CN113030905A

CN113030905A - Aerosol laser radar data quality control method and system

Info

Publication number: CN113030905A
Application number: CN202110376076.4A
Authority: CN
Inventors: 杨婷; 王海波; 王自发
Original assignee: Institute of Atmospheric Physics of CAS
Current assignee: Institute of Atmospheric Physics of CAS
Priority date: 2021-04-07
Filing date: 2021-04-07
Publication date: 2021-06-25

Abstract

The invention discloses a method for aerosol laser radar data quality control, which mainly comprises Monte Carlo uncertainty analysis, an image denoising algorithm and bilateral filtering. Wherein the Monte Carlo uncertainty analysis is based on random sampling of the lidar signal to distinguish between high and low uncertainty extinction coefficients; the image denoising algorithm is used for removing pulse noise in the laser radar data; bilateral filtering smoothes the lidar data and preserves the data edge structure. Correspondingly, the invention also discloses a system for controlling the data quality of the aerosol laser radar. The method shows good improvement effect on the laser radar data under long-time laser radar observation and pollution cases.

Description

Aerosol laser radar data quality control method and system

Technical Field

The invention relates to the technical field of aerosol laser radars, in particular to a method and a system for aerosol laser radar data quality control.

Background

The aerosol laser Radar (Lidar) is used as the extension of the traditional Radio or microwave Radar (Radio) to the optical frequency band, can provide the vertical distribution characteristic of high space-time resolution, And is a powerful tool for detecting the vertical structure of the aerosol.

As shown in fig. 1, the aerosol laser radar transmits laser to the atmosphere through a laser, receives backscattering signals of atmospheric molecules and atmospheric aerosol by using a telescope system, and carries out inversion of echo signals by using a laser radar equation to actively remotely sense the vertical structure of the atmospheric aerosol. The aerosol laser radar is based on an elastic scattering Mie theory, and the radar equation is as follows:

wherein r represents the height corresponding to the current echo signal, P (r) is the echo signal power at the height of r, P₀Is the laser radar initial echo power, C is the laser radar system parameter, beta_m(r) and alpha_m(r) represents the backscattering coefficient and extinction coefficient, beta, of atmospheric molecules at r height, respectively_a(r) and alpha_a(r) denotes the backscattering coefficient and extinction coefficient, respectively, of the atmospheric aerosol at the height r. The Fernald method is a single-wavelength inversion algorithm (Femald, 1984) that is mature and accurate at present, and is commonly used for inverting the vertical profile of the extinction coefficient by assuming the lidar ratio.

The method can be divided into space-based laser radars, aerial survey laser radars and foundation laser radars according to the deployment position of the laser radars. The air-based Lidar is mainly composed of CALIPO (Cloud-Aerosol Lidar and interferometric radar) and CATS (Cloud-Aerosol Transport System). CalipsO was developed by the cooperation of ESSP (earth System Science Pathof) of NASA (the National Aeronautics and Space administration) of the United states and CNES (Centre National d' animals spheres) in the republic of China, aiming to fill the gap in the observation of the global distribution and physicochemical properties of aerosols and clouds (Winker, 2003). In 9 months 2018, CALIPO fell about 16.5km from the "A-train" orbit, and the CloudSat satellite jointly constituted "C-train". CalipsO is a 16 day cyclic sun synchronous orbiting satellite. CALIPER (the cloud-Aerosol Lidar with Orthogonal Polarization) is a dual wavelength (532nm and 1064nm) elastic scattering Lidar mounted on CALIPO, with the vertical profile of total (Aerosol to air molecules) attenuation backscatter measured directly at the nadir view (Grigas et al, 2015). The CALIP moves at a speed of about 7km/s on the surface and the pulse repetition frequency is about 20Hz, so the horizontal resolution of the CALIP at the lower layer is about 333 m. In addition, the vertical resolution is 30m at the lower layer. CATS lidar is a dual wavelength (532nm and 1064nm) lidar largely at iss (the International Space station) (McGill et al, 2015). CATS have a height of about 415km with an inclination of 51 °, and are observed in tropical and medium latitudinal regions with a repeating cycle of about 4 days (Yorks et al, 2016). Currently, many ground-based Lidar networks have been constructed and put into use, such as AD-net (asian dual network) (Sugimoto et al, 2014), earlinet (the european airborne Research Lidar network) (Sicard et al, 2015), mplnet (the micro Pulse Lidar network) (Welton, 2002), and coralan (the terrestrial Operational Research Lidar network) (McKendry et al, 2011), among others. Aerosol lidar is widely used in atmospheric pollution studies such as characterization of aerosol vertical structures (Huang et al, 2012; Liu et al, 2013; Sun et al, 2013; Tian et al, 2018), assessment of pollution management effects (Yang et al, 2010), assessment of atmospheric chemical transport patterns (Li et al, 2012, 2013; Wang et al, 2014), determination of boundary layer heights (Liu et al, 2015; Yang et al, 2017b), inversion of aerosol components (Burton et al, 2012; sugimoo et al, 2014) and calculation of aerosol liquid water content (to Tan et al, 2020a, 2020 b).

