All-day sand and dust weather monitoring method

Abstract

The invention discloses an all-day sand and dust weather monitoring method. The method comprises the following steps: S1, sand and dust spectral index calculation; S2, time sequence feature extraction; S3, visibility inversion model construction; according to the method, a time sequence thermal infrared image of a static meteorological satellite is utilized, and a group of time sequence characteristics is constructed in combination with a spectral index and brightness temperature of each wave band to represent dynamic characteristics of a sand and dust event. By coupling a bidirectional long-short-term memory network model, the model is constructed based on the time sequence characteristics, and the inversion of the sand dust intensity is realized, so that the effect of dynamically monitoring the sand dust intensity at high precision all day long is achieved. In the process, the limitation that time sequence image information is difficult to fully utilize when the sand and dust intensity is inverted by a traditional spectral index method and a machine learning method is overcome, and the inversion precision of the all-day sand and dust intensity is remarkably improved.

Description

Technical Field

[0001] The present invention relates to a method for monitoring sand and dust weather, and particularly to a method for monitoring sand and dust weather all day long. Background Art

[0002] Due to the high uncertainty and complexity of the spatio-temporal distribution of sandstorms, traditional ground observation means often fail to meet the requirements of large-scale and real-time monitoring. Remote sensing technology, through sensors carried on satellite platforms, can continuously monitor the formation, propagation, and attenuation processes of sand and dust events at a macroscopic scale. Therefore, remote sensing technology has irreplaceable advantages in the field of sand and dust monitoring. Considering the satellite platforms for sand and dust monitoring based on remote sensing, they can be divided into two major categories: polar-orbiting satellites and geostationary satellites. Polar-orbiting satellite platforms operate along orbits at the two poles of the earth, can cover most regions of the globe, and provide images with high spatial resolution. However, due to the limitation of their revisit period, their temporal resolution is low, making it difficult to achieve dynamic monitoring of sand and dust weather. In contrast, geostationary satellites can continuously monitor a fixed area, providing great convenience for real-time dynamic monitoring of sand and dust events. Especially the Himawari-8 satellite, which can obtain high-precision data covering East Asia and the western Pacific region at 10-minute intervals, providing an important data source for monitoring the occurrence, development, and propagation processes of sandstorms. Developing a sand and dust all-day monitoring technology relying on Himawari-8 satellite data is of great significance.

[0003] Considering the existing remote sensing-based sand and dust monitoring technologies, there are mainly two methods: the method based on spectral indices and the method based on machine learning.

[0004] The spectral index method combines different bands according to the unique reflection characteristics of sand and dust in visible light, near-infrared, thermal infrared, and other bands, which are different from other ground objects, to improve the sensitivity to sand and dust and reduce the interference of background information, thereby realizing the identification of sand and dust. This method has the advantages of simple implementation and high computational efficiency. However, this method depends on the spectral information of limited bands, cannot describe complex non-linear relationships, has poor adaptability to different environmental backgrounds, and is easily affected by external factors such as surface types.

[0005] The machine learning method takes all band information of remote sensing images as input, and relies on a complex network structure to autonomously learn the characteristics of sand and dust based on a large number of training samples, and then realizes the monitoring of sand and dust based on the trained model. Commonly used methods include support vector machine method, random forest method, convolutional neural network method, etc. The machine learning method can not only make full use of all band information of remote sensing images, but also express complex non-linear relationships from input to output, thus enhancing the accuracy and robustness of sand and dust monitoring.

[0006] However, existing research has mostly focused on binary classification problems, distinguishing between "sand and dust" and "non-sand and dust". In production practice, it is difficult to accurately implement sand and dust weather disaster reduction measures without information on the intensity of sand and dust. A small amount of monitoring research on sand and dust intensity is mostly based on images at the observation moment, using spectral indices or machine learning methods to establish a quantitative relationship between spectral information and sand and dust intensity, and then using this relationship to detect the intensity of sand and dust. However, the above methods do not consider the use of temporal image information, and the monitoring accuracy of sand and dust intensity is limited. In fact, before the passage of sand and dust, the spectral information of the atmosphere will undergo a series of dynamic changes, and these temporal changes are of great significance for improving the accuracy of sand and dust identification. However, the existing sand and dust intensity monitoring methods cannot effectively utilize this temporal spectral information and are challenged in terms of sand and dust intensity monitoring accuracy.

