A cloud-aerosol state classification method based on max-dois observation dependent on cloud height

Abstract

The application discloses a cloud-aerosol state classification method based on cloud height dependence of MAX-DOAS observation, and belongs to the technical field of atmospheric remote sensing detection. The method comprises the following steps: collecting MAX-DOAS multi-elevation angle spectral data, calculating normalized zenith light intensity, zenith light intensity time change disturbance index, high-elevation angle O4 differential slant column concentration and aerosol optical thickness; determining an effective sensitive height as a low cloud and high cloud demarcation threshold based on an average kernel matrix; and through step-by-step discrimination, the observation is classified into eight states, i.e. clear sky, no cloud high aerosol, high cloud low aerosol, high cloud high aerosol, low cloud low aerosol, low cloud high aerosol, optical thick cloud and cloud hole. The application does not need to rely on external cloud detection equipment, and can realize automatic identification of cloud height dependent cloud-aerosol state only by using MAX-DOAS observation data, and the classification result can provide data screening and quality control basis for MAX-DOAS aerosol and trace gas inversion.

Description

Technical Field

[0001] This invention belongs to the field of atmospheric remote sensing technology, specifically relating to a cloud aerosol state classification method based on cloud height dependence of MAX-DOAS observations. It is applicable to the quality control and classification utilization of MAX-DOAS observation data in fields such as urban pollution monitoring, regional transport process analysis, boundary layer structure research, and satellite remote sensing product verification. Background Technology

[0002] Ground-based multiaxial differential absorption spectroscopy (MAX-DOAS) is a passive optical observation technique that uses scattered solar light for remote sensing of atmospheric composition. This technique receives scattered spectra from the sky at different elevation angles and utilizes the differences in sensitivity to the near-surface atmosphere through different observation paths to obtain information on the vertical distribution of aerosols and trace gases. Compared to point sampling instruments, MAX-DOAS offers advantages such as non-contact operation, continuous observation, high temporal resolution, sensitivity to the near-surface atmosphere, and suitability for long-term automated operation. Therefore, it is widely used in urban pollution monitoring, regional transport process analysis, boundary layer structure research, satellite remote sensing product validation, and trace gas and aerosol profile inversion.

[0003] A typical ground-based MAX-DOAS observation system usually consists of an outdoor telescope or optical receiver, an azimuth / elevation scanning mechanism, fiber optic transmission components, a spectrometer, a detector, a temperature control unit, a data acquisition and control unit, and back-end inversion processing software. The outdoor telescope or optical receiver receives scattered sunlight from different directions; the scanning mechanism controls the instrument to switch observations between multiple azimuth and elevation angles; the fiber optic cable transmits the received light signals to the indoor spectrometer; the spectrometer uses beam splitters such as gratings to decompose the incident light into spectra of different wavelengths; a CCD or CMOS detector converts the light signals into electrical signals; the data acquisition and control unit is responsible for integration time adjustment, dark current measurement, scan sequence control, and spectral data storage; and the back-end software performs spectral fitting, differential oblique column concentration calculation, radiative transfer simulation, and profile inversion.

[0004] The basic observation principle of MAX-DOAS is as follows: After solar radiation enters the atmosphere, it is scattered by air molecules, aerosols, and cloud particles. Ground-based instruments receive the scattered light from different elevation angles. Low elevation angle observation paths have a longer propagation distance in the near-surface atmosphere, making them more sensitive to near-surface aerosols and trace gases; high elevation angle or zenith direction observation paths are shorter and are more often used to construct reference spectra or characterize background radiation conditions. By performing differential absorption analysis on the observation spectra at different elevation angles, the differential slant column concentrations (dSCD) of absorbing components such as NO2, SO2, HCHO, and O4 can be obtained. Among them, the absorption signals of trace gases such as NO2, SO2, and HCHO can be used to invert the column concentrations and vertical profiles of pollutants; O4 is an oxygen molecule collision complex, and its concentration distribution is mainly determined by the atmospheric oxygen concentration, making it relatively stable. Therefore, changes in the O4 differential slant column concentration mainly reflect changes in the effective propagation path of photons and atmospheric scattering conditions.

[0005] In MAX-DOAS aerosol inversion, the O4 differential oblique column concentration is a crucial observation constraining the aerosol extinction profile. Typically, the O4 absorption signal is first used, combined with a radiative transfer model and optimal estimation methods, to invert parameters such as the aerosol extinction coefficient profile and aerosol optical thickness (AOD). Subsequently, the aerosol inversion results are used as the radiative transfer background for the inversion of trace gas profiles such as NO2, SO2, and HCHO. Therefore, the aerosol inversion results not only affect the monitoring of the aerosols themselves but also further influence the accuracy of subsequent trace gas inversions. In this inversion chain, whether the observation optical path is affected by both clouds and aerosols is one of the key factors determining the reliability of the inversion results.

[0006] Both clouds and aerosols alter the propagation path of solar radiation in the atmosphere, but their mechanisms of influence on MAX-DOAS observations differ. Aerosols, typically distributed in the near-surface and boundary layers, primarily affect the low-elevation O4 differential column concentration and its elevation angle distribution by enhancing scattering and altering the near-surface optical path. As aerosol loading increases, the optical path differences between different low-elevation observation paths may decrease, changing the dispersion of the low-elevation O4 differential column concentration. Clouds have stronger scattering, reflection, and blocking effects; their influence depends not only on the presence of clouds but also on cloud height, cloud optical thickness, cloud geometric thickness, and spatial inhomogeneity of the cloud field. Thin clouds may enhance scattered radiation, while low clouds or optically thick clouds may block incident and outgoing light, resulting in reduced radiance, abnormally low O4 differential column concentration, or inversion failure.