Along with the higher and higher requirements on the quantitative program of the aerosol laser radar data, the research on the quality control of the aerosol laser radar data is gradually developed. Item derivation (2018) is based on a laser radar hardware level, and discusses aspects such as signal acquisition modes and acquisition accumulation times in an optical-mechanical structure and an electronic module, selection of calibration height and radar ratio, calibration of an visibility meter, calibration of Rayleigh scattering, calibration of a sunshine photometer and the like. Even if hardware quality control is ensured, further software-level quality control is required. The data products of aerosol lidar are time series of vertical profiles of extinction coefficients, which have more complex information content and dimensionality than conventional ground observation, and therefore quality control schemes for conventional ground observation cannot be directly applied to lidar (Durre et al, 2010; Wu et al, 2018). Meanwhile, conventional outlier detection methods such as soliton (Liu et al, 2008) and K-order proximity (ramanswamy et al, 2000) cannot directly process aerosol lidar data products because the detected outliers are large values of the extinction coefficient representing the pollution plume. Scc (single Calculus chain) of EARLINET is a set of automated tools for laser radar data quality control, and includes hardware quality control module ELPP (EARLINET Lidar Pre-Processor) and software quality control module elda (EARLINET Lidar dataanalyzer) (D' Amico et al, 2015). The ELDA uses the laser radar signal after ELPP quality control as input to invert the aerosol optical properties, and performs software-level quality control after inversion (Mattis et al, 2016). The ELDA includes many classical algorithms such as monte carlo uncertainty analysis, error propagation, and also includes specially developed algorithms such as vertical smoothing and time averaging. Although some research has been done on the applicability and accuracy of ELDA (D' Amico et al, 2015), ELDA is only a lidar data software quality control scheme for europe, and the applicability to china with atmospheric pollution (He et al, 2002) remains to be studied. In addition, the ELDA currently has no smoothing scheme for elastic scattering lidar. This is probably because the only smoothing scheme in ELDA is for raman radar design, where the inversion requires derivation of the original signal. Besides the ELDA scheme applicable to Europe, the comprehensive software-level laser radar data quality control research is relatively few. Therefore, it is necessary to develop related research on quality control of the aerosol laser radar software level.

Disclosure of Invention

The invention aims to provide a method for controlling the quality of aerosol laser radar data, so as to optimize the laser radar data from a comprehensive software level.

In order to solve the technical problems, the technical scheme of the invention is as follows:

a method for aerosol laser radar data quality control comprises the following steps:

s1, acquiring laser radar data for detecting the atmospheric aerosol by using the aerosol laser radar system, and entering the step S2 when the numerical value is 0-1; when the value is > 1, the value is set to 1, and then the process proceeds to step S3; when the value is < 0, setting the value < 0 and the flag data to 0, and then proceeding to step S4;

s2, carrying out Monte Carlo uncertainty analysis on the laser radar data according to the radar ratio and the radar signals, when a high uncertainty value exists, rejecting the high uncertainty value, and entering the step S4; when there is a low uncertainty value, proceed directly to step S3;

s3, denoising the laser radar data by adopting an image denoising algorithm, and when the salt and pepper noise exists, rejecting the salt and pepper noise and entering the step S4; otherwise, directly entering step S4;

and S4, smoothing the laser radar data by adopting a bilateral filtering image processing algorithm, and further obtaining the extinction coefficient after quality control.