[0007] Based on the temporal thermal infrared images of geostationary meteorological satellites, this application constructs a set of temporal features using spectral indices and brightness temperatures of each band to characterize this phenomenon, and couples a bidirectional long short-term memory network to construct a model based on the temporal features to achieve all-weather dynamic high-precision monitoring of sand and dust intensity. Summary of the Invention

[0008] In order to solve the deficiencies of the above technologies, the present invention provides an all-weather sand and dust weather monitoring method.

[0009] In order to solve the above technical problems, the technical solution adopted by the present invention is: an all-weather sand and dust weather monitoring method, including the following steps:

[0010] S1. Calculation of sand and dust spectral indices. Cloud masking is performed based on the hourly thermal infrared brightness temperature images of geostationary meteorological satellites, and then the hourly brightness temperature difference spectral index, three-band volcanic ash product index, brightness temperature-adjusted dust index, and thermal infrared comprehensive dust index are calculated.

[0011] S2. Extraction of temporal features, including the brightness temperature difference spectral index, three-band volcanic ash product index, brightness temperature-adjusted dust index, thermal infrared comprehensive dust index, and clear sky state values, near-observation moment state values, and observation moment state values of the brightness temperature of each band.

[0012] S3. Construction of a visibility inversion model, which is based on a bidirectional long short-term memory network model and uses the temporal feature information in S2 as input.

[0013] S4. Sand and dust intensity mapping.

[0014] Furthermore, the brightness temperature difference spectral index includes three typical ones, and their respective calculation formulas are as follows:

[0015] BTD 11-12 = BT 11.2 - BT 12.4 ;

[0016] BTD 3-12 = BT 3.9 - BT 11.2 ;

[0017] BTD 8-11 = BT 8.6 - BT 11.2 ;

[0018] In the formula, BTD 11-12 , BTD 3-12 , BTD 8-11 respectively represent three typical brightness temperature difference spectral indices, and BT 3.9 , BT 8.6 , BT 11.2 , BT 12.4 correspond to the 3.9μm, 8.6μm, 11.2μm, and 12.4μm bands of the geostationary meteorological satellite respectively.

[0019] Furthermore, the calculation formula of the three-band volcanic ash product index is as follows:

[0020] TVAP = 60 + 10×(BT 12.4 - BT 11.2 ) + 3×(BT 3.9 - BT 11.2 ).

[0021] Furthermore, the calculation process of the brightness temperature-adjusted dust index is as follows:

[0022]

[0023] Among them, BDI = (BT 3.9 - BT 12.4 ) 2 ×(BT 11.2 - BT 12.4 );

[0024] In the formula, BADI is the brightness temperature-adjusted dust index, BDI is the brightness temperature dust index, and BDI 0.95 represents the 95% quantile of BDI.

[0025] Furthermore, the calculation process of the thermal infrared comprehensive dust index is as follows:

[0026]

[0027] Further, the clear-sky state values of the spectral indices and the brightness temperatures of each band refer to the values of the spectral indices and the brightness temperatures calculated using the clear-sky images synthesized from the images within 7 days before the observation time. When synthesizing, mainly based on the 11.2 μm band, the images at the observation time corresponding to the maximum value within 7 days are selected for synthesis of each band.

[0028] Further, the near-observation-time state values of the spectral indices and the brightness temperatures of each band refer to first calculating the spectral indices and the brightness temperatures of each band based on the images of each hour within 30 hours before the observation time, and then respectively calculating the time-series data of each index and the brightness temperatures of each band through the following formula:

[0029]

[0030] In the formula, F(x, y) is the final spectral index value and the brightness temperature value of each band of the pixel point (x, y) in the synthesized image; I i (x, y) represents the spectral index value or the brightness temperature value of each band of the i-th hour image at the pixel (x, y); w i is the weight of the i-th hour;

[0031]

[0032] In the formula, w i represents the weight of the i-th hour image, λ is a constant value that controls the weight increase rate and takes the value of 0.5, n is the total number of hours and takes the value of 30, and the image at the time farthest from the observation time takes the value of 1, and the image closest to the observation time takes the value of 30.

