Snowfall process snow depth prediction system and prediction method based on snow remote sensing and meteorological intelligent grid fusion

Abstract

The invention relates to the technical field of snow depth prediction, and discloses a snowfall process snow depth prediction method based on accumulated snow remote sensing and meteorological intelligent grid fusion, which comprises the following steps: constructing rain and snow phase state separation schemes under three time scales by using a dynamic threshold parameterization scheme and a frequency threshold scheme; analyzing a relationship between meteorological factors such as snow depth and precipitation in a snowfall process by using ground observation data so as to construct a water-snow quantitative conversion relationship model; remote sensing monitoring snow depth data fused by H8, MODISFSC and CheAMSR2 is used as a current snow distribution situation, meteorological factors such as meteorological intelligent grid weather forecast precipitation and air temperature are used as a future scene, and a'water-snow 'quantitative conversion relation mode is used for predicting the depth of accumulated snow which is gridded every 3 hours in the future snowfall process. Quantitative conversion between precipitation and actual ground snow depth in a point-to-surface weather forecast product is realized, the forecast accuracy of the snow depth is improved, and a basis is provided for high-temporal-spatial-precision quantitative snow depth forecast service on a regional surface.

Description

Technical Field

[0001] The present invention relates to the technical field of snow depth prediction, and in particular to a snow depth prediction system and method for a snowfall process based on the fusion of snow accumulation remote sensing and meteorological intelligent grid. Background Art

[0002] Snow remote sensing and weather forecast services are the basis for snow depth prediction. In previous snow depth forecasts, (1) there was a lack of a sophisticated rain-snow phase separation scheme. The snowfall phase identification threshold currently used in weather forecasts is an empirical value. Studies have shown that this empirical value is not fixed and may change with environmental conditions such as terrain, which can lead to incorrect identification of snowfall phases. (2) There is a lack of support for research on the mechanism of snowfall and snow accumulation processes. Simple empirical models are usually used to predict snow depth, and this empirical value may be different for different periods and regions. (3) The current status of snow depth is ignored. The current snow depth forecast product predicts the snow depth as an incremental forecast of the snow depth after precipitation occurs, but ignores the ground conditions before snow accumulation occurs, resulting in the predicted snow depth product being unable to well reflect the actual situation of snow depth on the ground. (4) The temporal resolution of the snow depth forecast product is relatively coarse. The snow depth forecast products released at present are usually for the entire precipitation period or daily precipitation conditions in the future, and there are no more refined forecast products, such as 1- or 3-hour rolling snow depth forecasts. Through refined snow depth forecasts, effective services can be provided for disaster prevention and mitigation. With the advancement of technical means, in the current meteorological services, the meteorological intelligent grid provides high-temporal resolution grid meteorological element forecast products for snow depth prediction, and provides data support for future scenarios for forecast products. Snow remote sensing can well solve the current status of snow monitoring on the surface. How to effectively and accurately integrate snow remote sensing and meteorological service products to achieve the connection problem from grid to surface snow depth forecast products requires the integration of meteorological forecast service products and snow remote sensing technology to achieve refined snow depth forecasts, so as to apply them to meteorological services such as snow monitoring.

[0003] In the existing rain and snow phase separation scheme, the model estimation of precipitation phase state is an important means to estimate precipitation phase state by simple empirical or semi-empirical schemes. At present, most of the precipitation phase identification schemes are based on high-altitude and ground meteorological conditions, and the method of coupling high-altitude data and ground meteorological data is used to capture the relationship between each precipitation phase and the temperature, potential thickness and 2m ground meteorological conditions at different altitudes in the atmosphere, as an important indicator for judging precipitation phase state. However, due to the small number of sounding observation sites, high-altitude data is limited and difficult to obtain, especially in high-altitude areas, where there are few observation stations and lack of data, the observation time, space, height and content of meteorological elements are very limited. Therefore, the application and promotion of this method is relatively difficult; the rain and snow identification scheme established using ground meteorological observation conditions is the most widely used one based on ground meteorological temperature thresholds, based on mathematical statistics schemes, using the frequency of occurrence of different precipitation phases, and statistically calculating the ground meteorological condition thresholds when precipitation occurs. Establish the critical temperature value for the transformation of rain / snow precipitation forms, and determine the temperature value corresponding to the point with the same probability of rain / snow occurrence as the critical temperature for the transformation of precipitation phase. The input data of this type of method is relatively easy to obtain, so it is widely used; including single critical and double critical temperature threshold methods. The single critical threshold method sets a critical temperature value to achieve the separation of solid and liquid precipitation, but it is difficult to distinguish the mixed phase of solid / liquid (sleet). Studies have shown that the single critical temperature method is not applicable to high-altitude areas; the double critical threshold method, that is, when the temperature is higher than the upper critical value, it is rain, and when it is lower than the lower critical temperature, it is snow. When it is between the two critical temperature values, it is calculated by linear interpolation; the double critical value threshold method can better distinguish the transition phases of rain and snow, but the critical temperature values ​​in different regions are different, and the fixed temperature threshold cannot identify the precipitation phase well.

[0004] The "water-snow" conversion relationship between precipitation and snow depth reflects the conversion of precipitation settling into snow depth, and its process also indicates its relationship with the meteorological environment. Snow depth and precipitation are two completely different concepts. The measured value of the rain gauge cannot represent the actual precipitation on the ground. When precipitation is converted into snow accumulation, it sublimates and melts. Different environmental temperature and humidity conditions affect the formation of ice and snow crystal shapes. In the process of ice and snow crystals colliding and aggregating to form snowflakes, air temperature and ice and snow crystal shapes play a leading role. The shape of snowfall depends on the air temperature and pressure near the ground. However, almost all studies have not explained the relationship between precipitation and snow depth. The research on long-term snowfall mainly focuses on the analysis of point data series of meteorological stations. Land surface models use snowfall density to convert snowfall into ground snow depth, but generally use a fixed ratio as a parameter, lacking the quantitative expression of meteorological fields in different areas of the actual ground. Therefore, due to the lack of more in-depth and comprehensive observations of high-frequency dynamic processes, there are few studies on the quantitative relationship between the two. Up to now, there is still no clear indicator system for the correspondence between the "water-snow" conversion process and the meteorological field during the same period. Summary of the invention

[0005] The present invention provides a snow depth prediction system and method for snowfall process based on the fusion of snow remote sensing and meteorological intelligent grid. The system combines the snow depth data of satellite remote sensing monitoring fused by H-8, MODIS_FSC and Che_AMSR2 as the current status of snow distribution, and uses meteorological elements such as precipitation and temperature predicted by meteorological intelligent grid weather forecast as future scenarios. The system uses a "water-snow" quantitative conversion relationship model to predict the 3-hourly gridded snow depth of future snowfall processes, and realizes the quantitative conversion of precipitation and actual ground snow depth in weather forecast products from point to surface, thereby improving the forecast accuracy of snow depth and providing a basis for quantitative snow depth forecast services with high temporal and spatial accuracy on regional surfaces.

[0006] The present invention provides a snow depth prediction method for snowfall process based on snow remote sensing and meteorological intelligent grid fusion, comprising:

[0007] Select precipitation phases and ground meteorological observation data at three time scales within a set time period, and construct a dynamic threshold parameterization scheme and a frequency threshold scheme to obtain two rain and snow phase separation models at three time scales; wherein the three time scales are minutes, hours and days;

[0008] Using data from field observation sites and national meteorological sites, the relationship between snow depth, precipitation and meteorological factors during snowfall is analyzed to construct a "water-snow" quantitative conversion relationship model; wherein the "water-snow" quantitative conversion relationship model includes a single-station model, a regional model and a provincial model;

[0009] The real-time data of snow depth inverted from different data sources are obtained by using H-8, microwave and optical fusion data;

[0010] The snow range and snow depth data monitored by snow remote sensing are used as the current status of snow distribution, and the meteorological elements of precipitation and temperature from the meteorological intelligent grid numerical weather forecast are used as the future scenario. The "water-snow" quantitative conversion relationship model is used to predict the snow depth at a 5km×5km grid every 3 hours during the future snowfall process. The rain-snow phase separation model is combined to achieve dynamic snow depth prediction.

[0011] Furthermore, the step of selecting precipitation phases at three time scales and ground meteorological observation data within a set time period, and constructing a dynamic threshold parameterization scheme and a frequency threshold scheme to obtain two rain and snow phase separation models at three time scales includes:

[0012] Select three time scale precipitation phases and ground meteorological observation data within a set time period;

[0013] Based on the wet-bulb temperature, the response relationship between altitude, relative humidity and different types of precipitation events in different wet-bulb temperature ranges is determined, and a dynamic threshold parameterization scheme is constructed;

[0014] Based on air temperature, the probability of different types of precipitation events occurring in different air temperature ranges is statistically analyzed, and the frequency threshold scheme is constructed using the frequency intersection value;

[0015] The parameterization scheme and frequency threshold scheme are extracted as two rain-snow phase separation models at three time scales.

[0016] Furthermore, the step of determining the response relationship between altitude, relative humidity and different types of precipitation events in different wet-bulb temperature ranges based on the wet-bulb temperature and constructing a dynamic threshold parameterization scheme includes:

[0017] The relative humidity and altitude of different precipitation phases under three time scales are divided into five intervals, and the relative humidity and altitude of each 0.2℃T under three time scales are calculated. W The ratio of the number of different precipitation phases to the total number of occurrences in the five relative humidity and altitude intervals within the interval; among them, the wet bulb temperature T W The calculation formula is:

[0018]

[0019] Among them, e s is the saturated water vapor pressure (hPa); T is the air temperature (℃); T W is the wet bulb temperature (℃); RH is the relative humidity (0-1); P is the atmospheric pressure (hPa);

[0020]

[0021] Among them, T0 represents the T of snowfall-rainfall conversion. W The value at the center of the interval, 2ΔT represents the main temperature range for sleet, ΔS is a parameter for adjusting the temperature scale, and as ΔS increases, the T at which snowfall-rainfall conversion occurs w The interval becomes wider, in which the parameterization schemes of T0, ΔT, and ΔS are calibrated based on the relationship between precipitation type, altitude, and relative humidity;

[0022] After analyzing the relationship between the occurrence frequency of different precipitation phases and the influence of relative humidity and altitude, the variation curves of the three parameters T0, ΔT, and ΔS in the parameterization scheme at the minute, hour, and day scales with relative humidity and altitude were fitted to obtain the fitting formulas of the three parameters at the three time scales respectively:

[0023] minute:

[0024] ΔT=0.0004×RH×RH-0.064×RH+3.2599

[0025] T0=0.1015×RH-9.8847

[0026] ΔS=-0.0073×RH+0.9328

[0027] Hour:

[0028] ΔT=0.0073×RH+0.2205

[0029] T0=0.1041×RH-9.9859

[0030] ΔS=0.0004×RH×RH-0.068×RH+3.1837

[0031] day:

[0032] ΔT=0.0012×RH×RH-0.1558×RH+5.8122

[0033] T0=0.142×RH-12.866

[0034] ΔS=-0.0111×RH+1.4138

[0035] The calculation formula for precipitation type is:

[0036]

[0037] Where type is the precipitation type, including rain, snow, and sleet; T min and T max These are the Snow-Sleet and Rain-Sleet temperature thresholds, respectively.