Further, the step S1 of acquiring lidar data for detecting atmospheric aerosol by using the aerosol lidar system includes:

s11, calculating the signal mean value and standard deviation of each height;

s12, establishing Gaussian distribution in each height layer based on the mean value and the standard deviation;

s13, randomly extracting N samples in Gaussian distribution of each height layer, and accordingly constructing N groups of laser radar signal vertical profiles;

s14, the N groups of radar signals and the original radar signals enter the aerosol laser radar system together to obtain N +1 groups of extinction coefficient vertical profiles;

and S15, obtaining the extinction coefficient mean value and the standard value of each height layer through N groups of extinction coefficient profiles, thereby constructing the extinction coefficient relative uncertainty vertical profile.

Further, in step S2: the atmospheric boundary layer height was set to 1 km.

The relative uncertainty threshold below the atmospheric boundary layer was set at 20% and the uncertainty threshold above was set at 30%.

Further, in step S2: the radar ratio is assumed to be 50 sr.

Further, the relative uncertainty of the radar ratio is set to 10%.

Further, in step S3: non-0 grid points less than or equal to 3 around each grid point are removed.

The invention also discloses a system for controlling the data quality of the aerosol laser radar, and the method is used.

Compared with the prior art, the invention has the beneficial effects that:

the method comprises Monte Carlo uncertainty analysis, an image denoising algorithm and bilateral filtering, wherein the Monte Carlo uncertainty analysis is based on random sampling of laser radar signals so as to distinguish extinction coefficients with high uncertainty and low uncertainty; the image denoising algorithm is used for removing pulse noise in the laser radar data; bilateral filtering smoothes the laser radar data and can keep a data edge structure; the invention mainly removes the abnormal value at the upper part of the aerosol, improves the smoothness of the aerosol layer and maintains the original vertical structure of the aerosol.

Drawings

FIG. 1 is a block diagram of aerosol lidar hardware;

FIG. 2 is a flow chart of a method for aerosol lidar data quality control according to the present invention;

FIG. 3 is a scatter plot of the Vertical Average Extinction Coefficient (VAEC) characterization of the present invention for

months

1, 4, 7 and 10 in 2017, where FIG. 3a shows the relationship between the VAEC before and after the quality control of the TRANSFER; FIG. 3b shows the relationship between VAEC before TRANSFER quality control and VAEC after DN quality control; FIG. 3c shows the relationship between VAEC before and after TRANSFER quality control; FIG. 3d shows the relationship between the MCA and the DN after the VAEC is controlled; FIG. 3e shows the relationship between the post-MCA quality control and the post-TRANSFER quality control VAEC; FIG. 3f shows the relationship between DN after quality control and VAEC after quality control of TRANSFER;

FIG. 4 is a graph showing the comparison of the vertical average extinction coefficients before and after TRANSFER quality control after wind vector coupling for 2017 in

months

1, 4, 7 and 10 and 4 months;

FIG. 5 is a graph showing a comparison of vertical average extinction coefficients before and after TRANSFER quality control after visibility results were coupled for 2017 for 1 month, 4 months, 7 months, 10 months, and 4 months in accordance with the present invention;

FIG. 6 is a graph showing comparison of vertical average extinction coefficients before and after TRANSFER quality control after RH coupling for 2017 in

months

1, 4, 7 and 10 and 4 months;

FIG. 7 shows that the invention has coupling PM of 1 month, 4 months, 7 months, 10 months and 4 months in 2017_2.5A schematic diagram for comparing vertical average extinction coefficients before and after TRANSFER quality control after mass concentration;

FIG. 8 shows the coupling O between the present invention and the coupling O for the present invention for the following months, 1 month, 4 months, 7 months, 10 months and 4 months₃A schematic diagram for comparing vertical average extinction coefficients before and after TRANSFER quality control after mass concentration;

FIG. 9 is an effect diagram of the original data of extinction coefficient time-height cross-section diagram inverted by the NIES laser radar system from 1 day to 3 days in 7 months in 2017 (a), after MCA quality control (b), DN quality control (c), BF quality control (d) and quality control increment (e);

FIG. 10 is a scatter plot of the extinction coefficient Vertical Average (VAEC) characterization of the present invention from 1 to 3 days 7/month 2017.

