A method and device for reconstructing sea ice data based on multi-source remote sensing images

Abstract

The present invention relates to the technical field of sea ice and ice condition data reconstruction, and in particular to a method and device for reconstructing sea ice and ice condition data based on multi-source remote sensing images. The method comprises: acquiring remote sensing image data of sea ice in a target sea area; preprocessing the remote sensing image data to obtain corrected remote sensing image data; obtaining sea ice area data and sea ice thickness data based on the corrected remote sensing image data; and obtaining a data set of the historical spatial distribution area and thickness of sea ice in the target sea area based on the sea ice area data and the sea ice thickness data. The above scheme of the present invention, based on remote sensing image data, extracts sea ice area data and sea ice thickness data, and then reconstructs a data set of the historical spatial distribution area and thickness of sea ice in the target sea area, which can provide technical support for analyzing the spatiotemporal distribution characteristics of sea ice and future sea ice risk assessment, and provide data support for the study of the dynamic ablation mechanism of long-term sea ice.

Description

Technical Field

[0001] The present invention relates to the technical field of sea ice and ice condition data reconstruction, and in particular to a method and device for reconstructing sea ice and ice condition data based on multi-source remote sensing images. Background Art

[0002] Traditional on-site observations of sea ice are easily restricted by site conditions, and the sea ice data and information obtained are limited. The development of satellite remote sensing technology has greatly enriched sea ice data and monitoring methods. The rapid development of satellite remote sensing observations and the increasing maturity of data extraction methods have greatly improved the temporal and spatial accuracy of sea ice monitoring. The extraction of ice data using remote sensing data has further promoted the progress of research related to sea ice observation. Visible light and infrared satellite sensors are widely used in the extraction of sea ice data due to their advantages such as high temporal and spatial resolution and easy reading. Sea ice data provide technical support for analyzing the temporal and spatial distribution characteristics of sea ice and future sea ice risk assessment, and provide data support for the study of long-term sea ice dynamic ablation mechanism. However, satellite sensors are easily obscured by clouds, etc., and there is a situation of poor image quality. The extracted sea ice daily-scale spatial data is missing, especially the sea ice data in early years is obviously missing. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a method and device for reconstructing sea ice condition data based on multi-source remote sensing images, so as to solve the problem of missing historical sea ice condition data.

[0004] In order to solve the above technical problems, the technical solution of the present invention is as follows:

[0005] According to one aspect of the present invention, a method for reconstructing sea ice condition data based on multi-source remote sensing images is provided, comprising:

[0006] Obtain remote sensing image data of sea ice in the target sea area;

[0007] Preprocessing the remote sensing image data to obtain corrected remote sensing image data;

[0008] Obtaining sea ice area data and sea ice thickness data based on the corrected remote sensing image data;

[0009] According to the sea ice area data and the sea ice thickness data, a data set of sea ice historical spatial distribution area and thickness in the target sea area is obtained.

[0010] Further, the remote sensing image data is preprocessed to obtain corrected remote sensing image data, including:

[0011] The remote sensing image data is subjected to geometric correction, format conversion and atmospheric correction to obtain corrected remote sensing image data.

[0012] Furthermore, sea ice area data and sea ice thickness data are obtained based on the corrected remote sensing image data, including:

[0013] Extracting the corrected remote sensing image data using a partition threshold method or a sample-based object-oriented classification method to obtain sea ice area data;

[0014] The corrected remote sensing image data is inverted using the logarithmic relationship of the sea ice albedo to obtain sea ice thickness data.

[0015] Furthermore, according to the sea ice area data and the sea ice thickness data, a data set of sea ice historical spatial distribution area and thickness in the target sea area is obtained, including:

[0016] Obtain meteorological data along the coast of the target sea area;

[0017] According to the sea ice area data, the sea ice thickness data and the meteorological data, a spatial autocorrelation model is used to obtain a data set of the historical spatial distribution area and thickness of sea ice in the target sea area.

[0018] Furthermore, according to the sea ice area data, the sea ice thickness data and the meteorological data, a spatial autocorrelation model is used to obtain a data set of sea ice historical spatial distribution area and thickness in the target sea area, including:

[0019] Obtaining a weight coefficient according to the sea ice area data, the sea ice thickness data and the meteorological data;

[0020] The sea ice area data, the sea ice thickness data, the weight coefficient and the meteorological data are used to obtain the weight coefficient and substitute it into the spatial autocorrelation model to obtain the historical spatial distribution area and thickness data set of sea ice in the target sea area.

[0021] Further, the sea ice area data, the sea ice thickness data, the weight coefficient and the meteorological data are substituted into the spatial autocorrelation model to obtain a data set of the historical spatial distribution area and thickness of sea ice in the target sea area, including:

[0022] Obtaining a kilometer grid of the target sea area, wherein the kilometer grid includes a plurality of pixel unit grids;

[0023] According to the sea ice area data, the sea ice thickness data and the kilometer grid, the sea ice area and the average sea ice thickness in each pixel unit grid are obtained;

[0024] According to the meteorological data and the kilometer grid, obtaining the temperature data in each pixel unit grid;

[0025] The sea ice area and average sea ice thickness in the pixel unit grid, the temperature data in the pixel unit grid, and the weight coefficient are input into the kilometer grid to obtain a data set of sea ice historical spatial distribution area and thickness in the target sea area.

[0026] Further, according to the meteorological data and the kilometer grid, the temperature data in each pixel unit grid is obtained, including:

[0027] The meteorological data is extracted into the kilometer grid using the inverse distance weighted method to obtain the temperature data in each pixel unit grid.

[0028] Furthermore, the method further comprises:

[0029] The receiver operating characteristic curve and the precision-recall curve are used to verify the sea ice historical spatial distribution area and thickness dataset of the target sea area to obtain the verification results.

