Ocean water color satellite sensor on-orbit rapid absolute radiometric calibration method

Abstract

The invention discloses an in-orbit rapid absolute radiometric calibration method for an ocean water color satellite sensor, and belongs to the technical field of radiometric calibration. The calibration method comprises the following specific steps: I, analyzing optical properties of site atmosphere and water by using 865nm aerosol optical thickness data and 443nm normalized water-leaving radiance data in AEROENT-OC historical observation data; iI, establishing a machine learning model, and taking on-site multispectral data as input and hyperspectral data as output; according to the method, the on-orbit absolute radiometric calibration coefficient of the sensor is calculated by using field site data at different positions instead of a calibration mode, and the on-satellite calibration result of the sensor is checked, so that the evaluation of the overall running state of the sensor can be completed within a relatively short time at the initial stage of launching; and the performance of the sensor can be fully reflected in a relatively wide dynamic range.

Description

Technical Field

[0001] The present invention relates to the technical field of radiometric calibration, and particularly to a method for on-orbit rapid absolute radiometric calibration of a marine water color satellite sensor. Background Art

[0002] Precise radiometric calibration of a water color satellite sensor is an important prerequisite for supporting its various applications (marine ecosystem, ecological-dynamic interaction, fishery, marine pollution, etc.) globally. In order to accurately characterize the radiation characteristics of each wavelength of the sensor, absolute radiometric calibration is performed on the ground using radiation with known characteristics before the sensor is launched. After the sensor is in orbit, due to factors such as vibration during launch, changes in the operating environment, and aging of its own components over time, the applicability of the absolute radiometric calibration data before launch is limited, resulting in a decline in the quality of sensor data. In order to accurately characterize the changes in the radiation response characteristics after the sensor is in orbit, on-orbit radiometric calibration of the sensor is usually carried out using different methods such as on-board calibration, vicarious calibration, and cross calibration.

[0003] On-board calibration mainly relies on the solar calibration method, which has been successfully applied to multiple on-orbit sensors such as (MODIS, VIIRS, OLCI), etc. However, it uses the sun as the radiation source and can obtain the radiation characteristics of the sensor at a single response intensity. Vicarious calibration takes in-situ measurements as the benchmark, such as the established MOBY and BOUSSOLE buoys. Cross calibration is actually an extension of the vicarious calibration method, using the observation data of another precisely calibrated reference payload as the benchmark, and can use observations in different regions of the world to ensure coverage of the dynamic range of the sensor. Due to the different atmospheric and water body compositions in different regions of the world, the top-of-atmosphere radiance received by the water color satellite sensor shows a relatively low magnitude, but there are also rich variations. However, the number of sites used is limited, and the advantages of using different sites are not reflected in terms of time and dynamic range; therefore, it is particularly important to invent a method for on-orbit rapid absolute radiometric calibration of a marine water color satellite sensor. Summary of the Invention

[0004] The purpose of the present invention is to solve the defects existing in the prior art, and to propose a method for on-orbit rapid absolute radiometric calibration of a marine water color satellite sensor.

[0005] In order to achieve the above purpose, the present invention adopts the following technical solutions: A method for on-orbit rapid absolute radiometric calibration of a marine water color satellite sensor, and the specific steps of this calibration method are as follows: Ⅰ. Analyze the optical properties of the on-site atmosphere and water body using the aerosol optical depth data at 865 nm and the normalized water-leaving radiance data at 443 nm in the AEROENT-OC historical observation data; Ⅱ. Establish a machine learning model with on-site multi-spectral data as input and hyperspectral data as output; Ⅲ. Use the established machine learning model to construct a hyperspectral on-site dataset. Then, with the reconstructed on-site water-leaving radiance as the basis and combined with various environmental conditions of the on-site atmosphere and water body as input parameters, calculate the theoretical value of the radiance at the top of the atmosphere that matches the satellite sensor by means of the OSOAA radiative transfer model; Ⅳ. Compare the calculation results with the calibration results on the sensor satellite and the current alternative calibration results using MOBY on a six-month cycle.