Detailed Description

The following further describes embodiments of the present invention with reference to the drawings. It should be noted that the description of the embodiments is provided to help understanding of the present invention, but the present invention is not limited thereto. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

The invention discloses a method for aerosol laser radar data quality control, which is a set of integration scheme specially aiming at aerosol laser radar data quality control and can also be called TRANSFER (a data acquisition and control scheme for aerosol radar), as shown in figure 2, the method mainly comprises Monte Carlo uncertainty analysis and bilateral filtering, and specifically comprises the following steps:

The monte carlo method can provide an approximate solution to mathematical problems through statistical sampling (fisherman, 1996), which is a common method for dealing with nonlinear inverse problems and simultaneously considering various uncertainty factors (dawn, 2010). The Monte Carlo Analysis (MCA) used in TRANSFER is based on random sampling of the lidar signal to distinguish between high and low uncertainty extinction coefficients. The radar signal has a vertical resolution of 6m and the output extinction coefficient vertical profile has a resolution of 30m by bilinear interpolation in the aerosol lidar system (Yang et al, 2017 a). To characterize the propagation of uncertainty from input to output, the signal mean and standard deviation at each height were calculated using a 5-grid sliding window corresponding to the vertical resolution of the lidar product. Based on the mean and standard deviation, a gaussian distribution is established at each height level. This way of establishing a gaussian distribution is also used in ELDA (Mattis et al, 2016). 50 samples were randomly drawn in the gaussian distribution of each height layer, thus constructing 50 sets of lidar signal vertical profiles. These 50 sets of radar signals entered the aerosol lidar system along with the original radar signal to yield 51 sets of vertical profiles of extinction coefficients. The extinction coefficient mean value and the standard deviation of each height layer can be obtained through 50 groups of extinction coefficient profiles, and therefore the extinction coefficient relative uncertainty vertical profile is constructed. Data is carried out on the extinction coefficient vertical profile obtained by inverting the original signal through the extinction coefficient relative uncertainty vertical profile constructed aboveAnd (4) quality screening. Considering that atmospheric pollution mainly occurs in the atmospheric boundary layer, different extinction coefficient relative uncertainty thresholds are respectively set above and below the boundary layer. In 2006-2017, the height of the atmospheric boundary layer in north china plains is about 1km (Su et al, 2018), and meanwhile, in order to reduce the height of the boundary layer calculated based on the laser radar, the height of the atmospheric boundary layer in the invention is set to 1 km. The relative uncertainty threshold below the atmospheric boundary layer was set at 20%, which is the same as the setting in ELDA (Mattis et al, 2016). Considering the relatively small integrated error below the scaled height using the Fernald inversion method and the relatively low boundary layer height setting of the present invention, the threshold above the boundary layer is set to 30%. In aerosol lidar systems, the radar ratio is assumed to be 50sr, a value typical for city aerosols and dust aerosols (Liu et al, 2002; Shimizu, 2004; Shimizu et al, 2016), and also a value that is not susceptible to relative humidity at 532nm (Takamura and Sasano, 1987). However, the radar ratio depends on the spatial inhomogeneity of the aerosol (Kovalev, 1995) and varies with height. The uncertainty of radar ratios has been studied extensively, including different aerosol compositions (Ackerman and Toon, 1981; Muller et al, 2007), different geographic locations (Tesche et al, 2007) and different heights (C: (A) (R))

et al, 2012). Since the focus of the TRANSFER is on the degree of improvement of the lidar data product by the data quality control scheme, the relative uncertainty of the radar ratio is set to 10% with reference to the recent correlation study (Xie et al, 2010).