[0038] Furthermore, the step of statistically analyzing the probability of occurrence of different precipitation phases in different air temperature ranges based on air temperature and constructing a frequency threshold scheme using frequency intersection values ​​includes:

[0039] The corresponding air temperature value when the frequency of rainfall, snowfall and sleet is equal can be used to obtain the frequency intersection value of the three cases of "rain-snow", "snow-sleet" and "rain-sleet", which is the frequency threshold;

[0040] The frequency thresholds obtained in the two cases of "rain-sleet" and "snow-sleet" within the temperature range of the set step are statistically used as the precipitation phase separation threshold, that is, when T>"rain-sleet" frequency threshold, it is judged as "rain", when T<"snow-sleet" frequency threshold, it is judged as "snow", and when it is between the two, it is judged as "sleet".

[0041] Furthermore, the step of using field observation site data and national meteorological site observation data to analyze the relationship between snow depth, precipitation, and meteorological factors during snowfall to construct a "water-snow" quantitative conversion relationship model includes:

[0042] The high-frequency field observation data of a complete snow season of the comprehensive snow observation experiment were obtained, and the outliers were filtered out. The data quality was controlled, and the relationship between snow depth and precipitation and meteorological factors during the snowfall process was analyzed, as well as the internal connection and confidence level of this relationship with different snow accumulation processes, different time periods, air pressure, and precipitation. The "water-snow" quantitative conversion relationship model based on the observation experiment was obtained, which is as follows:

[0043] Single-station "water-snow" conversion model: According to different seasons, establish autumn, winter and spring "water-snow" conversion models for a set number of meteorological stations, establish a model with "precipitation, temperature, wind direction and wind speed" meteorological elements as the dominant factors, and reserve data to test the model;

[0044] Regional "water-snow" conversion model: The set province and city are divided into three regions, namely the southern Qinghai region, the northern region and the eastern region. Different regional models are established using corresponding regional stations, and data are reserved to test the models. According to different seasons, the meteorological station data in the three regions are merged and processed, and the regional "water-snow" conversion models for autumn, winter and spring are established;

[0045] In southern Qinghai, the autumn model is SD = -1.068 + 0.767 × PRE_12 - 0.202 × TEM + 0.418 × PRE_24 - 0.26 × PRE_6, the winter model is SD = -0.011 + 0.579 × PRE_6 + 0.188 × PRE_12 - 0.051 × WIN_S, and the spring model is SD = 10.172 + 0.483 × PRE_12 - 0.111 × TEM + 0.288 × PRE_6 - 0.001 × ALT + (-4.576E-07) ×

[0046] WIN_D - 0.064 × longitude;

[0047] In the northern region, the autumn model is D = -2.224 + 0.486 × PRE_6 + 0.483 × WIN_S + 0.437 × PRE_24 + 0.765 × PRE_3, the winter model is SD = 7.519 + 0.59 × PRE_24 - 0.648 × PRE_1 - 0.189 × latitude, and the spring model is SD = -0.323 + 0.915 × PRE_6 + (2.077E-06) × WIN_D;

[0048] In the eastern region, the autumn model was SD = -0.246 + 1.088 × PRE_6 - 0.201 × PRE_12, the winter model was SD = 36.137 + 0.565 × PRE_24 - 0.351 × longitude, and the spring model was SD = -0.264 + 1.341 × PRE_6;

[0049] The provincial "water-snow" conversion model: According to different seasons, the autumn, winter and spring "water-snow" conversion models are established for the whole province, and data are reserved to test the models; among them, the autumn model is SD = -0.5 + 0.890 × PRE_6-0.032 × TEM, the winter model is SD = 0.012 + 0.515 × PRE_6 + 0.221 × PRE_12-0.046 × WIN_S, and the spring model is SD = 3.959 + 0.527 × PRE_6 + 0.343 × PRE_12-0.101 × TEM-0.001 × ALT-0.068 × latitude.

[0050] Furthermore, the step of using H-8, microwave and optical fusion data to obtain real-time snow depth data inverted from different data sources includes:

[0051] Acquire H-8 data on the information center platform and perform data preprocessing on the H-8 data; wherein the data preprocessing includes radiation calibration, data splicing, data projection conversion and geometric precision correction based on control points;

[0052] The Che_AMSR2 daily microwave snow depth data and the MODIS daily cloudless snow cover dataset (MODIS_FSC) in the Asian Water Tower region are used to fuse the two through fusion rules to invert the daily snow depth data.

[0053] Furthermore, the step of obtaining H-8 data on the information center platform and performing data preprocessing on the H-8 data includes:

[0054] Using the relationship between snow depth and NDSI index, a snow depth inversion model was established to obtain the snow depth data inverted from H-8 data. The NDSI index is calculated based on the combination of MODIS band 4 0.5 μm and band 6 1.64 μm. The formula is as follows:

[0055] NDSI=(R4-R6) / (R4+R6)

[0056] Among them, NDSI is the normalized difference snow index, R4 is the reflectivity of the 4th band, and R6 is the reflectivity of the 6th band. Corresponding to the band parameters of H-8, the NDSI index is transformed as follows:

[0057] NDSI=(R2-R5) / (R2+R5)

[0058] Among them, R2 is the reflectivity of the second band in H-8; R5 is the reflectivity of the fifth band in H-8;

[0059] Cloud recognition algorithm:

[0060]

[0061] Among them, B 4th is 20, B 2th is 20, B 5th 14,B 6th is 0.12.

[0062] Snow Cover Extent Extraction:

[0063]

[0064] B2>B 2th

[0065] B5>B 5th

[0066] Among them, NDSIth is 0.37, B 2th For 11, B 5th is 10;

[0067] Snow depth calculation model:

[0068] By establishing the relationship between snow depth and NDSI index, the following snow depth calculation model is obtained: SnowDepth = 32.186 × NDSI-9.42672;

[0069] After the H-8 data is preprocessed through reception, image stitching, radiation calibration, projection conversion, and spatial position correction, the normalized difference snow index NDSI is calculated. When NDSI>0.37, R2>0.11, and R5>0.1 are met at the same time, the pixel is judged to be snow accumulation; when R4>0.2, R2>0.2, and R5>0.14 are met at the same time, the pixel is judged to be cloud; the snow depth is inverted using the snow depth inversion algorithm.

[0070] Furthermore, the step of using Che_AMSR2 daily microwave snow depth data and the Asian Water Tower region MODIS daily cloudless snow cover dataset (MODIS_FSC) to fuse the two through a fusion rule to invert the daily snow depth real-time data includes:

[0071] The fusion rule matches and fuses MODIS_FSC data and Che_AMSR2 products in space, so that the snow depth value is redistributed according to the FSC value of its corresponding sub-pixel. The fused pixel includes a snow depth layer and a FSC layer. Each 25km Che product pixel corresponds to 50×50 500m FSC sub-pixels. The fusion rule includes two basic assumptions:

[0072] 1) The snow depth value of the low-resolution passive microwave pixel represents the average snow depth in the pixel area;

[0073] 2) The snow depth of a passive microwave pixel can be linearly redistributed according to the snow area size of its sub-pixel;

[0074] The snow depth of each sub-pixel can be calculated based on the snow depth value of the passive microwave pixel and the snow area ratio value of each sub-pixel. The passive microwave snow depth data is downscaled to a spatial resolution of 0.05° using the MODIS snow area. First, the MODIS snow area ratio data after cloud removal, i.e. 500m, is about 0.005°. The projection is converted into a latitude and longitude projection consistent with the passive microwave snow depth data. The pixel size is resampled to 0.05°. The calculation formula for the snow depth value of each 0.05° sub-pixel is:

[0075]

[0076] Among them, SD i is the snow depth value of the ith sub-pixel in the passive microwave pixel grid; FSC i is the snow area ratio of the i-th sub-pixel; SD grid is the snow depth value of the passive microwave pixel; n is the total number of sub-pixels corresponding to the passive microwave pixel.

[0077] Furthermore, the fusion algorithm also includes:

[0078] 1) The snow depth value of the passive microwave pixel is greater than 0, that is, SD grid >0, but the snow area values ​​of all sub-images monitored by MODIS are 0, that is, In this case, the snow depth values ​​of all sub-pixels were assigned to 0 based on the MODIS snow cover area judgment;

[0079] 2) The snow area ratio of the sub-pixel is greater than 0, that is, FSC i > 0, but the snow depth value of the passive microwave pixel is equal to 0, that is, SD grid =0; In this case, the snow attenuation curve function is used to convert the snow area ratio value of the sub-pixel into a snow depth value. The snow attenuation curve function is: SD_B tb =0.2912e 0.0388FSC .

[0080] The present invention also provides a snow depth prediction system for snowfall process based on the fusion of snow remote sensing and meteorological intelligent grid, comprising:

[0081] A construction module is used to select precipitation phases and ground meteorological observation data at three time scales within a set time period, and to construct a dynamic threshold parameterization scheme and a frequency threshold scheme to obtain two rain and snow phase separation models at three time scales; wherein the three time scales are minutes, hours and days;

[0082] An analysis module is used to analyze the relationship between snow depth, precipitation and meteorological factors during snowfall using data from field observation sites and national meteorological sites, so as to construct a "water-snow" quantitative conversion relationship model; wherein the "water-snow" quantitative conversion relationship model includes a single-station model, a regional model and a provincial model;

[0083] Inversion module, used to obtain the real-time snow depth data inverted from different data sources using H-8, microwave and optical fusion data;

[0084] The prediction module is used to use the snow remote sensing monitoring of snow range and snow depth data as the current status of snow distribution, and the meteorological elements of precipitation and temperature of the meteorological intelligent grid numerical weather forecast as the future scenario. The "water-snow" quantitative conversion relationship model is used to predict the snow depth at a 5km×5km grid every 3 hours during the future snowfall process, and the rain-snow phase separation model is combined to realize dynamic snow depth prediction.

