Detection and attribute characteristics of clouds and aerosols

Detection and attribute characteristics of clouds and aerosols

1. cloud layer
The main problem facing short-term solar forecasting is to predict the evolution of clouds. The evolution of the cloud layer itself is very sensitive to the ability to accurately describe the current cloud cover. This section discusses techniques and issues that accurately describe cloud cover through current satellite measurements. Once a cloud layer is detected, other techniques can be used to infer the optical properties of the cloud layer and the solar radiation passing through the cloud layer to the surface. Because clouds mainly play a role in regulating the direct radiation and scattering of solar radiation, accurate cloud detection is critical to solar energy applications.

Although the purpose of cloud detection is straightforward. However, compared with the steps used in all other cloud remote sensing processes, the methods available for cloud detection are more diverse. Many cloud attributes, including spectral attributes, can be used for cloud detection operations, such as cloud reflectance and thermal emission magnitude and spectral changes. In addition, cloudy scenes have higher temporal and spatial anisotropy than clear sky scenes, which can provide useful cloud detection metrics (such as spatial uniformity and total value comparison test).

Generally speaking, there are two types of cloud detection methods. One is the threshold-based method, which applies a predetermined threshold to all relevant cloud detection experiments. Several cloud detection experiments can infer the final cloud classification. Another method is to avoid the use of thresholds, and to use probability statistics techniques, that is, to replace the thresholds with a continuous function of cloud probability and each cloud detection metric.

In solar applications, cloud obscuration and snow-covered surfaces, as well as distinguishing cloudy weather and scenarios with increased aerosol concentration, pose challenges for cloud detection. The reason why snow cover can affect cloud detection is that snow cover has similar reflection and scattering properties to ice clouds, and will have short-term space/time effects.

Once a cloud layer is detected, since the reflection properties of the cloud layer are highly dependent on the thermodynamic phase state, knowledge of the thermodynamic phase state (liquid, ice or a mixture of the two) is required. When the mixed cloud exists, the current passive remote sensing satellite imager does not have the ability to directly and clearly infer whether the mixed phase exists, and the current phase detection is still strictly limited between the ice phase and the water phase. Cloud phase state is a strong function of cloud temperature; opaque clouds have a good approximation in this respect. For example, a brightness temperature of 11μm (window channel) can be used instead of cloud temperature. Very little liquid water exists in clouds with temperatures below 243K, and very little ice exists in clouds with temperatures above 263K. In addition to temperature, the reflectance of the cloud at 3.9μm can be used to distinguish ice clouds from water clouds. At this time, the reflectivity of water clouds is higher than that of ice clouds. It is important that most phase estimations derived from satellites only apply to the top of the clouds. Usually, very thick ice clouds will grow below freezing altitude and contain large amounts of liquid water that cannot be observed by satellites.

It is the most difficult to detect the phase state of optical thin ice clouds (cirrus clouds), because the temperature of cirrus clouds at 11μm and the reflectance information of 3.9μm are not enough to complete the phase state detection. Generally, most satellite technologies use the spectral characteristics of cirrus clouds in the infrared window and infrared water vapor bands to detect the presence of cirrus clouds. Although current satellite technology can detect the presence of cirrus clouds, it is generally difficult to detect lower clouds. The best phase state (liquid and ice) should be selected according to the actual application. For example, when a thin cirrus cloud appears above the lower water cloud, the lower water cloud has a much greater impact on the solar energy on the ground than the upper cirrus cloud, but the two have the opposite effect on the long-wave energy at the top of the atmosphere.

Although many satellite applications have adopted a clear cloud layer-phase state-detection scheme, with the enhancement of computing power, many applications have generated a set of cloud properties for the two phase states. The estimation of cloud phase state is determined by the degree of matching between the observed value and the entire set of cloud attributes. Some applications further develop this process to determine all cloud properties (including occlusion and phase state) at the same time. Although such technologies have very high flexibility, complex numerical calculations limit their practical application.

Usually, the next step after cloud remote sensing has completed occlusion and phase detection is to determine the height of the cloud. For solar applications, the driving factor of radiation transmission is not the vertical distribution of the clouds, but the shadows cast by the clouds on the ground. The most common method for estimating cloud height is through infrared channel sensors. As mentioned above, the physical temperature of opaque clouds is very close to their radiation temperature within a certain window channel. For non-opaque clouds, carbon dioxide or water vapor infrared absorption channels are needed. If the infrared absorption channel is lacking, multiple infrared window channels or visible light reflectance can be used together with the brightness temperature of the window channel to estimate cloud height (Heidinger and Pavolonis, 2009). The pressure and height of the cloud layer are derived by using the available temperature profile in the auxiliary data provided by the NWP model. If cloud views from multiple angles are available at the same time, stereo imaging can be used (for example, Hasler et al. 1991). In addition, cloud shadows can be used to estimate the height of cloud tops and cloud bases in some cases (for example, Simpson et al. 2000). These geometric methods can provide more direct measurement values, but their limitations in practical applications are far greater than spectroscopic techniques.