After monte carlo uncertainty analysis, some high uncertainty values still exist in the lidar data. The time sequence of the extinction coefficient vertical profile can be regarded as two-dimensional data, and the one-dimensional ground conventional observation quality control method cannot be directly applied. Therefore, the TRANSFER employs a simple but practical image DeNoising algorithm (DN) to remove impulse noise in the lidar data, i.e., based on the principle that non-0 grid points around each grid point are removed when the number of the grid points is less than or equal to 3.

In the aerosol laser radar system, 3000 echo signals within every 5min are averaged to obtain a radar signal vertical profile every 15 min. In order to prolong the service life and reduce the maintenance cost, the laser of the radar is stopped during the 10min (Shimizu, 2004). Although the time resolution of the aerosol lidar product is 15min, the changing characteristics of the aerosol within 10min of the laser being off cannot be captured. This results in an inverted vertical profile of extinction coefficients that in principle does not fully characterize the aerosol variations. Therefore, in this case, the inside of the plume needs to be smoothed while keeping the edge of the plume contaminated by the atmosphere observed by the conventional laser radar. Gaussian low pass filtering smoothes data by computing the weighted values of surrounding points, where the weighted values decay with distance. Gaussian low-pass filtering only takes into account the correlation over distance and ignores the similarity between the values themselves, which can cause the smoke plume edge structure in the lidar data to be destroyed. Therefore, TRANSFER smoothes lidar data using a Bilateral Filtering image processing algorithm (BF) that smoothes the data while preserving the data edge structure (Tomasi and Manduchi, 1998). This algorithm takes into account both distance proximity and value domain similarity. The filter calculation formula of BF to extinction coefficient f (x) is as follows:

where c (ξ, x) characterizes the proximity of the distance between the neighboring point ξ and the center point x, and s (f (ξ), f (x)) represents the degree of similarity in extinction coefficient values between two points. The distance proximity function c (ξ, x) and the value domain similarity function s (f (ξ), f (x)) are calculated as follows:

where d (ξ, x) represents the distance between two points and δ (f (ξ), f (x)) represents the difference in extinction coefficient values between two adjacent points. Sigma_dAnd σ_rRespectively representing the standard deviation of the distance and the extinction coefficient, and respectively setting the distances of 0.2 grid and 0.2km in the TRANSFER^-1. The discretized integration diameter is set to 5 pitches.

Taking an NIES lidar as an example, the actual effect and improvement degree of TRANSFER on lidar data processing are briefly described below. The observation periods for NIES lidar were 1, 4, 7 and 10 months in 2017. The reference data used the AERONET product. AODs with wavelengths of 440nm, 500nm and 670nm in AERONET products were converted to AODs of 532nm by quadratic polynomial interpolation (Eck et al, 1999) with higher accuracy and reliability than the Angstrom interpolation. And (4) integrating the original extinction coefficient and the extinction coefficient subjected to quality control by TRANSFER in the vertical direction to obtain the AOD based on the laser radar. The results of statistical tests on AOD products after conversion of extinction coefficients before and after quality control by TRANSFER to AOD and AERONET are shown in Table 1. As shown in Table 1, the RMSE (the Root Mean Square error) decreased from 0.87 to 0.51, by a relative decrease of 41%, while the MFE (the Mean Fractional error) decreased by 17%, while the Correlation Coefficient (R) improved only to a limited extent, by TRANSFER quality control. Therefore, it can be seen that the TRANSFER improves the absolute deviation of the extinction coefficient by the MCA, DN and BF integration method, and has little influence on the linear trend of the original data.

