A global sensitivity analysis method and system for ocean satellite radiometric benchmark transfer

Abstract

The application provides a global sensitivity analysis method for ocean satellite radiation reference transmission, which comprises the following steps: selecting initial parameters which preferably affect the atmospheric top reflectivity from four dimensions; constructing a monthly-grid radiation characteristic data set; performing sensitivity pre-screening on the initial parameters coupled with each ocean satellite band; constructing a radiation transmission simulation scene; performing low-difference sampling on the input variable parameter space; performing global sensitivity analysis on the input variable combination and the atmospheric top reflectivity output data set; adjusting the sampling scale through a double-threshold method; combining the global sensitivity index with the atmospheric top reflectivity index of the initial parameters coupled with each ocean satellite band to calculate the total error of each band reference transmission; and analyzing and judging the uncertainty sources in the ocean satellite radiation reference transmission process. The application can quantitatively identify the main error sources and the contribution of the coupling effect between variables to the atmospheric top reflectivity deviation, and improve the calculation efficiency while ensuring the analysis accuracy.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of satellite radiation reference transfer and uncertainty evaluation, and particularly relates to a global sensitivity analysis method and system for ocean satellite radiation reference transfer. BACKGROUND

[0002] Ocean satellite radiation reference transfer is a process of transferring the radiation observation results of a reference satellite sensor with higher spatial resolution, radiation resolution and spectral resolution to a target satellite sensor for providing a more accurate radiation reference value for the target instrument. Ideally, two satellite sensors should be sampled at the same time, same place and field of view, same observation geometry and same spatial spectral performance (including spectral resolution and spectral response function SRF) to eliminate the uncertainty caused by the differences of multiple sources as much as possible and ensure the radiation transfer accuracy. However, in actual application, it is difficult to realize strict matching. On the one hand, according to the definition of radiation reference transfer, the differences of the spectral response function SRF, spatial resolution and observation angle of the instrument in the transfer process exist and are usually determined. On the other hand, strict matching constraints will reduce the possibility of realizing the transfer and greatly reduce the transfer times, and the ocean water-atmosphere environment changes fast, and the atmospheric conditions, sea surface wind field and water color elements change in real time in the time and space dimensions.

[0003] Global sensitivity analysis can more comprehensively quantify the influence of time, space, spectrum, angle and other factors on the top-of-atmosphere radiation values received by the reference instrument and the target instrument in the radiation reference transfer process, and reveal the key uncertainty sources. On this basis, global sensitivity analysis can evaluate the influence of all parameters and their interaction effects on the results at one time, thereby guiding the optimization of observation matching constraint strategies and further correction of the influence of link transfer radiation, helping to reduce tedious single sensitivity analysis tests, and making a quantifiable prediction of the uncertainty that may be caused by expanding or relaxing the matching conditions. In this way, the ocean satellite radiation reference transfer process no longer relies only on the condition of “strict time-space-spectrum-angle synchronization”, but can reduce part of the matching constraint conditions within the appropriate link transfer uncertainty requirements, increase the effective transfer frequency and coverage. Traditional methods only focus on the differences of certain specific links and focus on the discussion of specific correction models or approximate conditions, but lack global sensitivity of systematic analysis of all uncertainty sources, resulting in a lack of complete understanding of the formation mechanism and magnitude of the transfer error. SUMMARY

[0004] The present application aims at the deficiencies of the prior art, and provides a global sensitivity analysis method for marine satellite radiation reference transmission.

[0005] To solve the above technical problems, the present application provides the following technical solutions.

[0006] A global sensitivity analysis method for marine satellite radiation reference transmission comprises the following steps:

[0007] Step 1. Based on the uncertainty sources in the marine satellite radiation reference transmission link, the initial parameters affecting the atmospheric top reflectivity are selected from the time domain, the spatial domain, the spectral domain and the angular domain respectively.

[0008] Step 2. A sea target area with stable time and space is selected, and a monthly-grid radiation characteristic data set is constructed through variation coefficient threshold screening.

[0009] Step 3. The sensitivity of the initial parameters is pre-screened in each marine satellite band, and the input variables are determined.

[0010] Step 4. According to the input variables and the monthly-grid radiation characteristic data set, and based on the sea-air coupled radiation transfer model, a radiation transfer simulation scene is constructed, and the input variable parameter space is set.

[0011] Step 5. The input variable parameter space is sampled with low difference, the input variable combination is generated, and radiation transfer simulation is performed to obtain an atmospheric top reflectivity output data set.

[0012] Step 6. The global sensitivity analysis is performed on the input variable combination and the atmospheric top reflectivity output data set to obtain a global sensitivity index.

[0013] Step 7. The sampling size is dynamically adjusted by the convergence global sensitivity index double threshold method.

[0014] Step 8. The global sensitivity index is combined with the atmospheric top reflectivity index of each marine satellite band coupled with the initial parameters to calculate the total error of each band reference transmission.

[0015] Step 9. According to the global sensitivity index and the total error of each band reference transmission, the uncertainty sources in the marine satellite radiation reference transmission process are analyzed and judged.

[0016] Further, the initial parameters in step 1 include:

[0017] Temporal dimension: chlorophyll concentration Chla, sediment concentration, 440 nm detritus absorption coefficient, 440 nm yellow substance absorption coefficient, aerosol optical depth AOD and air relative humidity Rh;

[0018] Spatial dimension: wind speed;

[0019] Spectral dimension: center wavelength difference

[0020] Angular dimension: solar zenith angle, observation zenith angle, relative azimuth angle.