[0033] Further, the observation-time state values of the spectral indices and the brightness temperatures of each band refer to the spectral index values and the brightness temperature values of each band extracted based on the observation-time image.

[0034] Further, the visibility inversion model includes an input layer, a hidden layer, and an output layer;

[0035] The input layer corresponds to the input vectors X t-2 、X t-1 、X t at three different times, X t-2 corresponds to the clear-sky state value of the brightness temperature of each band, X t-1 corresponds to the near-observation-time state value, and X t corresponds to the observation-time state value;

[0036] The output layer corresponds to the visibility value Y t ;

[0037] The hidden layer includes two sub-networks, and the two sub-networks are respectively responsible for processing the forward information flow and the reverse information flow to capture the forward feature information and the backward feature information of the input sequence.

[0038] Furthermore, based on the visibility inversion model in S3, hourly visibility data is inverted. According to the hourly visibility data, the dust intensity is divided into five levels, namely extreme severe sandstorm, severe sandstorm, sandstorm, floating dust / dust storm, and no dust, thus completing the hourly dust intensity mapping and realizing all-weather dust monitoring.

[0039] An all-weather dust weather monitoring method uses the sequential thermal infrared images of a geostationary meteorological satellite, combines spectral indices and brightness temperatures of each band, constructs a set of sequential features to characterize the dynamic characteristics of dust events. By coupling a bidirectional long short-term memory network model, a model is constructed based on these sequential features to realize the inversion of dust intensity, thereby achieving the effect of all-weather dynamic high-precision monitoring of dust intensity. In this process, the limitations of the traditional spectral index method and machine learning method in insufficiently using sequential image information when inverting dust intensity are overcome, and the inversion accuracy of all-weather dust intensity is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flowchart of the present invention.

[0041] Figure 2 It is the architecture of the visibility inversion model of the present invention.

[0042] Figure 3 It is the hourly dust intensity map disclosed in Embodiment 2. DETAILED DESCRIPTION OF THE INVENTION

[0043] The present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0044] Embodiment 1

[0045] This embodiment relates to an all-weather dust weather monitoring method, as Figure 1 shown, including the following steps:

[0046] S1. Calculation of dust spectral indices. Cloud masking is performed based on the hourly thermal infrared brightness temperature images of the geostationary meteorological satellite, and then the hourly brightness temperature difference spectral index, three-band volcanic ash product index, brightness temperature adjusted dust index, and thermal infrared comprehensive dust index are calculated. It should be noted that the specific model of the geostationary meteorological satellite in this embodiment is preferably Himawari-8, and the hourly thermal infrared brightness temperature images are obtained using the AHI sensor of Himawari-8; the cloud masking method is a prior art and is specifically carried out according to the method published by the author Yamamoto et al. in Geophysical Research Letters in 2016, Volume 43, Pages 5886–5894, which is not original to this application and will not be shown or described.

[0047] Preferably, the brightness temperature difference spectrum index (i.e., Brightness temperature difference, BTD) specifically includes three typical brightness temperature difference spectrum indices and the calculation formulas are as follows:

[0048] BTD 11-12 = BT 11.2 - BT 12.4 ;

[0049] BTD 3-12 = BT 3.9 - BT 11.2 ;

[0050] BTD 8-11 = BT 8.6 - BT 11.2 ;

[0051] In the formula, BTD 11-12 , BTD 3-12 , BTD 8-11 respectively represent three typical brightness temperature difference spectrum indices, and BT 3.9 , BT 8.6 , BT 11.2 , BT 12.4 correspond to the 3.9μm, 8.6μm, 11.2μm, and 12.4μm bands of the geostationary meteorological satellite respectively.