Table 1: RMSE, MEF and correlation coefficient of AOD of extinction coefficient conversion before and after TRANSFER quality control of 1 month, 4 months, 7 months and 10 months in 2017 with AOD of AERONET

Note: BT represents the absolute increment before and after TRANSFER quality control, AT represents the absolute increment after and after TRANSFER quality control, RI represents the relative increment before and after TRANSFER quality control

After proving that the TRANSFER quality control scheme positively contributes to the improvement of extinction coefficient in the whole observation period, next, pairwise comparison is carried out between sub-schemes in the TRANSFER scheme so as to quantitatively evaluate the influence degree of different sub-schemes on extinction data. FIG. 3 is a Vertical Average Extinction Coefficient (VAECs) scattergram before and after quality control of TRANSFER steps. The abscissa of each row in fig. 3 is the original VAEC, MCA-passed VAEC and DN-passed VAEC, respectively, and the ordinate of each column is the MCA-passed VAEC, DN-passed VAEC and whole TRANSFER fer scheme, respectively. From the perspective of the effect of each sub-scheme on the data alone (fig. 3a, 3d and 3f), the sub-scheme with the greatest effect on the original data is MCA, with the greatest RMSE (0.08 km)^-1) And the lowest correlation coefficient (0.91); secondly DN at RMSE of 0.04km^-1The correlation coefficient was 0.96; the impact of BF is minimal. The change in the TRANSFER neutron scheme from the raw data to the data can be seen along fig. 3a, 3b and 3 c. As shown, the scatter in the scatter plot is primarily centered on coordinates (0.05 km)^-1，0.045km^-1) In the vicinity, the extinction coefficient raw data as a whole shows a tendency of slight overestimation, close to a straight line with a slope of 1. There will still be several under-estimated scatter points in the graph, as will be discussed later. But at the coordinate (0.55 km)^-1，0.35km^-1) And coordinates (0.5 km)^-1，0.1km^-1) Two clusters of scattered dots are respectively present nearby. The second scatter cluster has a relatively large extinction coefficient value before quality control and a relatively low extinction coefficient value after quality control, which is mainly caused by the large uncertainty of these data points in the raw data. In the present invention, this range of scatter point clusters is called "virtual high area", and similarly, the first scatter point cluster is called "real high area".

VAEC before and after TRANSFER quality control in fig. 3c was coupled with meteorological factors to further investigate the performance of TRANSFER under different meteorological conditions (fig. 4a, 5a, 6 a). As shown in FIG. 4a, most of the wind vectors are centered on coordinates (0.05km-1, 0.045km-1) plusIn recent years, strong north wind is the main cause. The wind vector with relatively small wind speed and southward wind direction is mainly in the real high area, while the wind vector in the virtual high area has low wind speed and no uniform wind direction. As shown in fig. 5a, high visibility occurs mainly when there is strong north wind. The case where the visibility is less than 5km occurs in both the "real high zone" and the "virtual high zone". In addition, the "solid high zone" also includes some cases where the visibility is close to 15 km. Relative Humidity (RH) less than 40% occurs mainly at the coordinates (0.05 km)^-1，0.0045km^-1) Nearby. The "real high regions" present both cases with RH greater than 80% and cases with RH equal to about 60%, while the "dummy high regions" present both cases with RH greater than 80%. It can be seen by coupling meteorological factors that overestimation of VAECs, particularly in "real high regions", occurs mainly in conditions of weak south winds, low visibility, high RH, etc., which are characteristic of the typical adverse meteorological conditions in north china plains (Wang et al, 2014).

The absorption and scattering of gases and particulates contribute to the extinction coefficient together, with aerosols contributing significantly to the overall extinction coefficient (Singh and Dey, 2012). Recent research has found HO_xAnd NO_xThe presence of free radicals gives PM_2.5And O₃With a strong negative correlation (Li et al, 2019a, 2019 b). Therefore, the ground PM_2.5And O₃Direct and indirect effects on the extinction occur, respectively. These two pollution factors were assigned to VAEC scattergrams in the same way as the meteorological factors (fig. 7a and 8 a). PM in "real high zone_2.5The mass concentration is more than 250 mu g/m³Coordinate (0.05 km)^-1，0.045km^-1) PM of nearby and "virtual high zone_2.5The concentration is less than 75 mu g/m³。O₃And PM_2.5Having a similar distribution structure, which may be due to the ground O₃When the concentration is higher, the aerosol pollution smoke plume with higher contribution to the extinction is mainly positioned at high altitude, and when the PM is on the ground_2.5At higher concentrations, the aerosol pollution clusters are mainly concentrated on the ground.