[0021] Further, the step 2 specifically comprises:

[0022] Selecting 3-8 marine areas as spatiotemporal stable marine target areas; dividing 1 °x1 ° grid, respectively calculating the initial parameter variation coefficient threshold value in each grid, and finding the grid position with the long-time series monthly variation coefficient of each initial parameter being less than the variation coefficient threshold value to obtain the optimal grid;

[0023] Statistically analyzing the initial parameter radiation characteristic data set in the optimal grid.

[0024] Further, the step 3 of coupling each marine satellite band with the initial parameter respectively for sensitivity pre-screening to obtain the relative difference root mean square of the top-of-atmosphere reflectance and the sensitivity coefficient of the initial parameter coupled with each marine satellite band, judging whether each initial parameter has a significant impact on the radiation reference transmission of the marine target area under each marine satellite band, and determining the input variable;

[0025] The determined input variable includes: 550 nm aerosol optical depth and air relative humidity in the temporal dimension; wind speed in the spatial dimension; center wavelength in the spectral dimension; solar zenith angle, observation zenith angle and relative azimuth angle in the angular dimension.

[0026] Further, the step 4 uses the sea-air coupled radiation transfer model to construct the simulation scene, and sets the input variable parameter space according to the input variable determined in the step 3.

[0027] Further, the step 6 includes the first-order sensitivity index, the total effect sensitivity index and the second-order sensitivity index.

[0028] The first-order sensitivity index is: ;

[0029] The total effect sensitivity index is:

[0030] The second-order sensitivity index is:

[0031] wherein, to obtain the model with input variables by variance decomposition formula based on variance decomposition, the contribution of the input variable to the deviation of the result is obtained; the contribution of the interaction of the and the input variable to the deviation of the result; represents the amount of change in considering the output variance when the parameter is fixed, that is, the total population minus the part not containing the parameter ; represents the variance decomposition formula; , represents the number of input variables.

[0032] Further, the step 7 comprises:

[0033] The step 7 comprises: repeating steps 5 and 6 to perform multiple simulations, calculating the standard deviation of the first-order sensitivity index M , the total effect sensitivity index , , and the change value between adjacent two samplings , , , and the change value between adjacent two samplings , ,

[0034] Further, in the step 8, the global sensitivity index and the initial parameter in the step 7 are coupled with the relative difference root mean square of the atmospheric top reflectivity of each ocean series satellite band and the sensitivity coefficient , to obtain the quantitative influence of the coupling effect between parameters on the inter-satellite radiation reference transfer accuracy; including:

[0035] Step 8.1, for each input parameter , records the relative difference root mean square obtained by the initial parameter pre-screening in each ocean series satellite band , and converts it into absolute error ;

[0036] Step 8.2 calls the global sensitivity analysis result, and the total effect sensitivity index​ denoted as weight , the second order sensitivity denoted as coupling weight :

[0037]

[0038]

[0039] Step 8.3 Calculate the reference transfer total error of each ocean satellite band :

[0040]

[0041] where, is the new variable defined as the ocean satellite band , is the reflectance of the s-th sample of the ocean satellite band , is the normalized error variable of the reflectance denotes the top-of-atmosphere reflectance of the reference scene, i.e. the undisturbed scene, of the ocean satellite band , is the number of all global sensitivity analysis samples, is the second order raw moment of each model output , i.e. square each model output and then average over all samples.

[0042] Further, the step 9 comprises: ranking the main error sources according to the first order sensitivity index and the total effect index;

[0043] For the bands that exceed the radiation reference transfer accuracy requirements, implement matching threshold optimization or lookup table compensation for high contribution parameters;

[0044] If the parameter interaction effect is significant, establish a coupling correction model for overall correction.

[0045] On the other hand, the present application provides a global sensitivity analysis system for ocean satellite radiation reference transfer, comprising:

[0046] An initial parameter acquisition module, which is used to select initial parameters affecting the top-of-atmosphere reflectance from the time domain, the spatial domain, the spectral domain and the angular domain based on the uncertainty sources in the ocean satellite radiation reference transfer link;

[0047] A radiation characteristic data set construction module, which is used to select a stable ocean target area in time and space, and construct a monthly-grid radiation characteristic data set through variation coefficient threshold screening;

[0048] The input variable determination module is used for sensitivity pre-screening of coupling of each ocean series satellite band with the initial parameters respectively, and determining the input variable;

[0049] The parameter space setting module is used for setting the input variable parameter space according to the input variable and the monthly-grid radiation characteristic data set, and based on the sea-air coupled radiation transmission model to construct a radiation transmission simulation scene;

[0050] The output data set acquisition module is used for low-difference sampling of the input variable parameter space, generating an input variable combination and performing radiation transmission simulation, and acquiring the atmospheric top reflectivity output data set;

[0051] The global sensitivity index calculation module is used for global sensitivity analysis of the input variable combination and the atmospheric top reflectivity output data set, and acquiring the global sensitivity index;

[0052] The sampling scale adjustment module is used for dynamically adjusting the sampling scale through the convergence global sensitivity index double threshold method;

[0053] The total error calculation module is used for combining the global sensitivity index with the atmospheric top reflectivity index of coupling of each ocean series satellite band with the initial parameters respectively, and calculating the total error of each band benchmark transmission;

[0054] The uncertainty source analysis module is used for analyzing and judging the uncertainty source in the ocean satellite radiation benchmark transmission process according to the global sensitivity index and the total error of each band benchmark transmission.

