A multi-scale fusion method and system based on satellite sea surface temperature

Abstract

This application discloses a multi-scale fusion method and system based on satellite sea surface temperature, relating to the field of satellite remote sensing technology. The method comprises obtaining a sea surface temperature dataset; performing data preprocessing on the sea surface temperature dataset to obtain a processed sea surface temperature dataset; performing a quality accuracy evaluation on each data point in the processed sea surface temperature dataset based on the measured sea surface temperature; the quality accuracy evaluation is evaluated using an evaluation index consisting of mean deviation, absolute deviation, root mean square error, standard deviation, and correlation coefficient; performing deviation correction on each data point in the processed sea surface temperature dataset based on the quality accuracy evaluation results of each data point using a piecewise regression method to obtain a corrected sea surface temperature dataset; and constructing a multi-scale fusion model based on the corrected sea surface temperature dataset using a multi-scale optimal interpolation method. This application can realize multi-scale information fusion of sea surface temperature.

Description

Technical Field

[0001] The present application relates to the field of satellite remote sensing technology, and in particular to a multi-scale fusion method and system based on satellite sea surface temperature. Background Art

[0002] Sea surface temperature data are mainly obtained through two channels: on-site field measurements and satellite remote sensing observations. Traditionally, on-site field measurements rely on conventional observation systems such as survey ships, coastal stations and buoy sites, but the number of observation points of these methods is limited, resulting in a small amount of data collected. In contrast, ocean satellite remote sensing observations, with their all-weather, near-real-time, wide coverage and long-term repeated observation capabilities, effectively supplement the shortcomings of on-site measurement data. Satellite remote sensing mainly relies on infrared and microwave sensors to monitor sea surface temperature. Remote sensing technology can achieve large-scale synchronous observations, which not only significantly reduces the cost of sampling data, but also increases the observation speed and has high temporal and spatial resolution. In recent years, satellite ocean remote sensing technology has become one of the key means of detecting sea surface temperature, and it can provide a large amount of ocean data.

[0003] Currently, the most widely used and effective method for processing ocean in-situ measured data and satellite remote sensing data is data fusion to produce sea temperature analysis products with high temporal and spatial resolution. However, these methods have some limitations when processing satellite remote sensing data, such as the loss of some important information during the fusion process. Summary of the Invention

[0004] The purpose of this application is to provide a multi-scale fusion method and system based on satellite sea surface temperature, which can accurately and effectively extract multi-scale information from satellite remote sensing observations of sea surface temperature and realize multi-scale information fusion of sea surface temperature.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a multi-scale fusion method based on satellite sea surface temperature, comprising:

[0007] A sea surface temperature dataset is obtained; data in the sea surface temperature dataset is obtained by multiple satellite sensors.

[0008] Data preprocessing is performed on the sea surface temperature dataset to obtain a processed sea surface temperature dataset.

[0009] Based on the measured sea surface temperature, a quality accuracy evaluation is performed on each data in the processed sea surface temperature dataset; the evaluation indicators of the quality accuracy evaluation are composed of mean deviation, absolute deviation, root mean square error, standard deviation and correlation coefficient.

[0010] According to the quality accuracy evaluation results of each data, based on the segmented regression method, deviation correction is performed on each data in the processed sea surface temperature dataset to obtain a corrected sea surface temperature dataset.

[0011] Based on the corrected sea surface temperature dataset, a multi-scale fusion model is constructed using the multi-scale optimal interpolation method.

[0012] Optionally, the data in the sea surface temperature dataset include Fengyun satellite data, ocean satellite data, iQuam measured data, Argo buoy data, OISST data and OSTIA data.

[0013] Optionally, performing data preprocessing on the sea surface temperature dataset to obtain a processed sea surface temperature dataset specifically includes:

[0014] For the Fengyun satellite data, the sea mark temperature data, satellite zenith angle and quality flag bits in the Fengyun satellite data are transposed or rotated, and the invalid land values, other invalid data, sea mark temperature points without valid zenith angle data, and data with a difference of more than 5°C in the OISST monthly average data are eliminated according to the mask data to obtain the processed Fengyun satellite data.

[0015] For the ocean satellite data, the sea surface temperature data in the ocean satellite data are transposed, and the invalid land values are eliminated according to the mask data, the remaining invalid data points are filled with NaN, and the data with the OISST monthly average data difference greater than 5℃ are obtained to obtain the processed ocean satellite data.

[0016] For the iQuam measured data, the daily average values of the latitude and longitude data, quality mark data, water depth data, observation time data, and sea surface temperature data in the iQuam measured data are solved to obtain the processed iQuam measured data.

[0017] For the Argo float data, the Argo data with a depth range of 5-6m from the sea surface are selected as the processed Argo float data.

[0018] For the OISST data, the sea mark temperature data and latitude and longitude information in the OISST data are transposed or rotated, and the invalid sea mark temperature data are eliminated according to the mask information to obtain the processed OISST data.

[0019] For OSTIA data, the sea mark temperature data and latitude and longitude information in the OSTIA data are transposed or rotated, and invalid sea mark temperature data are eliminated according to the mask information to obtain the processed OSTIA data.

[0020] Optionally, the formula expressions for the mean deviation, absolute deviation, root mean square error, standard deviation, and correlation coefficient in the evaluation indicators are:

[0021]

[0022]

[0023]

[0024]

[0025]

[0026] Among them, Bias is the average bias, Abs_Bias is the absolute bias, RMSE is the root mean square error, STDE is the standard deviation, R is the correlation coefficient, S i , I i represent the sea mark temperature data value inverted from infrared data and the measured sea mark temperature data of the i-th matching point, respectively, and n is the total number of matching points between satellite and measured data; All matching points S i , I i The mean of .