[0030] According to another aspect of the present invention, there is provided a device for reconstructing sea ice data based on multi-source remote sensing images, comprising:

[0031] The acquisition module is used to obtain the remote sensing image data of sea ice in the target sea area and send it to the preprocessing module;

[0032] A preprocessing module is used to preprocess the remote sensing image data to obtain corrected remote sensing image data and send the corrected remote sensing image data to the extraction module;

[0033] An extraction module, used to obtain sea ice area data and sea ice thickness data according to the corrected remote sensing image data, and send them to a reconstruction module;

[0034] The reconstruction module is used to obtain a data set of the historical spatial distribution area and thickness of sea ice in the target sea area based on the sea ice area data and the sea ice thickness data.

[0035] The above solution of the present invention includes at least the following beneficial effects:

[0036] The above scheme of the present invention, based on the remote sensing image data of sea ice in the target sea area, extracts sea ice area data and sea ice thickness data, and then reconstructs the historical spatial distribution area and thickness data set of sea ice in the target sea area, which can provide technical support for analyzing the spatiotemporal distribution characteristics of sea ice and future sea ice risk assessment, and provide data support for the study of long-term sea ice dynamic ablation mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a step diagram of the method for reconstructing sea ice condition data based on multi-source remote sensing images of the present invention;

[0038] Figure 2It is a workflow diagram of the reconstruction method of sea ice condition data based on multi-source remote sensing images;

[0039] Figure 3 It is a representation diagram of the spatial autocorrelation model based on thermodynamic and kinetic factors;

[0040] Figure 4 This is a schematic diagram of kilometer grid extraction for sea ice reconstruction;

[0041] Figure 5 is the meteorological indicator T ave Schematic diagram of spatial interpolation;

[0042] Figure 6 This is an example of a connectivity grid of 8 given target kilometer grid points in the sea ice grid data;

[0043] Figure 7 It is a device connection diagram of the device for reconstructing sea ice condition data based on multi-source remote sensing images of the present invention.

[0044] Description of reference numerals:

[0045] Figure 4 Middle: 41- sea ice, 42- weather station, 43- target sea area;

[0046] Figure 5 Middle: 51 – area with higher average daily temperature, 52 – area with lower average daily temperature. DETAILED DESCRIPTION

[0047] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0048] like Figure 1 As shown, an embodiment of the present invention proposes a method for reconstructing sea ice data based on multi-source remote sensing images, comprising:

[0049] Step S1, obtaining remote sensing image data of sea ice in the target sea area;

[0050] Step S2, preprocessing the remote sensing image data to obtain corrected remote sensing image data;

[0051] Step S3, obtaining sea ice area data and sea ice thickness data according to the corrected remote sensing image data;

[0052] Step S4: obtaining a data set of historical spatial distribution area and thickness of sea ice in the target sea area based on the sea ice area data and the sea ice thickness data.

[0053] The above scheme of this embodiment, based on the remote sensing image data of sea ice in the target sea area, extracts sea ice area data and sea ice thickness data, and then reconstructs the historical spatial distribution area and thickness data set of sea ice in the target sea area, which can provide technical support for analyzing the spatiotemporal distribution characteristics of sea ice and future sea ice risk assessment, and provide data support for the study of long-term sea ice dynamic ablation mechanism.

[0054] In an optional embodiment of the present invention, step S2 includes:

[0055] The remote sensing image data is subjected to geometric correction, format conversion and atmospheric correction to obtain corrected remote sensing image data.

[0056] In an optional embodiment of the present invention, step S3 includes:

[0057] Extracting the corrected remote sensing image data using a partition threshold method or a sample-based object-oriented classification method to obtain sea ice area data;

[0058] The corrected remote sensing image data is inverted using the logarithmic relationship of the sea ice albedo to obtain sea ice thickness data.

[0059] In an optional embodiment of the present invention, step S4 includes:

[0060] Step S41, obtaining meteorological data along the coast of the target sea area;

[0061] Step S42: according to the sea ice area data, the sea ice thickness data and the meteorological data, a spatial autocorrelation model is used to obtain a data set of the historical spatial distribution area and thickness of sea ice in the target sea area.

[0062] In an optional embodiment of the present invention, step S42 includes:

[0063] Step S421, obtaining a weight coefficient according to the sea ice area data, the sea ice thickness data and the meteorological data;

[0064] Step S422: Substitute the sea ice area data, the sea ice thickness data, the weight coefficient and the meteorological data to obtain the weight coefficient into the spatial autocorrelation model to obtain the sea ice historical spatial distribution area and thickness data set of the target sea area.

[0065] In an optional embodiment of the present invention, step S422 includes:

[0066] Step S4221, obtaining a kilometer grid of the target sea area, wherein the kilometer grid includes a plurality of pixel unit grids;

[0067] Step S4222: obtaining the sea ice area and average sea ice thickness in each pixel unit grid according to the sea ice area data, the sea ice thickness data and the kilometer grid;

[0068] Step S4223, obtaining the temperature data in each pixel unit grid according to the meteorological data and the kilometer grid;

[0069] Step S4224, input the sea ice area and average sea ice thickness in the pixel unit grid, the temperature data in the pixel unit grid, and the weight coefficient into the kilometer grid to obtain a data set of the historical spatial distribution area and thickness of sea ice in the target sea area.

[0070] In an optional embodiment of the present invention, step S4223 includes:

[0071] The meteorological data is extracted into the kilometer grid using the inverse distance weighted method to obtain the temperature data in each pixel unit grid.

[0072] In an optional embodiment of the present invention, the method further comprises:

[0073] Step S5: Use the receiver operating characteristic curve and the precision-recall curve to verify the sea ice historical spatial distribution area and thickness dataset of the target sea area to obtain a verification result.