[0007] Because MAX-DOAS is primarily sensitive to the near-surface atmosphere, and its vertical sensitivity decreases rapidly with increasing altitude, the impact of cloud layers at different altitudes on MAX-DOAS observations and inversions varies significantly. When low clouds are within the effective sensitive altitude range of MAX-DOAS, they tend to directly truncate the optical path at high elevation angles, resulting in a significant reduction in the O4 differential oblique column concentration at the 30° isohyet elevation angle. Under the clear-sky assumption, when performing optimal aerosol estimation and inversion, this type of optical path change caused by low clouds may be misattributed by the algorithm to enhanced aerosol extinction, leading to an overestimation of AOD or near-surface extinction coefficient. In contrast, when high clouds are above the instrument's main sensitive layer, their impact on the effective near-surface optical path is relatively weak, and observations under some high-cloud conditions may still retain value for aerosol or trace gas inversion. Therefore, in long-term MAX-DOAS observations, simple cloud screening alone cannot fully reflect the usability of observational data under different cloud conditions.

[0008] The implementation schemes most similar to this invention currently include the following categories.

[0009] The first category comprises cloud detection and screening methods based on MAX-DOAS observations. These methods typically utilize color index (CI), zenith radiance, radiance temporal variation, O4 differential column concentration, or combinations thereof, to determine whether the observation is under clear skies, cloud cover, or disturbance. The advantage of these methods is that they require no additional observation equipment and maintain temporal consistency with MAX-DOAS data; however, most methods primarily serve for cloud screening or cloud cover / clearness assessment, focusing on eliminating data affected by clouds rather than further describing relative cloud height, aerosol loading, and cloud-aerosol mixing conditions. Therefore, these methods struggle to determine the differential impact of low clouds, high clouds, cloud holes, optically thick clouds, and high aerosol events on the inversion results.

[0010] The second category relies on external active or imaging observation equipment to acquire cloud attributes. For example, lidar and cloud radar can provide high vertical resolution information on cloud height and structure; all-sky cameras can acquire cloud cover, cloud movement, and sky coverage; and satellite cloud products can provide information on cloud top height, cloud optical thickness, and cloud cover over a wide range. These methods can obtain cloud structure parameters relatively directly, but require additional equipment or external data sources, resulting in high construction and maintenance costs, difficulty in long-term continuous operation, inconsistencies with MAX-DOAS observation paths, and significant temporal and spatial matching errors. For most long-term observation sites that only deploy MAX-DOAS, external cloud profiling equipment is not a standard configuration, making it difficult to use as a necessary input for operational classification processes.

[0011] Specifically, the existing technology has the following shortcomings: (1) The traditional MAX-DOAS cloud identification method mainly determines whether the observation is affected by clouds, but cannot fully distinguish between low clouds and high clouds. Since low clouds and high clouds have different effects on the effective optical path of MAX-DOAS, the concentration of O4 differential oblique columns and the aerosol inversion results, cloud screening alone will mask the differences in data availability under different cloud conditions.

[0012] (2) The color index CI is affected by clouds, aerosols, solar zenith angle, observation geometry and instrument spectral response. Under high aerosol load and thin cloud conditions, it is easy to have a problem of not being unique in interpretation, and it is difficult to use it as a basis for classifying complex cloud aerosol states alone.

[0013] (3) Although external lidar, cloud radar or satellite cloud products can provide cloud structure information, they are costly and require high maintenance. They are not entirely consistent with the MAX-DOAS observation path, temporal resolution and spatial representativeness, and are difficult to serve as the necessary input for the long-term operational classification of most MAX-DOAS sites.

[0014] (4) When performing MAX-DOAS aerosol inversion based on the optimal estimation algorithm under the clear sky assumption, changes in the optical path caused by low clouds may be misinterpreted by the inversion algorithm as changes in aerosol extinction, resulting in an overestimation of AOD or near-surface extinction coefficient. Existing methods lack state marking for this type of cloud-induced inversion bias risk.

[0015] (5) Complex scenarios such as cloud holes, optical thick clouds, and low clouds with high aerosols are common in coastal urban areas. Traditional binary or few-category cloud screening methods cannot meet the needs of refined utilization of long-term data. Summary of the Invention

[0016] Existing MAX-DOAS cloud identification methods primarily focus on whether observations are affected by clouds, often simply categorizing observations into two states: cloud-covered and cloudless. These methods cannot adequately distinguish the position of clouds relative to the effective sensitive height of MAX-DOAS, nor can they simultaneously describe complex conditions such as cloud height, aerosol load, optically thick clouds, and cloud holes. For MAX-DOAS inversion, the impact of low and high clouds differs: when low clouds are located within the instrument's main sensitive layer, they significantly shorten the effective optical path, reducing the concentration of the high-elevation-angle O4 differential oblique column, and may be misattributed to enhanced aerosol extinction in clear-sky assumption inversions; high clouds have a relatively weaker impact on the near-surface sensitive optical path, and some observations still have inversion value.

[0017] The technical problem this invention aims to solve is to establish a cloud-aerosol state classification method that can distinguish between relative cloud height and aerosol load, without relying on external cloud observation equipment such as lidar, cloud radar, all-sky cameras, or satellite cloud products, using only MAX-DOAS's own observations and inversion derivations. This method enables automatic identification of states such as clear sky, cloudless high aerosol, high cloud low aerosol, high cloud high aerosol, low cloud low aerosol, low cloud high aerosol, optically thick cloud, and cloud hole, providing a basis for observation condition discrimination, data screening, and quality control for MAX-DOAS aerosol and trace gas inversion.