[0055] Compared with the prior art, the present application has the following beneficial effects:

[0056] 1) The present application sorts out the uncertainty sources of the ocean satellite radiation benchmark transmission from the time domain, the space domain, the spectral domain and the angle domain, and quantitatively identifies the independent and coupled effects of each factor through the global sensitivity analysis method, which is more comprehensive and scientific than the traditional local sensitivity analysis;

[0057] 2) By optimizing the ocean stable target grid, extracting the radiation characteristics of the real ocean target, constructing the monthly radiation characteristic data set of 47 high applicability grids from 2012 to 2022, limiting the simulation input range accordingly, improving the consistency between simulation and actual application, reducing the transmission error caused by the space-time stability of the water-atmosphere condition of the benchmark transmission target, and improving the precision of the ocean satellite radiation benchmark transmission;

[0058] 3) The sampling strategy is combined with the convergence double threshold method, which takes into account the efficiency and accuracy, and provides a feasible technical path for actual large-scale satellite radiation benchmark transmission and satellite calibration, cross-calibration engineering;

[0059] 4) Combined with the results of global sensitivity analysis, the two-stage coupling discrimination mechanism of single parameter coupling satellite series of ocean sub-band atmospheric top reflectivity relative difference root mean square and sensitivity coefficient, the sensitivity index (first-order sensitivity index, total effect sensitivity index and second-order sensitivity index) obtained by global sensitivity analysis, and the direct correlation of radiation reference transfer accuracy and The quantitative influence of the coupling between parameters on the accuracy of radiation reference transfer is obtained.

[0060] 5) According to the influence mechanism of key factors such as observation geometry, aerosol distribution, wind speed on atmospheric top reflectivity and the actual situation of real ocean stable target, the reference transfer uncertainty evaluation scheme, matching constraint condition optimization and further radiation reference transfer influence correction scheme are proposed according to the sensitivity analysis results, so as to improve the accuracy and applicability range of ocean satellite radiation reference transfer. BRIEF DESCRIPTION OF DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0062] Figure 1 The flow chart of the embodiment of the present application.

[0063] Figure 2 The total distribution diagram of Chla in five fields of the embodiment of the present application, (a) is the total distribution of Chla in five fields from 2012 to 2022, (b) is the total distribution of Chla in 47 preferred grids from 2012 to 2022. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical scheme and advantages of the present application more clear, the technical scheme in the present application will be described clearly and completely in the following combined with the drawings in the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the present application.

[0065] Embodiment 1

[0066] As Figure 1 shown, the flow of the embodiment of the present application includes the following steps:

[0067] Step 1, analysis of the uncertainty sources of the satellite radiometric reference transfer link, and selection of the input parameter items for global sensitivity analysis.

[0068] According to the error sources in time, space, spectrum, and angle.

[0069] (1) In the transfer link, different transfer time intervals will cause time domain differences in water and atmosphere. For the optimal area of the marine stable target, the main influencing parameters of the water body include chlorophyll concentration Chla, sediment concentration, 440 nm detritus absorption coefficient, and 440 nm absorption coefficient of yellow substances, and the main influencing parameters of the atmosphere include aerosol optical depth AOD and air relative humidity Rh.

[0070] (2) The spatial influence mainly includes the scale effect caused by the inconsistency of the spatial resolution of the reference instrument and the target instrument and the spatial positioning error. For the optimal area of the marine stable target, the spatial homogeneity is high, the positioning difference has little influence, and the influence of spatial coverage can be ignored. The spatial uncertainty of the radiation change of the optimal target mainly reflects the difference in solar glint of the rough water-air interface under different observation scales (spatial resolution). According to the Cox-Munk theory, higher wind speed will cause larger waves on the sea surface, thereby enhancing the specular reflection effect of the sea surface, that is, the solar glint phenomenon, which will cause the radiation signal of part of the wave band to be excessively amplified, increasing the uncertainty of the radiation transmission. Therefore, the influence of the spatial condition difference is converted into the main parameter of the sea surface roughness: wind speed.

[0071] (3) The spectral domain difference is introduced by the difference in the spectral response function SRF of the target instrument and the reference instrument, the difference in the center wavelength, and the spectral characteristics of the observed target. For the known two sensors, the spectral response function SRF is a determined two-function or sequence. For the selected reference and target sensors, the difference in SRF finally reflects the difference in the D / N value of the same atmospheric radiation received by the sensor at the same wavelength. Therefore, the output item of the global sensitivity analysis of the present application is the atmospheric reflectivity, and the center wavelength difference is selected as the input parameter item of the spectral domain influence.

[0072] (4) The angle domain difference is introduced by the observation geometry (solar zenith angle, observation zenith angle, relative azimuth angle) difference between the two instruments and the directional characteristic difference of the target radiation. For the marine stable target optimal area with high spatial homogeneity, the main influencing parameters of the angle domain are the observation geometry (solar zenith angle, observation zenith angle, relative azimuth angle).

[0073] Step 2, selection of the marine stable target, and formation of a monthly-grid radiation characteristic data set suitable for the marine satellite radiometric reference transfer.