[0074] like Figure 2 The specific process of the method for reconstructing sea ice data based on multi-source remote sensing images proposed in the present invention is as follows:

[0075] Considering the availability and accessibility of data, the sea ice data used in this embodiment is derived from satellite remote sensing images. After preprocessing the satellite remote sensing data, the sea ice area and sea ice thickness can be extracted respectively. Among them, the satellite images used for remote sensing extraction and inversion of sea ice data mainly include AVHRR (very high resolution radiometer), MODIS (moderate resolution imaging spectrometer), and GOCI (geostationary ocean color imager), with a time range from December to March of the following year in winter, and a spatial range covering the entire target sea area. Preprocessing of satellite remote sensing data can control the quality of remote sensing images, eliminate data with poor quality or cloud cover, and retain high-quality images for subsequent preprocessing. Preprocessing mainly includes operations such as data format conversion, atmospheric correction, and geometric correction. Then the sea ice area is extracted, mainly based on threshold method and sample-based object-oriented classification method, supplemented by visual interpretation to correct classification errors. Sea ice thickness inversion is based on the albedo semi-empirical logarithmic model. Finally, with the help of sea ice area data obtained from visual interpretation of Landsat TM satellite (Landsat) images and sea ice thickness data from field observations, a dataset of historical non-continuous daily-scale spatial distribution area and thickness of the target sea area over the past N years was obtained.

[0076] The preprocessing of AVHRR data first converts the format through the NOAA data analysis and processing system, then eliminates the error of the atmosphere on the reflectivity of the remote sensing image through atmospheric correction, and finally uses geometric correction to eliminate geometric deformation. The preprocessing of MOSDIS data includes geometric correction, format conversion and atmospheric correction. Geometric correction is performed using the ENVI5.1 (remote sensing image processing platform) tool. Finally, the radiative transfer model module of ENVI is used for atmospheric correction. GOCI preprocessing mainly includes radiation calibration, geometric correction and atmospheric correction. Radiation calibration is completed by ENVI software by converting the original image value DN into radiation brightness. The geometric correction method is the geographic location lookup table file, and the atmospheric module is completed using the FLAASH module (atmospheric correction module) of ENVI software. L = Gain*DN+Offset, where Gain is the magnification factor, DN is the original image value, and Offset is the offset. Among them, Gain and Offset are 10-6 and 0 respectively. TM satellite image data undergoes preprocessing such as geometric correction, format conversion and atmospheric correction.

[0077] After preprocessing the satellite remote sensing data, area extraction and thickness inversion can be performed. The sea ice area extraction of AVHRR is mainly based on the partition threshold method, that is, setting thresholds in different partitions to distinguish sea ice information. This method can reduce the impact of sediment suspension areas on the extraction of sea ice area information. First, the difference between land and sea is distinguished by near-infrared spectroscopy, and then the albedo (band 2) and brightness temperature (band 4) are used to distinguish sea ice and sea water. The threshold partition formula for sea ice and sea water is as follows: Sea ice: band 2>T2 and band 4<T4, where T2 and T4 are ice-water distinction thresholds, and the range of T2 in different regions is 0.08 to 0.14, and the range of T4 is 266.0 to 269.0. The sea ice area extraction of MODIS and GOCI is based on sample-oriented object-oriented classification, with an accuracy of 90-95%. With the help of ENVI5.1 software, the remote sensing image is segmented into block units, and sea ice, sea water, land and clouds are selected as learning samples. The K-nearest neighbor method is used for image classification because it has less anomalies and noise. Finally, each image is completely corrected by visual interpretation to obtain the sea ice area. For the Landsat TM image used to verify the accuracy, since its spatial resolution is nearly 30m, the distinction between sea ice and sea water is extremely high, and the sea ice area can be extracted using the visual interpretation method. The inversion principle of sea ice thickness is based on the good logarithmic relationship of sea ice albedo. The semi-empirical logarithmic model expression of ice thickness albedo is as follows:

[0078]

[0079] k=1-α sea / α max

[0080] Where h is the thickness of sea ice; α h is the sea ice albedo; α max is the albedo when the ice thickness is infinite; k is the correlation coefficient, μ α is the observed albedo attenuation coefficient; α max , α sea , μ α Select 0.700, 0.100 and 1.209 respectively.

[0081] The relationship between sea ice thickness and albedo from AVHRR, MODIS and GOCI can be expressed as:

[0082] α h =0.423α1+0.577α2

[0083] α h =0.160β1+0.291β2+0.243β3+0.116β4+0.112β5+0.0.081β7-0.0015

[0084] α h=-0.1356γ1-0.2704γ2+1.4087γ3-0.3284γ4-0.0812γ5+0.6204γ6-0.1474γ7-0.0268γ8-0.0464

[0085] Where α1 and α2 are the values ​​of the first and second channels of AVHRR, β1-β5, β7 are the sea ice albedo of MODIS bands 1-5 and 7, and γ1-γ8 are the albedo of GOCI image bands 1-8.

[0086] The generation and disappearance process of sea ice in the target sea area during the winter ice age is significantly affected by thermodynamic and kinetic factors. This embodiment establishes a reconstruction model based on thermodynamics and kinetics, such as Figure 3 . First, from a thermodynamic perspective, commonly used meteorological indicators are selected to try to characterize thermodynamic influencing factors. Secondly, from a dynamic perspective, there is no significant correlation with wind speed, which may be because the wind speed measured at the station is the average of multiple times on the day, and the instantaneous wind speed in the wind speed characterization is more in line with the law of sea ice formation and drift. Since the types of high-precision long-time series data that can be obtained are limited, and the spatial distribution of sea ice is relatively greatly affected by neighboring sea ice, considering the number of model calculations and computing power, the 8-connectivity kilometer grid sea ice situation of the target kilometer grid itself and its surrounding neighbors is selected to characterize the spatial dynamic factors, that is, the interaction between the sea ice in the target grid and the surrounding 1 km sea ice, or the interaction with the sea water when there is no ice, such as friction and collision with neighboring sea ice.