[0027] Optionally, according to the quality accuracy evaluation results of each data, based on a piecewise regression method, bias correction is performed on each data in the processed sea surface temperature dataset to obtain a corrected sea surface temperature dataset, specifically including:

[0028] According to the formula T xz =b0+b1T s +b2(T s -T c ), bias correction is performed on each data in the sea surface temperature dataset.

[0029] Among them, T xz is the corrected sea surface temperature, b0~b2 are regression coefficients, T s and (T s -T c ) represent the sea surface temperature inversion value and the sea surface temperature anomaly value, respectively.

[0030] Optionally, a multi-scale fusion model is constructed based on the corrected sea surface temperature dataset using a multi-scale optimal interpolation method, specifically including:

[0031] Obtain a background field error covariance matrix; the background field error covariance matrix is composed of a diagonal matrix D composed of background field errors and a horizontal correlation matrix ρ of the predicted background field.

[0032] A spatial correlation function model is constructed based on the background field error covariance matrix and the corrected sea surface temperature dataset.

[0033] When fusing sea surface temperature data, the background error correlation function of the observation points of different data is calculated.

[0034] The correlation coefficient between the observation points is calculated based on the correlation distribution parameter fitting formula of the background error.

[0035] Optionally, the background field error covariance matrix is expressed as follows:

[0036] B=D 0.5 ρD 0.5 .

[0037] Among them, D is the diagonal matrix composed of background field errors, ρ is the horizontal correlation matrix of the predicted background field, and the matrix R is calculated using the same processing method as the background field error covariance matrix B.

[0038] Optionally, when fusing sea surface temperature data, background error correlation functions of observation points of different data are calculated, specifically including:

[0039] According to the formula Calculate the background error correlation function for observation points of different data.

[0040] Among them, R ij represents the correlation coefficient, S i 、S j represent the SST values of sites i and j, respectively, and 〈〉 represents the vector average.

[0041] Optionally, the correlation coefficient between the observation points is calculated according to a correlation distribution parameter fitting formula of the background error, specifically including:

[0042] According to the correlation distribution parameter fitting formula Calculates the correlation coefficient between observations.

[0043] Among them, ρ ij represents the correlation coefficient between points i and j, r1 and r2 represent the distance in the latitude and longitude directions respectively, and λ1 and λ2 represent the latitude lines respectively.

[0044] In the second aspect, the present application provides a multi-scale fusion system based on satellite sea surface temperature, including

[0045] The data acquisition module is used to acquire a sea surface temperature data set; the data in the sea surface temperature data set is acquired by a variety of satellite sensors.

[0046] The preprocessing module is used to perform data preprocessing on the sea surface temperature dataset to obtain a processed sea surface temperature dataset.

[0047] An evaluation module is used to perform quality accuracy evaluation on each data in the processed sea surface temperature dataset based on the measured sea surface temperature; the evaluation indicators of the quality accuracy evaluation are composed of mean deviation, absolute deviation, root mean square error, standard deviation and correlation coefficient.

[0048] The correction module is used to perform deviation correction on each data in the processed sea surface temperature dataset based on the quality accuracy evaluation result of each data and the segmented regression method to obtain a corrected sea surface temperature dataset.

[0049] The fusion module is used to construct a multi-scale fusion model based on the corrected sea surface temperature dataset using a multi-scale optimal interpolation method.

[0050] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0051] The present application provides a multi-scale fusion method and system based on satellite sea surface temperature. First, a sea surface temperature data set is collected and then preprocessed. Next, the actual measured sea surface temperature data is used to evaluate the quality and accuracy of each data point in the preprocessed data set. The evaluation indicators include mean deviation, absolute deviation, root mean square error, standard deviation and correlation coefficient. Based on these evaluation results, a piecewise regression method is used to perform deviation correction on each data point in the data set to obtain a corrected sea surface temperature data set. Finally, using this corrected data set, a model is constructed through a multi-scale optimal interpolation method to achieve multi-scale fusion of sea surface temperature information. The present application can use satellite sea surface data as the input source, establish a model system from data quality control, deviation correction and data fusion, and construct a multi-scale data fusion model, which can effectively extract multi-scale information from satellite remote sensing observations of sea surface temperature. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0053] Figure 1 A flowchart of a multi-scale fusion method based on satellite sea surface temperature is provided in one embodiment of the present application.

[0054] Figure 2 A structural schematic diagram of a fusion method based on domestic satellite sea surface temperature provided in one embodiment of the present application.

[0055] Figure 3A deviation correction flow chart is provided for an embodiment of the present application.

[0056] Figure 4 A schematic diagram of a satellite sea surface temperature fusion process provided in one embodiment of the present application.

[0057] Figure 5 A schematic diagram of the structure of a multi-scale fusion system based on satellite sea surface temperature provided in one embodiment of the present application. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0059] Over two-thirds of the Earth's surface is ocean, and seawater has a high specific heat capacity. Over 70% of solar radiation is absorbed by the ocean each year, primarily stored in its surface layer. The ocean surface is the direct interface between the ocean and the atmosphere. Sea surface temperature (SST), a fundamental parameter of the ocean surface, is crucial for studying ocean-atmosphere material and energy exchange. SST can intuitively characterize changes in ocean thermal and dynamic conditions and air-sea interactions. It is a fundamental parameter of thermodynamic circulation at various scales and a key indicator of physical oceanography, such as ocean fronts, ocean water masses, and ocean currents. SST is also a major factor influencing long-term weather processes and global climate change. Many climate phenomena and meteorological disasters, such as El Niño-La Niña, hurricanes, and severe tropical storms, are inextricably linked to it. SST directly influences the life processes of marine organisms, such as population distribution, reproduction, and migration, and plays a crucial role in determining the habitats of different fish species and in marine fishery development. For example, in marine fisheries, each species has its own optimal temperature range for survival. Therefore, SST can be used to forecast fish stocks, providing important guidance for marine fishery development. In particular, the location of central fishing grounds and analysis of development status require fast, accurate, and comprehensive SST structured products. Furthermore, various marine science research and other operational marine applications are placing increasing demands on SST products. In summary, providing high temporal and spatial resolution, high-precision SST structured products is crucial for ocean research and addressing global climate change.

