily life factor into a convolutional neural network;

[0114] Spatial feature maps are extracted using several convolutional and pooling layers to obtain a spatial feature matrix. Its definition is:

[0115]

[0116] in, This represents the feature extraction function of a convolutional neural network.

[0117] Long Short-Term Memory (LSTM) networks are used to capture the changing patterns of time-series factors, including weekly, monthly, and seasonal fluctuations. The steps include:

[0118] will sequence Input the Long Short-Term Memory (LSTM) network;

[0119] Obtain the time feature vector Defined as:

[0120]

[0121] in To represent the length of historical days used for modeling time dependencies; This is the state transition function for the Long Short-Term Memory (LSTM) network model.

[0122] Furthermore, during the snowmelt forecasting process, based on the meteorological forecast values, underlying surface parameters, and topographic information within the forecast period, a similarity matching algorithm is used to retrieve historical samples from the database that have the smallest Euclidean distance to the feature vector of the forecast period. The distribution of degree-day factors under the most similar spatiotemporal background is obtained as the degree-day factor input value for the forecast period, thus obtaining the snowmelt amount and uncertainty.

[0123] When forecasting snow melt for future periods, the input variables include statistical characteristics of meteorological elements for the current day and several past days (such as daily average temperature, accumulated temperature, radiation, wind speed, humidity, etc.), underlying surface information (such as surface type, elevation, slope, aspect, etc.), spatial feature values ​​extracted by a convolutional neural network, and the mean and variance of the degree-daily factor over the past n days. Based on the above forecast information, a feature vector for the forecast period is constructed. It then retrieves the most similar historical state from the "historical sample feature database," using a similarity matching algorithm based on feature vectors. With the eigenvector of the prediction period Euclidean distance between :

[0124]

[0125] in Represents the eigenvector of the th element. One dimension; This represents the total dimension of the feature vectors.

[0126] The snow melting calculation is based on the following formula:

[0127]

[0128] in, The lifespan factor of the most similar sample in history; Forecast positive accumulated temperature during the forecast period; This represents the amount of snow melted in that pixel on day t. The uncertainty of snow melt can be estimated by the variance of similar sample groups.

[0129] Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention, such as... Figure 2 As shown, the present invention also provides a system for retrieving daily factors based on snowmelt equivalent difference, comprising:

[0130] The data acquisition and preprocessing unit is used to acquire snow water equivalent data for multiple consecutive days in the study area. It unifies the snow water equivalent data from multi-source remote sensing products, ground observation data, and numerical model simulation results with meteorological element data to the same spatial resolution and to a daily scale, and merges them into a standardized input dataset. The meteorological element data includes air temperature, radiation, wind speed, and relative humidity.

[0131] The data assimilation and uncertainty calculation unit is used to fuse multi-source snow water equivalent data in a standardized input dataset using data assimilation technology to obtain assimilated snow water equivalent data and calculate the corresponding uncertainty.

[0132] The snow melting process discrimination unit is used to detect changes in the assimilated snow water equivalent data between two adjacent days. When a decrease occurs, it combines meteorological conditions such as temperature, radiation, wind speed and relative humidity, and applies reliability constraints to the difference in snow water equivalent data based on uncertainty, in order to determine whether it is a snow melting process.

[0133] The day factor calculation unit is used to calculate the day factor at the pixel scale when it is determined to be snow melting, based on the difference between the assimilated snow water equivalent data of two adjacent days and the positive cumulative temperature difference between the two days.

[0134] The daily life factor database construction unit is used to store the spatial distribution of daily life factors and the corresponding meteorological elements, underlying surface type, elevation and slope aspect information, and to construct a spatiotemporal database of daily life factors.

[0135] The feature extraction unit is used to extract the spatial distribution features of the degree-day factor using a convolutional neural network, and to analyze the time series changes of the spatial distribution features using a long short-term memory network to generate a multi-dimensional feature vector of the degree-day factor.

[0136] The similarity matching and snowmelt amount prediction unit is used in the snowmelt forecasting process to retrieve historical samples with the smallest Euclidean distance to the feature vector of the forecast period from the spatiotemporal database based on the meteorological forecast value, underlying surface parameters and topographic information of the forecast period, and obtain the corresponding degree-day factor distribution as the prediction input, calculate the snowmelt amount, and estimate the snowmelt uncertainty based on the variance of similar sample groups.