[0087] When selecting meteorological indicators, the temperature daily data of the coastal meteorological station 42 are used. According to the distribution of meteorological stations along the coast of the target sea area 43 and the time series of meteorological elements, 17 coastal stations are selected, such as Figure 4 The time series of data from each meteorological station is consistent with the time series of sea ice age data, which is the winter ice age (December to March of the following year) in the past N years, and the data type is daily average temperature data. Eight temperature indicators are selected, namely daily average temperature (Tave), cumulative freezing degree days (CFDD), cumulative melting degree days (CTDD), the difference between CFDD and CTDD (θ), light ice year freezing temperature comprehensive index (Light_freezing), light ice year melting temperature comprehensive index (Light_melting), heavy ice year freezing temperature comprehensive index (Heavy_freezing), and heavy ice year melting temperature comprehensive index (Heavy_melting). Among them, CFDD and CTDD are the cumulative values ​​of the temperature difference between the daily average temperature below or above the freezing point, respectively, and e is the difference between the two. Both are important parameter indicators for measuring how cold the weather is and how long it lasts; the other four light (heavy) ice year freezing (melting) indices refer to Yan et al., and the calculation formula is as follows:

[0088]

[0089]

[0090] θ=CFDD-CTDD

[0091] Light_-Freezing

[0092] =-0.2290T ave -0.0919T min +1.2008CFDD-0.8186θ+0.4963CFDD 2 +0.3619θ 2 +0.4977

[0093] Light-_Melting

[0094] =0.1179T ave -0.9126T min -0.7242CTDD+1.0260CFDD 2 +0.7655

[0095] Heavy_Freezing

[0096] =0.2109T ave -0.4219T min +1.1318CFDD-0.5428θ+1.4826CFDD 2 -1.4811θ 2 +0.1476

[0097] Heavy_Melting

[0098] =0.2277T ave -1.0621T min -0.0987CTDD+0.0444CFDD 2 +0.7277

[0099] Among them, CFDD is the cumulative freezing degree day (degree / day), CTDD is the cumulative melting degree day (degree / day), and θ is the difference between the two. These indicators are the cumulative values ​​of the average temperature below or above the freezing point temperature in a certain period of time. They are important parameters for measuring the degree and duration of cold weather; T ave is the daily average temperature (degrees), T min The lowest temperature of the day (degrees), calculated based on the average daily temperature of 17 meteorological stations along the coast of the target sea area; T f is the critical temperature of sea ice freezing and melting in the target sea area, selected as -2 degrees; t is the natural day from the first ice day; d s and d eare the start and end dates, respectively.

[0100] Sea ice data reconstruction is based on the pixel unit grid, and a 1km×1km grid is drawn for the target sea area, such as Figure 4 , the area and thickness data obtained from the remote sensing images of sea ice 41 in the past N years were extracted into kilometer grids, and the sea ice area and average sea ice thickness in each grid pixel were obtained. In addition, the temperature data of 8 selected meteorological stations in multiple countries along the coast of the target sea area were calculated respectively, and the station data were interpolated into spatial surface data with a resolution of 1km by using the inverse distance weighted method of ArcMap10.6 (geographic information management software), as shown in Figure 5 T ave Schematic diagram of spatial interpolation, then extract the temperature index data to the kilometer grid to obtain the temperature data value in each grid unit.

[0101] Construct a kilometer grid with 500 rows and 500 columns, totaling 250,000 grid points, to ensure that the grid covers the target sea area. There are 76,784 kilometer grids in the target sea area. Write serial numbers, row numbers, and column numbers for the 250,000 grids, which will be used to mark and extract grid points in the subsequent model. The sea ice data is N years in total, recorded as Y. td，ij is the sea ice area (or thickness) value in the kilometer grid of year t, date d, row i, column j. Here: t∈T is the set of historical sea ice years; d∈D is the set of dates in the winter sea ice period, usually the ice period of the current year, such as December 1 to March 20 of the following year; i, j∈{1, 2, ..., 500} are the row and column numbers of the kilometer grid. are the date sets corresponding to the sea ice remote sensing observation and the sea ice-free remote sensing observation in year t. Let t0∈T be the given sea ice year to be reconstructed, d0∈D 0，mis is a date with no sea ice remote sensing observations in the given year to be reconstructed. The spatiotemporal autocorrelation reconstruction model formula is as follows:

[0102]

[0103] Where t is the year, d is the date, i is the row number, j is the column number, 0≤w td，ij ≤1 is the weight coefficient, satisfying

[0104] Indicates the historical year set of sea ice data used to fill the target year t0; for example, when t0 = n, you can take or It is also possible to take an asymmetric year combination or all years of the target year and use all years to reconstruct the data.

[0105] Indicates that the target date d0 is For the adjacent dates with sea ice observations in the year, the algorithm with at most one adjacent date before and after is used. The algorithm with two or more adjacent dates before and after can be considered. For example, d0 = year a, month b, day c, t = a-2 / a-1, and in year a-2 / a-1, the closest dates with sea ice observations before and after month b, day c, are c-2 and c+5 respectively. Then N a-2，a年b月c日 ={a-1 year b month c-2 day, a-1 year b month c+5 day}. It represents the adjacent grid corresponding to the target grid i0j0 for a given year t and date d. Usually it takes "1st order", that is, a circle of adjacent grid points with 8 connectivity, that is, If i0j0 is the interior grid, then There are a total of 9 grid cells, including 8 neighboring grids whose intersection with the target grid is not empty and the target grid itself, such as Figure 6 In order to speed up the calculation and avoid the extra judgment and calculation required when the i0j0 grid point is at the edge or the four corners, only the grid points with l, j∈{2,…,499} in the entire grid are selected for calculation. At the same time, the data of all dates are spliced ​​into a matrix, and the matrix multiplication is used for unified operation, and finally converted into a 498×498 kilometer grid.