[0018] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A cloud aerosol state classification method based on MAX-DOAS observations with high cloud dependence includes the following steps: Data acquisition steps: Collect ground-based MAX-DOAS multi-elevation angle solar scattering spectral data, and perform preprocessing such as dark current subtraction and electron bias subtraction on the spectral data to obtain multi-elevation angle corrected spectra and observation auxiliary information of effective scanning sequences; The derived quantity calculation steps are as follows: Based on the multi-elevation angle corrected spectrum, calculate the normalized zenith intensity R, the zenith intensity time variation perturbation index, the high elevation angle O4 differential oblique column concentration D, the low elevation angle O4 differential oblique column concentration dispersion Sp_low, and the aerosol optical thickness AOD; preferably, the zenith intensity time variation perturbation index TSI is a zenith intensity time smoothness index, sliding window standard deviation, adjacent scan difference, normalized residual, or a perturbation index based on the residual fitted by the solar zenith angle; the high elevation angle O4 differential oblique column concentration D is the 15° elevation angle O4 differential oblique column concentration, the 20° elevation angle O4 differential oblique column concentration, the 30° elevation angle O4 differential oblique column concentration, or a weighted combination thereof. Threshold calibration steps: Determine the cloud height boundary threshold and the zenith light intensity reference threshold R. 0.01 And R1, zenith light intensity TSI threshold T0, AOD threshold, high elevation angle O4 differential oblique column concentration threshold D0, and cloud hole identification time window W; preferably, the zenith light intensity reference threshold R 0.01 R1 and R2 were determined through radiative transfer simulation: Under cloudless conditions, two scenarios were set up, AOD = 0.01 and AOD = 1, respectively, to simulate the normalized zenith intensity at the 400nm band under different solar zenith angles. A reference intensity curve consistent with the instrument detector count value and integration time was obtained. In actual classification, the corresponding R2 was obtained from the reference curve according to the solar zenith angle corresponding to the observation time. 0.01 and R1; Classification steps: Based on the comparison results between the derived quantity and the threshold, each scan sequence is classified step by step into one of eight states: clear sky, cloudless high aerosol, high cloud low aerosol, high cloud high aerosol, low cloud low aerosol, low cloud high aerosol, optical thick cloud, and cloud hole.

[0019] As a preferred technical solution, the normalized zenith intensity R is calculated according to the following formula:

[0020] Among them, DN 400 t represents the detector count value at 400 nm in the zenith direction. exp Indicates the integration time for spectral acquisition; The zenith intensity temporal variation perturbation index is the zenith intensity temporal smoothness index (TSI), which is calculated according to the following formula:

[0021] Where TSI(R,n) is the temporal smoothness index of zenith intensity corresponding to the nth scan sequence, R n R is the normalized zenith intensity of the nth scan sequence. n-1 R is the normalized zenith intensity of the previous sequence. n+1 Δt1 represents the normalized zenith intensity of the next sequence, Δt2 represents the time interval between the current sequence and the previous sequence, and Δt3 represents the time interval between the next sequence and the current sequence.

[0022] As a preferred technical solution, the dispersion of O4 differential oblique column concentration at low elevation angle, Sp_low, is defined as the difference between the maximum and minimum values ​​of the O4 differential oblique column concentration within the elevation angle range of 1° to 10°:

[0023] Wherein, dSCD(O4, 1°-10°) represents the O4 differential oblique column concentration corresponding to low elevation angle observations of 1°, 2°, 3°, 4°, 6°, 8° and 10°.

[0024] As a preferred technical solution, the cloud height boundary threshold is determined based on the average kernel matrix of MAX-DOAS aerosol inversion. Specifically, the variation characteristics of the diagonal elements of the average kernel matrix with height are extracted and its vertical envelope is constructed. The height corresponding to reducing the envelope to 10% of its maximum value is defined as the upper limit of the effective sensitive height H of MAX-DOAS aerosol inversion. m , with H m As the dividing threshold between low clouds and high clouds.

[0025] As a preferred technical solution, the classification step specifically includes: S1: If the zenith radiance is lower than R 0.01If AOD is greater than 1, or if D is abnormally low at the same time, the scanning sequence is judged to be in a state of thick optical clouds or low-quality observation. S2: If the zenith radiance is lower than R1 and the TSI is lower than T0, the scan sequence is determined to be in a state without obvious cloud disturbance; further, if the AOD is less than 0.15, it is determined to be clear sky; if the AOD is greater than 0.15, it is determined to be cloudless high aerosol. S3: For scan sequences that are not classified in S1 and S2, if D is greater than D0, they are initially identified as high clouds; if D is less than D0, they are initially identified as low clouds. S4: For high cloud conditions, if AOD is less than 0.15, it is determined to be high cloud with low aerosols; if AOD is greater than 0.15, it is determined to be high cloud with high aerosols. For low cloud conditions, if AOD is less than 0.3, it is determined to be low cloud with low aerosols; if AOD is greater than 0.3, it is determined to be low cloud with high aerosols. S5: For a scan sequence initially identified as a high cloud, search for a low cloud scan sequence within the time window W before and after it; if it exists, remark the scan sequence as a cloud hole.

[0026] The present invention has the following advantages over the prior art: (1) This invention does not require external lidar, cloud radar, all-sky camera or satellite cloud products, but only relies on the observation and inversion results of MAX-DOAS itself, making it suitable for long-term site operational applications.

[0027] (2) This invention does not only perform cloud / no cloud screening, but further distinguishes high cloud, low cloud, cloud hole, optical thick cloud and aerosol load state. The classification results are more suitable for guiding subsequent inversion and quality control.

[0028] (3) This invention links the concentration of the high elevation angle O4 differential oblique column with the effective sensitive height of MAX-DOAS, so that cloud height discrimination has a clear optical path physical basis.