[0006] As a further solution of the present invention, the specific steps of analyzing the optical properties of the on-site atmosphere and water body using the aerosol optical depth data at 865 nm and the normalized water-leaving radiance data at 443 nm in step Ⅰ are as follows: S1.1: Extract the aerosol optical depth data at 865 nm and the normalized water-leaving radiance data at 443 nm from the selected AEROENT-OC historical observation data, namely AOD_865 and Lwn_443, and screen the records with complete data and high spatial coverage consistency within the time range according to user requirements; S1.2: Clean the extracted AOD_865 data and Lwn_443 data, unify the data with different time or spatial resolutions into the target format, and then align AOD_865 and Lwn_443 according to the time stamp and spatial position to form an analysis data pair; S1.3: According to the spatial distribution of AOD_865, identify the high-concentration aerosol regions, determine the temporal variation law of aerosols, and then plot the spectral characteristic curves of different water body types through Lwn_443. Analyze the differences in Lwn_443 in different water body environments through multi-point comparison, and record each influencing factor.

[0007] As a further solution of the present invention, the specific steps of establishing the machine learning model in step Ⅱ are as follows: S2.1: Normalize the on-site collected multi-spectral data to the range of [0, 1], divide the processed multi-spectral data into a training set and a test set, then improve the MLP model to obtain the MRH model, use ReLU as the activation function of the model, and perform a weighted combination of MSE and MAPE to construct the combined loss function of the model to complete the construction of the machine learning model; S2.2: Input the training set data into the machine learning model in batches according to a training batch size of 64 and 100 training epochs. After the forward propagation of the machine learning model for each batch of training set data, the predicted values are output, and the loss value between the predicted values and the actual values is calculated through the combined loss function. The Adam optimizer adjusts the model weights based on the generated loss value; S2.3: After each round of training, input the test set into the machine learning model and calculate the loss value through the combined loss function to evaluate the accuracy of the model in reconstructing hyperspectral data. Use the spectral curve comparison chart to check the matching degree between the predicted hyperspectral data and the actual values within the entire wavelength range. Repeat the training and testing until the preset training epochs are reached or the model is overfitted, then stop the training and output the final machine learning model.

[0008] As a further solution of the present invention, the machine learning model described in S2.1 includes two hidden layers. The first layer has 64 neurons, and the second layer has 32 neurons. The number of neurons in each layer is optimized according to the complexity of the input and output. Moreover, the number of neurons in the output layer of the machine learning model is the same as the number of wavelengths of the target hyperspectral data. Each neuron corresponds to the radiance value of a hyperspectral band. At the same time, the output layer does not use an activation function and directly outputs a linear result to match the continuous spectral characteristics.

[0009] As a further solution of the present invention, the specific steps for calculating the theoretical value of the top-of-atmosphere radiance in step IV are as follows: S3.1: Use the hyperspectral data generated by the machine learning model and combine the atmospheric and water environmental conditions recorded by AEROENT-OC to construct a hyperspectral field dataset. Then, input each group of data of the hyperspectral water-leaving radiance, on-site atmospheric conditions, and water optical properties in the hyperspectral field dataset into the OSOAA radiative transfer model; S3.2: Based on the input groups of data and according to multiple environmental factors, the OSOAA radiative transfer model simulates the radiative transfer process from the ocean surface to the top of the atmosphere and calculates the top-of-atmosphere radiance. Then, compare the simulated TOA radiance with the actually measured radiance on-site for spectral correction.

[0010] As a further solution of the present invention, the environmental factors described in S3.2 specifically include solar and ground conditions, seasonal changes, sunshine cycles, wind speed and ocean waves, the vertical structure of the atmosphere, precipitation and clouds, and terrain factors.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: The on-orbit rapid absolute radiometric calibration method for a marine color satellite sensor analyzes the optical properties of the in-situ atmosphere and water body using aerosol optical depth data at 865 nm and normalized water-leaving radiance data at 443 nm from AEROENT-OC historical observation data. Then, the MLP model is improved to obtain the MRH model, with ReLU used as the activation function of the model, and MSE and MAPE are weighted and combined to construct the combined loss function of the model to complete the construction of the machine learning model. The machine learning model is trained and tested with a training batch size of 64 and 100 training epochs. Using the established machine learning model, a hyperspectral in-situ dataset is constructed. Then, with the help of the OSOAA radiative transfer model, based on the reconstructed in-situ water-leaving radiance, combined with various environmental conditions of the in-situ atmosphere and water body as input parameters, the theoretical value of the top-of-atmosphere radiance matching the satellite sensor is calculated. Taking a 6-month period, the calculated results are compared with the on-board calibration results of the sensor satellite and the current alternative calibration results using MOBY. Using in-situ site data at different locations, the on-orbit absolute radiometric calibration coefficient of the sensor is calculated by the method of alternative calibration, and the on-board calibration results of the sensor are verified. It can complete the evaluation of the overall operating state of the sensor in a relatively short time at the initial stage of launch and can fully reflect the performance of the sensor within a wide dynamic range. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification, and are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention.