[0052] Preferably, the calculation formula of the three-band volcanic ash product index (i.e., Three-band volcanic ash product, TVAP) is as follows:

[0053] TVAP = 60 + 10×(BT 12.4 - BT 11.2 ) + 3×(BT 3.9 - BT 11.2 ).

[0054] Preferably, the calculation process of the brightness temperature adjusted dust index (i.e., Brightness temperature adjusted dustindex, BADI) is as follows:

[0055]

[0056] Among them, BDI = (BT 3.9 - BT 12.4 ) 2 ×(BT 11.2 - BT 12.4 );

[0057] In the formula, BADI is the brightness temperature adjusted dust index as described above, BDI is the brightness temperature dust index (i.e., Brightness temperature dust index, BDI), and BDI 0.95 represents the 95th percentile of BDI.

[0058] Preferably, the calculation process of the thermal infrared integrated dust index (i.e., Thermal Infrared Integrated Dust Index, TIIDI) is as follows:

[0059]

[0060] S2. Temporal feature extraction, including the brightness temperature difference spectral index and the clear sky state values, the state values at the adjacent observation time, and the state values at the observation time for each band.

[0061] Preferably, the clear sky state values of the spectral index and the brightness temperature of each band refer to the values of each spectral index and the brightness temperature calculated using the clear sky image synthesized from the images in the 7 days before the observation time. When synthesizing, mainly based on the 11.2μm band, the images at the observation time corresponding to the maximum value in the 7 days are selected for synthesis of each band.

[0062] Preferably, the state values of the spectral index and the brightness temperature of each band at the adjacent observation time refer to first calculating the spectral index and the brightness temperature values of each band based on the images of each hour within 30 hours before the observation time, and then calculating the temporal data of each index and the brightness temperature of each band through the following formula:

[0063]

[0064] In the formula, F(x,y) is the final spectral index value and the brightness temperature value of the pixel point (x,y) in the synthesized image; I i (x,y) represents the spectral index value or the brightness temperature value of each band of the i-th hour image at the pixel (x,y); w i is the weight of the i-th hour;

[0065]

[0066] In the formula, w i represents the weight of the i-th hour image, λ is a constant that controls the weight increase rate and takes a value of 0.5, n is the total number of hours and takes a value of 30, and the image at the time farthest from the observation time takes a value of 1, and the image adjacent to the observation time takes a value of 30.

[0067] The state values of the spectral index and the brightness temperature of each band at the observation time refer to the spectral index values and the brightness temperature values of each band extracted from the observation time image.

[0068] S3. Visibility inversion model construction: Based on the bidirectional long short-term memory network model, using the temporal feature information in S2 as input, and then building a visibility inversion model;

[0069] As Figure 2 shown, the visibility inversion model includes an input layer, a hidden layer, and an output layer;

[0070] The input layer corresponds to input vectors X t-2 、X t-1 、X t at three different times. X t-2 corresponds to the clear sky state value of the brightness temperature of each band, X t-1 corresponds to the state value at the adjacent observation time, and X t corresponds to the state value at the observation time;

[0071] The hidden layer includes two sub-networks, which are responsible for processing the forward information flow and the backward information flow respectively to capture the forward feature information and the backward feature information of the input sequence.

[0072] Preferably, each sub-network includes three LSTM basic modules, which respectively process the input information of X t-2 、X t-1 、X t . They are connected in series, so that information can be transmitted in time series. It should be noted that the LSTM basic module is well-known, and those skilled in the art can clarify its content, so no detailed explanation will be given here.

[0073] The output layer corresponds to the visibility value Y t at the observation time t, and is connected to the LSTM basic module that processes the X t time in the hidden layer.

[0074] S4. Dust intensity mapping: Using the visibility inversion model in S3 to invert the hourly visibility data, and based on the national standard "Dust Weather Grade (GB / T 20480-2017)", the dust intensity is divided into five grades: extreme severe sandstorm, severe sandstorm, sandstorm, floating dust / blowing sand, no dust, then complete the hourly dust intensity mapping and realize all-day dust monitoring.