[0106] The selection of weight coefficients in the autocorrelation reconstruction model directly affects the results of sea ice data completion, considering that sea ice changes are mainly directly affected by thermodynamics and dynamics, and the sea ice characteristics of each unit grid are similar to those of the surrounding neighboring grids. When calculating the weight coefficient, it is mainly defined by the two elements of input meteorological indicators and ice level. It depends on the absolute value of the difference between the characteristic indicators of the grid unit to be reconstructed and the input neighboring remote sensing sea ice data, and the absolute value of the difference between the ice level of the year to which the reconstructed grid belongs and the year of the input borrowed data. In addition, the exponential function is ideal for sea ice fitting. Therefore, the weight calculation formula is as follows:

[0107]

[0108] Where t is the year, d is the date, i is the row number, j is the column number, g is the ice level, T is the meteorological index, and λ is the model parameter.

[0109] It can be seen from the weight formula that when λ is a positive value and the temperature index difference and other characteristics remain unchanged, the closer the ice level is to the year to be reconstructed, the greater the weight of the neighboring grid, and vice versa. When the ice level is fixed, the smaller the difference between the meteorological index of the grid to be reconstructed and the input data, the greater the corresponding weight, and vice versa, that is, the smaller the relative difference of the temperature index and the closer the ice level, the greater the weight coefficient. Finally, based on the sea ice remote sensing observation data and meteorological index data, the different weight coefficients obtained are normalized and characterized, and brought into the model formula to obtain the simulated sea ice data results.

[0110] The autocorrelation model used in this embodiment involves the mediation parameter λ, which is used to characterize the proportion between the relative difference of meteorological indicators and the relative difference of ice level. The local optimality of the parameter λ is determined by grid search, and the parameter λ with different values ​​is substituted. The simulation results of the model are obtained by simulating the model with the help of curve values ​​and trends, and the λ value corresponding to the optimal evaluation index is selected as the result model. From the sea ice data of the target sea area in the past N years, the sea ice area and thickness data that can be extracted and inverted with the help of remote sensing satellite data are limited, only more than one thousand days, but the complete ice age data involved in the past 32 years is about more than three thousand days. In order to avoid the large deviation of the model simulation results caused by too few data sets available for training, the accuracy should be guaranteed as much as possible when selecting parameter values. In the limited remote sensing data training set, as much training as possible is used with the help of existing data. Therefore, the Leave-One-Out method in the cross-validation method is adopted, that is, only one day of data is used as test data in each simulation, and all other remote sensing observation data are included in the training sample set to achieve the simulation results closest to the expected value of the entire test set.

[0111] Parameter selection and model evaluation mainly rely on the receiver operating characteristic curve (ROC curve), which is a tool for analysis and verification using coordinate mapping and a common statistical method for intuitive evaluation. In practical applications, the ROC curve can be used to select the best prediction model, that is, to judge and select the optimal parameters through the curve, thereby obtaining the optimal parameter values ​​of the model.

[0112] From the definition of the ROC curve, we can know that each point on the curve represents a threshold. Each pixel in the prediction result map corresponds to the probability or area size of sea ice predicted by a model. The prediction results are filtered using this threshold. When it is higher than this threshold, the results are retained, and when it is lower than this threshold, it is set to 0. By selecting the threshold using the Youden index method, the optimal screening threshold for the prediction results can be obtained, so that the model prediction is more accurate. The Youden index method is a method that calculates the vertical distance from each point on the ROC curve to the y=x line and selects its maximum value. The formula is: Youdenindex=TPR+FPR-1. Usually, when the number of positive and negative samples is sufficient, the ROC curve can effectively reflect the prediction ability of the model. However, since the X-axis of the ROC curve uses FPR, in the case of imbalanced categories, the distinction between the model curves is small, resulting in poor results when the ROC curve is used for model evaluation. The grid cells with sea ice are taken as positive samples, and the grid cells without sea ice are taken as negative samples. The average number of positive samples over the years is about 10436 grids. In comparison, the average number of positive samples is generally much smaller than the 66348 grids of negative samples, and the category distribution is uneven. Therefore, the PR curve (Precision-Recall) is introduced as another evaluation index to select the optimal parameters and thresholds to ensure the accuracy of parameter selection and model evaluation. The AUC value of PR is used as the standard for evaluating the classification effect to obtain the optimal parameters of various artificial features, and then its ROC curve is calculated. The Youden index method is used to calculate the optimal threshold. The optimal threshold is selected to make the FPR as small as possible and the TPR as large as possible, and then the optimal threshold in data prediction is obtained. The grids with prediction results higher than the optimal threshold are judged to have ice.

[0113] The model for sea ice area and sea ice thickness was reconstructed based on eight meteorological indicators, namely, daily average temperature (Tave), cumulative freezing degree days (CFDD), cumulative melting degree days (CTDD), the difference between CFDD and CTDD (θ), the comprehensive index of freezing temperature in light ice years (Light_freezing), the comprehensive index of melting temperature in light ice years (Light_melting), the comprehensive index of freezing temperature in heavy ice years (Heavy_freezing), and the comprehensive index of melting temperature in heavy ice years (Heavy_melting). For each different meteorological indicator, different values ​​of parameter λ were also substituted, and the model simulation and result evaluation were carried out for the annual sea ice area and sea ice thickness data, and the PR curve and ROC curve were obtained respectively, and the area under their curves AUC was calculated to select the optimal value. When selecting the parameter range, first, based on the preliminary simulation and parameter evaluation results, the parameter values ​​were determined to be from 0.001 to 80. When the parameter was less than 0.5, a total of five values ​​were selected: 0.001, 0.0025, 0.005, 0.01, and 0.25; when the parameter was greater than or equal to 0.5, 0.5, 1.0, 1.5, 2.0, etc. were selected with a span unit of 0.5. As the parameter values ​​changed, the AUC values ​​also showed a "single peak" curve that first increased and then decreased. Therefore, when selecting the parameter range for different meteorological indicators in different years, the parameter range was further narrowed with the help of the basic AUC value change trend.