When determining solar energy reaching the surface, the most important cloud attribute is the integrated vertical optical depth (τ) of the cloud. The optical depth is usually defined at a certain reference wavelength in the visible spectrum, and the optical depth can also be scaled to other wavelengths. In a given cloud phase, the size of cloud particles controls the spectral change of cloud optical depth. The most common way to describe particle size is to divide the third moment of the particle size distribution by the second moment to get the effective radius. The advantage of using re is that other details of the size distribution can be ignored. The sphericity of the droplets in the water cloud can satisfy the application of Mie scattering theory. However, for ice crystals with complex shapes, other computationally intensive solutions are needed to generate the required properties. Although the computing power in this area has made great progress, there are still great uncertainties in the process of determining the best particle shape and cloud properties in a certain cloud layer.

Figure 1 lists the most common basic principles for determining cloud optical depth and its spectral dependence. In the past 30 years, this method has been developed through the use of satellites, aircraft, and on-site observations. The basis of this method is the use of two solar reflectance channels, so it is also called dual spectroscopy. One of the channels is the spectral window area, which contains negligible cloud particle radiation absorption (for example, 0.65μm). The other channel must be located in a spectral window area where a cloud particle can fully absorb the radiation, and the absorption is also required to be sensitive to the particle size commonly found in the cloud (for example, 1.6μm, 2.2μm, or 3.9μm). In Figure 1, the Y-axis is the absorption channel, and the X-axis is the non-absorption channel. The solid line represents the constant τ, and the broken line represents the constant re. This figure shows the τ-re of most parameter spaces. The curve is orthogonal, and the sum and re values ​​of most reflectance can be determined. But when the cloud layer is thin, it is difficult to measure the re value; when the particle size is very small (<3μm), multiple methods can be used, but the re value of cloud particles is rarely less than 5μm. Except that the difference in scattering properties changes the curve style in Figure 1, the application of this method to ice clouds and water clouds is the same. When the reflectance is outside the range of the τ-re table, most technologies will return to the climatic value. The best evaluation technique is a universal numerical framework in which dual-spectral inversion can be performed. The advantages of this technique are that errors can be estimated and constraints can be used in a timely manner.

Detection and attribute characteristics of clouds and aerosols
Figure 1 The dual spectral technology of cloud optical depth

Most dual spectroscopy methods assume that the cloud layer has a uniform phase state and particle size and disappears in the vertical direction. Both measured values ​​and theories show that the truth of this hypothesis is low. Many sensors use absorption channels of different or multiple frequencies. The depth of the clouds that can be seen by different channels depends on the intensity of particle absorption. When there are fewer absorption channels, deeper clouds can be seen. The vertical change of the actual cloud particle size is coupled with different sensitivities to obtain the measured value of re. Re has nothing to do with the spectrum, but varies according to the specific channel used. The current absorption channel used by the geosynchronous sensor is 3.9μm, which is only sensitive to particles in the top area of ​​most clouds. In the common adiabatic growth type clouds, the particle size will increase as the height in the cloud layer rises, r derived from the measured value of 3.9μm. It is much larger than the results derived from the 1.6μm and 2.1μm measurement values. The smallest particles in ice clouds are often located at the top of the clouds, which is the opposite of the former.

When using dual-spectrum method for τ and re inversion, three challenging scenarios will be encountered. The first is snowfall. Snowfall will severely reduce the ability to extract τ and re information from most currently available channels, and reduce the sensitivity of non-absorbent channels to τ. Newer sensors (such as MODIS and VIIRS) overcome this problem by using non-absorption channels in the snow-sensitive spectral region. The second case is the appearance of multi-layer clouds with different phases. Studies have shown that a large proportion of thin cirrus clouds will lie above the lower water clouds. If thin cirrus clouds do exist but are not detected, the absorption of ice clouds is stronger than that of water clouds, so thin cirrus clouds will significantly affect the inversion of re. The last case is the most complicated and common. All current dual-spectrum applications assume that the clouds are parallel to the plane, that is, they are considered to be uniform and infinitely extending in the horizontal direction. In fact, there are very few uniform clouds in the observation range of current sensors. Their three-dimensional structure will enhance the reflection on the side of the cloud, increase shadows, direct sunlight through the cloud area and increase the average number of cloudy days. Although most of the effects can be offset in average space and time, the three-dimensional effect can greatly affect the inversion of cloud properties and the distribution of solar energy at any location. Currently, scholars are trying to use dual spectroscopy to explain these effects.

2 aerosol
Since the optical thickness and particle size of aerosols are an order of magnitude smaller than the optical thickness and particle size of clouds, aerosol remote sensing is different from cloud remote sensing. In addition, unlike water droplets and ice particles, aerosols do not have a clear absorption band related to particle size, and their particle size is estimated by the spectral change of optical thickness. That is, the particle size of the aerosol will decrease as the optical thickness (and therefore related to the reflection spectrum) spectrum changes increase. Generally, aerosols have little effect on infrared radiation, and the technology used to assess the height of aerosols at the top of clouds cannot assess the vertical distribution of aerosols.

Related Posts