[0085] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0086] The present invention also provides a computer-readable storage medium on which a computer program is stored, and the computer program implements the steps of the above method when executed by a processor.

[0087] The beneficial effects of the present invention are:

[0088] 1. The present invention adopts rain-snow phase separation models of different scales to realize snowfall forecasts at different time and space scales; based on a new observation method, a more refined indicator system is obtained to realize refined snow monitoring in time and space, which will not only play an important social benefit in the field of ecological environment protection and construction engineering, but will also produce significant economic benefits in animal husbandry production and disaster prevention and mitigation.

[0089] 2. Construct a rain / snow phase separation scheme to achieve rainfall / snow phase forecasts on different time scales of minutes, hours and days, and provide support for weather forecasts on different time scales. The minute-scale precipitation phase separation model significantly improves the accuracy and practicality of the forecast by providing higher temporal and spatial resolution in meteorological forecasts. The minute-scale precipitation phase separation model can provide extremely high temporal resolution, allowing meteorological forecasts to capture weather changes in a short period of time. By refining the time scale, these models can more accurately predict the start and end time of precipitation. On the basis of high temporal resolution, the minute-scale precipitation phase separation model can usually also provide more detailed spatial distribution information, helping meteorologists understand the meteorological characteristics of a specific area.

[0090] 3. Construct the "water-snow" conversion relationship, combine the snow area and snow depth monitoring status of satellite remote sensing inversion using H-8, microwave and optical fusion, integrate the numerical weather forecast products of meteorological intelligent grid, and realize the quantitative conversion of precipitation and actual ground snow depth in the 3-hour regional surface weather forecast products, improve the forecast accuracy of snow depth, and provide a basis for snow depth weather forecast services.

[0091] 4. The conversion between snow and water can help us understand the occurrence and development of snow conditions from a mechanistic perspective. At present, the conversion between snow and water mainly focuses on the conversion from snow to water, which is called "snow water equivalent", while the conversion from water to snow is insufficient, especially for plateau areas. This invention is unique in terms of research area, research object and method targeting.

[0092] 5. The actual snow monitoring situation inverted from different data sources can realize the dynamic monitoring of the "snowfall-snow accumulation" process, and provide data basis for timely grasping the snow disaster situation and formulating reasonable disaster relief plans. It provides a quantitative basis for estimating whether supplementary feeding is needed and how much supplementary feeding is needed. The present invention will provide important support for the improvement of production economic benefits in industries such as animal husbandry and their management decisions to seek benefits and avoid harm, thereby generating considerable economic benefits. When facing snow disasters, improving the pertinence of disaster relief and disaster resistance, improving the accuracy of disaster degree estimation, improving the effectiveness of disaster resistance preparations, and reducing the waste of reserve feed will generate direct economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] Figure 1 The figure is a schematic diagram of a method flow according to an embodiment of the present invention.

[0094] Figure 2 This is a technical flow chart of the inversion of actual snow depth products in the present invention.

[0095] Figure 3It is a schematic diagram of the ratio of the number of occurrences of different precipitation phases in the relative humidity and altitude ranges to the total number of occurrences on the scales of minutes (a)-(j), hours (a')-(j') and days (a'')-(j'') in the present invention.

[0096] Figure 4 It is a schematic diagram of wet-bulb temperature parameter calculation in the present invention.

[0097] Figure 5 Schematic diagrams of D and H are shown in Figures 1 to 3. The relationship between T0, ΔT and ΔS and relative humidity (RH) in each relative humidity (RH) zone in the present invention (a)-(c), and the relationship between T0, ΔT and ΔS and average altitude (ALT) in each altitude (ALT) zone (d)-(f).

[0098] Figure 6 Schematic diagram of the frequency threshold method in the present invention.

[0099] Figure 7 Schematic diagram of cross-validation results of minute-, hour-, and day-scale models in different separation schemes of the present invention.

[0100] Figure 8 This is a comparison chart of the simulated values ​​and measured values ​​of snow depth of the three models in the present invention.

[0101] Fig. 9 This is a comparison chart of the simulated values ​​and measured values ​​of snow depth in the provincial and regional models in the present invention.

[0102] Fig.10 Schematic diagram of the H-8 data preprocessing process in the present invention.

[0103] Fig.11 It is a schematic diagram of the snow depth reconstruction technology process in the present invention.

[0104] Fig.12 This is a schematic diagram of the fusion of MODIS / passive microwave snow products in the present invention.

[0105] Fig.13 This is a flow chart of the H-8 satellite remote sensing inversion of the daily maximum snow depth in the present invention.

[0106] Fig.14 FIG. 1 is a schematic diagram of a device structure according to an embodiment of the present invention.

[0107] Fig.15 The figure is a schematic diagram of the internal structure of a computer device according to an embodiment of the present invention.

[0108] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0109] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0110] The present invention constructs a rain, snow and sleet phase separation scheme suitable for the daily, hourly and minute time scales of the region, and better realizes the snowfall phase identification on different time scales; combines the satellite remote sensing monitoring snow depth data fused by H-8, MODIS_FSC and Che_AMSR2 as the current status of snow distribution, takes the meteorological elements such as precipitation and temperature of the meteorological intelligent grid weather forecast as the future scenario, and uses the "water-snow" quantitative conversion relationship model to predict the 3-hour gridded snow depth of the future snowfall process, so as to realize the quantitative conversion of precipitation and actual ground snow depth in the point-to-surface weather forecast products, improve the forecast accuracy of snow depth, and provide a basis for the quantitative snow depth forecast service with high temporal and spatial accuracy on the regional surface.

[0111] like Figure 1 As shown, the present invention provides a snow depth prediction method for snowfall process based on the fusion of snow remote sensing and meteorological intelligent grid, comprising:

[0112] S1. Construction of rain-snow phase separation schemes at minute, hour and day time scales: Select precipitation phases at three time scales and ground meteorological observation data within a set time period, and construct a dynamic threshold parameterization scheme and a frequency threshold scheme to obtain two rain-snow phase separation models at three time scales.

[0113] Based on precipitation phase observation data and combined with rainfall / snow ground meteorological conditions, the rain / snow phase identification model under different time scales is studied; for example, precipitation phase and ground meteorological observation data at three time scales of minutes, hours and days in spring (March-May) and autumn (September-November) at 18 stations in the Sanjiangyuan area in southern Qinghai from 2012 to 2021 were selected. Based on wet-bulb temperature, the response relationship between altitude, relative humidity and different types of precipitation events in different wet-bulb temperature ranges was explored, and a dynamic threshold parameterization scheme was constructed; based on air temperature, the probability of occurrence of different precipitation phases in different air temperature ranges was statistically analyzed, and the frequency intersection value was used to construct the frequency threshold method. Finally, two rain and snow phase separation models at different time scales (minutes, hours and days) in the Sanjiangyuan area were obtained.

[0114] S2. Utilize the data from field observation stations and national meteorological stations to analyze the relationship between snow depth, precipitation and meteorological factors during snowfall, so as to construct a "water-snow" quantitative conversion relationship model; wherein, the "water-snow" quantitative conversion relationship model includes a single-station model, a regional model and a provincial model.

[0115] (1) The "water-snow" conversion relationship between precipitation and snow depth reflects the conversion of precipitation into snow depth, and its process also indicates its relationship with the meteorological environment. Using the data from field observation stations, the relationship between snow depth and ground meteorological factors such as precipitation and temperature was preliminarily analyzed. For example, on this basis, the observation data of 22 national meteorological stations in Qinghai Province were used to establish a regression model of hourly snow depth and meteorological factors using the stepwise regression method. During the study period, high-frequency field observation data of a complete snow season of the comprehensive snow observation experiment were obtained, and the wild values ​​were filtered out. The data quality was controlled to reduce errors. The relationship between snow depth and precipitation and meteorological factors during the snowfall process was analyzed, and the internal connection and confidence level of this relationship with different snow accumulation processes, different time periods, air pressure, and precipitation were obtained. The "water-snow" conversion comprehensive model based on the full-factor observation experiment was obtained.

[0116] Single-station model: According to different seasons, single-station autumn, winter and spring "water-snow" conversion models were established for 22 meteorological stations in the study area. Considering the business application of the model and the meteorological elements currently forecasted in numerical weather forecasts, a model with "precipitation, temperature, wind direction, wind speed" meteorological elements as the leading factors was established. In order to test the model, some snow accumulation process data of the corresponding season was reserved as verification data when the model was established, and the model accuracy was better evaluated through the correlation test level and RMSE.

[0117] Regional model: Due to the regional differences in precipitation, considering the business promotion and application, it is difficult to effectively promote the "water-snow" conversion model for each of the 22 meteorological stations in the province in business operations. According to the model analysis of the 22 meteorological stations, it can be seen that the sensitivity factors corresponding to the models in different geographical locations are different. For example, Qinghai Province is roughly divided into three regions, namely the northern region of the province, the southern region of Qinghai, and the eastern region. The southern region of Qinghai includes: Dari, Gande, Henan, Jiuzhi, Maduo, Maqin, Nangqian, Qingshuihe, Qumalai, Tuotuohe, Wudaoliang, Zaduo, Zeku, and Zhiduo; the northern region of the province includes: Chaka, Qilian, Toler, and Yeniugou; the eastern region of the province includes: Datong, Hualong, Ledu, Guide, and Xunhua. Different regional models are established using the corresponding regional stations, and data are reserved for verification. According to different seasons, the meteorological station data in the three regions are merged and processed, and the "water-snow" conversion models for autumn, winter, and spring are established. The model establishment and model evaluation methods are consistent with the site model.

[0118] Provincial model: According to different seasons, the autumn, winter and spring "water-snow" conversion models are established for the whole province. The model establishment and model evaluation methods are consistent with the site model.

[0119] (2) Model verification:

[0120] Typical snowfall processes in different seasons were selected to test the provincial, regional and single-station models. The test results show that both the regional model and the provincial model can better describe the occurrence and development process of snow depth.

[0121] S3. Use H-8, microwave and optical fusion data to obtain real-time snow depth data inverted from different data sources.