[0029] (4) The present invention describes the short-term non-smooth changes in zenith radiance using TSI, which can improve the ability to identify rapidly changing cloud fields and cloud hole states.

[0030] (5) This invention takes into account the problem that AOD inversion may be too high under low cloud conditions. Different AOD load classification standards are adopted for low cloud conditions, which can reduce the classification error caused by the confusion between aerosols and cloud effects.

[0031] (6) The present invention can retain some cloud impact data but still have utilization value MAX-DOAS observation data, avoiding the waste of data caused by simply removing all cloud impact data in the traditional method.

[0032] (7) The classification results of this invention can be further used for inversion applicability judgment, quality control mark generation, long-term observation statistics and data screening under complex cloud conditions. Attached Figure Description

[0033] Figure 1 A flowchart of the method provided by the present invention; Figure 2 This is a schematic diagram illustrating the determination of the cloud relative height threshold based on the MAX-DOAS average kernel in this invention.

[0034] Figure 3 This is a schematic diagram illustrating the determination of the zenith reference radiation through radiative transfer simulation in this invention.

[0035] Figure 4 This is an example image of a cloudless, low-aerosol weather measurement in one embodiment of the present invention, wherein: (a) the aerosol state of the day measured by radar is marked, and the AOD (white dot) below 2km on the right Y-axis of the image. (b) the result of smoothing the aerosol profile measured by radar using the kernel function of MAX-DOAS. (c) the aerosol profile inverted by MAX-DOAS. (d) MAX-DOAS zenith radiation (DN value) and TSI. (e) the O4 differential oblique column concentration measured by MAX-DOAS. (f) the color index distribution calculated based on zenith radiation.

[0036] Figure 5 This is a measurement example of a cloudless high aerosol weather event according to an embodiment of the present invention.

[0037] Figure 6 This is a measurement example of high-cloud, low-aerosol weather in one embodiment of the present invention.

[0038] Figure 7 This is an example diagram of high-cloud, high-aerosol weather measurement in one embodiment of the present invention.

[0039] Figure 8 This is an example diagram of a measurement of low cloud and low aerosol weather in one embodiment of the present invention.

[0040] Figure 9 This is an example diagram of a measurement of low cloud and high aerosol weather in one embodiment of the present invention.

[0041] Figure 10 This is an example diagram of cloud hole weather measurement in one embodiment of the present invention.

[0042] Figure 11 This is an example diagram of optical thick cloud weather measurement in one embodiment of the present invention.

[0043] Figure 12 This is a flowchart illustrating the automatic classification method for cloud aerosols according to the present invention.

[0044] Figure 13 This is an example diagram of the cloud classification results measured by MAX-DOAS in Panyu District, Guangdong Province, according to an embodiment of the present invention. Detailed Implementation

[0045] The present invention will be further described below with reference to embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments are not intended to limit the present invention. In addition, unless otherwise specified, the preparation processes in the following embodiments are all conventional methods in the prior art, and therefore will not be described in detail.

[0046] Example 1 A cloud aerosol state classification method based on MAX-DOAS observations with high cloud dependence is described in the flowchart below. Figure 1 It includes six parts: data acquisition, data preprocessing, derived quantity calculation, threshold calibration, hierarchical classification, and result output. Specifically, it includes the following steps: 1. Raw Spectrum Acquisition The purpose of this step is to obtain the ground-based MAX-DOAS raw observation spectra that meet the requirements of subsequent cloud-aerosol state classification, and to perform basic quality control and signal correction on the raw spectra to form an effective scanning sequence that can be used for subsequent derived quantity calculations.

[0047] First, solar scattering spectral data are acquired using a ground-based MAX-DOAS instrument. Following a pre-set scanning program, the instrument sequentially completes observations at the zenith and multiple low elevation angles within a single scan sequence. In one embodiment, the MAX-DOAS instrument observes in the 290-430 nm wavelength range, with a spectral resolution of approximately 0.5 nm; the observation elevation angles include 90°, 1°, 2°, 3°, 4°, 6°, 8°, 10°, 15°, and 30°. The 90° zenith direction characterizes variations in zenith scattered light intensity, the low elevation angles characterize near-surface long-path observation information, and the 30° isohyet elevation angles characterize variations in high-elevation observation light paths. During acquisition, the instrument automatically adjusts the integration time based on the solar scattered light intensity to maintain the detector signal within a reasonable dynamic range, and records auxiliary information such as observation time, solar zenith angle, solar azimuth angle, instrument azimuth angle, elevation angle, integration time, and detector status during the scan.

[0048] Secondly, the acquired raw spectra are preprocessed. Since the detector still exhibits dark current and electronic bias under no-light conditions, the raw spectra need to be subtracted for dark current and electronic bias using dark signals obtained synchronously or at nearby times, resulting in corrected sky scattering spectra. Spectral data with detector saturation, abnormal integration time, excessively low light intensity, missing scan sequences, low signal-to-noise ratio, or significant instrument malfunctions are discarded or marked for quality. This step primarily ensures that the data entering subsequent calculations has stable spectral quality and a complete scan structure.

[0049] After the above processing, the multi-elevation angle corrected spectrum and its observational auxiliary information corresponding to each valid scan sequence are output. This output serves as the basic input for subsequent zenith radiance calculation, O4 differential oblique column concentration fitting, AOD inversion, and cloud-aerosol state classification.

[0050] 2. Derivative Calculation The purpose of this step is to calculate key observational derivatives for cloud-aerosol state classification based on preprocessed MAX-DOAS multi-elevation angle spectral data. These derivatives include normalized zenith intensity, zenith intensity temporal smoothness index, O4 differential oblique column concentration correlation index, and aerosol optical thickness (AOD) obtained by inversion based on the optimal estimation method.