[0074] To address the error sources in the above time-space-spectrum-angle domain, first, the stable target field of the sea is selected, which has stable optical properties and high spatial homogeneity. As shown in Table 1, five internationally recommended stable target fields of the sea with clean water and stable atmospheric conditions are selected [North Pacific Gyre (NPG), South Pacific Gyre (SPG), North Atlantic Gyre (NAG), South Atlantic Gyre (SAG), and South Indian Ocean Gyre (SIG)]. These target fields are located in open ocean gyre areas away from continental pollution sources, with clean atmospheric conditions. The main components and sources of aerosols in the atmosphere are relatively simple, reducing the complexity caused by aerosols from different sources; at the same time, the typical oligotrophic characteristics are used, and the concentration of yellow substance and other parameters are set to zero, simplifying the analysis process. On this basis, smaller grid ranges are divided, and the variation coefficient (CV) threshold of formula (1) (Rrs555, Rrs488, Rrs443, and Chla variation coefficient thresholds are set to 20%, 28%, respectively) is set. The multi-source measured data from January 1, 2012 to January 1, 2022 are selected for screening (the basic information of the data is shown in Table 2), and the preferred stable target grid of the sea is obtained. The extreme value range of the radiation characteristics of the above preferred grid is counted, and the parameter with larger variation range is selected as the input variable for subsequent global sensitivity analysis. The variation coefficient threshold is set to 28%).

[0075]

[0076] wherein, is the monthly standard deviation of each parameter; is the monthly average value of each parameter.

[0077] Table 1 Location of internationally commonly used stable target fields of the sea

[0078]

[0079] Table 2 Basic information of data

[0080]

[0081] Step 3, according to the uncertainty sources of the radiation reference transfer link of marine satellites, all parameters are coupled with each marine satellite band to do sensitivity pre-screening and calculate the corresponding sensitivity coefficient, and the input variables for global sensitivity analysis are selected accordingly.

[0082] Taking HY-1C-COCTS as the target sensor (bandwidth 20 nm, each channel center wavelength 412, 443, 490, 520, 565, 670, 750, 865 nm, relative spectral response function RSR sequence interval 1 nm) as an example, under each wave band, based on the key radiation characteristic data set in the 1°*1° grid with higher applicability selected from the earlier marine stable target candidate area, and in view of the distribution frequency of each time domain parameter in the marine stable target optimization area from 2012 to 2022, the interval closest to 90% and 95% coverage is found by symmetrically expanding the peak position of the normal distribution fitted, as shown in Figure 2 Fig. 2, wherein Fig. 2(a) is the total distribution of Chla in the five field areas from 2012 to 2022, and Fig. 2(b) is the total distribution of Chla in the 47 optimization grids from 2012 to 2022. Taking the distribution of chlorophyll a concentration Chla in (b) of Figure 2 as an example, under the premise that the center of the maximum probability density is the reference value, the univariate offset is calculated by formula (1) and formula (2) respectively. The relative difference root mean square of the top-of-atmosphere reflectance and the sensitivity coefficient .

[0083]

[0084]

[0085] wherein, is the top-of-atmosphere reflectance without offset, the offset top-of-atmosphere reflectance is obtained after adding different offsets (including positive offset top-of-atmosphere reflectance and negative offset top-of-atmosphere reflectance ), F represents the number of samples under the offset , represents the traversal index, is the equal amplitude perturbation of the variable in the positive and negative directions.

[0086] According to the relative difference root mean square of the top-of-atmosphere reflectance and the sensitivity coefficient of each parameter coupled with the wave bands of each marine series satellite, it is determined whether each parameter has a significant impact on the radiation reference transfer based on the marine stable target optimization area under each wave band, and the input variables of global sensitivity analysis are selected accordingly.

[0087] Step 4, determination of global sensitivity analysis input variables and construction of radiation transfer simulation scene.

[0088] Based on the above preferred marine stable target field parameter extreme value range, the parameter space of radiation transfer simulation is constructed. The OSOAA sea-air coupled radiation transfer model is used as a simulation tool, and the basic scene conditions of the simulation are set as clear sky and no cloud, and the aerosol type is selected as marine aerosol. The main input variables include: aerosol optical thickness in time domain, wind speed in time domain, central wavelength in spectral domain, observation relative azimuth angle in angular domain, solar zenith angle, and observation zenith angle. The value range of the main input variables of OSOAA (as shown in Table 3) and the reference value of other input parameters (as shown in Table 4) are mainly obtained from the following three aspects: combining the radiation characteristic extreme value range extracted based on the marine stable target field in step 2, and comprehensively considering the analysis range of 0-16 m / s (of which the frequency is higher) determined by the wind speed distribution frequency analysis; the observation geometric condition is set according to the possible differences between the reference star and the target star, the relative azimuth angle considers 0° and 180°, which are two extreme cases to cover the strongest and weakest polarization glare conditions, and because when the solar zenith angle is greater than 70°, the performance of the chlorophyll-a extraction algorithm in satellite data verification is significantly reduced, resulting in a significant increase in error. Therefore, the present application is set in the range of 0-70° for the solar zenith angle, 0-90° for the observation zenith angle, and 0-180° for the relative azimuth angle. According to the needs of the reference transfer, the wavelength range is set to 380-2300 nm with 1 nm interval for the zenith reflectance, so as to cover the typical visible and near-infrared wavebands of marine remote sensing. Through the above setting, the radiation transfer simulation scheme covering various possible difference conditions is established.

[0089] Table 3 Variation range of input variables for global sensitivity analysis

[0090]

[0091] Table 4 Reference state value of other input parameters of OSOAA radiation transfer simulation

[0092]

[0093] Step 5, sampling of each variable parameter combination and radiation transfer simulation calculation.