Although the overall performance of TRANSFER during observation was analyzed in combination with meteorological factors and pollution factors, the dominant factors in different seasonsAre different from each other. As can be seen from Table 1, RMSE exhibits a decreasing trend in summer, winter, spring and fall in absolute increments and a completely reverse order in relative changes. The above results depend on the degree of deviation of the AOD of AERONET and may vary depending on the different calibration data chosen. In order to study the influence of TRANSFER on extinction coefficient data itself in different seasons, RMSE and correlation coefficients were used to characterize the degree of change of data before and after quality control, rather than the degree of conformance to the AOD of AERONET. As shown in fig. 4b, in winter, there is a significant difference in wind vector: strong northern wind appears at the coordinates (0.05 km)^-1，0.045km^-1) Nearby, weak winds occur in the high-realized zone. The visibility and RH distribution laws in winter are similar (fig. 5b and 6 b). At coordinate (0.05 km)^-1，0.045km^-1) The situation that the visibility is more than 10km, even reaches 30km, and the RH is less than 20% occurs nearby, while the situation that the visibility is lower than 5km and the RH is higher than 80% occurs in a 'real high region'. In the "real high region" PM_2.5The mass concentration reaches 250 mu g/m³Above, and at the coordinate (0.05 km)^-1，0.045km^-1) Nearby PM_2.5The mass concentration is less than 75 mu g/m³. Winter, O₃The mass concentration was low in each of the scattered-point cluster regions (fig. 8 b). Four seasons, winter VAECs with minimal variation, maximum correlation coefficient (0.93) and minimum RMSE (0.07. mu.g/m)³) Whereas summer VAECs vary the most with the smallest correlation coefficient (0.8) and the largest RMSE (0.15. mu.g/m)³). At coordinate (0.05 km)^-1，0.045km^-1) Visibility of more than 15km occurs, while visibility of less than 5km occurs in both "real high zones" and "virtual high zones". The distribution structure of RH in summer is similar to visibility. However, part of the high visibility and low RH scatter occurs in the "real high zone", mainly due to the different vertical distribution of the aerosol. Summer O₃Scatter distribution and PM in winter_2.5The scattered points are distributed similarly, and the ozone mass concentration of the solid high area is more than 200 mu g/m³And at the coordinate (0.05 km)^-1，0.045km^-1) Nearby O₃The concentration is less than 50 mu g/m³. The large "solid high zone" in FIGS. 7a and 8a was found in comparison with spring and summerValues are respectively PM_2.5High concentration of oxygen and oxygen in winter₃High concentration contribution in summer.

If VAECs are able to provide rich information about the atmosphere column, the horizontal average of the extinction coefficients can give a positive contribution of TRANSFER to the vertical structure of the aerosol. As shown in fig. 9, the average profile of extinction coefficient after TRANSFER fer quality control is negative, which indicates that systematic overestimation of the extinction coefficient of the aerosol lidar system is corrected, which is consistent with the foregoing conclusion of the vertical average of extinction coefficients. The absolute value of the increment before and after the quality control of the TRANSFER in summer is the maximum in four seasons within the height range of 0.5-6 km, which indicates that the improvement degree of the extinction coefficient of the TRANSFER to summer is the maximum. Below 2km, the absolute value of the increment is greater in winter than in spring, and vice versa above 2 km. The extinction coefficient time-averaged profiles except for spring, summer, autumn and winter have maxima in the atmospheric boundary layer of 0.09km each^-1、0.09km^-1And 0.05km^-1The heights appeared to be 0.95km, 0.6km and 0.4km, respectively. The difference arises because changes in the height of the atmospheric boundary layer are effected by thermal convection, mechanical turbulence, and meteorological conditions (Guinot et al, 2006). As shown in FIG. 9a, the average vertical profile of extinction coefficient in summer has a maximum value between 0.5km and 2km due to aerosol zone transport (He et al, 2009) and high level PM occurring in summer_2.5Has a high hygroscopicity (Sun et al, 2013).