[0122] like Figure 2 As shown in the figure, the real-time snow depth images inverted from different data sources are obtained by using H-8, microwave and optical fusion data. After downloading H-8 data from the information center platform and completing preprocessing such as positioning and calibration, the relationship between snow depth and snow accumulation index (NDSI) is used to establish a snow depth inversion model, and the snow depth data inverted from H-8 data is obtained. In order to reduce cloud pollution, the maximum snow depth data with a frequency of 10 minutes from 11:00 to 14:00 is synthesized as the real-time data of daily snow depth; the Che_AMSR2 daily microwave snow depth data and the MODIS daily cloudless snow cover dataset (2000-2022) (MODIS_FSC) in the Asian Water Tower region are used, and the two are fused through fusion rules to invert the real-time data of daily snow depth.

[0123] S4. The snow range and snow depth data monitored by remote sensing are used as the current status of snow distribution, and the meteorological elements of precipitation and temperature from the meteorological intelligent grid numerical weather forecast are used as future scenarios. The "water-snow" quantitative conversion relationship model is used to predict the snow depth at a 5km×5km grid every 3 hours during future snowfall processes. The rain-snow phase separation model is combined to achieve dynamic snow depth prediction.

[0124] In one embodiment, the two rain and snow phase separation models in step S1 include:

[0125] (1) Dynamic Threshold Parameterization Scheme

[0126] like Figure 3 It is the ratio of the number of occurrences of different precipitation phases in each relative humidity (RH) and altitude (ALT) interval to the total number of occurrences under the three time scales. The relative humidity (RH) of different precipitation phases under the three time scales is divided into five intervals: 0-60%, 60-70%, 70-80%, 80-90% and 90-100%; the altitude (ALT) is divided into five intervals: 3100-3400, 3400-3700, 3700-4000, 4000-4300 and 4300-4600m; and the three time scales are calculated for each 0.2℃T W The ratio of the number of occurrences of different precipitation phases to the total number of occurrences in the five RH and ALT intervals.

[0127] like Figure 4As shown in the figure, T0 represents the T of snowfall-rainfall conversion. W The value at the center of the interval, 2ΔT represents the main temperature range for sleet, ΔS is a parameter for adjusting the temperature scale, and as ΔS increases, the T at which snowfall-rainfall conversion occurs w The interval becomes wider, and the parameterization schemes of T0, ΔT, and ΔS are calibrated based on the relationship between precipitation type, altitude, and relative humidity.

[0128] Wet bulb temperature T W The calculation formula is:

[0129]

[0130] Among them, e s is the saturated water vapor pressure (hPa); T is the air temperature (℃); T W is the wet bulb temperature (℃); RH is the relative humidity (0-1); P is the atmospheric pressure (hPa);

[0131]

[0132] After analyzing the relationship between the frequency of different precipitation phases and the influence of relative humidity and altitude, the variation curves of the three parameters T0, ΔT, and ΔS in the parameterization scheme at the minute, hour, and day scales with relative humidity and altitude are fitted as follows: Figure 5 As shown, the fitting formulas of the three parameters at three time scales are obtained respectively:

[0133] The fitting formula for minute parameters is:

[0134] ΔT=0.0004×RH×RH-0.064×RH+3.2599

[0135] T0=0.1015×RH-9.8847

[0136] ΔS=-0.0073×RH+0.9328

[0137] Fitting formula for hourly parameters:

[0138] ΔT=0.0073×RH+0.2205

[0139] T0=0.1041×RH-9.9859

[0140] ΔS=0.0004×RH×RH-0.068×RH+3.1837

[0141] Fitting formula for daily parameters:

[0142] ΔT=0.0012×RH×RH-0.1558×RH+5.8122

[0143] T0=0.142×RH-12.866

[0144] ΔS=-0.0111×RH+1.4138

[0145] The calculation formula for precipitation type is:

[0146]

[0147] Where type is the precipitation type, including rain, snow, and sleet; T min and T max These are the Snow-Sleet and Rain-Sleet temperature thresholds, respectively.

[0148] (2) Frequency Threshold Scheme

[0149] like Figure 6 As shown in the figure, the corresponding air temperature values ​​when the frequencies of rainfall, snowfall and sleet are equal can obtain the frequency intersection values ​​of the three cases of "rain-snow", "snow-sleet" and "rain-sleet", which are the frequency thresholds. The frequency thresholds obtained in the two cases of "rain-sleet" and "snow-sleet" within the temperature range of 0.5℃ are used as the precipitation phase separation threshold, that is, when T>"rain-sleet" frequency threshold, it is judged as "rain", when T<"snow-sleet" frequency threshold, it is judged as "snow", and when it is between the two, it is judged as "sleet".

[0150] Precipitation phase test method:

[0151] The confusion matrix is ​​used to test the precipitation phase. This method is a commonly used tool in fields such as machine learning and pattern recognition to evaluate the performance of classification models. Specifically, it is a matrix in which each row represents the predicted category, each column represents the actual category, and each element in the matrix represents the number of samples correctly or incorrectly classified by the classifier. The confusion matrix is ​​used as a method for precipitation testing, and the recall rate, precision rate, accuracy rate and F1 value are used as evaluation indicators for solution testing.

[0152] The recall rate is also called the recall rate, which is expressed as:

[0153]

[0154] Precision and accuracy are expressed as:

[0155]

[0156] The accuracy and correctness are expressed as:

[0157]

[0158] The F1 value is the harmonic mean of precision and recall. Since precision and recall are contradictory, we introduce the F1 value. When the F1 value is high, both precision and recall will be high. Its expression is as follows:

[0159]

[0160] Precipitation phase test results and analysis of separation method:

[0161] like Figure 7 As shown in the figure, the models of the two separation schemes at three scales, namely minute, hour and day, were cross-validated on the minute, hour and day data in the spring and autumn of 2020-2021. The confusion matrix method was used to test the precipitation type, and the test factors included recall, precision, correctness and F1.

[0162] After comprehensive analysis of the accuracy and F1 value in the test results, it was found that in the parameterized scheme, the minute and hourly models have similar and good discrimination effects on minute rain, snow and sleet; the daily model has good discrimination effects on hourly rain, daily snow and minute sleet. (The daily model has relatively good discrimination effects on hourly rain and sleet, but the F1 value is relatively low.)

[0163] For minute data, the hourly model has a better effect on distinguishing rain and sleet, and the minute model has a better effect on distinguishing snow; for hourly data, the minute model and the hourly model have equivalent and better effects on distinguishing snow, and the hourly model has a relatively better effect on distinguishing rain and sleet; for daily data, the daily model has a relatively better effect on distinguishing the three precipitation phases, but the F1 values ​​are all low.

[0164] In the frequency threshold scheme, the minute and hourly models have good discrimination effects on minute rain, snow and sleet; the daily model has good discrimination effects on daily-scale rain, hourly-scale snow and sleet, but the F1 value of sleet is low. For minute data, the minute model has good discrimination effects on rain and sleet, and the hourly model has good discrimination effects on snow; for hourly data, the hourly model has good discrimination effects on rain and sleet, and the daily model has good discrimination effects on snow; for daily data, the daily model has good discrimination effects on rain and snow, and the discrimination accuracy of sleet is high at 0.82, but the F1 value is low at 0.23.

[0165] The two schemes have slightly different effects on the recognition of precipitation phases in minutes, hours and days. The recognition accuracy rates of rain, snow and sleet are above 0.88, 0.82 and 0.77 respectively, and the F1 values ​​are above 0.85, 0.78 and 0.18 respectively. The recognition effect of each model on different precipitation phases is "rain>snow>sleet" from high to low. The recognition accuracy rate of sleet is above 0.77, but its F1 value is low, which indicates that the two schemes have more missed and wrong judgments of sleet. For the identification of rainfall phase, the hourly model in the parameterized scheme has the best identification effect for minute rain, and the minute model in the frequency threshold scheme has the best identification effect for minute rain. In general, the parameterized scheme is slightly better than the frequency threshold scheme. For the identification of snowfall phase, the minute, hourly and daily models in the frequency threshold scheme are slightly better than the parameterized scheme in identifying snow. For the identification of sleet phase, the frequency threshold scheme has a slightly better accuracy in identifying sleet than the parameterized scheme, but the F1 value is lower than that of the parameterized scheme.

[0166] Comparing the two schemes, for the identification of precipitation phases at different time scales, the smaller the time scale, the better the identification effect; both schemes have good separation effects on rain and snow, with the best identification effect on rain and poor identification effect on sleet; the parameterized scheme is slightly better than the frequency threshold scheme for rain identification, and the frequency threshold scheme is slightly better than the parameterized scheme for snow identification. The accuracy of sleet identification is relatively high, but the recall and precision values ​​are low, and there are many missed and wrong identifications of sleet.

[0167] In one embodiment, in step S2, the “precipitation-snow depth” conversion relationship is:

[0168] The conversion of "precipitation-snow depth" is the conversion of "water-snow". In order to quantitatively obtain a comprehensive model of the "water-snow" conversion relationship, during the study period, high-frequency field observation data of a complete snow season of the comprehensive snow observation test were obtained, and the wild values ​​were filtered out. The data quality was controlled to reduce errors. The relationship between snow depth and precipitation and meteorological factors during the snowfall process was analyzed, as well as the internal connection and confidence level of this relationship with different snow accumulation processes, different time periods, air pressure, and precipitation. The "water-snow" conversion relationship model based on the full-factor observation test was obtained.

[0169] (1) Data selection

[0170] Snow depth data: To ensure the validity of the data, the dynamic snow depth monitoring data of 22 national ground meteorological stations in Qinghai Province that were put into use in November 2018 can be selected to obtain the snow depth data of 22 meteorological stations from November 2018 to May 2019 and from October 2019 to January 2020;

[0171] Meteorological element data: select meteorological elements such as temperature, wind direction, wind speed, etc., and keep them synchronized with snow depth monitoring data;

[0172] Manual quality checks were conducted one by one for possible data quality issues. During the inspection, due to the noise caused by the instrument and the environment, the snow depth data often had a sudden increase in abnormal values ​​or null values. The confirmed abnormal data were deleted. Based on the snow depth data of the 22 selected monitoring stations, the monitoring period data with no snowfall or small snow depth changes were deleted, and the retained monitoring data were used to establish the model. All monitoring data were measured at an interval of one hour, that is, full-hour monitoring data.