[0060] Sea surface temperature (SST) data are primarily obtained through in situ measurements and satellite remote sensing. Traditional in situ measurements rely primarily on conventional observation systems such as survey vessels, coastal stations, and buoys. These observation points are very limited, resulting in small amounts of data. Furthermore, data quality is often unsatisfactory due to instrument limitations and other factors, such as human factors. In recent years, with the continuous improvement of temperature measurement instrument precision and the development of new SST measurement platforms, such as Argo floats and underwater gliders, the quality and quantity of in situ measurements have greatly improved. However, compared to actual needs, the amount of SST data obtained through in situ measurements remains very limited, with high measurement costs, slow collection speeds, and suboptimal temporal synchronization and spatial continuity. Compared to in situ measurements, ocean satellite remote sensing offers the advantages of all-weather, near-real-time, wide coverage, and long-term repeatability, effectively compensating for the shortcomings of in situ measurements. Satellite remote sensing of SST primarily relies on infrared and microwave sensors. Among them, the high temporal and spatial resolution of the observation results is the advantage of infrared sensors, but its disadvantage is that it is easily affected by aerosols, atmospheric cloud conditions, etc., which leads to a low inverted SST; microwaves have extremely strong penetrability, so microwave sensors can generally conduct observations around the clock and in all weather conditions, but the spatial resolution of the observation results is slightly worse than that of infrared sensors.

[0061] In satellite remote sensing observations, controllable observation conditions combined with diverse observational data ensure the multi-source nature of the observed remote sensing data, encompassing multiple temporal phases, multiple bands, multiple polarizations, and multiple platforms. In research and application, it is ideal to understand, study, and apply this multi-source nature of satellite remote sensing data from diverse perspectives and at multiple levels. Striving to identify errors within observational data, correct them, and eliminate them, thereby potentially improving the quality of satellite remote sensing data, is one area we must continuously explore.

[0062] Remote sensing allows for simultaneous observations across large scales, significantly reducing data sampling costs, while also increasing speed and providing high temporal and spatial resolution. In recent years, satellite ocean remote sensing has become a primary method for detecting sea surface temperature. Satellite remote sensing yields vast amounts of ocean data, tens or even hundreds of times greater than the combined total of field observations collected over the past century. Optimizing the utilization of this vast amount of data has long been a worthy research question.

[0063] The most common and effective approach for ocean in-situ measurements and satellite remote sensing data is to fuse them, leveraging their strengths and weaknesses to maximize their advantages and generate high-temporal and spatial resolution sea temperature analysis products. Data fusion is a dynamic process that requires continuously incorporating new observations into the model, taking into account factors such as the background and observation field errors, as well as the data's temporal and geographic distribution. As new observations are continuously incorporated into the model, the model's simulated prediction trajectory is gradually optimized and corrected, allowing the model data to continuously approach the real-world data trajectory. This ultimately improves model prediction accuracy and achieves greater consistency between model data and measured data. Numerous data processing methods have been developed over the course of data fusion, including stepwise correction algorithms, hybrid analysis algorithms, optimal interpolation algorithms, kriging, variational methods, and Kalman filtering. Each method has its own unique characteristics, advantages, and disadvantages. When selecting a research method, one can choose one based on specific research objectives.

[0064] SST is an extremely important parameter for the ocean, and obtaining high-temporal, high-spectral, and high-precision sea surface temperature products is crucial. Furthermore, the independently developed Fengyun (FY-3) and Ocean (HY-2) satellite series provide infrared and microwave remote sensing observations of global SST. Using this independently developed satellite data to produce high-precision SST reanalysis products and promote the development of domestic sea surface temperature data integration is a topic worthy of research.

[0065] Multi-source remote sensing data fusion requires addressing the theoretical framework, algorithms, and models for fusion of remote sensing data from multiple sensors with varying spectral and spatiotemporal resolutions, as well as establishing a performance evaluation index system for the fusion results. Currently, research on multi-source data fusion is growing, but most of this research focuses on specific application areas. Each approach establishes its own fusion principles based on the specific problem, achieving its own optimal solution. However, there is no consistent evaluation criteria for fusion results, making it impossible to accurately measure the final fusion level of the fusion product.

[0066] Overall, ocean remote sensing data inversion technology has made significant progress, and its products have been applied to a certain extent. However, compared with similar foreign satellites, the sea surface temperature products generated by domestic satellite sensors are still less mature and lack international recognition and application, which is also the gap with foreign satellite sensors. At the same time, although many scholars have conducted data fusion research, research on domestic independent satellite remote sensing data is still insufficient, which limits the further promotion and application of domestic satellite data. This application utilizes multi-scale assimilation technology, uses domestic satellite data as the input source, establishes a model system from data quality control, bias correction and data fusion, constructs a data fusion system, and produces multi-scale satellite sea surface temperature fusion products, which can provide technical and data support for remote sensing data application and scientific research.

[0067] The purpose of this application is to provide a multi-scale fusion method and system based on satellite sea surface temperature, which can accurately and effectively extract multi-scale information from satellite remote sensing observations of sea surface temperature and realize multi-scale information fusion of sea surface temperature.