[0013] Figure 1 It is a flowchart of a method for on-orbit rapid absolute radiometric calibration of a marine color satellite sensor proposed by the present invention; Figure 2 It is an algorithm block diagram of a method for on-orbit rapid absolute radiometric calibration of a marine color satellite sensor proposed by the present invention; Figure 3 It is a diagram comparing the errors between the machine learning model and the BOM model of a method for on-orbit rapid absolute radiometric calibration of a marine color satellite sensor proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] Referring to Figures 1-3 , a method for on-orbit rapid absolute radiometric calibration of a marine color satellite sensor, the specific steps of the calibration method are as follows: Analyze the optical properties of the in-situ atmosphere and water body using aerosol optical depth data at 865 nm and normalized water-leaving radiance data at 443 nm from AEROENT-OC historical observation data.

[0015] Specifically, extract the aerosol optical depth data at 865 nm and the normalized water-leaving radiance data at 443 nm from the selected AEROENT-OC historical observation data, namely AOD_865 and Lwn_443. Then, screen the records with complete data and high spatial coverage consistency within the time range according to user requirements. Clean the extracted AOD_865 data and Lwn_443 data, unify the data with different time or spatial resolutions into the target format, and then align AOD_865 and Lwn_443 according to the time stamp and spatial position to form an analysis data pair. According to the spatial distribution of AOD_865, identify the high-concentration aerosol regions, determine the temporal variation law of aerosols, and then plot the spectral characteristic curves of different water body types through Lwn_443. Analyze the differences in Lwn_443 in different water body environments through multi-point comparison, and record each influencing factor.

[0016] Build a machine learning model with in-situ multi-spectral data as the input and hyperspectral data as the output.

[0017] Specifically, normalize the in-situ collected multi-spectral data to the range of [0, 1], and divide the processed multi-spectral data into a training set and a test set. Then, improve the MLP model to obtain the MRH model, use ReLU as the activation function of the model, and perform a weighted combination of MSE and MAPE to construct the combined loss function of the model to complete the construction of the machine learning model. According to the training batch size of 64 and the number of training epochs of 100, input the training set data into the machine learning model in batches. The predicted values are output after the forward propagation of the machine learning model for each batch of training set data, and the loss value between the predicted value and the actual value is calculated through the combined loss function. The Adam optimizer adjusts the model weights according to the generated loss value. After each round of training, input the test set into the machine learning model and calculate the loss value through the combined loss function to evaluate the accuracy of the model in reconstructing hyperspectral data, and use the spectral curve comparison chart to check the matching degree between the predicted hyperspectral data and the actual value in the entire wavelength range. Repeat the training and testing until the preset number of training epochs is reached or the model is overfitted, then stop the training and output the final machine learning model.

[0018] It should be further noted that the machine learning model contains two hidden layers. The first layer has 64 neurons, and the second layer has 32 neurons. The number of neurons in each layer is optimized according to the complexity of the input and output. Moreover, the number of neurons in the output layer of the machine learning model is the same as the number of wavelengths of the target hyperspectral data. Each neuron corresponds to the radiance value of a hyperspectral band. At the same time, the output layer does not use an activation function and directly outputs a linear result to match the continuous spectral characteristics.