[0114] It can be seen that as the parameter value increases, the AUC results calculated by the ROC curve and the PR curve show a "single peak" change, first increasing to the highest value and then gradually decreasing. Therefore, when simulating the sea ice area and sea ice thickness of the target sea area, the parameter threshold of the index simulation is selected in the range of 0 to 20, and the corresponding different parameter λ values ​​are 0.001, 0.01, 0.5, 1.0, 1.5...19.5, 20.0, respectively. There are 42 models for sea ice area and thickness. For the 42 simulation results, the ROC curve and PR curve and their area under the curve AUC are calculated and plotted to evaluate the parameter goodness. Among them, the parameter λ value is 2.0, and the maximum AUC corresponding to the ROC curve and PR curve of sea ice area are 0.9892 and 0.7864 respectively. The parameter value is 2.0, and the maximum AUC corresponding to the ROC curve and PR curve of sea ice thickness are 0.9892 and 0.7864 respectively. Therefore, the optimal parameter for the Tave index corresponding to the sea ice area and the Light_freezing index corresponding to the sea ice thickness is 2.0.

[0115] For the sea ice area and sea ice thickness of the target sea area in recent N years, 8 meteorological indicators and their corresponding different parameter values ​​are substituted respectively to simulate the reconstruction model. Then, the PR curve, ROC curve and AUC value obtained by the optimal parameters under each indicator are compared to select the optimal meteorological indicator. Specifically, the evaluation results of different parameters under a certain meteorological indicator are compared first, and the parameter corresponding to the maximum AUC value is selected as the optimal parameter under the meteorological indicator. Then, the AUC values ​​of the optimal parameters of the 8 meteorological indicators are further compared, and the meteorological indicator corresponding to the maximum AUC value is selected, which is the optimal meteorological indicator for the reconstruction model of that year. Considering that the ice conditions in the target sea area vary greatly in different years, the optimal meteorological indicator and its optimal parameter are selected for each year, which can better improve the accuracy of the simulation results and characterize the differences between different years. The simulation process is carried out for the 8 meteorological indicators in turn. There are about 670 simulation processes under different parameter values ​​corresponding to the single meteorological indicator of sea ice area and thickness, and each simulation obtains about 3360 images of sea ice area and sea ice thickness output results. All eight indicators output a total of approximately 2.226 million image results.

[0116] Since the overall AUC value of the ROC curve is relatively high, the average value of all indicators is above 0.95. The optimal meteorological indicators corresponding to each year were further determined. When the optimal meteorological indicator for sea ice area was selected, the average AUC values ​​of the ROC curve and the PR curve were 0.8706 and 0.9611, respectively. When the indicator was selected, the most years for data reconstruction using CTDD were 9 years, followed by the Tave index, which was 8 years; Light_freezing was 7 years; CFDD and Heavy_freezing were 3 years, and Light_melting was 2 years. When the optimal meteorological indicator for sea ice thickness was selected, the average AUC values ​​of the ROC curve and the PR curve were 0.7793 and 0.9587, respectively. The most years for data reconstruction using the Light_freezing index were 13 years, followed by the Heavy_freezing index, which was 9 years; CTDD and Tave were 3 years, CFDD was 2 years, and Heavy_melting and Heavy_Lighting were 1 year.

[0117] The sea ice area data of the target sea area extracted by remote sensing in the past N years were compared with the model simulation results corresponding to the same date. The total sea ice area obtained by model simulation is in good agreement with the remote sensing inversion data. The model simulation results of the average sea ice thickness are consistent with the inversion results, and with the influence of climate warming, the floating range of the average sea ice thickness gradually becomes smaller, specifically, the average sea ice thickness of AVHRR data basically fluctuates within 30cm, the average sea ice thickness of MODIS data basically fluctuates within 25cm, and the GOCI data in the past 10 years show that the average sea ice thickness has decreased significantly, basically fluctuating within 20cm. In contrast, the correlation of thickness simulation is much lower than that of area simulation. The correlation coefficient between AVHRR, MODIS and GOCI simulation results and observations is 0.42 (p<0.01), but the overall ice thickness fluctuation range is consistent. For some years, the annual variation characteristics and trends are similar, and the date of occurrence of the maximum average thickness is also partially consistent with or close to the observed date. The spatial distribution of sea ice obtained by model simulation is generally consistent with that obtained by remote sensing inversion.

[0118] The method for reconstructing sea ice and ice condition data based on remote sensing images of this embodiment extracts N years of non-continuous spatiotemporal distribution data of ice conditions in the target sea area based on multi-source satellite remote sensing data, and reconstructs N years of continuous spatiotemporal distribution data of ice conditions with the help of coastal meteorological data of the target sea area and a spatial autocorrelation model. In the case of a lack of on-site observation data and a large number of missing satellite remote sensing data, the spatial autocorrelation model has a good overall simulation effect on sea ice area and thickness, and has high reliability. It can provide continuous spatiotemporal distribution data of sea ice in the target sea area on a daily scale, thereby helping to construct a continuous daily time series of ice conditions, which is used to study the sea ice formation and disappearance process, long-term spatiotemporal dynamic changes and refined spatiotemporal distribution characteristics in the target sea area.

[0119] like Figure 7 One embodiment of the present invention provides a device for reconstructing sea ice data based on multi-source remote sensing images, comprising:

[0120] The acquisition module is used to obtain the remote sensing image data of sea ice in the target sea area and send it to the preprocessing module;

[0121] A preprocessing module is used to preprocess the remote sensing image data to obtain corrected remote sensing image data and send the corrected remote sensing image data to the extraction module;

[0122] An extraction module, used to obtain sea ice area data and sea ice thickness data according to the corrected remote sensing image data, and send them to a reconstruction module;

[0123] The reconstruction module is used to obtain a data set of the historical spatial distribution area and thickness of sea ice in the target sea area based on the sea ice area data and the sea ice thickness data.