[0068] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0069] Example 1

[0070] like Figure 1 This embodiment provides a multi-scale fusion method based on satellite sea surface temperature, including:

[0071] Step 101: Acquire a sea surface temperature dataset; data in the sea surface temperature dataset is acquired by multiple satellite sensors.

[0072] Step 102: performing data preprocessing on the sea surface temperature dataset to obtain a processed sea surface temperature dataset.

[0073] Step 103: Based on the measured sea surface temperature, perform a quality accuracy evaluation on each data in the processed sea surface temperature dataset; the evaluation indicators of the quality accuracy evaluation are composed of mean deviation, absolute deviation, root mean square error, standard deviation and correlation coefficient.

[0074] Step 104: Based on the quality accuracy evaluation results of each data and using a piecewise regression method, bias correction is performed on each data in the processed sea surface temperature dataset to obtain a corrected sea surface temperature dataset.

[0075] Step 105: Based on the corrected sea surface temperature dataset, a multi-scale fusion model is constructed using a multi-scale optimal interpolation method.

[0076] In some embodiments, when executing step 101, the specific steps may be as follows:

[0077] The acquired sea surface temperature dataset includes Fengyun satellite data, ocean satellite data, iQuam measured data, Argo buoy data, OISST data and OSTIA data.

[0078] In some embodiments, when executing step 102, the specific steps may be as follows:

[0079] For the Fengyun satellite data, the sea mark temperature data, satellite zenith angle and quality flag bits in the Fengyun satellite data are transposed or rotated, and the invalid land values, other invalid data, sea mark temperature points without valid zenith angle data, and data with a difference of more than 5°C in the OISST monthly average data are eliminated according to the mask data to obtain the processed Fengyun satellite data.

[0080] For the ocean satellite data, the sea surface temperature data in the ocean satellite data are transposed, and the invalid land values are eliminated according to the mask data, the remaining invalid data points are filled with NaN, and the data with the OISST monthly average data difference greater than 5℃ are obtained to obtain the processed ocean satellite data.

[0081] For the iQuam measured data, the daily average values of the latitude and longitude data, quality mark data, water depth data, observation time data, and sea surface temperature data in the iQuam measured data are solved to obtain the processed iQuam measured data.

[0082] For the Argo float data, the Argo data with a depth range of 5-6m from the sea surface are selected as the processed Argo float data.

[0083] For the OISST data, the sea mark temperature data and latitude and longitude information in the OISST data are transposed or rotated, and the invalid sea mark temperature data are eliminated according to the mask information to obtain the processed OISST data.

[0084] For OSTIA data, the sea mark temperature data and latitude and longitude information in the OSTIA data are transposed or rotated, and invalid sea mark temperature data are eliminated according to the mask information to obtain the processed OSTIA data.

[0085] Specifically, before using single-satellite sensor data, this embodiment requires a good understanding of it, clarifying the characteristics of the data, performing certain preprocessing and quality control on the data's outliers, filler values, and invalid values, and then matching statistics with the measured data to evaluate and analyze the accuracy of the SST.

[0086] Therefore, preprocessing focuses on understanding the basic information of individual stars, separating and reading the different elements in the file. Quality control uses separated land and sea masks, quality flags, and other necessary control screening criteria to remove filler values from the SST data and identify, judge, and eliminate outliers, ultimately retaining the data of good quality that meets the requirements.

[0087] (1) Fengyun satellite data preprocessing and quality control:

[0088] Because daytime SST is susceptible to warming due to sunlight, creating a diurnal thermocline, while nighttime SST does not warm, potentially leading to a large temperature difference between day and night, the daytime and nighttime data are processed separately. The FY-3CVIRR data file first reads the SST, satellite zenith angle, and quality flag within the study range. The data is then transposed or rotated to obtain a grid with the upper left corner at (100.025°E, 49.975°N) and the lower right corner at (159.975°E, 0.025°N), resulting in a grid resolution of 0.041667°. The valid range of SST values is -2°C to 35°C, and the scaling factor for conversion to actual SST is 0.01. The binary quality mark is converted to decimal, and data with the last two digits of 00, 01, and 10, i.e. excellent, good, and poor data, are selected. Invalid land values are eliminated based on the mask data, and NaN is filled in for the remaining invalid data or SST points without valid zenith angle data. Then, data with a difference of more than 5°C from the monthly average OISST data are eliminated. The daytime and nighttime data are processed in the same way, and finally the preliminary satellite data set is obtained.

[0089] To obtain FY-3C MWRI data, first read the SST and quality flags for the study area within the file and transpose the data. The grid is positioned at (100.125°E, 49.875°N) in the upper left corner and (159.875°E, 0.125°N) in the lower right corner, with a grid resolution of 0.25°. The valid SST range is -2°C to 35°C, and the conversion to actual SST is scaled by 0.01. Data with quality flags of 50 and 51 (excellent and poor) are selected. Invalid land values are removed using a mask, and remaining invalid data points are filled with NaN. Data with a difference of more than 5°C from the monthly mean OISST data are also removed. Data processing is identical for descending (daytime) and ascending (nighttime) orbits. This yields the final, preliminarily read satellite dataset. (Fengyun Data User Manual)

[0090] (2) Ocean satellite data preprocessing and quality control:

[0091] To retrieve the HY-2B RM data, you first need to read the SST data for the study area within the file and transpose the data. The upper left corner of the grid is (100.125°E, 49.875°N) and the lower right corner is (159.875°E, 0.125°N). The grid resolution is 0.25°. The valid range of SST values is -2°C to 35°C. The conversion to actual SST is scaled by a factor of 0.01. All valid observations are selected, and invalid land values are removed based on the mask data. The remaining invalid data points are filled with NaN. Data with a difference of more than 5°C from the monthly mean of OISST are also removed. The data processing methods for descending (daytime) and ascending (nighttime) are the same. Finally, the initial satellite dataset is obtained.