Although the data quality control performance of the TRANSFER quality control scheme observed by the laser radar for a long time is verified, the improvement degree of the TRANSFER on the data quality of the laser radar needs to be researched under the pollution condition. Contamination cases from 7/1 to 7/3 in 2017 were used to examine the quality control effect of TRANSFER. The results show that after TRANSFER fer quality control, the RMSE of the AOD scaled by extinction coefficient during this contamination example and the AOD of AERONET decreased from 1.4 to 0.84, and the MFE also decreased from 0.58 to 0.43.

Fig. 9 shows the quality control effect of the TRANSFER sub-scheme in the quality control scheme during this contamination example, and fig. 10 is similar to fig. 3 and shows the comparison result of VAEC in each quality control sub-scheme during the contamination example. As shown in the figure, the laser radar successfully captures the pollution processMultiple extinction coefficient of (4) is large. FIG. 9a shows that in the range of 2-6 km, there is much impulse noise. Although these impulse noises can be removed artificially, the removal situation may vary from operator to operator. Meanwhile, some impulse noises cannot be directly distinguished. FIG. 9b shows the result of MCA quality control on the original data, which shows that a large number of high uncertainty values are removed, some of the values can be directly distinguished and some of the values cannot be distinguished, wherein the effect in the range of 2-4 km is most remarkable. The blind areas detected by the lidar are also removed to some extent because of the high relative uncertainty. As shown in FIG. 10a, the extinction coefficient data after MCA quality control shows that a large number of VAECs are centered on the coordinates (0.1 km)^-1，0.09km^-1) Nearby, followed by "real high regions", only a few scatters appear in "virtual high regions". The DN algorithm adopted in the invention is a supplement of MCA, and can continuously remove the residual impulse noise after MCA quality control. Above 4km, the effect of the DN algorithm is more pronounced. The complement of DN can also be shown in FIG. 10d, and the RMSE and correlation coefficient of the MCA-controlled VAEC and the DN-controlled VAEC were 0.05km each^-1And 0.96, less than and greater than the RMSE and correlation coefficient of the original data and the MCA quality-controlled data, respectively. BF is the last step of the TRANSFER quality control scheme, and the extinction coefficient data after MCA quality control and DN quality control is subjected to edge-preserving denoising, namely the edge of the extinction coefficient large-value area is not damaged and the inside of the large-value area is smoothed. As can be seen from fig. 10f, after BF quality control, the extinction coefficient data trend is not changed, and the correlation coefficient is about 1.0. As can be seen from fig. 9e, the TRANSFER mainly removes the outlier located at the upper part of the aerosol layer, improves the smoothness of the aerosol layer, and maintains the original aerosol vertical structure. Overestimation of the raw data is mainly due to the uncertainty present in some parameter settings (e.g. threshold to reject clouds).

The protection scope of the invention is not limited by the invention, and the invention also discloses a method for controlling the data quality of the aerosol laser radar.

The embodiments of the present invention have been described in detail with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It will be apparent to those skilled in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, and the scope of protection is still within the scope of the invention.

Claims

1. A method for aerosol laser radar data quality control is characterized by comprising the following steps:

2. The method of claim 1, wherein the step S1 of acquiring lidar data for detecting atmospheric aerosol with an aerosol lidar system comprises:

s11, calculating the signal mean value and standard deviation of each height;

3. The method for aerosol lidar data quality control according to claim 2, wherein in step S2: the atmospheric boundary layer height was set to 1 km.

4. The aerosol lidar data quality control method according to claim 3, wherein: the relative uncertainty threshold below the atmospheric boundary layer was set at 20% and the uncertainty threshold above was set at 30%.

5. The method for aerosol lidar data quality control according to claim 1, wherein in step S2: the radar ratio is assumed to be 50 sr.

6. The aerosol lidar data quality control method according to claim 5, wherein: the relative uncertainty of the radar ratio was set to 10%.

7. The method for aerosol lidar data quality control according to claim 1, wherein in step S3: non-0 grid points less than or equal to 3 around each grid point are removed.

8. A system for controlling data quality of aerosol laser radar, comprising: use of a method according to any preceding claim.