[0051] The purpose of this step is to calculate key observational derivatives for cloud-aerosol state classification based on preprocessed MAX-DOAS multi-elevation angle spectral data. These derivatives include normalized zenith intensity, zenith intensity temporal smoothness index, O4 differential oblique column concentration correlation index, and aerosol optical thickness (AOD) obtained by inversion based on the optimal estimation method.

[0052] (1) Normalized zenith intensity R.

[0053] The visible light intensity in the MAX-DOAS zenith-direction observation spectrum is selected as a characterizing quantity of zenith radiation variation. In one embodiment, the zenith-direction detector count value DN at 400 nm is selected, and this DN value is divided by the corresponding integration time to obtain the normalized zenith intensity R:

[0054] Among them, DN 400 t represents the zenith detector count at 400 nm. exp This represents the integration time for this spectral acquisition. Since the MAX-DOAS instrument may not be absolutely radiometrically calibrated, this invention does not require obtaining absolute irradiance or absolute radiance; instead, it uses normalized detector light intensity as a proxy for actual changes in zenith radiation. This quantity can be used to characterize the influence of clouds and aerosols on the intensity of scattered light from the sky.

[0055] (2) Time smoothness index of zenith light intensity (TSI).

[0056] To characterize the smoothness of zenith intensity changes over time, a temporal smoothness index (TSI) is introduced. Under clear sky or stable aerosol conditions, zenith intensity mainly changes slowly with the solar zenith angle, and the changes between adjacent scan sequences are relatively smooth. When clouds move rapidly, cloud cover changes abruptly, or cloud holes appear, zenith intensity exhibits short-term non-smooth changes, and the TSI increases accordingly.

[0057] For the normalized zenith intensity R of the nth scan sequence n Let its preceding sequence be R. n-1 The next sequence is R. n+1 If the time interval between the current sequence and the previous sequence is Δt1, and the time interval between the next sequence and the current sequence is Δt2, then the TSI can be calculated using the normalized second-order time variation:

[0058] Here, TSI(R,n) represents the temporal smoothness index of zenith intensity corresponding to the nth scan sequence. This index essentially reflects the second-order variation characteristics of the zenith intensity time series and is used to identify non-smooth disturbances caused by rapid changes in the cloud field. In actual classification, its absolute value or comparison with a reference threshold can be used to determine whether the observation is affected by cloud disturbances.

[0059] (3) O4 differential slant column concentration.

[0060] Based on the preprocessed MAX-DOAS multi-elevation angle spectra, DOAS spectral fitting of O4 absorption was performed using QDOAS software. In one embodiment, the fitting band was selected as 338-370 nm, obtaining the O4 differential column concentration near 360 nm. For each scan sequence, the 90° zenith spectrum in the same scan sequence was used as the reference spectrum, and the O4 differential column concentrations of each low-elevation and high-elevation angle observation spectrum relative to the zenith reference spectrum were calculated.

[0061] After obtaining the O4 differential oblique column concentrations at multiple elevation angles, two classification indicators were further constructed: First, the concentration D of O4 differential oblique column at a 30° elevation angle. 30 This index is used alone to characterize whether the optical path for high elevation angle observations is affected by low cloud truncation. When low clouds are located within the main sensitive altitude range of MAX-DOAS, the optical path for 30° elevation angle observations is easily affected by cloud scattering and obstruction. 30 Typically, it decreases significantly; however, high cloud cover has a relatively weaker impact on this indicator. Therefore, D 30 It can serve as an important indicator for distinguishing between low clouds and high clouds.

[0062] Second, the dispersion of O4 differential column concentration at low elevation angles, Sp_low. This index describes the distribution difference of O4 differential column concentration within a low elevation angle range. In one embodiment, Sp_low is defined as the difference between the maximum and minimum O4 differential column concentrations within an elevation angle range of 1°-10°.

[0063] Wherein, dSCD(O4, 1°-10°) represents the O4 differential oblique column concentration corresponding to low elevation angle observations of 1°, 2°, 3°, 4°, 6°, 8° and 10°. Sp_low can be used to help characterize near-surface aerosol load and differences in optical path during low elevation angle observations.

[0064] (4) Aerosol optical thickness (AOD).

[0065] Based on the aforementioned O4 differential oblique column concentrations, a MAX-DOAS aerosol inversion was performed using the optimal estimation method to obtain the aerosol extinction profile and aerosol optical thickness (AOD). In this process, the O4 differential oblique column concentrations under different aerosol profile conditions were calculated using a radiative transfer model. Through optimal estimation iterations, the difference between the simulated and observed O4 differential oblique column concentrations was minimized, thus obtaining the optimal aerosol extinction profile. AOD can be obtained by integrating the inverted aerosol extinction coefficient in the vertical direction.

[0066] In this invention, AOD is mainly used to distinguish aerosol load states, and also to identify anomalous inversion states that may occur under low cloud or optically thick cloud conditions. Since low clouds may alter the effective optical path and be misattributed to aerosol extinction enhancement in clear-sky hypothetical inversions, AOD not only reflects the true aerosol load but can also serve as an auxiliary classification index for cloud-induced optical path changes and inversion anomalies.

[0067] 3. Threshold Calibration The purpose of this step is to determine the key discrimination thresholds required for cloud-aerosol state classification and to explain the physical meaning and basis for determining each threshold. These thresholds include the cloud height boundary threshold, the zenithal hourly light intensity (ZAM) reference threshold, the ZAM temporal smoothness threshold, the air-dwelling density (AOD) threshold, the O4 differential oblique column concentration threshold at a 30° elevation angle, and the cloud hole identification time window. These thresholds can be calibrated or adjusted based on instrument response, observation bands, regional climate characteristics, historical observation samples, and radiative transfer simulation results.