[0092] (3) iQuam measured data preprocessing and quality control:

[0093] As previously mentioned, this embodiment utilizes three types of measured data from iQuam: coastal anchored buoys, Argo buoys, and ship surveys. These data undergo quality verification by the system. For Argo buoy data, information such as latitude and longitude, quality flag, water depth, observation time, and sea surface temperature (SST) is extracted. SST data with a quality flag of 5 or higher, covering a valid SST range of -2°C to 40°C, are selected and stored separately during the day and at night. Coastal anchored buoy and ship survey data are also read and stored separately during the day and at night. Each data set is then tested by repeated observations at the same location on the same day. If multiple observations exist, the average of these observations is used to represent the observation at that point. The order of selected measured data is then set to Argo, coastal anchored buoy, and ship survey data. That is, if two or more types of measured data are available at the same location on the same day, they are selected in that order. Otherwise, the only valid observation at that point is selected. This results in a measured SST dataset.

[0094] (4) Argo float data preprocessing:

[0095] Since the amount of SST data from the Argo float at 0-1m is small and cannot meet the research needs, referring to the previous practices, this embodiment selects Argo data with a depth range of 5-6m, removes SST values exceeding -2℃-40℃, and generates a data set for backup.

[0096] (5) OISST data reading and preprocessing:

[0097] To create OISST data, you first need to read the latitude and longitude information and SST of the study area from the file and transpose the data so that the upper left corner of the grid is (100.125°E, 49.875°N) and the lower right corner is (159.875°E, 0.125°N), with a grid resolution of 0.25°. Invalid SST values are removed based on mask information, and the 30-year daily average climatological state from 1989 to 2018 and the monthly average from June to December 2019 are calculated to generate the dataset.

[0098] (6) OSTIA data reading and preprocessing:

[0099] The OSTIA data first reads the latitude, longitude and SST data of the study area in the file, and transposes the data to obtain the upper left corner of the grid (100.025°E, 49.975°N) and the lower right corner (159.975°E, 0.025°N). Invalid SST is eliminated based on mask information, and the monthly average values from June to September 2019 are calculated to generate a dataset for backup.

[0100] In some embodiments, when executing step 103, the specific steps may be as follows:

[0101] Accuracy evaluation requires selecting an appropriate spatiotemporal window, matching the measured iQuam data with the satellite data, using statistical methods, calculating certain evaluation indicators, and analyzing the differences between the satellite data and the measured SST, so as to achieve a certain evaluation effect on the satellite data.

[0102] When the single-star data is temporally and spatially matched with the measured data, the time nodes separating daytime and nighttime are 06:00-18:00 for the daytime period and 18:00-06:00 for the nighttime period.

[0103] The evaluation indicators are the following five items: mean deviation (Bias), absolute deviation (Abs_Bias), root mean square error (RMSE), standard deviation (STDE) and correlation coefficient (R), and their calculation formulas are:

[0104]

[0105]

[0106]

[0107]

[0108]

[0109] Among them, Bias is the average bias, Abs_Bias is the absolute bias, RMSE is the root mean square error, STDE is the standard deviation, R is the correlation coefficient, S i , I i represent the sea mark temperature data value inverted from infrared data and the measured sea mark temperature data of the i-th matching point, respectively, and n is the total number of matching points between satellite and measured data; All matching points S i , I i The mean of .

[0110] In some embodiments, when executing step 104, the specific steps may be as follows:

[0111] This embodiment adopts a segmented regression method to correct the SST deviation, and its error-related variables are obtained by analyzing the SST inversion algorithm.

[0112] The VIRR sea surface temperature operational product is based on a linear relationship model between the brightness temperature at the central wavelengths of 10.8 μm (CH4) and 12 μm (CH5) and the measured SST. The calculation formula is as follows:

[0113] T x =a0+a1T 11 +a2(T 11 -T 12 )+a3(T 11 -T 12 )(secθ-1).

[0114] Where: T x is the sea mark temperature retrieved from remote sensing, T 11 、T 12 represent the brightness temperatures at the central wavelength of 10.8 μm and 12 μm respectively, a0-a3 represent the regression coefficients, and θ represents the observation zenith angle.

[0115] In the infrared sensor sea mark temperature inversion principle, the product results are mainly related to the thermal infrared brightness temperature of 10.8μm and 12μm and the observation zenith angle θ. However, the inverted sea mark temperature product no longer contains brightness temperature data, so the inverted sea surface temperature T is used for bias correction. S Instead of the original brightness temperature CH4, the climatological daily mean sea mark temperature (T C ) instead of CH5, where T C It is obtained by averaging the 30-year (1989-2018) OISST daily analysis products with a resolution of 25 km. In the VIRR bias correction, the bilinear interpolation method is required to convert T CThe grid resolution is converted to 5 km, which matches the infrared SST. Then, a sample dataset (MD_all) for the entire study area is established for daytime and nighttime. MD_all contains the temporal and spatial information of sensors in the entire study area and the corresponding measured SST, T S 、T C and θ, the deviation correction formula is as follows:

[0116] T xz =b0+b1T s +b2(T s -T c )+b3(T s -T c )(sec θ-1).

[0117] Where: T xz is the corrected SST, b0~b3 are regression coefficients, T xz It can be regarded as [T s , T s -T c ,(T s -T c )(secθ-1)] is estimated by the linear combination of the three regression operators, which represent the original inverted SST, SST anomalies, and variables related to the observed zenith angle, respectively.

[0118] When performing bias correction, the estimation of the regression coefficient requires appropriate matching samples. This application first divides the study area into four sub-areas according to latitude. For each area, this application uses the time-related samples of the 15 days before the experiment to generate the training dataset for the bias correction pixels (Local Matchup Datasets, MD local ).