[0124] The above scheme of this embodiment, based on the remote sensing image data of sea ice in the target sea area, extracts sea ice area data and sea ice thickness data, and then reconstructs the historical spatial distribution area and thickness data set of sea ice in the target sea area, which can provide technical support for analyzing the spatiotemporal distribution characteristics of sea ice and future sea ice risk assessment, and provide data support for the study of long-term sea ice dynamic ablation mechanism.

[0125] In an optional embodiment of the present invention, the preprocessing module is specifically used for:

[0126] The remote sensing image data is subjected to geometric correction, format conversion and atmospheric correction to obtain corrected remote sensing image data.

[0127] In an optional embodiment of the present invention, the preprocessing module is specifically used for:

[0128] Extracting the corrected remote sensing image data using a partition threshold method or a sample-based object-oriented classification method to obtain sea ice area data;

[0129] The corrected remote sensing image data is inverted using the logarithmic relationship of the sea ice albedo to obtain sea ice thickness data.

[0130] In an optional embodiment of the present invention, the reconstruction module is specifically used to:

[0131] Obtain meteorological data along the coast of the target sea area;

[0132] According to the sea ice area data, the sea ice thickness data and the meteorological data, a spatial autocorrelation model is used to obtain a data set of the historical spatial distribution area and thickness of sea ice in the target sea area.

[0133] In an optional embodiment of the present invention, the reconstruction module is specifically used to:

[0134] Obtaining a weight coefficient according to the sea ice area data, the sea ice thickness data and the meteorological data;

[0135] The sea ice area data, the sea ice thickness data, the weight coefficient and the meteorological data are used to obtain the weight coefficient and substitute it into the spatial autocorrelation model to obtain the historical spatial distribution area and thickness data set of sea ice in the target sea area.

[0136] In an optional embodiment of the present invention, the reconstruction module is specifically used to:

[0137] Obtaining a kilometer grid of the target sea area, wherein the kilometer grid includes a plurality of pixel unit grids;

[0138] According to the sea ice area data, the sea ice thickness data and the kilometer grid, the sea ice area and the average sea ice thickness in each pixel unit grid are obtained;

[0139] According to the meteorological data and the kilometer grid, obtaining the temperature data in each pixel unit grid;

[0140] The sea ice area and average sea ice thickness in the pixel unit grid, the temperature data in the pixel unit grid, and the weight coefficient are input into the kilometer grid to obtain a data set of sea ice historical spatial distribution area and thickness in the target sea area.

[0141] In an optional embodiment of the present invention, the reconstruction module is specifically used to:

[0142] The meteorological data is extracted into the kilometer grid using the inverse distance weighted method to obtain the temperature data in each pixel unit grid.

[0143] like Figure 7 In an optional embodiment of the present invention, the device further comprises:

[0144] The verification module is used to verify the sea ice historical spatial distribution area and thickness data set of the target sea area by using the receiver operating characteristic curve and the precision-recall curve to obtain the verification result.

[0145] One embodiment of the present invention also provides a computer-readable storage medium, on which instructions are stored. When the instructions are executed on a computer, the computer executes the following Figure 1 The method described.

[0146] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for reconstructing sea ice data based on multi-source remote sensing images, characterized in that: include: Obtain remote sensing image data of sea ice in the target sea area; Preprocessing the remote sensing image data to obtain corrected remote sensing image data; Obtaining sea ice area data and sea ice thickness data based on the corrected remote sensing image data; According to the sea ice area data and the sea ice thickness data, a data set of sea ice historical spatial distribution area and thickness in the target sea area is obtained; Wherein, the sea ice area data and sea ice thickness data are obtained according to the corrected remote sensing image data, including: Extracting the corrected remote sensing image data using a partition threshold method or a sample-based object-oriented classification method to obtain sea ice area data; The corrected remote sensing image data is inverted using the logarithmic relationship of the sea ice albedo to obtain the sea ice thickness data; specifically, the sea ice thickness data is obtained by h = -ln[(1-α h α max -1 )·k -1 ]·μ α -1 and k = 1-α sea / α max Get; where h is the sea ice thickness; α h is the sea ice albedo; α max is the albedo when the ice thickness is infinite; k is the correlation coefficient, μ α is the observed albedo attenuation coefficient; α max , α sea , μ α They are 0.700, 0.100 and 1.209 respectively; α max -1 is α max The reciprocal of μ α -1 μ α The reciprocal of Wherein, according to the sea ice area data and the sea ice thickness data, a data set of sea ice historical spatial distribution area and thickness in the target sea area is obtained, including: Obtain meteorological data along the coast of the target sea area; According to the sea ice area data, the sea ice thickness data and the meteorological data, a spatial autocorrelation model is used to obtain a data set of sea ice historical spatial distribution area and thickness in the target sea area; Wherein, according to the sea ice area data, the sea ice thickness data and the meteorological data, a spatial autocorrelation model is used to obtain a data set of sea ice historical spatial distribution area and thickness in the target sea area, including: Obtaining a weight coefficient according to the sea ice area data, the sea ice thickness data and the meteorological data; Substituting the sea ice area data, the sea ice thickness data, the weight coefficient and the meteorological data into a spatial autocorrelation model to obtain a data set of the historical spatial distribution area and thickness of sea ice in the target sea area; The sea ice area data, the sea ice thickness data, the weight coefficient and the meteorological data are substituted into the spatial autocorrelation model to obtain the historical spatial distribution area and thickness data set of sea ice in the target sea area, including: Obtaining a kilometer grid of the target sea area, wherein the kilometer grid includes a plurality of pixel unit grids; According to the sea ice area data, the sea ice thickness data and the kilometer grid, the sea ice area and the average sea ice thickness in each pixel unit grid are obtained; According to the meteorological data and the kilometer grid, obtaining the temperature data in each pixel unit grid; The sea ice area and average sea ice thickness in the pixel unit grid, the temperature data in the pixel unit grid, and the weight coefficient are input into the kilometer grid to obtain a data set of sea ice historical spatial distribution area and thickness in the target sea area; Wherein, obtaining the temperature data in each pixel unit grid according to the meteorological data and the kilometer grid includes: Extracting the meteorological data into the kilometer grid using the inverse distance weighted method to obtain the temperature data in each pixel unit grid; The spatial autocorrelation model is: in, Where t is the year, d is the date, i is the row number, j is the column number, g is the ice level, T is the meteorological index, 0≤w td,ij ≤1 is the weight coefficient, satisfying λ is the tuning parameter, is the sea ice area or thickness value in the simulated kilometer grid at year t0, date d0, row i0, column j0, td,ij is the sea ice area or thickness value in the kilometer grid in year t, date d, row i, column j, To fill the target year t0, the historical year set of sea ice data is used. is the date adjacent to the target date d0 with sea ice observations in year t, Given the year t and date d, the adjacent grid corresponding to the target grid i0j0.