[0137] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0138] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for inverting daily degree factors based on snow water equivalent difference, characterized in that, include: Acquire snow water equivalent data for the study area over several consecutive days, and unify the snow water equivalent data from multi-source remote sensing products, ground observation data, and numerical model simulation results with meteorological element data to the same spatial resolution and to a daily scale, merging them into a standardized input dataset; the meteorological element data includes air temperature, radiation, wind speed, and relative humidity; The multi-source snow water equivalent data in the standardized input dataset are fused using data assimilation technology to obtain assimilated snow water equivalent data, and the corresponding uncertainty is calculated. The change in snow water equivalent data after assimilation between two adjacent days is detected. When a decrease occurs, the meteorological conditions of temperature, radiation, wind speed and relative humidity are combined, and the reliability of the difference in snow water equivalent data is constrained according to the uncertainty, so as to determine whether it is a snow melting process. When snowmelt is identified, the degree-day factor at the pixel scale is calculated based on the difference in snow water equivalent data after assimilation between two adjacent days and the positive cumulative temperature difference between the two days. The spatial distribution of the daily time factor and the corresponding meteorological elements, underlying surface type, elevation and slope aspect information are stored to construct a spatiotemporal database of the time factor. A convolutional neural network is used to extract the spatial distribution features of the day-of-day factor, and a long short-term memory network is combined to analyze the time series changes of the spatial distribution features to generate a multidimensional feature vector of the day-of-day factor. In the snowmelt forecasting process, based on the meteorological forecast values, underlying surface parameters and topographic information of the forecast period, a similarity matching algorithm is used to retrieve historical samples with the smallest Euclidean distance between the multidimensional feature vectors corresponding to the feature vectors of the forecast period in the spatiotemporal database, obtain the corresponding degree-day factor distribution as the forecast input, calculate the snowmelt amount, and estimate the snowmelt uncertainty based on the variance of similar sample groups.

2. The method for retrieving daily degree factors based on snow water equivalent difference according to claim 1, characterized in that, Data assimilation techniques are used to fuse multi-source snowmelt equivalent data in the standardized input dataset to obtain assimilated snowmelt equivalent data, and the corresponding uncertainty is calculated, including: The snow water equivalent obtained from the numerical model is set as the state variable, and the ground observation data and multi-source remote sensing data are set as the observation variables. An observation operator is established to map the state variables to the observation space, and the observation error covariance matrix and system process noise are set according to the accuracy experience of different data sources. An ensemble model is used to generate a set of predicted states; The predicted state set is updated based on the Kalman gain to obtain assimilated snow water equivalent data, and the process is carried out in parallel at each pixel point in the study area according to time steps. The uncertainty at the cell level is calculated based on the deviation of observations from each data source relative to the multi-source average snow water equivalent data.

3. The method for retrieving daily degree factors based on snow water equivalent difference according to claim 2, characterized in that, The uncertainty The calculation formula is: ； Where x and y represent the horizontal and vertical coordinates of a spatial location, respectively; t represents a specific day or time. The snow water equivalent data value of the i-th data source; denoted as the average snow water equivalent of a pixel across multiple sources; N represents the number of data sources participating in the fusion.

4. The method for retrieving daily degree factors based on snow water equivalent difference according to claim 1, characterized in that, The system detects changes in the assimilated snow water equivalent data over two consecutive days. When a decrease occurs, it combines meteorological conditions such as air temperature, radiation, wind speed, and relative humidity, and applies reliability constraints to the snow water equivalent data difference based on the uncertainty, to determine whether it is a snowmelt process. This includes: a) Calculate the difference between the assimilated snow water equivalent data on day t and day t-1 at the pixel scale, and proceed to the subsequent discrimination when the difference is negative; b) Determine that the air temperature on day t exceeds a preset temperature threshold; c) Determine that the radiation on day t exceeds a preset radiation threshold; d) Determine that the wind speed and relative humidity on day t satisfy their respective preset threshold conditions; e) Compare the absolute value of the difference with the uncertainty, and determine that the absolute value is greater than twice the uncertainty; When the conditions in steps b) to e) are met simultaneously, it is determined that the snow melting process occurs in the pixel on day t.

5. The method for retrieving daily degree factors based on snow water equivalent difference according to claim 1, characterized in that, When snowmelt is identified, the degree-day factor at the pixel scale is calculated based on the difference in assimilated snow water equivalent data between two adjacent days and the positive cumulative temperature difference between the two days, including: Only for pixels identified as part of the snow melting process, the snow water equivalent difference between day t and day t-1 is calculated based on the assimilated snow water equivalent data. Calculate the positive cumulative temperature difference between two adjacent days; the positive cumulative temperature difference is the cumulative value of the portion of the daily average temperature of the pixel on day t-1 and day t that exceeds the threshold temperature; The degree-day factor of the pixel on day t is obtained by the ratio of the obtained snow water equivalent difference to the positive temperature difference.