[0119] R=[T s , T s -T c ,(T s -T c )(secθ-1)] T .

[0120] D local =<(R- <r> )(R- <r>) T >.

[0121] Where: R represents the regression operator in vector form, D local is the covariance of the regression operator, and <·> is the mean of the vector. SST estimation error δT xz And the corresponding error variance V(δT xz ) can be expressed as:

[0122]

[0123] Where: c is the estimated regression coefficient, ρ represents the estimated standard deviation of the estimation error, which can be seen as R and <r>Distance in multidimensional space, for each independent sub-region, by selecting each pixel (i, j) corresponding to MD local The dataset {ρ local }|with a single ρ i，j The samples whose difference is less than a certain threshold S are regarded as the optimal associated matching samples of the pixel point (MD optimal ). This paper sets the initial S = 0.1, when the optimal sample size N optimal If it is less than 50, S is increased by 0.01 until the optimal matching point is all matching points. Then, the optimal sample (MD optimal ), estimate the regression coefficient of each pixel by the least square method, and finally use the calculated regression coefficient to combine the T corresponding to each pixel S 、T C and θ, and finally the bias-corrected sea surface temperature T xz .

[0124] The bias correction principle for microwave data is exactly the same as that for infrared data. The difference is that microwave data does not have a term related to the satellite zenith angle. The microwave sea surface temperature bias correction formula is as follows:

[0125] T xz =b0+b1T s +b2(T s -T c ).

[0126] Where: T xz , the physical meanings of b0~b2 are the same as above, T xz Can be used as a regression operator T s and (T s -T c ), T s and (T s -T c ) represent the SST inversion value and SST anomaly value respectively. s , T s -T c ] T Except for the difference in the infrared data deviation correction formula, the rest of the processing methods, threshold settings, and processes are exactly the same as those for infrared SST deviation correction, so I will not describe them here in detail. Figure 3 As shown, Figure 3 Noptical is the optimal number of samples, and Nlocal is the number of local samples.

[0127] In some embodiments, when executing step 104, the specific steps may be as follows:

[0128] This application uses a multi-scale optimal interpolation method. The analysis value of a spatial grid point is obtained by adding the modified SST increment to the background field value at that point. The modified value increment is obtained by weighted summation of the deviations between the valid observation values and the background field value around the spatial grid point. The weight coefficient here satisfies the condition of minimizing the analysis error of the grid point, rather than weights determined empirically or based on distance. Multi-scale information in the observation data is proposed through multiple iterations. The expression is as follows:

[0129]

[0130] Where: the subscripts m and s represent the spatial grid point and observation grid point locations, respectively; the superscripts a, b, and o represent the analysis value, background field value, and observation value, respectively. When the spatial grid resolutions of the observation point, background field point, and grid point are different, the bilinear interpolation operator is used to interpolate to the grid point resolution, then:

[0131]

[0132] H1 and H2 are bilinear interpolation operators. After data preprocessing, the weight coefficient matrix (gain matrix) K can be derived by solving equation (5-2):

[0133] K=BH T (HBH T +R)- 1 .

[0134] Where, the superscript T is the matrix transpose; B is the background field error covariance matrix, and R is the observation error covariance matrix.

[0135] When determining the covariance matrix B of the background error, the following two basic assumptions are often made: (1) B is constant; (2) the horizontal correlation of the background field error decreases exponentially with the increase of horizontal distance. Then:

[0136] B=D 0.5 ρD 0.5 .

[0137] Where: D is the diagonal matrix composed of background field errors, ρ is the horizontal correlation matrix of the predicted background field, and the matrix R is calculated using the same processing method as the background field error covariance matrix B.

[0138] When fusing SST data, it is assumed that the SST background error is anisotropic and the correlation decays faster in the longitude direction than in the latitude direction. The background error correlation function represents the error correlation between grid points i and j, as shown in the following formula:

[0139]

[0140] Where R ij represents the correlation coefficient, S i 、S j Represent the SST values of sites i and j respectively, and <> represents the vector average.

[0141] In this paper, the correlation distribution of background error uses the following formula for parameter fitting:

[0142]

[0143] Where, ρ ij represents the correlation coefficient between points i and j, r1 and r2 represent the distance in the latitude and longitude directions respectively, λ1 and λ2 represent the correlation scale in the latitude and longitude directions respectively. In this paper, λ1 = 200 km and λ2 = 150 km are selected.

[0144] Since the distribution of observation points is uneven, when calculating the weight coefficient K, (HBH T +R) -1 The inversion of often fails. Therefore, to avoid the inversion process, multiply both sides of Equation (13) by (HBH T + R), we get:

[0145] K(HBH T + R)=BH T .

[0146] Then transpose the matrices on both sides to get:

[0147] (HBH T +R) T K T =HB T .

[0148] After the above transformation processing, the problem of solving K is transformed into solving its transposed matrix, which can avoid the error in the matrix inversion. However, since the observation matrix is sparse and the distribution of elements in the matrix is irregular, the transposed matrix may still be ill-conditioned. To address this practical problem, this paper uses the singular value decomposition (SVD) method to calculate the least squares solution of the linear equations. The grid size of the fusion analysis field of this application is 500×600. When performing the optimal interpolation, it faces problems such as insufficient computer memory and insufficient SST data in some areas. In order to effectively solve the above problems, the grid-by-grid interpolation method is finally adopted. Point-by-point interpolation uses the idea of ​​point as the basic calculation unit to seek relevant points within the relevant radius, calculate the correlation between them, and obtain the fusion value of the point. It is then expanded in space row by row and column by column to calculate the analysis value of the analysis grid.