[0019] Reference Figure 3It can be seen that a machine learning model is established to reconstruct the multi-spectral on-site observation data. By comparing with the results obtained by the existing empirical relationship method, the performance of the model is verified. For the error of the hyperspectral reconstruction results of the on-site data in the visible light domain by the MRH and BOM models, except for the 17.42% error in the 670-nm band, the error of the MRH model is less than 5% in the 400 - 580-nm band, less than 9% in the 580 - 670-nm band, and the error gradually increases after the 670-nm band. This indicates that in the top-of-atmosphere radiance it reflects, the uncertainty in the blue-green band is 0.5%, and the uncertainty in the red band is between 1% and 2%. In contrast, the error generated by the BOM model is always higher than that of the MRH model, and the maximum reconstruction error is as high as 34.19%.

[0020] Using the established machine learning model, a hyperspectral on-site dataset is constructed. Then, with the help of the OSOAA radiative transfer model, based on the reconstructed on-site water-leaving radiance, combined with the on-site atmospheric and water environmental conditions as input parameters, the calculation of the theoretical value of the top-of-atmosphere radiance matching the satellite sensor is completed.

[0021] Specifically, using the hyperspectral data generated by the machine learning model, combined with the atmospheric and water environmental conditions recorded by AEROENT-OC, a hyperspectral on-site dataset is constructed. Then, the hyperspectral water-leaving radiance, on-site atmospheric conditions, and each group of data of the water optical properties in the hyperspectral on-site dataset are input into the OSOAA radiative transfer model. The OSOAA radiative transfer model, based on the input groups of data and according to multiple environmental factors, simulates the radiative transfer process from the ocean surface to the top of the atmosphere and calculates the top-of-atmosphere radiance. Then, the simulated TOA radiance is compared with the actually measured radiance on-site for spectral correction.

[0022] It should be further noted that the environmental factors specifically include solar and ground conditions, seasonal changes, sunshine duration, wind speed and ocean waves, the vertical structure of the atmosphere, precipitation and clouds, and terrain factors.

[0023] Taking a 6-month cycle, the calculation results are compared with the calibration results on the sensor satellite and the alternative calibration results currently carried out using MOBY.

[0024] In addition, it should be noted that by comparing the calculation results with the actual satellite observation results and with the results obtained by the existing MOBY, the advantages of the present invention are highlighted in terms of the number of matches and the sensor dynamic range. The comparison of the top-of-atmosphere radiances obtained by substituting the MODIS-Aqua sensor with the hyperspectral in-situ dataset reconstructed by the machine learning model and its on-board calibration at the wavelengths of red (678 nm), green (547 nm), and blue (412 nm). And taking a 6-month period as an example, the comparison of the top-of-atmosphere radiance coverage ranges reflected by AERONET-OC (red dots) and MOBY (blue dots) is presented. From the perspective of the substitution calibration and the on-board calibration, the correlation coefficients of all the comparison results are greater than 0.97, and the absolute mean percentage error is less than 3.70%. From the perspective of the top-of-atmosphere radiance coverage ranges reflected by AEROENT-OC and MOBY, within the set period, the number of matches between AERONET-OC and MODIS-Aqua is 56 times, while the number of matches of MOBY is only 3 times. At 412 nm, the top-of-atmosphere radiance coverage range reflected by AERONET-OC is 4 - 12 mW / cm2 / μm / sr, while the coverage range of MOBY is only 9 - 10.5 mW / cm2 / μm / sr. At 555 nm, the top-of-atmosphere radiance coverage range reflected by AERONET-OC is 1.5 - 5 mW / cm2 / μm / sr, while the corresponding coverage range of MOBY is only 3 - 3.5 mW / cm2 / μm / sr. At 678 nm, the top-of-atmosphere radiance coverage range reflected by AERONET-OC is 0.7 - 2.2 mW / cm2 / μm / sr, while the corresponding coverage range of MOBY is only 1.1 - 1.4 mW / cm2 / μm / sr.