6. The method for retrieving daily degree factors based on snow water equivalent difference according to claim 1, characterized in that, A convolutional neural network is used to extract the spatial distribution features of the day-of-day factor, and a long short-term memory network is combined to analyze the time series changes of the spatial distribution features, generating a multi-dimensional feature vector of the day-of-day factor, including: The two-dimensional matrix of the daily life factor is used as input to the convolutional neural network; Several convolutional and pooling layers are used to extract spatial feature maps, resulting in a spatial feature matrix that characterizes spatial texture, local extremum distribution, and edge features. The spatial feature matrix arranged in chronological order, with the historical days length as a preset value, is input into the long short-term memory network to capture the weekly, monthly, and seasonal changes of the degree-day factor in the time series, thus obtaining a time feature vector. The multidimensional feature vector of the degree-day factor is generated based on the spatial feature matrix and the temporal feature vector.

7. The method for retrieving degree-day factors based on snow water equivalent difference according to claim 1, characterized in that, In the snowmelt forecasting process, based on the meteorological forecast values, underlying surface parameters, and topographic information for the forecast period, a similarity matching algorithm is used to retrieve historical samples from the spatiotemporal database that have the smallest Euclidean distance between the multidimensional feature vectors corresponding to the feature vectors of the forecast period. The corresponding degree-day factor distribution is then obtained as the forecast input to calculate the snowmelt amount. Finally, the snowmelt uncertainty is estimated based on the variance of the similar sample groups, including: Based on the meteorological forecasts, underlying surface parameters, and topographic information for the forecast period, a feature vector for the forecast period is constructed. Using Euclidean distance as a similarity metric, the multidimensional feature vectors corresponding to historical samples in the spatiotemporal database are sorted from smallest to largest distance, and similar sample groups are selected. The degree-day factor distribution corresponding to the historical sample with the smallest Euclidean distance in the similar sample group is used as the degree-day factor input value for the prediction period; The snow melt amount for the forecast period is calculated per pixel based on the positive accumulated temperature for the forecast period and the input value of the degree-day factor. The uncertainty of the predicted snowmelt amount is estimated based on the variance of the similar sample groups, thus obtaining the snowmelt uncertainty.

8. The method for retrieving daily degree factors based on snow water equivalent difference according to claim 7, characterized in that, The prediction period feature vector consists of the statistical characteristics of the daily average temperature, accumulated temperature, radiation, wind speed, and relative humidity of the current day and several past days, as well as the land surface type, elevation, slope, and aspect, the spatial feature values ​​extracted by the convolutional neural network, and the mean and variance of the degree-day factor over the past n days.

9. A system for retrieving daily degree factors based on snow water equivalent difference, characterized in that, include: The data acquisition and preprocessing unit is used to acquire snow water equivalent data for multiple consecutive days in the study area, unify the snow water equivalent data from multi-source remote sensing products, ground observation data and numerical model simulation results with meteorological element data to the same spatial resolution, unify them to a daily scale, and merge them into a standardized input dataset; the meteorological element data includes air temperature, radiation, wind speed and relative humidity; The data assimilation and uncertainty calculation unit is used to fuse the multi-source snow water equivalent data in the standardized input dataset using data assimilation technology to obtain assimilated snow water equivalent data and calculate the corresponding uncertainty. The snow melting process discrimination unit is used to detect changes in the assimilated snow water equivalent data between two adjacent days. When a decrease occurs, it combines the meteorological conditions of the air temperature, radiation, wind speed and relative humidity, and applies a reliability constraint on the difference in snow water equivalent data based on the uncertainty, in order to determine whether it is a snow melting process. The day factor calculation unit is used to calculate the day factor at the pixel scale when it is determined to be snow melting, based on the difference between the assimilated snow water equivalent data of two adjacent days and the positive cumulative temperature difference between the two days. The daily life factor database construction unit is used to store the spatial distribution of the daily life factors and the corresponding meteorological elements, underlying surface type, elevation and slope aspect information, and to construct the spatiotemporal database of the daily life factors. The feature extraction unit is used to extract the spatial distribution features of the day-of-day factor using a convolutional neural network, and to analyze the time series changes of the spatial distribution features using a long short-term memory network to generate a multidimensional feature vector of the day-of-day factor. The similarity matching and snowmelt amount prediction unit is used in the snowmelt forecasting process to retrieve historical samples with the smallest Euclidean distance between the multidimensional feature vectors corresponding to the feature vectors of the forecast period and the meteorological forecast values, underlying surface parameters and topographic information of the forecast period using a similarity matching algorithm. The corresponding degree-day factor distribution is obtained as the prediction input, the snowmelt amount is calculated, and the snowmelt uncertainty is estimated based on the variance of the similar sample group.

Images