[0068] (1) Cloud height boundary threshold The cloud height threshold is used to distinguish between low clouds and high clouds. This invention does not simply adopt a fixed meteorological cloud height classification standard, but determines the relative boundary between low clouds and high clouds based on the effective sensitivity height of the aerosol extinction profile obtained by the MAX-DOAS instrument.

[0069] Specifically, based on the average kernel matrix retrieved by MAX-DOAS aerosol retrieval, the variation characteristics of the diagonal elements of the average kernel matrix with altitude are extracted, and its vertical envelope is constructed. The diagonal elements of the average kernel reflect the sensitivity of the retrieval results to the actual atmospheric conditions. When this envelope decreases to 10% of its maximum value, the corresponding altitude is defined as the upper limit H of the effective sensitive altitude for MAX-DOAS aerosol retrieval. m Located in H m The following cloud layers directly affect the main sensitive optical path of MAX-DOAS, and are therefore defined as low clouds; located in H m The cloud layers mentioned above have a relatively weak impact on the near-surface sensitive optical path, and are therefore defined as high clouds.

[0070] In one embodiment, based on the observation geometry, band, and aerosol inversion settings adopted in this study, the average kernel analysis yields H. m The distance is approximately 2 km, therefore 2 km is used as the threshold for distinguishing between low and high clouds. For example... Figure 2 As shown.

[0071] (2) Zenith light intensity reference threshold The zenith intensity reference threshold is used to determine whether the observed zenith intensity deviates significantly from the clear-sky reference state, and to assist in identifying states such as cloudless, high aerosol, cloud disturbance, and optically thick clouds. This invention sets two zenith intensity reference thresholds, denoted as R... 0.01 And R1.

[0072] Among them, R 0.01 R1 represents the zenith light intensity reference value under low AOD clear sky conditions, which can be used to characterize the zenith light intensity level under near-clean clear sky backgrounds; R2 represents the zenith light intensity reference value under high AOD clear sky conditions, which can be used to characterize the upper limit or reference range of zenith light intensity under conditions of high aerosol loading but no clouds. When the zenith light intensity is lower than R... 0.01 This usually indicates an abnormal decrease in observed light intensity relative to clear-sky, low AOD conditions, which may correspond to thick optical clouds, near-ground fog, precipitation-related cloud systems, or low-light, low-quality observations; when the zenith light intensity is located at R... 0.01 When the zenith light intensity is near or below R1 and the time variation is relatively smooth, it can help to determine that there is no obvious cloud disturbance. When the zenith light intensity is higher than the corresponding clear sky reference and is accompanied by a high TSI, it can be used as an indication of the presence of cloud disturbance.

[0073] In one embodiment, R 0.01R1 and R2 can be determined through radiative transfer simulation. Specifically, different AOD scenarios can be set under cloudless conditions, such as AOD = 0.01 and AOD = 1, to simulate the 400 nm zenith intensity or normalized zenith intensity under different solar zenith angles. The simulation results are then converted into a reference intensity curve consistent with the instrument's DN / integration time. In actual classification, the corresponding R1 is obtained from the reference curve based on the solar zenith angle corresponding to the observation time. 0.01 And R1. For example Figure 3 As shown.

[0074] (3) Zenith intensity TSI threshold The zenith intensity TSI threshold is used to identify non-smooth perturbations in zenith intensity time series. Under clear sky or stable aerosol conditions, zenith intensity mainly changes slowly with the solar zenith angle, and the TSI is relatively small. When clouds move rapidly, cloud cover changes, or cloud holes appear, the zenith intensity changes between adjacent scan sequences are no longer smooth, and the TSI increases significantly.

[0075] This invention sets a TSI threshold T0 to distinguish between smooth change states and cloud disturbance states. In one embodiment, T0 is 1 × 10⁻⁶. -4 When TSI is less than T0, the temporal variation of zenith light intensity is considered to be relatively smooth, which can be used as one of the criteria for no obvious cloud disturbance; when TSI is greater than T0, the zenith light intensity is considered to have obvious short-term disturbance, which can be used as one of the criteria for cloud influence or rapidly changing states such as cloud holes.

[0076] (4) AOD threshold The AOD threshold is used to distinguish aerosol load states and assist in identifying optically thick clouds or anomalous inversion states. This invention sets AOD thresholds of 0.15, 0.3, and 1.

[0077] AOD=0.15 is used to distinguish between low and high aerosol loads under cloudless or high cloud conditions. When there is no significant cloud influence or the influence of high clouds is weak, AOD mainly reflects the near-surface aerosol load, so 0.15 can be used as the boundary between low and high aerosol loads.

[0078] An AOD of 0.3 is used to distinguish between low and high aerosol loads under low cloud conditions. Since low clouds may truncate the effective MAX-DOAS optical path and be misattributed to enhanced aerosol extinction in clear-sky aerosol inversion, the retrieved AOD under low cloud conditions may be inflated. To avoid misclassifying all slight increases in AOD caused by low clouds as high aerosol loads, this invention uses a higher value of 0.3 as the low / high aerosol load boundary in the low cloud branch.

[0079] AOD = 1 is used to identify optically thick clouds or anomalous inversion conditions. When the inverted AOD is greater than 1, accompanied by a significant decrease in zenith intensity or an anomalous decrease in O4 differential column concentration, it usually indicates that the observation may be affected by optically thick clouds, near-ground fog, precipitation-related cloud systems, or severe optical path anomalies. Such observations should not be directly used as conventional MAX-DOAS aerosol profile inversion results.