[0173] (2) Data processing

[0174] Snowfall on the Qinghai Plateau occurs in autumn, winter and spring, and the precipitation phase in summer is not considered. Since the precipitation phase in most areas in September is rain or sleet, and there is little snowfall, the data in September are not included in the model calculation. When the model was established, the monitoring data of the complete snow season was used to classify the data into autumn, winter and spring. Autumn is defined as October-November, winter is December-February of the following year, and spring is March-May. The data was obtained from the CIMISS platform of the Qinghai Provincial Information Center. Since the downloaded data is in world time, the data was converted to Beijing time for analysis.

[0175] Since the snow depth data obtained on the big data platform is an integer (1cm), the precipitation and snow depth data at one-hour intervals change little. Therefore, the data collected by the snow depth monitoring system of each meteorological station (0.1cm) was used. By calculating the ratio of "new snowfall depth / precipitation", the snowfall water content ratio SLR was obtained. A preliminary analysis was performed. The results showed that the range of SLR in the same area and the same period varies greatly, and it is not applicable in business operations. Therefore, SPSS software is used in the present invention to analyze the interaction between snow depth and meteorological conditions during snowfall. After analyzing the progressive regression method, it was found that the stepwise regression method can be used to establish a relationship model between snow depth and meteorological elements.

[0176] (3) “Water-Snow” Conversion Model

[0177] The "water-snow" conversion relationship between precipitation and snow depth reflects the conversion of precipitation deposition into snow depth. The process also indicates its relationship with the meteorological environment. The stepwise regression method is used to establish a regression model of hourly snow depth and meteorological elements.

[0178] 1) Establishment and verification of single-station “water-snow” conversion model

[0179] According to different seasons, as shown in Table 1-3, the "water-snow" conversion model for autumn, winter and spring was established for 22 meteorological stations in the Qinghai study area. Considering the business application of the model and the meteorological elements predicted in the current numerical weather forecast, a model with "precipitation, temperature, wind direction, wind speed" meteorological elements as the leading factors was established. In order to test the model, some snow accumulation process data of the corresponding season was reserved as verification data when the model was established, and the model accuracy was better evaluated through the correlation test level and RMSE.

[0180] Table 1 Winter models and verification information of each meteorological station

[0181]

[0182] *. Significantly correlated at the 0.05 level (bilateral); **. Significantly correlated at the 0.01 level (bilateral)

[0183] Table 2 Autumn models and verification information of each meteorological station

[0184]

[0185]

[0186] *. Significantly correlated at the 0.05 level (bilateral); **. Significantly correlated at the 0.01 level (bilateral)

[0187] Table 3 Spring models and verification information of each meteorological station

[0188]

[0189] *. Significantly correlated at the 0.05 level (bilateral); **. Significantly correlated at the 0.01 level (bilateral)

[0190] The model establishment and related information of each meteorological station in winter, spring and autumn are shown in Tables 1, 2 and 3. Since some stations did not experience snowfall during the snow season, the model could not be established.

[0191] Tables 1, 2, and 3 give the specific information of the model test for each season, including the data used in the model test, correlation, and RMSE. The data used in the model test are reserved for the corresponding season. Without considering the prediction accuracy of the weather forecast, real-time weather element data such as temperature, precipitation, wind direction, and wind speed are used for model testing. From the test results, it can be seen that among the seasonal models of the 22 meteorological stations, the correlation R of most models is 2Above 0.7, with significant correlation at the 0.01 level, RMSE is mostly below 2 cm. In winter, except for Chaka, Maqin and Hualong, which are significantly correlated at the 0.05 level, the rest of the regions passed the 0.01 significance level test, and the RMSE is between 0.3-2 cm; in autumn, except for Zeku, which did not pass the significance level test, and Maduo, which was significantly correlated at the 0.05 level, the rest of the regions passed the 0.01 significance level test, and the RMSE is between 0.5-2.4 cm; in spring, except for Maqin and Nangqian, which are significantly correlated at the 0.05 level, the rest of the regions passed the 0.01 significance level test, and the RMSE is between 0.3-2.7 cm.

[0192] 2) Establishment and verification of regional “water-snow” conversion model

[0193] Due to the regional differences in precipitation, it is difficult to effectively promote the establishment of a "water-snow" conversion model for each of the 22 meteorological stations in the province in business operations. According to the model analysis of the 22 meteorological stations, it can be seen that the sensitivity factors corresponding to the models in different geographical locations are different. Therefore, Qinghai Province is roughly divided into three regions, namely the northern region, the southern region and the eastern region.

[0194] The southern Qinghai region includes: Dari, Gande, Henan, Jiuzhi, Maduo, Maqin, Nangqian, Qingshuihe, Qumalai, Tuotuohe, Wudaoliang, Zaduo, Zeku, Zhiduo; the northern region of the province includes: Chaka, Qilian, Toler, Yeniugou; the eastern region of the province includes: Datong, Hualong, Ledu, Guide, Xunhua. The corresponding regional stations were used to establish different regional models, and the data were reserved for verification. According to different seasons, the meteorological station data in the three regions were merged and processed, and the "water-snow" conversion model for autumn, winter and spring was established. The model establishment and model evaluation methods are consistent with the site model.

[0195] As shown in Table 4, in the southern Qinghai region, the autumn model is SD = -1.068 + 0.767 × PRE_12 - 0.202 × TEM + 0.418 × PRE_24 - 0.26 × PRE_6, the winter model is SD = -0.011 + 0.579 × PRE_6 + 0.188 × PRE_12 - 0.051 × WIN_S, and the spring model is SD = 10.172 + 0.483 × PRE_12 - 0.111 × TEM + 0.288 × PRE_6 - 0.001 × ALT + (-4.576E-07) × WIN_D - 0.064 × longitude;

[0196] In the northern region, the autumn model is D = -2.224 + 0.486 × PRE_6 + 0.483 × WIN_S + 0.437 × PRE_24 + 0.765 × PRE_3, the winter model is SD = 7.519 + 0.59 × PRE_24 - 0.648 × PRE_1 - 0.189 × latitude, and the spring model is SD = -0.323 + 0.915 × PRE_6 + (2.077E-06) × WIN_D;

[0197] In the eastern region, the autumn model was SD = -0.246 + 1.088 × PRE_6 - 0.201 × PRE_12, the winter model was SD = 36.137 + 0.565 × PRE_24 - 0.351 × longitude, and the spring model was SD = -0.264 + 1.341 × PRE_6;

[0198] The provincial "water-snow" conversion model: According to different seasons, the autumn, winter and spring "water-snow" conversion models are established for the whole province, and data are reserved to test the models; among them, the autumn model is SD = -0.5 + 0.890 × PRE_6-0.032 × TEM, the winter model is SD = 0.012 + 0.515 × PRE_6 + 0.221 × PRE_12-0.046 × WIN_S, and the spring model is SD = 3.959 + 0.527 × PRE_6 + 0.343 × PRE_12-0.101 × TEM-0.001 × ALT-0.068 × latitude.

[0199] Table 4 Establishment and verification of regional model of Qinghai Province

[0200]

[0201] *. Significantly correlated at the 0.05 level (bilateral); **. Significantly correlated at the 0.01 level (bilateral)

[0202] The establishment of winter, spring and autumn models and related information for southern, northern and eastern Qinghai are shown in Table 5. The table also gives the specific information of the model test for each quarter in the three regions, including the number of samples, data, correlation and RMSE used in the model test. The data used for model testing are reserved seasonal data for the corresponding regions. Without considering the prediction accuracy of weather forecasts, real-time weather element data such as temperature, precipitation, wind direction and wind speed are used for model testing. From the test results, it can be seen that in the regional seasonal models, the correlation R of most models is 2 Above 0.7, except for the eastern region in spring, which was not tested due to limited data and stations, the test levels in other regions were significantly correlated at the 0.01 level, and the RMSE was below 2 cm.

[0203] 3) Establishment and verification of the provincial “water-snow” conversion model

[0204] According to different seasons, the autumn, winter and spring "water-snow" conversion models were established for the whole province. The model establishment and model evaluation methods are consistent with the site model.

[0205] The establishment of winter, spring and autumn models and related information are shown in Table 5. The table also gives the specific information of the model test for each season in the province, including the number of samples, data, correlation and RMSE used in the model test. From the test results, it can be seen that in the provincial seasonal model, the correlation R 2 All of them were above 0.7, all were significantly correlated at the 0.01 level, and all of them had RMSE below 1.7 cm.

[0206] Table 5 Establishment and verification of snow depth model in Qinghai Province

[0207]

[0208] *. Significantly correlated at the 0.05 level (bilateral); **. Significantly correlated at the 0.01 level (bilateral)

[0209] (4) Model verification

[0210] Without considering the temperature, precipitation and other meteorological elements predicted by the weather forecast, the actual meteorological data of the set date (February 27) was used to test the model in three different ways. The precipitation on February 27 was mainly concentrated in the southern and eastern regions of Qinghai Province. In order to better verify the accuracy of the model, a total of 131 data from 3:00 to 22:00 on the 27th were selected throughout the province and the model was tested separately.

[0211] Table 6 Correlation between model-simulated snow depth and measured snow depth in the province and the region on February 27

[0212]

[0213] *. Significantly correlated at the 0.05 level (bilateral); **. Significantly correlated at the 0.01 level (bilateral)

[0214] Provincial model verification: The established provincial model was used to simulate the snow depth of 131 data points, and the bilateral correlation between the measured snow depth and the simulated snow depth was obtained using SPSS analysis software (Table 6). 2 It is 0.903, which is significantly correlated at the 0.01 level. The snow depth data simulated by the model and the measured snow depth data in the whole province are of good accuracy.

[0215] Regional model test: The model test is carried out in a unified way with the provincial model. The specific test information is shown in Table 6. As can be seen from the table, except for the northern region, the model test correlation in the eastern and southern Qinghai regions is relatively good. In the northern region, only Gangcha and Wulan had a small amount of precipitation on February 27, the number of samples was small, and the snow depth duration at a single station was short, so the accuracy of the regional snow depth model did not reach the significant correlation level.

[0216] Single-station snowfall process model verification: In order to verify the accuracy of the provincial model, regional model and single-station model at the same time, three meteorological stations in Henan, Gande and Hualong, where the snowfall time is relatively long, were selected from the 22 established single-station models to verify the snowfall process of the three models. The snowfall process lasted for 4, 8 and 9 hours respectively. Figure 8 (Comparison of snow depth simulation values ​​and measured values ​​of three models in Henan, Gande and Hualong on February 27) It can be seen that the simulated snow depth in Henan and Gande areas is consistent with the measured snow depth change trend, the simulation effect is good, and the correlation between the three models is also good; at the Hualong meteorological station, the simulation effects of the three models are relatively poor. After 10 o'clock, the snow depth simulated by the provincial model and the single-station model showed a weak downward trend. The snow depth simulated by the regional model is relatively consistent with the measured snow depth change trend, but the deviation in the high-value area is large.