[0094] Step 3. Sampling in the multi-dimensional parameter space to generate a large number of simulation scenarios. A quasi-Monte Carlo method (e.g. Sobol sequence) is used to sample the parameter space uniformly and efficiently. N sets of input variable combinations are obtained, where N must be a power of 2 to ensure the convergence in the sensitivity index calculation. For each set of input variable combinations, the OSOAA model is called to simulate the top-of-atmosphere reflectance as the observed result under the conditions of the reference star or target star. The large number of simulation outputs are stored as the data basis for the sensitivity analysis. Due to the use of the Sobol sequence quasi-Monte Carlo method sampling strategy and the OSOAA implementation of the simulation considering the sea-atmosphere interface coupling, this step can obtain a database of top-of-atmosphere reflectance outputs covering the real change conditions of the preferred ocean stable target.

[0095] Step 6. Global sensitivity analysis calculation.

[0096] The input variable combination-top-of-atmosphere reflectance output data set obtained by simulation is subjected to global sensitivity analysis. The Sobol-based global sensitivity analysis (GSA) method is used to decompose the model with d input variables into :

[0097]

[0098] where is the contribution of the th input variable, is the interaction contribution of the th and th input variables, is the output variance produced by the simultaneous interaction of all variables. Based on this decomposition, the first-order sensitivity index , total effect sensitivity index and second-order sensitivity index are defined as:

[0099] Step 6.1. Calculate the first-order sensitivity index .

[0100] The contribution rate of each input variable to the output top-of-atmosphere reflectance deviation when the variable changes alone is evaluated. Using the variance decomposition principle, the proportion of the output variance caused by the variable when other variables are randomly taken is calculated.

[0101] ​

[0102] Step 6.2, Calculate the total effect sensitivity index .

[0103] Evaluation variables The total effect of the interaction on the output. Consider the amount of variance of the output when the parameter is fixed, i.e. subtract from the population the part that does not contain the parameter . .

[0104]

[0105] Step 6.3, Calculate the high order (mainly second order) interaction effects .

[0106] On the basis of steps 6.1 and 6.2, calculate the second order interaction sensitivity index between the important parameters , , for identifying significant parameter coupling effects.

[0107]

[0108] Step 7, Automatically expand the sample size by the convergence double threshold method.

[0109] Increasing the number of random sampling can theoretically increase the analysis accuracy, but when it increases to a certain number, it is no longer possible to distinguish whether the difference in the result value is due to the selection of random numbers or the increase in the number of random sampling, i.e. at this time the increase in the number of random sampling is meaningless. In order to balance the accuracy and efficiency of the result calculation, according to the above step 5 non-interfering 5 times (1st, …, m, …, 5th simulation) repeated experiments, the standard deviation of the first order sensitivity index , the total effect sensitivity index between the same simulation (m is a fixed value M, which means a certain time in the above 1-5 simulations, M can be 1, 2, 3, 4, 5) is calculated , , the change value between adjacent two samplings and , and the average value of and corresponding to all N values (2 power, the upper limit of N is set to 2 14 =16384, the experimental results verify that: usually in 2 10~12 has reached the best) in the same experiment m=M is calculated respectively and , when the standard deviation of and obtained by formula (7), (8) and the change value between adjacent two samplings , When all values ​​are less than the threshold, the judgment result has converged; if it has not yet converged, the number of samples should be increased further. N Until the above threshold requirements are met, the specific calculation method is as follows, where Indicate the input variable number (e.g., aerosol optical thickness AOD, wind speed, etc.); according to step 5, The following table serves as a counter to represent the current variable. A sampling in the Mth experiment (therefore) To indicate the previous time, for example: It is 2 9 The second sampling, then It is 2 10 (second sampling)

[0110]

[0111]

[0112]

[0113]

[0114] The specific threshold requirement is the standard deviation of a single simulation (m=M). , Less than 0.05; and in the 1st, ..., mth, ..., 5th simulations, the input variable in the nth sampling of the 5 groups (n is a constant N-1) corresponding , Input variables in the nth sampling of 5 groups (n is a constant N) corresponding The difference is taken as the average. , Threshold for change:

[0115]

[0116]

[0117] Step 8, combine the root mean square of the relative differences in atmospheric top reflectivity obtained in Step 3. and sensitivity coefficient Step 6 (after verification in step 7) Global sensitivity index (first-order sensitivity index) Total effect sensitivity index and second-order sensitivity index Through a two-level error quantization coupling evaluation method, the global sensitivity index is linked to the band-specific radiometric accuracy directly. and Quantitative indicators are unified to the same degree of measurement framework, realizing the traceable mapping of "parameter coupling-band accuracy", and providing quantitative transfer matching constraint optimization and error evaluation link closed loop for the satellite-target satellite radiation reference transfer.

[0118] Step 8.1, for each input parameter , record each determined band center wavelength in the marine series satellite band (with values) within the range Step 8.2, call the global sensitivity analysis results, and set the total effect index as the weight , and the second-order interaction index as the coupling weight .

[0119] Step 8.3, calculate the expected total error of each band reference transfer . First, define a new variable by formula (13), which represents the normalized error variable of the reference scene (no disturbance) atmospheric top reflectivity and the reflectivity generated by the Sobol global sensitivity analysis of the s-th sample . Therefore, is the relative difference between the "disturbance scene-reference scene":

[0120] Step 8.4, calculate the expected total error of each band reference transfer . First, define a new variable by formula (13), which represents the normalized error variable of the reference scene (no disturbance) atmospheric top reflectivity and the reflectivity generated by the Sobol global sensitivity analysis of the s-th sample . Therefore, is the relative difference between the "disturbance scene-reference scene":

[0121]

[0122] Note that in a certain determined band (i.e. take a certain value), it is equivalent to an observation sample of the model in formula (3).