[0075] Example 2

[0076] On the basis of Example 1, Example 2 discloses the practical application of Example 1.

[0077] In this embodiment, hourly Himawari-H8 satellite AHI sensor thermal infrared band images, hourly visibility data, and hourly wind speed data of 119 meteorological observation stations in Inner Mongolia from March to May (spring) in 2020 - 2022 were collected. The following steps were used to generate an hourly sand-dust intensity spatial distribution map of Inner Mongolia from 2020 to 2022. The specific steps are as follows:

[0078] A1. Referring to the method in step S1 of Embodiment 1, the collected satellite images were first cloud masked according to the method published by Yamamoto et al. in Geophysical Research Letters, Volume 43, Pages 5886 - 5894 in 2016. Then, three typical brightness temperature difference spectral indices were calculated based on the formula disclosed in step S1 of Embodiment 1, the three-band volcanic ash product index was calculated, the brightness temperature adjusted dust index was calculated, and the thermal infrared comprehensive dust index was calculated.

[0079] A2. Referring to the method in step S2 of Embodiment 1, the temporal features of each observation moment were extracted, including the clear sky state values of the above spectral indices and the brightness temperatures of each band, the state values of adjacent observation moments, and the state values of the observation moment.

[0080] A3. The observed visibility values of stations with sand-dust occurrences in Inner Mongolia from 2020 to 2022 were extracted. 70% of the data was randomly selected as the training data set, and the other 30% of the data was used as the validation data set. The temporal feature values of the images corresponding to the ground observed visibility in the training data set were extracted, and they were used as training data to Figure 2 train the visibility inversion model shown.

[0081] A4. Based on the trained visibility inversion model, the temporal features calculated in A2 were used as inputs to generate an hourly visibility data product of Inner Mongolia in spring from 2020 to 2022.

[0082] A5. Using the visibility data product generated in A4, based on the national standard "Sand-dust Weather Grade (GB / T 20480 - 2017)" and combined with the collected wind speed data, a sand-dust intensity data product was generated.

[0083] Based on the above steps, all-day sand-dust weather can be monitored. The visibility data of the validation data set was used to verify the visibility inversion model constructed in step A4, and the average relative error was only 23.83%. An hourly sand-dust intensity map during the sandstorm occurrence in Inner Mongolia on March 15, 2021, as shown in Figure 3 was generated, which truly reflects the actual situation.

[0084] In summary, the method disclosed in Embodiment 1 also has the potential to be extended to other geostationary satellite images, which is hereby noted.

[0085] This application discloses a method for monitoring sand and dust weather all day long. By using the sequential thermal infrared images of geostationary meteorological satellites and combining spectral indices and brightness temperatures of each band, a set of sequential features is constructed to characterize the dynamic characteristics of sand and dust events. By coupling a bidirectional long short-term memory network model, a model is constructed based on these sequential features to achieve the inversion of sand and dust intensity, thus achieving the effect of all-weather dynamic high-precision monitoring of sand and dust intensity. In this process, the limitations of the traditional spectral index method and machine learning method in fully utilizing the information of sequential images when inverting sand and dust intensity are overcome, and the inversion accuracy of sand and dust intensity all day long is significantly improved.

[0086] The above embodiments are not limitations to the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by those skilled in the art within the scope of the technical solution of the present invention also fall within the protection scope of the present invention.

Claims

1. A method for monitoring dust weather throughout the day, characterized in that, It includes the following steps: S1. Calculation of dust spectral index: Perform cloud masking based on the hourly thermal infrared brightness temperature images of geostationary meteorological satellites, and then calculate the hourly brightness temperature difference spectral index, three-band volcanic ash product index, brightness temperature adjusted dust index, and thermal infrared comprehensive dust index; S2. Temporal feature extraction, including the brightness temperature difference spectral index, three-band volcanic ash product index, brightness temperature adjusted dust index, thermal infrared comprehensive dust index, and clear sky state values, near-observation moment state values, and observation moment state values of the brightness temperature of each band; S3. Construction of visibility inversion model: Based on the bidirectional long short-term memory network model, use the temporal feature information in S2 as input; S4. Dust intensity mapping.