Claims

1. A method for on-orbit rapid absolute radiometric calibration of ocean color satellite sensors, characterized in that, The specific steps of this calibration method are as follows: Ⅰ. Analyze the optical properties of the on-site atmosphere and water body by using the aerosol optical depth data at 865 nm and the normalized water-leaving radiance data at 443 nm in the AEROENT-OC historical observation data; Ⅱ. Establish a machine learning model with the on-site multi-spectral data as the input and the hyperspectral data as the output; Ⅲ. Use the established machine learning model to construct a hyperspectral on-site data set, and then, with the help of the OSOAA radiative transfer model, based on the reconstructed on-site water-leaving radiance, combined with the on-site atmospheric and water environmental conditions as input parameters, complete the calculation of the theoretical value of the radiance at the top of the atmosphere that matches the satellite sensor; Ⅳ. Take 6 months as a cycle, and compare the calculation results with the calibration results on the sensor satellite and the current alternative calibration results using MOBY; The specific steps of establishing the machine learning model described in step Ⅱ are as follows: S2.1: Normalize the on-site collected multi-spectral data to the range of [0, 1], divide the processed multi-spectral data into a training set and a test set, then improve the MLP model to obtain the MRH model, use ReLU as the activation function of this model, and perform a weighted combination of MSE and MAPE to construct the combined loss function of this model, so as to complete the construction of the machine learning model; S2.2: According to the training batch size of 64 and the number of training epochs of 100, input the training set data into the machine learning model in batches. After the forward propagation of the machine learning model for each batch of training set data, the predicted value is output, and the loss value between the predicted value and the actual value is calculated through the combined loss function. The Adam optimizer adjusts the model weights according to the generated loss value; S2.3: After each round of training, input the test set into the machine learning model, and calculate the loss value through the combined loss function to evaluate the accuracy of the model in reconstructing the hyperspectral data. Use the spectral curve comparison chart to check the matching degree of the predicted hyperspectral data and the actual value in the entire wavelength range. Repeat the training and testing until the preset number of training epochs is reached or the model is overfitted, then stop the training and output the final machine learning model; The machine learning model described in S2.1 contains two hidden layers. The first layer has 64 neurons, and the second layer has 32 neurons. The number of neurons in each layer is optimized according to the complexity of the input and output. The number of neurons in the output layer of this machine learning model is the same as the number of wavelengths of the target hyperspectral data. Each neuron corresponds to the radiance value of a hyperspectral band. At the same time, the output layer does not use an activation function and directly outputs a linear result to match the continuous spectral characteristics.

2. A method for on-orbit fast absolute radiometric calibration of a marine water color satellite sensor according to claim 1, characterized in that The specific steps of analyzing the optical properties of the on-site atmosphere and water body by using the aerosol optical depth data at 865 nm and the normalized water-leaving radiance data at 443 nm described in step Ⅰ are as follows: S1.1: Extract the aerosol optical depth data at 865 nm and the normalized water-leaving radiance data at 443 nm from the selected AEROENT-OC historical observation data, namely AOD_865 and Lwn_443, and screen the records with complete data and high spatial coverage consistency within the time range according to user requirements; S1.2: Clean the extracted AOD_865 data and Lwn_443 data, unify the data with different time or spatial resolutions into the target format, and then align AOD_865 and Lwn_443 according to the time stamp and spatial position to form an analysis data pair; S1.3: According to the spatial distribution of AOD_865, identify the high-concentration aerosol regions, determine the temporal variation law of aerosols, then plot the spectral characteristic curves of different water body types through Lwn_443, analyze the differences of Lwn_443 in different water body environments through multi-point comparison, and record each influencing factor.

3. A method for on-orbit rapid absolute radiometric calibration of a marine water color satellite sensor according to claim 1, characterized in that The specific steps for calculating the theoretical value of the top-of-atmosphere radiance described in Step Ⅳ are as follows: S3.1: Use the hyperspectral data generated by the machine learning model, combine the atmospheric and water body environmental conditions recorded by AEROENT-OC, construct a hyperspectral in-situ dataset, and then input each group of data of hyperspectral water-leaving radiance, in-situ atmospheric conditions and water body optical properties in the hyperspectral in-situ dataset into the OSOAA radiative transfer model; S3.2: Based on the input data of each group and according to multiple environmental factors, the OSOAA radiative transfer model simulates the radiative transfer process from the ocean surface to the top of the atmosphere, calculates the top-of-atmosphere radiance, and then compares the simulated TOA radiance with the actually measured radiance in-situ for spectral correction.

4. A method for on-orbit fast absolute radiometric calibration of a marine water color satellite sensor according to claim 3, characterized in that The environmental factors described in S3.2 specifically include solar and ground conditions, seasonal changes, sunshine duration, wind speed and ocean waves, vertical structure of the atmosphere, precipitation and clouds, and terrain factors.

Images