[0217] Hourly model verification: Snow depth simulation was performed on the area where snowfall occurred in a certain period of time. The precipitation time period was relatively concentrated and the snowfall area was relatively wide. Continuous hourly snow depth simulation was performed. Fig. 9 The comparative analysis chart of the simulated snow depth of the provincial and regional models and the measured snow depth for four hours from 8:00 to 11:00 on February 27 and the correlation comparison table 7 are given. It can be seen from the figure that the provincial model and the regional model are consistent with the change trend of the measured snow depth value. When the snow depth is small, the simulation effect is high, and when the snow depth is large, the model snow depth is relatively poor. It may be because when the model was established, the data used was limited and the high values ​​of snow depth were few, so the model was not sensitive to high values.

[0218] Table 7 Correlation between the simulated and measured snow depth values ​​of the provincial and regional models from 8:00 to 11:00 on February 27

[0219]

[0220]

[0221] *. Significantly correlated at the 0.05 level (bilateral); **. Significantly correlated at the 0.01 level (bilateral)

[0222] It can be seen from the correlation table 7 that the correlation between the provincial model and the regional model is above 0.8, and the RMSE is about 1cm through the significance test at the 0.01 level. The model accuracy is good. Since the precipitation process is concentrated in the southern Qinghai region and the precipitation stations are relatively concentrated, the test of the southern Qinghai regional model is more favorable. By comparing the two models, it is found that the provincial model is slightly better than the regional model.

[0223] In one embodiment, in step S3, remote sensing inversion:

[0224] The Himawari-8 (Himawari-8, hereinafter referred to as H-8) meteorological satellite is one of the Sunflower series satellites. The satellite is equipped with the AHI (Advanced Himawari Imager), a visible light and infrared scanning radiometer, with a minimum temporal resolution of 10 minutes and a maximum visible light resolution of 500 meters. It has a total of 16 spectral channels, distributed in visible light, near infrared, and thermal infrared. The Himawari-8 meteorological satellite has increased the number of observation channels from 5 to 16 (HimawariCAST 14). It has 3 visible light, 3 near infrared, and 10 infrared channels, which can meet the needs of civil aviation, water conservancy, agriculture, transportation, and other fields for monitoring and forecasting of disastrous weather, urban flood warning, and large-scale environmental monitoring, and can also be applied to military meteorology.

[0225] (1) H-8 snow depth inversion

[0226] The H8 satellite data selected in the present invention can be downloaded using FTP on the website. The monitoring methods for snow accumulation in the Qinghai-Tibet Plateau often vary greatly due to the data source, algorithm and human factors used. In combination with the actual situation, a study on remote sensing monitoring technology of snow accumulation in Qinghai Province based on geostationary meteorological satellite H8 data was carried out. The snow depth inversion method of H8 satellite data is similar to the snow depth inversion method of MODIS data.

[0227] 1) Data preprocessing: H8 data preprocessing includes radiation calibration, data splicing, data projection conversion and geometric correction based on control points. The preprocessing technology process is as follows: Fig.10 shown.

[0228] 2) Snow monitoring model: The monitoring methods of snow in the Qinghai-Tibet Plateau often vary greatly due to the data sources, algorithms and human factors used. Based on the actual situation, a study on remote sensing monitoring technology of snow in Qinghai Province based on the geostationary meteorological satellite H-8 data was carried out. The remote sensing monitoring model of snow coverage and depth based on H-8 data is as follows:

[0229] NDSI index calculation: The NDSI index is a measurement method based on the reflectivity of snow to the visible and short-wave infrared bands and the relative size of the reflectivity difference. In the visible light band, the reflectivity of clouds is higher than that of land, soil, vegetation, and water in the clear sky, and can be accurately distinguished. The reflectivity of clouds and snow with large optical thickness is similar, and both have high reflectivity, but in the near-infrared band, there is a big difference between clouds and snow. Between 1.55-1.65μm and near 2.1μm, due to the absorption of solar radiation by snow, it shows the characteristics of low reflectivity; in the same spectral region, clouds reflect solar radiation, and the reflectivity is much greater than that of snow, so this band is an ideal band for distinguishing between clouds and snow. The normalized difference snow index NDSI established based on these band characteristics can distinguish most clouds from snow. The NDSI index is calculated based on the combination of MODIS's 4th band (0.5μm) and 6th band (1.64μm), and the formula is as follows:

[0230] NDSI=(R4-R6) / (R4+R6)

[0231] Among them, NDSI is the normalized difference snow index, R4 is the reflectance of the 4th band, and R6 is the reflectance of the 6th band, corresponding to the band parameters of H-8. The algorithm is also applicable to any image data of multispectral sensors with a green band of 0.5-0.6μm and a short-wave infrared band of 1.5-1.75μm. Corresponding to the band parameters of H-8, the NDSI index is transformed as follows:

[0232] NDSI=(R2-R5) / (R2+R5)

[0233] Among them, R2 is the reflectivity of the second band in H-8; R5 is the reflectivity of the fifth band in H-8;

[0234] Cloud recognition algorithm:

[0235]

[0236] Among them, B 4th is 20, B 2th is 20, B 5th 14,B 6th is 0.12.

[0237] Snow Cover Extent Extraction:

[0238]

[0239] B2>B 2th

[0240] B5>B 5th

[0241] Among them, NDSIth is 0.37, B 2th For 11, B 5th is 10;

[0242] Snow depth calculation model:

[0243] By establishing the relationship between snow depth and NDSI index, the following snow depth calculation model is obtained:

[0244] SnowDepth=32.186×NDSI-9.42672;

[0245] After the H-8 data is preprocessed through reception, image stitching, radiation calibration, projection conversion, and spatial position correction, the normalized difference snow index NDSI is calculated. When NDSI>0.37, R2>0.11, and R5>0.1 are met at the same time, the pixel is judged to be snow accumulation; when R4>0.2, R2>0.2, and R5>0.14 are met at the same time, the pixel is judged to be cloud; the snow depth is inverted using the snow depth inversion algorithm.

[0246] (2) Inversion of real-time snow depth by combining optical and microwave fusion

[0247] Visible light, near infrared and microwave bands are commonly used in snow remote sensing. Visible light and near infrared bands have certain advantages in extracting snow range. Since its application, MODIS data has played an important role in snow monitoring due to its high temporal and spatial resolution. The mature MODIS snow mapping algorithm and global MODIS snow area products have been widely verified and applied. However, due to its susceptibility to cloud pollution, it has disadvantages in the inversion of snow depth. Passive microwave remote sensing has the characteristics of all-weather and strong penetration, and has strong advantages in the inversion of snow depth. After years of development, microwave inversion snow depth algorithms and related remote sensing products have gradually matured and have been well proven in application. Introducing MODIS FSC products in passive microwave snow depth products can improve the downscaling accuracy of snow depth in mountainous areas. High-quality MODIS snow area products can better express the heterogeneity within the pixels of low-resolution passive microwave remote sensing snow depth data, thereby improving the accuracy of snow depth data.

[0248] The main data sources used for optical and microwave fusion are the passive microwave snow depth product Che_AMSR2 product and the daily snow cover dataset, which is the MODIS daily cloud-free snow cover dataset (MODIS_FSC) in the Asian Water Tower region. Specifically:

[0249] 1) Dataset

[0250] The microwave snow depth set used in the present invention is a long-term series data set of snow depth in China, which is inverted based on satellite-borne passive microwave brightness temperature data. The daily snow cover data set used is the MODIS daily cloudless snow cover data set (MODIS_FSC) in the Asian water tower region. Based on the MODIS global surface reflectance product MO / YD09GA product, the multi-endmember spectral hybrid analysis algorithm based on endmember automatic extraction (MESMA-AGE) and the multi-step joint spatiotemporal interpolation algorithm (MSTI) are used to construct the MODIS daily cloudless snow cover data set in the Asian water tower region from 2000 to 2022.

[0251] 2) Snow depth downscaling algorithm based on multi-source data fusion

[0252] The snow depth reconstruction process is as follows Fig.11 As shown in the figure, based on the passive microwave snow depth product downscaling method, combined with MODIS snow cover data, the Che product is spatially downscaled according to the empirical fusion rules and snow decay curve to obtain MODIS / passive microwave fusion snow depth data.

[0253] The core of the snow depth product downscaling fusion algorithm is to spatially match and fuse MODIS_FSC data and Che_AMSR2 products, so that the snow depth values ​​are redistributed according to the FSC values ​​of their corresponding sub-pixels. The fused pixels include snow depth layer and FSC layer, and each 25km Che product pixel corresponds to 50×50 500m FSC sub-pixels.

[0254] The fusion algorithm contains two basic assumptions:

[0255] (1) The snow depth value of a low-resolution passive microwave pixel represents the average snow depth in the pixel area;

[0256] (2) The snow depth of a passive microwave pixel can be linearly redistributed according to the snow area of ​​its sub-pixel.

[0257] Therefore, the snow depth of each sub-pixel can be calculated based on the snow depth value of the passive microwave pixel and the snow area ratio value of each sub-pixel. Fig.12 This is a schematic diagram of the fusion algorithm.

[0258] Taking the use of MODIS snow area to downscale passive microwave snow depth data to a spatial resolution of 0.05° as an example, the cloud-removed MODIS snow area ratio data (500m, about 0.005°) is first projected and resampled, and the projection is converted into a latitude and longitude projection consistent with the passive microwave snow depth data, and the pixel size is resampled to 0.05°. The calculation formula for the snow depth value of each 0.05° sub-pixel is:

[0259]

[0260] Among them, SD i is the snow depth value of the ith sub-pixel in the passive microwave pixel grid; FSC i is the snow area ratio of the i-th sub-pixel; SD grid is the snow depth value of the passive microwave pixel; n is the total number of sub-pixels corresponding to the passive microwave pixel (taking the downscaling from 0.25° to 0.05° as an example, the value of n is 25).