2. The all-day dust weather monitoring method according to claim 1, wherein The brightness temperature difference spectral index includes three typical ones and their respective calculation formulas are as follows: BTD 11-12 = BT 11.2 - BT 12.4 ; BTD 3-12 = BT 3.9 - BT 11.2 ; BTD 8-11 = BT 8.6 - BT 11.2 ; where BTD 11-12 , BTD 3-12 , and BTD 8-11 represent three typical brightness temperature difference spectral indices respectively, and BT 3.9 , BT 8.6 , BT 11.2 , and BT 12.4 correspond to the 3.9μm, 8.6μm, 11.2μm, and 12.4μm bands of the geostationary meteorological satellite respectively.

3. The all-day dust weather monitoring method according to claim 2, wherein The calculation formula of the three-band volcanic ash product index is as follows: TVAP = 60 + 10×(BT 12.4 - BT 11.2 ) + 3×(BT 3.9 - BT 11.2 )。 4. The all-day dust weather monitoring method according to claim 2, characterized in that , The calculation process of the brightness temperature adjusted dust index is as follows: where BDI = (BT 3.9 - BT 12.4 ) 2 × (BT 11.2 - BT 12.4 ); Wherein, BADI is the brightness temperature adjusted dust index, BDI is the brightness temperature dust index, and BDI 0.95 represents the 95th percentile of BDI.

5. The all-day dust weather monitoring method according to claim 2, wherein The calculation process of the thermal infrared comprehensive dust index is as follows:

6. The all-day dust weather monitoring method according to claim 1, characterized in that: The clear sky state values of the spectral index and the brightness temperature of each band refer to the values of each spectral index and brightness temperature calculated using the clear sky image synthesized from the images in the 7 days before the observation moment. When synthesizing, mainly based on the 11.2μm band, select the image corresponding to the maximum value in the 7 days for each band synthesis.

7. The all-day dust weather monitoring method according to claim 6, characterized in that The near-observation moment state values of the spectral index and the brightness temperature of each band refer to first calculating the spectral index and the brightness temperature values of each band based on the images of each hour within 30 hours before the observation moment, and then calculating the temporal sequence data of each index and the brightness temperature of each band through the following formula: where \(F(x,y)\) is the final spectral index value and the brightness temperature values of each band of the pixel \((x,y)\) in the synthetic image; \(I\) i (x,y) represents the spectral index value or the brightness temperature value of each band of the \(i\)-th hour image at the pixel \((x,y)\); \(w\) i is the weight of the \(i\)-th hour; where w i represents the weight of the image at the i-th hour, λ is a constant controlling the weight increasing rate with a value of 0.5, n is the total number of hours with a value of 30, and the image at the farthest time from the observation time has a value of 1, and the image adjacent to the observation time has a value of 30.

8. The all-day dust weather monitoring method according to claim 7, wherein: The observation moment state values of the spectral index and the brightness temperature of each band refer to the spectral index values and the brightness temperature values of each band extracted based on the observation moment image.

9. The all-day dust weather monitoring method according to claim 1, characterized in that: The visibility inversion model includes an input layer, a hidden layer, and an output layer; The input layer corresponds to input vectors X at three different times t-2 , X t-1 , X t , X t-2 corresponds to the clear sky state value of the brightness temperature of each band, X t-1 corresponds to the state value at the adjacent observation time, X t corresponds to the state value at the observation time; The output layer corresponds to the visibility value Y at the observation time t t ; The hidden layer includes two sub-networks, and the two sub-networks are respectively responsible for processing the forward information flow and the backward information flow to capture the forward feature information and backward feature information of the input sequence.

10. The all-day dust weather monitoring method according to claim 1, wherein: Invert the hourly visibility data based on the visibility inversion model in S3, and divide the dust intensity into five levels according to the hourly visibility data, namely extreme strong sandstorm, strong sandstorm, sandstorm, floating dust / dust storm, and no dust, then complete the hourly dust intensity mapping and realize all-weather dust monitoring.

Images