[0123] Based on the radiation reference transfer of the marine stable target preferred area, the parameter disturbance range is small, and the first-order Taylor expansion of the radiation transfer model under the premise of "small offset disturbance + orthogonal decomposition" can be written as formula (14), and the local sensitivity coefficients and ( and Index of input variable, value range 1- For each determined wave band :

[0124]

[0125] According to the definition in step 8.1, The small deviation (can be positive or negative) of the input variable relative to the reference value, Similarly, the average of all Sobol global sensitivity analysis samples (the total number of samples is ) (that is, the average of the deviation of the model for multiple random observations, the result is equivalent to the definition of expectation in probability theory), the present application constructs a random sequence :

[0126]

[0127] The variance of the random variable is the second order central moment defined in probability theory, so the other variance decomposition form of the model equivalent to equation (3) is:

[0128]

[0129] Since step 5 uses Sobol sequence in low-discrepancy sequence for sampling, the average error (numerically, that is, the expectation ) can quickly approach 0 as the sample size increases. According to the accuracy requirement of interstellar radiation reference transfer, if cannot be ignored (strictly =0), then mirror expansion is done on the sample set, and the sensitivity coefficient is updated:

[0130]

[0131]

[0132] Equation (14) and equation (15) are combined, squared on both sides to obtain equation (16). Refer to the variance decomposition method of equation (5) and equation (6), decompose equation (16) which is approximately equivalent to equation (3) to obtain equation (17) and equation (18). Combine the absolute error in step 8.1 to further derive the total error synthesis equation (19) to obtain the total error of reference transfer :

[0133]

[0134]

[0135]

[0136]

[0137] Combining the preceding equations (3) - (15), is the covariance operation, is the variance decomposition operation, represents the individual input contribution to the output variance, represents the joint contribution of variables to .

[0138] By step 8, the list of parameters and their upper limits that need to be constrained or compensated for each band can be given; at the same time, if another ocean satellite is replaced as the target star, only the RSR (Relative Spectral Response function) curve and the single parameter are replaced, and the method can be repeated.

[0139] Step 9, global sensitivity result analysis and uncertainty evaluation.

[0140] According to the calculated sensitivity indices, the sources of uncertainty in the radiometric reference transfer process of the ocean satellite are analyzed and judged. Factors with high first-order sensitivity indices (indicating that the independent effect of the factor is significant) and factors with high total effect indices (indicating that the influence including interaction is significant) are focused on. The sensitivity analysis results are converted into guidance for the reference transfer: on the one hand, the main sources of uncertainty are identified, and their contribution to the difference in the top-of-atmosphere radiation is quantified; on the other hand, for secondary factors, it can be considered to appropriately relax the strict matching requirement or simplify the processing in actual calibration. If necessary, the reference transfer method can be improved based on the sensitivity analysis results, for example, an empirical correction model is established for the top several factors with the highest total effect to reduce the system bias introduced by them. In addition, the rationality of the selection of the ocean stable target can also be verified through sensitivity analysis: if the analysis shows that some environmental parameters (such as aerosols) still produce greater uncertainty in the selected target area, it is necessary to reselect a more stable area or further constrain the observation conditions. The specific judgment basis is:

[0141] Through the first-order sensitivity index, the influence of each input variable on the output result is sorted. The more sensitive the influence is, the more stringent the constraint condition of the variable in the transfer process needs to be;

[0142] If the total combined error exceeds the accuracy requirement of the radiometric reference transfer, the most contributing or Tighten the matching threshold or use lookup table compensation until... The requirements are met. Specifically, for the determined center wavelength of the band... Each input parameter The corresponding error Coupling error The relative root mean square difference obtained from the single-parameter pre-screening in step 8.1, combined with step 3. The relationship between them satisfies equations (20) and (21):

[0143]

[0144]

[0145] Based on the relative magnitudes of the total effect sensitivity index and the first-order sensitivity index, the strength of the coupling effect between the variable and other variables can be determined. If the coupling effect of the variable has a small impact on the output atmospheric top reflectivity, much smaller than the impact of a single variable (more than two orders of magnitude different), the interaction effect between the parameters can be considered negligible. The influence of the atmospheric top reflectivity introduced by the differences of all input variables during the transmission process (the resulting uncertainty) can be simplified to a function between the total effect sensitivity index of equation (22) and the difference in the change of a single input, and then summed.

[0146]

[0147] in, This represents the final synthesized reflectance error, which is the overall root mean square (RMS) error of reflectance obtained after the combined effect of the uncertainties of all input variables. This indicates the number of parameters that affect the accuracy of radiation reference transfer; Indicates the counting subscript, used to iterate through numbers from 1 to... Each input variable; Indicates the first The baseline values ​​used in the benchmark analysis for each input variable. It is the first The disturbance of each input variable, representing The magnitude of deviation from the benchmark value; It is the first The sensitivity coefficient of each variable represents " The percentage of the root mean square (RMS) deviation of the atmospheric top reflectivity when the unit change is (e.g., 1%, 1 mg / m²). 3 wait).

[0148] The second-order sensitivity index matrix of the parameters with strong coupling effect is analyzed to determine the variables with strong coupling effect. The parameters with strong coupling effect need to be considered as a whole in the uncertainty analysis. Thus, the embodiments have been described in detail with reference to the drawings, and based on the above description, researchers in the field should have a clear understanding of the technical solutions of the present application for utilizing global sensitivity analysis to quantitatively evaluate the multi-factor uncertainty of the marine satellite radiation reference transfer link.