[0261] The following two special cases are also considered in the fusion algorithm:

[0262] a. The snow depth value of the passive microwave pixel is greater than 0, that is, SD grid >0, but the snow area values ​​of all sub-images monitored by MODIS are 0, that is, In this case, the snow depth values ​​of all sub-pixels were assigned to 0 based on the MODIS snow cover area judgment;

[0263] b. The snow area ratio of the sub-pixel is greater than 0, that is, FSC i > 0, but the snow depth value of the passive microwave pixel is equal to 0, that is, SD grid =0; In this case, the snow attenuation curve function is used to convert the snow area ratio value of the sub-pixel into a snow depth value. The snow attenuation curve function is: SD_B tb =0.2912e 0.0388FSC .

[0264] In one embodiment, in step S4, the snow depth is predicted every 3 hours during the snowfall process:

[0265] The current status of snow distribution is determined by satellite remote sensing monitoring of snow coverage and snow depth data, and future scenarios are determined by meteorological intelligent grid forecasts of precipitation, temperature and other meteorological elements. The "snowfall-snow depth" quantitative conversion relationship model is used to achieve a 3-hourly snow depth forecast for future snowfall processes.

[0266] The realization of snow depth prediction requires the integration of 3-hourly meteorological intelligent forecast grid data and actual snow distribution data. The actual snow distribution data is inverted by the fusion algorithm of H-8, MODIS_FSC and Che_AMSR2 respectively. Among them, when using H-8 to invert snow depth data, in order to solve the data gap of snow data contaminated by clouds, multi-time synthesis is used to remove non-persistent clouds. The H-8 satellite data in the Qinghai-Tibet Plateau has less cloud pollution before 14:00. In order to reduce cloud pollution, the 10-minute frequency snow depth data from 11:00 to 14:00 every day is synthesized, and the maximum daily snow depth is obtained as the actual snow depth of the day. Fig.13 This is the flow chart of H-8 inversion of daily maximum snow depth.

[0267] The future scenario forecast data uses the three-hour precipitation forecast, temperature forecast, wind forecast and other products of the Qinghai Meteorological Intelligent Grid. H-8 data and related meteorological intelligent grid forecast data are pre-processed and segmented by the National Meteorological Administration, and then pushed to the resource pool of the Qinghai Meteorological Information Center by the National Meteorological Administration Information Center.

[0268] It can be seen from the models of various meteorological stations that the meteorological factors that affect the snow depth in different regions are mainly temperature, snowfall and wind. The accumulation of snow depth varies with the period of accumulated snowfall. In general, the accumulated snowfall in the past 12 hours will affect the accumulation of snow depth. Therefore, when determining the quantitative conversion model, the precipitation in the past 12 hours needs to be considered comprehensively. According to the comparative verification and analysis results of the hourly "snowfall-snow depth" quantitative conversion model in different seasons, the model effects are different in different periods. Therefore, the Qinghai model in autumn, spring and winter is selected for calculation respectively. Since the "water-snow" quantitative conversion model requires the precipitation in the past 6 hours and the accumulated precipitation in the past 12 hours, the accumulated precipitation data from 5:00 to 17:00 is required when calculating the snow depth forecast for the next 3 hours at 14:00, of which the precipitation from 8:00 to 17:00 is obtained from the smart grid forecast data, and the precipitation from 5:00 to 8:00 is obtained from the 3-hour accumulated precipitation actual data of the smart grid. The snow depth forecast data for the next three hours at 14:00 will be used as the actual snow depth in the "water-snow" model formula for the next three hours at 17:00, and the subsequent forecasts will be similar.

[0269] The present invention is based on a large amount of field observation data. From the perspective of mechanism research, a "rain / snow phase separation model" and a "water-snow" quantitative conversion model are established. Through the existing numerous field snow automatic observation stations on the plateau, it is extended to the entire plateau area. Combined with satellite remote sensing monitoring data and refined grid numerical forecast products, a quantitative snow depth forecast product for 3 hours is obtained, which is significantly different from most methods based on large-scale statistics and empirical estimation. The key link in the actual application needs is solved. There are no accurate forecast indicators of "snowfall phase" and "snow depth" in the current weather forecast products, and only forecast products based on empirical models can be relied on. There are obvious differences in the snow depth in different forecasted areas, and the lack of a "water-snow" conversion relationship model, especially a regional application model, is a technical key to achieving this conversion. The establishment of the "water-snow" relationship model at different levels of the present invention has built a relatively complete chain from mechanism to application, which can provide technical support for solving the technical problems of this key link. The study on the prediction of snow depth is carried out from different levels of mechanism-simplification-application, providing a more comprehensive perspective on the relationship between snowfall, snow accumulation, snow morphology and actual precipitation in plateau areas and their spatial differences.

[0270] like Fig.14As shown, the present invention also provides a snow depth prediction system for snowfall process based on the fusion of snow remote sensing and meteorological intelligent grid, comprising:

[0271] Construction module 1 is used to select precipitation phases and ground meteorological observation data at three time scales within a set time period, and to construct a dynamic threshold parameterization scheme and a frequency threshold scheme to obtain two rain and snow phase separation models at three time scales;

[0272] Analysis module 2 is used to analyze the relationship between snow depth, precipitation and meteorological factors during snowfall using field observation station data and national meteorological station observation data, so as to construct a "water-snow" quantitative conversion relationship model; wherein the "water-snow" quantitative conversion relationship model includes a single-station model, a regional model and a provincial model;

[0273] Inversion module 3 is used to obtain the real-time snow depth data inverted from different data sources using H-8, microwave and optical fusion data;

[0274] Prediction module 4 is used to use the snow remote sensing monitoring snow range and snow depth data as the current status of snow distribution, and the meteorological elements of precipitation and temperature of the meteorological intelligent grid numerical weather forecast as the future scenario. The "water-snow" quantitative conversion relationship model is used to predict the snow depth of 5km×5km grid points every 3 hours during the future snowfall process, and the rain-snow phase separation model is combined to realize dynamic snow depth prediction.

[0275] The above modules are used to execute the corresponding steps in the above-mentioned snow depth prediction method for snowfall process based on the fusion of snow remote sensing and meteorological intelligent grid. The specific implementation method thereof is described in the above-mentioned method embodiment and will not be repeated here.

[0276] like Fig.15 As shown, the present invention also provides a computer device, which can be a server, and its internal structure can be as shown in Fig.15 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store all data required for the process of the snow depth prediction method for the snowfall process based on the fusion of snow remote sensing and meteorological intelligent grid. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the snow depth prediction method for the snowfall process based on the fusion of snow remote sensing and meteorological intelligent grid is implemented.

[0277] Those skilled in the art will understand that Fig.15 The structure shown in is merely a block diagram of a portion of the structure related to the present application solution and does not constitute a limitation on the computer device to which the present application solution is applied.

[0278] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, any one of the above-mentioned methods for predicting snow depth during snowfall based on the fusion of snow remote sensing and meteorological intelligent grid is implemented.

[0279] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0280] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0281] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A snow depth prediction method for snowfall process based on the fusion of snow remote sensing and meteorological intelligent grid, characterized in that: include: Select precipitation phases and ground meteorological observation data at three time scales within a set time period, and construct a dynamic threshold parameterization scheme and a frequency threshold scheme to obtain two rain and snow phase separation models at three time scales; wherein the three time scales are minutes, hours and days; Using data from field observation sites and national meteorological sites, the relationship between snow depth, precipitation and meteorological factors during snowfall is analyzed to construct a "water-snow" quantitative conversion relationship model; the "water-snow" quantitative conversion relationship model includes a single-station model, a regional model and a provincial model; The real-time data of snow depth inverted from different data sources are obtained by using H-8, microwave and optical fusion data; The snow range and snow depth data monitored by snow remote sensing are used as the current status of snow distribution, and the meteorological elements of precipitation and temperature in the meteorological intelligent grid numerical weather forecast are used as the future scenario. The "water-snow" quantitative conversion relationship model is used to predict the snow depth at a 5km×5km grid every 3 hours during the future snowfall process. The dynamic snow depth prediction is achieved by combining the rain-snow phase separation model.

2. The snow depth prediction method for snowfall process based on snow remote sensing and meteorological intelligent grid fusion according to claim 1 is characterized in that: The step of selecting precipitation phases at three time scales and ground meteorological observation data within a set time period, and constructing a dynamic threshold parameterization scheme and a frequency threshold scheme to obtain two rain and snow phase separation models at three time scales includes: Select three time scale precipitation phases and ground meteorological observation data within a set time period; Based on the wet-bulb temperature, the response relationship between altitude, relative humidity and different types of precipitation events in different wet-bulb temperature ranges is determined, and a dynamic threshold parameterization scheme is constructed; Based on air temperature, the probability of different types of precipitation events occurring in different air temperature ranges is statistically analyzed, and the frequency threshold scheme is constructed using the frequency intersection value; The parameterization scheme and frequency threshold scheme are extracted as two rain-snow phase separation models at three time scales.

3. The snow depth prediction method for snowfall process based on snow remote sensing and meteorological intelligent grid fusion according to claim 1 is characterized in that: The step of determining the response relationship between altitude, relative humidity and different types of precipitation events in different wet-bulb temperature ranges based on the wet-bulb temperature and constructing a dynamic threshold parameterization scheme includes: The relative humidity and altitude of different precipitation phases under three time scales are divided into five intervals, and the relative humidity and altitude of each 0.2℃T under three time scales are calculated. W The ratio of the number of different precipitation phases to the total number of occurrences in the five relative humidity and altitude intervals within the interval; among them, the wet bulb temperature T W The calculation formula is: Among them, e s is the saturated water vapor pressure (hPa); T is the air temperature (℃); T W is the wet bulb temperature (℃); RH is the relative humidity (0-1); P is the atmospheric pressure (hPa); Among them, T0 represents the T of snowfall-rainfall conversion. W The value at the center of the interval, 2ΔT represents the main temperature range for sleet, ΔS is a parameter for adjusting the temperature scale, and as ΔS increases, the T at which snowfall-rainfall conversion occurs w The interval becomes wider, in which the parameterization schemes of T0, ΔT, and ΔS are calibrated based on the relationship between precipitation type, altitude, and relative humidity; After analyzing the relationship between the occurrence frequency of different precipitation phases and the influence of relative humidity and altitude, the variation curves of the three parameters T0, ΔT, and ΔS in the parameterization scheme at the minute, hour, and day scales with relative humidity and altitude were fitted to obtain the fitting formulas of the three parameters at the three time scales respectively: minute: ΔT=0.0004×RH×RH-0.064×RH+3.2599 T0=0.1015×RH-9.8847 ΔS=-0.0073×RH+0.9328 Hour: ΔT=0.0073×RH+0.2205 T0=0.1041×RH-9.9859 ΔS=0.0004×RH×RH-0.068×RH+3.1837 day: ΔT=0.0012×RH×RH-0.1558×RH+5.8122 t0=0.142×RH-12.866 ΔS=-0.0111×RH+1.4138 The calculation formula for precipitation type is: Where type is the precipitation type, including rain, snow, and sleet; T min and T max These are the "snow-sleet" and "rain-sleet" temperature thresholds, respectively.