[0080] (5) Concentration threshold of O4 differential oblique column at 30° elevation angle A 30° elevation angle O4 differential oblique column concentration threshold is used to distinguish between low clouds and high clouds. This invention sets a 30° elevation angle O4 differential oblique column concentration threshold D0. In one embodiment, D0 is 5 × 10⁻⁶. 42 molecules 2 cm -5 .

[0081] Observations at a 30° elevation angle are considered relatively high, and their effective optical path is highly sensitive to the truncation effect of low clouds. When the cloud layer is below the MAX-DOAS effective sensitive altitude, cloud scattering and obstruction significantly shorten the effective optical path of 30° elevation angle observations, reducing the O4 differential oblique column concentration. When the cloud layer is at a higher altitude, the O4 differential oblique column concentration at a 30° elevation angle typically remains at a high level. Therefore, D 30 A value less than D0 can serve as an important criterion for determining the state of low clouds. 30 A value greater than D0 can be used as a criterion for determining whether a cloud formation is high or has no significant low cloud influence.

[0082] In one embodiment, D0 can be determined jointly based on radiative transfer simulations and actual observation samples. The radiative transfer simulation sets different cloud base heights, cloud optical thicknesses, and aerosol loads, analyzing the response of O4 differential oblique column concentration at a 30° elevation angle to changes in cloud height. In actual observations, D0 can be verified under low cloud conditions by combining cloud base heights identified by lidar or typical cases. 30 Whether it remains consistently below this threshold. Based on a combination of simulation and observation results, 5 × 10 42 molecules 2 cm -5 The reference threshold was determined to distinguish between low clouds and high clouds.

[0083] (6) Threshold for cloud hole recognition time window The cloud hole identification time window is used to identify short-term openings or localized light transmission during rapid changes in low clouds. Under cloud hole conditions, the MAX-DOAS derivation at a given moment may appear close to a high-cloud or cloudless state, but the time period before and after it still belongs to a low-cloud process. Therefore, relying solely on the D of a single scan sequence... 30 AOD (Advanced Distance Analysis) may misclassify cloud holes as high clouds or no clouds.

[0084] This invention sets a cloud hole identification time window W. In one embodiment, W is ±30 min. Specifically, for an observation initially determined to be a high cloud, if a low cloud observation exists within 30 minutes before and after it, the observation is remarked as a cloud hole. This rule is used to reflect the temporal continuity characteristic of cloud holes as short-term openings in a low cloud field or as a transitional state of rapid changes in low clouds.

[0085] Figures 4 to 11 Examples of derived quantity distributions under different weather classifications are shown.

[0086] 4. Automatic classification Each scan sequence is classified according to the following procedure: Step S1: Identify optically thick clouds. If the zenith radiance is lower than the low AOD clear sky reference threshold R... 0.01 And AOD is greater than 1, or D appears at the same time. 30 If the value is abnormally low, the scan sequence is determined to be in a state of thick optical clouds or low-quality observation.

[0087] Step S2: Identify a state with no significant cloud disturbance. If the zenith radiance is lower than the high AOD clear sky reference threshold R1 and the TSI is lower than T0, the scan sequence is considered to have no significant cloud disturbance. Further, if the AOD is less than 0.15, it is determined to be clear sky; if the AOD is greater than 0.15, it is determined to be cloudless high aerosol.

[0088] Step S3: Identify cloud impact status. Scan sequences not classified in steps S1 and S2 are considered to have cloud disturbance or cloud aerosol mixing effects. If D 30 If D is greater than D0, it is initially determined to be a high cloud; if D 30 The value is less than D0, so it is preliminarily determined to be low cloud.

[0089] Step S4: Differentiate aerosol loads under cloud conditions. For high cloud conditions, if AOD is less than 0.15, it is determined to be high cloud with low aerosol loads; if AOD is greater than 0.15, it is determined to be high cloud with high aerosol loads. For low cloud conditions, if AOD is less than 0.3, it is determined to be low cloud with low aerosol loads; if AOD is greater than 0.3, it is determined to be low cloud with high aerosol loads.

[0090] Step S5: Identify cloud hole states. For scan sequences initially identified as high clouds, check whether low cloud scan sequences exist within the preceding and following time windows W; if so, remark the scan sequence as a cloud hole. A cloud hole represents a short-term opening or localized light transmission state during rapid changes in the low cloud field. Its derivative may be close to the high cloud or cloudless state, but physically it belongs to the dynamic transition process of low clouds.

[0091] The complete process is as follows Figure 12 As shown, Figure 12The example classification results are shown, applied to a MAX-DOAS site in Panyu, Guangzhou, for one year of cloud aerosol classification.

[0092] Figure 13 This is a set of cloud classification results applied to the Panyu area of ​​Guangdong Province in 2025 using MAX-DOAS measurements. The classification results of this application example are consistent with the cloud results measured by a lidar unit located approximately 3 km away for 70% of the time period.

[0093] Example 2 This embodiment illustrates alternative solutions for each parameter in the method of the present invention.

[0094] In another embodiment, the concentration D of the O4 differential oblique column at a 30° elevation angle is... 30 It can be replaced by a weighted combination index of O4 differential oblique column concentration at elevation angles of 15° and 30°, which can also reflect the truncation effect of clouds on the effective optical path at high elevation angles.

[0095] The temporal smoothness index of zenith intensity (TSI) can be replaced by the sliding window standard deviation, adjacent scan difference, normalized residual, or perturbation index based on the residual fitted by the solar zenith angle. All of these can achieve the function of identifying short-term non-smooth changes in zenith intensity.

[0096] The AOD threshold can be adaptively determined based on the observation area, instrument type, season, and historical samples. Alternatively, the aerosol extinction profile integral, near-surface extinction coefficient, or low-elevation-angle O4 differential oblique column concentration dispersion Sp_low can be used to replace AOD for determining aerosol load.