[0149] Unless specifically described or steps must occur in sequence, the order of the steps described above is not limited to the above list, and can be changed, combined or rearranged according to the design required by different transfer targets, references or target star states. And the above embodiments can select different global sensitivity analysis input variables, ranges according to the analysis method of step 1, select the sampling number considering the accuracy and efficiency according to the evaluation method of step 6, and also can select the recommended preferred marine stable target in the embodiments and the selected target satellite to form more embodiments.

[0150] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other apparatus. The time required to simulate the OSOAA in step 4 is related to the configuration of the computer device. In addition, the present application is not directed to any particular programming language, and researchers can use various programming languages to implement the content of the present application described herein.

[0151] Embodiment 2

[0152] The present embodiment provides a global sensitivity analysis system for marine satellite radiation reference transfer, comprising:

[0153] An initial parameter acquisition module is used to select initial parameters affecting the atmospheric top reflectivity from time domain, space domain, spectral domain and angular domain based on the uncertainty sources in the marine satellite radiation reference transfer link;

[0154] A radiation characteristic data set construction module is used to select a marine target area with stable time and space, and construct a monthly-grid radiation characteristic data set through variation coefficient threshold screening;

[0155] An input variable determination module is used to pre-screen the sensitivity of the initial parameters coupled with each marine satellite band to determine the input variables;

[0156] A parameter space setting module is used to construct a radiation transfer simulation scene based on the input variables and the monthly-grid radiation characteristic data set, and set the input variable parameter space based on the sea-air coupled radiation transfer model;

[0157] An output data set acquisition module, which is configured to perform low-discrepancy sampling on an input variable parameter space, generate input variable combinations, perform radiation transfer simulation, and acquire an atmospheric top reflectivity output data set;

[0158] A global sensitivity index calculation module, which is configured to perform global sensitivity analysis on the input variable combinations and the atmospheric top reflectivity output data set, and acquire a global sensitivity index;

[0159] A sampling scale adjustment module, which is configured to dynamically adjust the sampling scale by using a double-threshold method of converging the global sensitivity index;

[0160] A total error calculation module, which is configured to couple the global sensitivity index and the initial parameters with the atmospheric top reflectivity indexes of each satellite wave band of the marine series, and calculate the total error of each wave band reference transfer;

[0161] An uncertainty source analysis module, which is configured to analyze and determine the uncertainty sources in the radiation reference transfer process of the marine satellite according to the global sensitivity index and the total error of each wave band reference transfer.

[0162] It should be understood that the parts not elaborated in the specification are all prior art.

[0163] It should be understood that the above description of the preferred embodiments is detailed, and therefore should not be considered as a limitation on the scope of patent protection of the present application. It is not necessary or possible to enumerate all the embodiments here. Those skilled in the art can make substitutions or modifications without departing from the scope of the claims of the present application, and all such substitutions or modifications fall within the scope of the present application. The scope of protection of the present application should be subject to the appended claims.

Claims

1. A method for global sensitivity analysis of marine satellite radiometric reference transfer, characterized in that, The method comprises the following steps: Step 1. Based on the uncertainty sources in the satellite radiation reference transmission link, the initial parameters affecting the atmospheric top reflectivity are selected from the time domain, the spatial domain, the spectral domain and the angular domain respectively; Step 2. Selecting a spatiotemporally stable marine target area, constructing a monthly-grid radiation characteristic data set through a variation coefficient threshold screening; Step 3. Sensitivity pre-screening of the initial parameters coupled with each marine satellite band respectively to determine the input variables; Step 4. According to the input variables and the monthly-grid radiation characteristic data set, and based on the sea-air coupled radiation transfer model, a radiation transfer simulation scene is constructed, and the input variable parameter space is set; Step 5. Low-discrepancy sampling is performed on the input variable parameter space to generate input variable combinations and perform radiation transfer simulation to obtain an atmospheric top reflectivity output data set; Step 6. Global sensitivity analysis is performed on the input variable combinations and the atmospheric top reflectivity output data set to obtain a global sensitivity index; Step 7. The sampling size is dynamically adjusted through a convergence global sensitivity index double threshold method; Step 8. The global sensitivity index is combined with the atmospheric top reflectivity index of each marine satellite band coupled with the initial parameters to calculate the total error of each band reference transmission; Step 9. According to the global sensitivity index and the total error of each band, the uncertainty sources in the marine satellite radiation reference transmission process are analyzed and judged.

2. A global sensitivity analysis method for ocean satellite radiance benchmark transfer according to claim 1, characterized in that, The initial parameters in step 1 include: Time domain: chlorophyll concentration Chla, sediment concentration, 440 nm detritus absorption coefficient, 440 nm yellow substance absorption coefficient, aerosol optical depth AOD and air relative humidity Rh; Spatial domain: wind speed; Spectral domain: center wavelength difference; Angular domain: solar zenith angle, observation zenith angle and relative azimuth angle.

3. A global sensitivity analysis method for ocean satellite radiance benchmark transfer according to claim 1, characterized in that, The step 2 specifically includes: Selecting 3-8 marine areas as spatiotemporally stable marine target areas; dividing 1°x1° grid, calculating the variation coefficient threshold of each initial parameter in each grid, and finding the grid position with a long-time series monthly variation coefficient smaller than the variation coefficient threshold to obtain an optimized grid; Statistical analysis of the initial parameter radiation characteristic data set in the optimized grid.