4. The snow depth prediction method for snowfall process based on snow remote sensing and meteorological intelligent grid fusion according to claim 3 is characterized in that: The step of statistically analyzing the probability of occurrence of different precipitation phases in different air temperature ranges based on air temperature and constructing a frequency threshold scheme using frequency intersection values ​​includes: The corresponding air temperature value when the frequency of rainfall, snowfall and sleet is equal can be used to obtain the frequency intersection value of the three cases of "rain-snow", "snow-sleet" and "rain-sleet", which is the frequency threshold; The frequency thresholds obtained in the two cases of "rain-sleet" and "snow-sleet" within the temperature range of the set step length are statistically used as the precipitation phase separation threshold, that is, when T>"rain-sleet" frequency threshold, it is judged as "rain", when T<"snow-sleet" frequency threshold, it is judged as "snow", and when it is between the two, it is judged as "sleet".

5. The snow depth prediction method for snowfall process based on snow remote sensing and meteorological intelligent grid fusion according to claim 1 is characterized in that: The step of using field observation site data and national meteorological site observation data to analyze the relationship between snow depth, precipitation, and meteorological factors during snowfall to construct a "water-snow" quantitative conversion relationship model includes: The high-frequency field observation data of a complete snow season of the comprehensive snow observation experiment were obtained, and the outliers were filtered out. The data quality was controlled, and the relationship between snow depth and precipitation and meteorological factors during the snowfall process was analyzed, as well as the internal connection and confidence level of this relationship with different snow accumulation processes, different time periods, air pressure, and precipitation. The "water-snow" quantitative conversion relationship model based on the observation experiment was obtained, which is as follows: Single-station "water-snow" conversion model: According to different seasons, establish autumn, winter and spring "water-snow" conversion models for a set number of meteorological stations, establish a model with "precipitation, temperature, wind direction and wind speed" meteorological elements as the leading factors, and reserve data to test the model; Regional "water-snow" conversion model: The province is divided into three regions, namely the southern Qinghai region, the northern region and the eastern region. Different regional models are established using corresponding regional stations, and data are reserved to test the models. According to different seasons, the meteorological station data in the three regions are merged and processed to establish the regional "water-snow" conversion models for autumn, winter and spring. In southern Qinghai, the autumn model is SD = -1.068 + 0.767 × PRE_12 - 0.202 × TEM + 0.418 × PRE_24 - 0.26 × PRE_6, the winter model is SD = -0.011 + 0.579 × PRE_6 + 0.188 × PRE_12 - 0.051 × WIN_S, and the spring model is SD = 10.172 + 0.483 × PRE_12 - 0.111 × TEM + 0.288 × PRE_6 - 0.001 × ALT + (-4.576E-07) × WIN_D - 0.064 × longitude; In the northern region, the autumn model is D = -2.224 + 0.486 × PRE_6 + 0.483 × WIN_S + 0.437 × PRE_24 + 0.765 × PRE_3, the winter model is SD = 7.519 + 0.59 × PRE_24 - 0.648 × PRE_1 - 0.189 × latitude, and the spring model is SD = -0.323 + 0.915 × PRE_6 + (2.077E-06) × WIN_D; In the eastern region, the autumn model was SD = -0.246 + 1.088 × PRE_6 - 0.201 × PRE_12, the winter model was SD = 36.137 + 0.565 × PRE_24 - 0.351 × longitude, and the spring model was SD = -0.264 + 1.341 × PRE_6; "Water-snow" conversion model for the whole province: According to different seasons, the autumn, winter and spring "water-snow" conversion models are established for the whole province, and data are reserved to test the models; among them, the autumn model is SD = -0.5 + 0.890 × PRE_6-0.032 × TEM, the winter model is SD = 0.012 + 0.515 × PRE_6 + 0.221 × PRE_12-0.046 × WIN_S, and the spring model is SD = 3.959 + 0.527 × PRE_6 + 0.343 × PRE_12-0.101 × TEM-0.001 × ALT-0.068 × latitude.

6. The snow depth prediction method for snowfall process based on snow remote sensing and meteorological intelligent grid fusion according to claim 1 is characterized in that: The steps of using H-8, microwave and optical fusion data to obtain real-time snow depth data inverted from different data sources include: Acquire H-8 data on the information center platform and perform data preprocessing on the H-8 data; wherein the data preprocessing includes radiation calibration, data splicing, data projection conversion and geometric precision correction based on control points; The Che_AMSR2 daily microwave snow depth data and the MODIS daily cloudless snow cover dataset in the Asian water tower region are used to fuse the two through fusion rules to invert the daily snow depth data.

7. The snow depth prediction method for snowfall process based on snow remote sensing and meteorological intelligent grid fusion according to claim 6 is characterized in that: The step of obtaining H-8 data on the information center platform and performing data preprocessing on the H-8 data includes: Using the relationship between snow depth and NDSI index, a snow depth inversion model was established to obtain the snow depth data inverted from H-8 data. The NDSI index is calculated based on the combination of MODIS band 4 0.5 μm and band 6 1.64 μm. The formula is as follows: NDSI=(R4-R6) / (R4+R6) Among them, NDSI is the normalized difference snow index, R4 is the reflectivity of the 4th band, and R6 is the reflectivity of the 6th band. Corresponding to the band parameters of H-8, the NDSI index is transformed as follows: NDSI=(R2-R5) / (R2+R5) Among them, R2 is the reflectivity of the second band in H-8; R5 is the reflectivity of the fifth band in H-8; Cloud recognition algorithm: Among them, B 4th is 20, B 2th is 20, B 5th 14,B 6th is 0.

12. Snow Cover Extent Extraction: B2>B 2th B5>B 5th Among them, NDSIth is 0.37, B 2th is 11, B 5th is 10; Snow depth calculation model: By establishing the relationship between snow depth and NDSI index, the following snow depth calculation model is obtained: SnowDepth = 32.186 × NDSI -9.42672; After the H-8 data is preprocessed through reception, image stitching, radiation calibration, projection conversion, and spatial position correction, the normalized difference snow index NDSI is calculated. When NDSI>0.37, R2>0.11, and R5>0.1 are met at the same time, the pixel is judged to be snow accumulation; when R4>0.2, R2>0.2, and R5>0.14 are met at the same time, the pixel is judged to be cloud; the snow depth is inverted using the snow depth inversion algorithm.

8. The snow depth prediction method for snowfall process based on snow remote sensing and meteorological intelligent grid fusion according to claim 6 is characterized in that: The step of using Che_AMSR2 daily microwave snow depth data and the Asian Water Tower area MODIS daily cloudless snow cover dataset to fuse and invert the daily snow depth real-time data through a fusion rule includes: The fusion rule matches and fuses MODIS_FSC data and Che_AMSR2 products in space, so that the snow depth value is redistributed according to the FSC value of its corresponding sub-pixel. The fused pixel includes a snow depth layer and a FSC layer. Each 25km Che product pixel corresponds to 50×50 500m FSC sub-pixels. The fusion rule includes two basic assumptions: 1) The snow depth value of the low-resolution passive microwave pixel represents the average snow depth in the pixel area; 2) The snow depth of passive microwave pixels can be linearly redistributed according to the snow area size of its sub-pixels; The snow depth of each sub-pixel can be calculated based on the snow depth value of the passive microwave pixel and the snow area ratio value of each sub-pixel. The passive microwave snow depth data is downscaled to a spatial resolution of 0.05° using the MODIS snow area. First, the MODIS snow area ratio data after cloud removal, i.e. 500m, is about 0.005°. The projection is converted into a latitude and longitude projection consistent with the passive microwave snow depth data. The pixel size is resampled to 0.05°. The calculation formula for the snow depth value of each 0.05° sub-pixel is: Among them, SD i is the snow depth value of the ith sub-pixel in the passive microwave pixel grid; FSC i is the snow area ratio of the i-th sub-pixel; SD grid is the snow depth value of the passive microwave pixel; n is the total number of sub-pixels corresponding to the passive microwave pixel.

9. The snow depth prediction method for snowfall process based on snow remote sensing and meteorological intelligent grid fusion according to claim 8 is characterized in that: The fusion algorithm also includes: 1) The snow depth value of the passive microwave pixel is greater than 0, that is, SD grid >0, but the snow area values ​​of all sub-images monitored by MODIS are 0, that is, In this case, the snow depth values ​​of all sub-pixels were assigned to 0 based on the MODIS snow cover area judgment; 2) The snow area ratio of the sub-pixel is greater than 0, that is, FSC i > 0, but the snow depth value of the passive microwave pixel is equal to 0, that is, SD grid =0; In this case, the snow attenuation curve function is used to convert the snow area ratio value of the sub-pixel into a snow depth value. The snow attenuation curve function is: SD_B tb =0.2912e 0.0388FSC .

10. A snow depth prediction system for snowfall process based on the fusion of snow remote sensing and meteorological intelligent grid, characterized in that: include: A construction module is used to select precipitation phases and ground meteorological observation data at three time scales within a set time period, and to construct a dynamic threshold parameterization scheme and a frequency threshold scheme to obtain two rain and snow phase separation models at three time scales; wherein the three time scales are minutes, hours and days; An analysis module is used to analyze the relationship between snow depth, precipitation and meteorological factors during snowfall using data from field observation sites and national meteorological sites, so as to construct a "water-snow" quantitative conversion relationship model; wherein the "water-snow" quantitative conversion relationship model includes a single-station model, a regional model and a provincial model; Inversion module, used to obtain the real-time snow depth data inverted from different data sources using H-8, microwave and optical fusion data; The prediction module is used to use the snow accumulation range and snow depth data monitored by snow remote sensing as the current status of snow distribution, and the meteorological elements of precipitation and temperature of the meteorological intelligent grid numerical weather forecast as the future scenario. The "water-snow" quantitative conversion relationship model is used to predict the snow depth at a 5km×5km grid every 3 hours during the future snowfall process, and the rain-snow phase separation model is combined to realize dynamic snow depth prediction.

Images