[0149] The singular value decomposition method is as follows:

[0150]

[0151] Among them, S m×n is a diagonal matrix. The SVD method can effectively eliminate abnormal solutions and meet the accuracy of equation solution.

[0152] After the data has been quality controlled and processed as above, the satellite sea surface temperature melting process is as follows: Figure 4 shown.

[0153] Example 2

[0154] like Figure 5 As shown, this embodiment provides a multi-scale fusion system based on satellite sea surface temperature, including:

[0155] The data acquisition module 501 is used to acquire a sea surface temperature dataset; the data in the sea surface temperature dataset is acquired by a variety of satellite sensors.

[0156] The preprocessing module 502 is configured to perform data preprocessing on the sea surface temperature dataset to obtain a processed sea surface temperature dataset.

[0157] The evaluation module 503 is used to perform quality accuracy evaluation on each data in the processed sea surface temperature dataset based on the measured sea surface temperature; the evaluation indicators of the quality accuracy evaluation are composed of mean deviation, absolute deviation, root mean square error, standard deviation and correlation coefficient.

[0158] The correction module 504 is configured to perform deviation correction on each data in the processed sea surface temperature dataset based on the quality accuracy evaluation results of each data and a piecewise regression method to obtain a corrected sea surface temperature dataset.

[0159] The fusion module 505 is used to construct a multi-scale fusion model based on the corrected sea surface temperature dataset using a multi-scale optimal interpolation method.

[0160] In summary, this application has the following technical effects:

[0161] 1) This application can comprehensively realize the quality control and deviation correction of domestically produced independent satellite remote sensing sea surface temperature. Due to the susceptibility to uncontrollable factors such as bad weather conditions and aging equipment, the obtained satellite remote sensing data is very different from the actual SST. In order to more comprehensively understand the accuracy of independent satellite remote sensing sea surface temperature observations, the outliers, filler values, and invalid values of the data must be pre-processed and quality controlled before use for single-star sensor data, and then matched with the measured data for statistics to evaluate and analyze the accuracy of SST. This application separates different elements, deletes the filler values in the SST data based on the separated sea and land masks, quality tags and other necessary control screening conditions, and identifies, judges and eliminates the outliers, and finally retains the data with better quality that meets the requirements. In addition, the SST inversion model does not fully and correctly reflect the relationship between the measured brightness temperature and the actual SST, and during the inversion process, the infrared is susceptible to incomplete cloud detection, and the microwave is susceptible to aerosols and land radiation. The inverted products are often not accurate and are very prone to outliers. This application adopts a segmented sample regression method to process the estimated error in segments, uses different matching samples according to the error size interval, calculates different regression coefficients using the least squares method, and performs bias correction. This method can effectively eliminate sensor statistical errors in sea surface temperature products based on the matched optimal measured samples.

[0162] 2) The method based on this application can effectively extract multi-scale information from satellite sea surface temperature and achieve multi-scale information fusion of sea surface temperature. Satellite remote sensing sea surface temperature data from different sources has different sampling resolutions and accuracies. This application proposes a multi-scale optimal interpolation method. By setting multiple search radii to cyclically extract multi-scale information from sea surface temperature, and setting influence weights based on the observation accuracy of different satellites, it achieves multi-scale information fusion of sea surface temperature, improving the effective resolution and accuracy of the fused sea surface temperature product.

[0163] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0164] The present embodiment uses specific examples to illustrate the principles and implementation methods of the present application. The description of the above examples is only intended to help understand the method and core concept of the present application. At the same time, for those skilled in the art, according to the concept of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present application.< / r> < / r> < / r>

Claims

1. A multi-scale fusion method based on satellite sea surface temperature, characterized in that: The multi-scale fusion method based on satellite sea surface temperature includes: Acquire a sea surface temperature dataset; data in the sea surface temperature dataset is acquired by multiple satellite sensors; performing data preprocessing on the sea surface temperature dataset to obtain a processed sea surface temperature dataset; Based on the measured sea surface temperature, performing a quality accuracy evaluation on each data in the processed sea surface temperature dataset; the evaluation indicators of the quality accuracy evaluation are composed of mean deviation, absolute deviation, root mean square error, standard deviation and correlation coefficient; According to the quality accuracy evaluation results of each data, based on the segmented regression method, each data in the processed sea surface temperature dataset is subjected to deviation correction to obtain a corrected sea surface temperature dataset; Based on the corrected sea surface temperature dataset, a multi-scale fusion model is constructed using the multi-scale optimal interpolation method. Among them, based on the corrected sea surface temperature dataset, a multi-scale optimal interpolation method is used to construct a multi-scale fusion model, which specifically includes: Obtaining a background field error covariance matrix; the background field error covariance matrix is composed of a diagonal matrix D composed of background field errors and a horizontal correlation matrix ρ of the predicted background field; Based on the background field error covariance matrix and the corrected sea surface temperature dataset, a spatial correlation function model is constructed; When fusing sea surface temperature data, calculate the background error correlation function of observation points of different data; Calculate the correlation coefficient between observation points based on the correlation distribution parameter fitting formula of background error; The formula expression of the background field error covariance matrix is: B=D 0.5 ρD 0.5 ; Where D is a diagonal matrix composed of background field errors, ρ is the horizontal correlation matrix of the predicted background field, and the matrix R is calculated using the same processing method as the background field error covariance matrix B; When fusing sea surface temperature data, the background error correlation function of the observation points of different data is calculated, including: According to the formula Calculate the background error correlation function of observation points of different data; Among them, R ij represents the correlation coefficient, S i 、S j denote the sea surface temperature values at stations i and j, respectively, and 〈〉 denotes the vector average; According to the background error correlation distribution parameter fitting formula, the correlation coefficient between the observation points is calculated, including: According to the correlation distribution parameter fitting formula Calculate the correlation coefficient between observation points; Among them, ρ ij represents the correlation coefficient between points i and j, r1 and r2 represent the distance in the latitude and longitude directions, respectively, and λ1 and λ2 represent the correlation scale in the latitude and longitude directions, respectively.