4. A global sensitivity analysis method for ocean satellite radiance benchmark transfer according to claim 1, characterized in that, The step 3 is sensitive pre-screening of coupling each ocean series satellite band with the initial parameters respectively, to obtain the relative difference root mean square of the atmosphere top reflectivity of coupling each ocean series satellite band with the initial parameters respectively and the sensitivity coefficient , to determine whether each initial parameter has a significant impact on the radiation reference transmission of the marine target area under each ocean series satellite band, and to determine the input variable; The determined input variables include: time domain 550 nm aerosol optical depth and air relative humidity; spatial domain wind speed; Spectral domain center wavelength; angular domain solar zenith angle, observation zenith angle and relative azimuth angle.

5. A global sensitivity analysis method for ocean satellite radiance benchmark transfer according to claim 1, characterized in that, In step 4, the simulation scene is constructed using the sea-air coupled radiation transfer model, and the input variable parameter space is set according to the input variables determined in step 3.

6. A global sensitivity analysis method for ocean satellite radiance benchmark transfer according to claim 4, characterized in that, In step 6, the global sensitivity index includes the first-order sensitivity index, the total effect sensitivity index and the second-order sensitivity index; First order sensitivity index is: ; Total effect sensitivity index is: Second order sensitivity index is: in, In order to have Input variables model Through variance decomposition Based on variance decomposition, the first... The contribution of each input variable to the bias of the result; For the first and the The contribution of the interaction of the input variables to the bias of the results; Indicates considering when the parameter The change in output variance when fixed is the total variance minus the variance excluding parameters. Part of; Represents the variance decomposition formula; , This indicates the number of input variables.

7. A global sensitivity analysis method for marine satellite radiometric benchmark transfer according to claim 6, characterized in that, The step 7 comprises: repeating the steps 5 and 6 to perform multiple simulations, calculating the first-order sensitivity index of the nth simulation M , , , , , , , , , , , when the first-order sensitivity index of the nth simulation, the total effect sensitivity index, the standard deviation of the nth simulation, the change value between the adjacent two simulations, the change amount between the adjacent two simulations are all less than the preset threshold value respectively, it is determined that the result has converged; if it is still not converged, the sampling number is continuously increased until the preset threshold value requirement is met.

8. A global sensitivity analysis method for ocean satellite radiance benchmark transfer according to claim 6, wherein, In step 8, the global sensitivity index and the initial parameters in step 7 are coupled with the relative difference of the atmospheric top reflectivity of each satellite band of the ocean series to obtain the quantitative influence of the coupling effect between parameters on the inter-satellite radiation reference transfer accuracy. and the sensitivity coefficient , including: Step 8.1 for each input parameter The relative difference root mean square of the initial parameter pre-screening is recorded in each ocean series satellite band , converted into absolute error ;​ Step 8.2 calls the global sensitivity analysis results to get the total effect sensitivity index denoted as weight , the second order sensitivity denoted as coupling weight : Step 8.3 Calculation of the total error of the reference transfer for each ocean series satellite band : where, is the defined ocean color satellite band is a new variable, is the ocean color satellite band is the reflectance produced by the s-th sample is the normalized error variable, is the ocean color satellite band is the top-of-atmosphere reflectance of the reference scene, i.e., the undisturbed scene is the total number of global sensitivity analysis samples, is the second-order raw moment of each model output i.e., square each model output and average over all samples.

9. A global sensitivity analysis method for ocean satellite radiance benchmark transfer according to claim 6, wherein, Step 9 includes: sorting the main error sources according to the first-order sensitivity index and the total effect index; For the bands that exceed the radiation reference transmission accuracy requirement, the matching threshold optimization or lookup table compensation is implemented for the high-contribution parameters; if the parameter interaction effect is significant, a coupled correction model is established for overall correction.

10. A system for global sensitivity analysis of marine satellite radiometric reference transfer, characterized in that, It comprises: The initial parameter acquisition module is used for selecting initial parameters affecting the atmospheric top reflectivity from time domain, space domain, spectral domain and angle domain based on uncertainty sources in the ocean satellite radiation reference transfer link; The radiation characteristic data set construction module is used for selecting a marine target area with stable time and space, and constructing a monthly-grid radiation characteristic data set through variation coefficient threshold screening; The input variable determination module is used for coupling each ocean series satellite band with the initial parameters respectively to perform sensitivity pre-screening and determine input variables; The parameter space setting module is used for constructing a radiation transfer simulation scene based on the input variables, the monthly-grid radiation characteristic data set and the ocean-atmosphere coupled radiation transfer model, and setting an input variable parameter space; The output data set acquisition module is used for low-discrepancy sampling of the input variable parameter space, generating input variable combinations, performing radiation transfer simulation, and acquiring an atmospheric top reflectivity output data set; The global sensitivity index calculation module is used for global sensitivity analysis of the input variable combinations and the atmospheric top reflectivity output data set, and acquiring a global sensitivity index; The sampling scale adjustment module is used for dynamically adjusting the sampling scale through a convergence global sensitivity index double-threshold method; The total error calculation module is used for combining the global sensitivity index with the atmospheric top reflectivity index of each ocean series satellite band coupled with the initial parameters, and calculating the total error of each band reference transfer; The uncertainty source analysis module is used for analyzing and judging the uncertainty sources in the ocean satellite radiation reference transfer process according to the global sensitivity index and the total error of each band reference transfer; The global sensitivity analysis system for the ocean satellite radiation reference transfer is used to perform the steps in the global sensitivity analysis method for the ocean satellite radiation reference transfer according to any one of claims 1-9.

Images