[0097] The cloud hole time window can be set to ±20 min, ±40 min, or other time ranges according to the MAX-DOAS scanning cycle and the rate of change of the regional cloud field. It can also be implemented using a dynamic window or a continuous state transition model.

[0098] The boundary between high and low clouds is not fixed at 2 km, but is dynamically determined based on the average kernel, weighting function, observation band, instrument geometry, and inversion settings.

[0099] The classification method can be replaced by a threshold discrimination process with machine learning models such as decision trees, random forests, support vector machines, or neural networks. The input features still include zenith radiance, time-varying indicators, O4 differential column concentration, and AOD, which are derived from MAX-DOAS itself.

[0100] Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

Claims

1. A cloud aerosol state classification method based on MAX-DOAS observations with high cloud dependence, characterized in that, Includes the following steps: Data acquisition steps: Collect ground-based MAX-DOAS multi-elevation angle solar scattering spectral data, and preprocess the spectral data to obtain multi-elevation angle corrected spectra and observation auxiliary information of effective scanning sequences; Derivative calculation steps: Based on the multi-elevation angle corrected spectrum, calculate the normalized zenith intensity R, the zenith intensity time variation perturbation index, the high elevation angle O4 differential oblique column concentration D, the low elevation angle O4 differential oblique column concentration dispersion Sp_low, and the aerosol optical thickness AOD. Threshold calibration steps: Determine the cloud height boundary threshold and the zenith light intensity reference threshold R. 0.01 And R1, zenith light intensity TSI threshold T0, AOD threshold, high elevation angle O4 differential oblique column concentration threshold D0, and cloud hole identification time window W; Classification steps: Based on the comparison results between the derived quantity and the threshold, each scan sequence is classified step by step into one of eight states: clear sky, cloudless high aerosol, high cloud low aerosol, high cloud high aerosol, low cloud low aerosol, low cloud high aerosol, optical thick cloud, and cloud hole.

2. The cloud aerosol state classification method according to claim 1, characterized in that, The preprocessing includes dark current subtraction and electronic bias subtraction.

3. The cloud aerosol state classification method according to claim 1, characterized in that, The normalized zenith intensity R is calculated according to the following formula: Among them, DN 400 t represents the detector count value at 400 nm in the zenith direction. exp This represents the integration time for spectral acquisition.

4. The cloud aerosol state classification method according to claim 1, characterized in that, The low-elevation-angle O4 differential column concentration dispersion Sp_low is defined as the difference between the maximum and minimum values ​​of the O4 differential column concentration within the elevation angle range of 1° to 10°: Wherein, dSCD(O4, 1°-10°) represents the O4 differential oblique column concentration corresponding to low elevation angle observations of 1°, 2°, 3°, 4°, 6°, 8° and 10°.

5. The cloud aerosol state classification method according to claim 1, characterized in that, The cloud height threshold is determined based on the average kernel matrix obtained from MAX-DOAS aerosol inversion. Specifically, the variation characteristics of the diagonal elements of the average kernel matrix with height are extracted and its vertical envelope is constructed. The height corresponding to reducing the envelope to 10% of its maximum value is defined as the upper limit of the effective sensitive height H of MAX-DOAS aerosol inversion. m , with H m As the dividing threshold between low clouds and high clouds.

6. The cloud aerosol state classification method according to claim 1, characterized in that, The zenith light intensity reference threshold R 0.01 R1 and R2 were determined through radiative transfer simulation: Under cloudless conditions, two scenarios were set up, AOD = 0.01 and AOD = 1, respectively, to simulate the normalized zenith intensity at the 400nm band under different solar zenith angles. A reference intensity curve consistent with the instrument detector count value and integration time was obtained. In actual classification, the corresponding R2 was obtained from the reference curve according to the solar zenith angle corresponding to the observation time. 0.01 And R1.

7. The cloud aerosol state classification method according to claim 1, characterized in that, The classification steps specifically include: S1: If the zenith radiance is lower than R 0.01 If AOD is greater than 1, or if D is abnormally low at the same time, the scanning sequence is judged to be in a state of thick optical clouds or low-quality observation. S2: If the zenith radiance is lower than R1 and the TSI is lower than T0, the scan sequence is determined to be in a state without obvious cloud disturbance; further, if the AOD is less than 0.15, it is determined to be clear sky; if the AOD is greater than 0.15, it is determined to be cloudless high aerosol. S3: For scan sequences that are not classified in S1 and S2, if D is greater than D0, they are initially identified as high clouds; if D is less than D0, they are initially identified as low clouds. S4: For high cloud conditions, if AOD is less than 0.15, it is determined to be high cloud with low aerosols; if AOD is greater than 0.15, it is determined to be high cloud with high aerosols. For low cloud conditions, if AOD is less than 0.3, it is determined to be low cloud with low aerosols; if AOD is greater than 0.3, it is determined to be low cloud with high aerosols. S5: For a scan sequence initially identified as a high cloud, search for a low cloud scan sequence within the time window W before and after it; if it exists, remark the scan sequence as a cloud hole.

8. The cloud aerosol state classification method according to any one of claims 1 to 7, characterized in that, The high elevation angle O4 differential oblique column concentration D is the 15° elevation angle O4 differential oblique column concentration, 20° elevation angle O4 differential oblique column concentration, 30° elevation angle O4 differential oblique column concentration or a weighted combination thereof.

9. The cloud aerosol state classification method according to any one of claims 1 to 7, characterized in that, The disturbance index of temporal variation of zenith intensity is a temporal smoothness index of zenith intensity, sliding window standard deviation, adjacent scan difference, normalized residual, or disturbance index based on the fitting residual of solar zenith angle.

Images