2. The method for reconstructing sea ice data based on multi-source remote sensing images according to claim 1, characterized in that: Preprocessing the remote sensing image data to obtain corrected remote sensing image data includes: The remote sensing image data is subjected to geometric correction, format conversion and atmospheric correction to obtain corrected remote sensing image data.

3. The method for reconstructing sea ice data based on multi-source remote sensing images according to claim 1, characterized in that: The method further comprises: The receiver operating characteristic curve and the precision-recall curve are used to verify the sea ice historical spatial distribution area and thickness dataset of the target sea area to obtain the verification results.

4. A device for reconstructing sea ice data based on multi-source remote sensing images, characterized in that: include: The acquisition module is used to obtain the remote sensing image data of sea ice in the target sea area and send it to the preprocessing module; A preprocessing module is used to preprocess the remote sensing image data to obtain corrected remote sensing image data and send the corrected remote sensing image data to the extraction module; An extraction module, used to obtain sea ice area data and sea ice thickness data according to the corrected remote sensing image data, and send them to a reconstruction module; A reconstruction module, used to obtain a data set of historical spatial distribution area and thickness of sea ice in a target sea area according to the sea ice area data and the sea ice thickness data; Wherein, the sea ice area data and sea ice thickness data are obtained according to the corrected remote sensing image data, including: Extracting the corrected remote sensing image data using a partition threshold method or a sample-based object-oriented classification method to obtain sea ice area data; The corrected remote sensing image data is inverted using the logarithmic relationship of the sea ice albedo to obtain the sea ice thickness data; specifically, the sea ice thickness data is obtained by h = -ln[(1-α h α max -1 )·k -1 ]·μ α -1 and k = 1-α sea / α max Get; where h is the sea ice thickness; α h is the sea ice albedo; α max is the albedo when the ice thickness is infinite; k is the correlation coefficient, μ α is the observed albedo attenuation coefficient; α max , α sea , μ α They are 0.700, 0.100 and 1.209 respectively; α max -1 is α max The reciprocal of μ α -1 μ α The reciprocal of Wherein, according to the sea ice area data and the sea ice thickness data, a data set of sea ice historical spatial distribution area and thickness in the target sea area is obtained, including: Obtain meteorological data along the coast of the target sea area; According to the sea ice area data, the sea ice thickness data and the meteorological data, a spatial autocorrelation model is used to obtain a data set of sea ice historical spatial distribution area and thickness in the target sea area; Wherein, according to the sea ice area data, the sea ice thickness data and the meteorological data, a spatial autocorrelation model is used to obtain a data set of sea ice historical spatial distribution area and thickness in the target sea area, including: Obtaining a weight coefficient according to the sea ice area data, the sea ice thickness data and the meteorological data; Substituting the sea ice area data, the sea ice thickness data, the weight coefficient and the meteorological data into a spatial autocorrelation model to obtain a data set of the historical spatial distribution area and thickness of sea ice in the target sea area; The sea ice area data, the sea ice thickness data, the weight coefficient and the meteorological data are substituted into the spatial autocorrelation model to obtain the historical spatial distribution area and thickness data set of sea ice in the target sea area, including: Obtaining a kilometer grid of the target sea area, wherein the kilometer grid includes a plurality of pixel unit grids; According to the sea ice area data, the sea ice thickness data and the kilometer grid, the sea ice area and the average sea ice thickness in each pixel unit grid are obtained; According to the meteorological data and the kilometer grid, obtaining the temperature data in each pixel unit grid; The sea ice area and average sea ice thickness in the pixel unit grid, the temperature data in the pixel unit grid, and the weight coefficient are input into the kilometer grid to obtain a data set of sea ice historical spatial distribution area and thickness in the target sea area; Wherein, obtaining the temperature data in each pixel unit grid according to the meteorological data and the kilometer grid includes: Extracting the meteorological data into the kilometer grid using the inverse distance weighted method to obtain the temperature data in each pixel unit grid; The spatial autocorrelation model is: in, Where t is the year, d is the date, i is the row number, j is the column number, g is the ice level, T is the meteorological index, 0≤w td,ij ≤1 is the weight coefficient, satisfying λ is the tuning parameter, is the sea ice area or thickness value in the simulated kilometer grid at year t0, date d0, row i0, column j0, td,ij is the sea ice area or thickness value in the kilometer grid in year t, date d, row i, column j, To fill the target year t0, the historical year set of sea ice data is used. is the date adjacent to the target date d0 with sea ice observations in year t, Given the year t and date d, the adjacent grid corresponding to the target grid i0j0.

5. A computer-readable storage medium, characterized in that: Instructions are stored thereon, and when the instructions are executed on a computer, the computer is caused to execute the method according to any one of claims 1 to 3.

Images