2. The multi-scale fusion method based on satellite sea surface temperature according to claim 1, characterized in that: The sea surface temperature data set includes Fengyun satellite data, ocean satellite data, iQuam measured data, Argo buoy data, OISST data and OSTIA data.

3. The multi-scale fusion method based on satellite sea surface temperature according to claim 2, characterized in that: Performing data preprocessing on the sea surface temperature dataset to obtain a processed sea surface temperature dataset specifically includes: For the Fengyun satellite data, the sea surface temperature data, satellite zenith angle and quality flag bits in the Fengyun satellite data are transposed or rotated, and invalid land values, other invalid data, sea surface temperature points without valid zenith angle data, and data with a difference of more than 5°C in the OISST monthly average data are eliminated based on the mask data to obtain the processed Fengyun satellite data; For ocean satellite data, the sea surface temperature data in the ocean satellite data is transposed, and invalid land values are eliminated according to the mask data, and the remaining invalid data points are filled with NaN, and the data with the difference of OISST monthly average data greater than 5℃ are obtained to obtain the processed ocean satellite data; For the iQuam measured data, solving the daily average value of the latitude and longitude data, quality mark data, water depth data, observation time data, and sea surface temperature data in the iQuam measured data to obtain the processed iQuam measured data; For Argo float data, the Argo data with a depth range of 5-6m from the sea surface were selected as the processed Argo float data; For OISST data, the sea surface temperature data and latitude and longitude information in the OISST data are transposed or rotated, and invalid sea surface temperature data are eliminated according to the mask information to obtain the processed OISST data; For OSTIA data, the sea surface temperature data and latitude and longitude information in the OSTIA data are transposed or rotated, and invalid sea surface temperature data are eliminated according to the mask information to obtain the processed OSTIA data.

4. The multi-scale fusion method based on satellite sea surface temperature according to claim 3, characterized in that: The formula expressions for the mean deviation, absolute deviation, root mean square error, standard deviation and correlation coefficient in the evaluation indicators are as follows: Among them, Bias is the average bias, Abs_Bias is the absolute bias, RMSE is the root mean square error, STDE is the standard deviation, R is the correlation coefficient, S i , I i represent the sea surface temperature data value inverted from infrared data and the measured sea surface temperature data of the i-th matching point, respectively, and n is the total number of matching points between satellite and measured data; All matching points S i , I i The mean of .

5. The multi-scale fusion method based on satellite sea surface temperature according to claim 1, characterized in that: According to the quality and accuracy evaluation results of each data, based on the segmented regression method, each data in the processed sea surface temperature dataset is subjected to deviation correction to obtain a corrected sea surface temperature dataset, specifically including: According to the formula T xz =b0+b1T s +b2(T s -T c ), performing bias correction on each data in the sea surface temperature dataset; Among them, T xz is the corrected sea surface temperature, b0, b1, b2 are regression coefficients, T s and (T s -T c ) are the sea surface temperature inversion value and the sea surface temperature anomaly value, respectively.

6. A multi-scale fusion system based on satellite sea surface temperature, characterized by: include: A data acquisition module is used to acquire a sea surface temperature data set; the data in the sea surface temperature data set is acquired by a variety of satellite sensors; a preprocessing module, configured to perform data preprocessing on the sea surface temperature dataset to obtain a processed sea surface temperature dataset; An evaluation module is configured to perform a quality accuracy evaluation on each data in the processed sea surface temperature dataset based on the measured sea surface temperature; the evaluation indicators of the quality accuracy evaluation are composed of mean deviation, absolute deviation, root mean square error, standard deviation and correlation coefficient; a correction module for performing deviation correction on each data in the processed sea surface temperature dataset based on a piecewise regression method according to a quality accuracy evaluation result of each data, to obtain a corrected sea surface temperature dataset; The fusion module is used to construct a multi-scale fusion model based on the corrected sea surface temperature dataset using a multi-scale optimal interpolation method; Among them, based on the corrected sea surface temperature dataset, a multi-scale optimal interpolation method is used to construct a multi-scale fusion model, which specifically includes: Obtaining a background field error covariance matrix; the background field error covariance matrix is composed of a diagonal matrix D composed of background field errors and a horizontal correlation matrix ρ of the predicted background field; Based on the background field error covariance matrix and the corrected sea surface temperature dataset, a spatial correlation function model is constructed; When fusing sea surface temperature data, calculate the background error correlation function of observation points of different data; Calculate the correlation coefficient between observation points based on the correlation distribution parameter fitting formula of background error; The formula expression of the background field error covariance matrix is: B=D 0.5 ρD 0.5 ; Where D is a diagonal matrix composed of background field errors, ρ is the horizontal correlation matrix of the predicted background field, and the matrix R is calculated using the same processing method as the background field error covariance matrix B; When fusing sea surface temperature data, the background error correlation function of the observation points of different data is calculated, including: According to the formula Calculate the background error correlation function of observation points of different data; Among them, R ij represents the correlation coefficient, S i 、S j denote the sea surface temperature values at stations i and j, respectively, and 〈〉 denotes the vector average; According to the background error correlation distribution parameter fitting formula, the correlation coefficient between the observation points is calculated, including: According to the correlation distribution parameter fitting formula Calculate the correlation coefficient between observation points; Among them, ρ ij represents the correlation coefficient between points i and j, r1 and r2 represent the distance in the latitude and longitude directions, respectively, and λ1 and λ2 represent the correlation scale in the latitude and longitude directions, respectively.

Images