| Clouds are formed by the condensation of water droplets and ice crystals in the atmosphere when they get cooler,covering approximately 50 to 70%of the earth's surface.Lots of clouds will directly affect the quality of remote sensing imagery data and reduce the availability of data.Meanwhile,cloud also affects the global atmospheric radiation.It has an important influence on the earth's atmospheric radiation balance and climate change mainly through the interaction with solar radiation,and is an important component of the earth's radiation budget balance.Among them,cloud parameters are important input data in climate change,environmental monitoring and meteorological forecast.Accurate retrieval of cloud parameters will directly affect the credibility of the different application results.China successfully launched Gaofen-5(GF-5)satellite in Taiyuan Satellite Launch Center on May 09,2018.It is equipped with the atmospheric aerosol Directional Polarimetric Camera(DPC),which is the first spaceborne detector with multi-angle polarization observation capability in China and is currently the only on-orbit multi-angle polarization detector in the world.DPC obtains the intensity radiation and polarized radiation signals of the cloud through multi-angle polarization observation of the cloud,which provides more effective observation information for cloud monitoring and parameters retrieval,and improves the capability of cloud parameters retrieval.In this paper,based on DPC imagery data,combined with the characteristics of the DPC instrument itself and the polarization characteristics of the cloud,which are used to study cloud identification,cloud phase identification and cloud optical thickness retrieval research.The key of cloud parameter retrieval lies in the accuracy of cloud identification.Based on the multispectral and polarization characteristics of DPC imagery data,a cloud identification algorithm model for DPC data is constructed.In DPC cloud identification algorithm model,the surface is firstly divided into land and ocean surface,and cloud identification is carried out separately.Meanwhile,compared with the original cloud identification algorithm based on the fixed thresholds,it is found that there are great differences in cloud identification thresholds in blue band over the land,near infrared band over the ocean and oxygen A absorption band under different regions and times according to statistics.This study will build a dynamic cloud identification threshold library for the above three cloud identification algorithms,and build a desert,bare soil bright surface library and snow-covered surface library over land and ocean for special surface,which can improve the accuracy of cloud identification results over special surface to a certain extent.For cloud identification over the land,bright surface identification,blue band dynamic cloud identification,oxygen A apparent pressure dynamic cloud identification and polarized cloud identification are respectively carried out to obtain cloud,clear sky and uncertain results.For cloud identification over the ocean,the dynamic cloud identification based on the oxygen A apparent pressure,near infrared dynamic cloud identification and polarized cloud identification are respectively carried out,and cloud,clear sky and uncertain results are also obtained.In the above DPC cloud identification results,since the snow and ice surface and the cloud have similar spectral reflection information,the snow/ice-covered surface correction is performed for the DPC cloud identification results.The method is mainly based on the differences in the reflectance of the apparent oxygen A pressure,blue,infrared and near-infrared band between the cloud and the snow/ice,supplemented by the snow and ice surface library,which will eventually be rejected as cloud pixels of ice and snow pixels,to obtain more accurate cloud identification result.At the same time,based on the dynamic cloud identification algorithm,a cloud confidence evaluation criterion based on DPC data is constructed,and the final DPC cloud identification results are divided into four categories:100%high confidence cloud,clear sky,less than 100%low confidence cloud and clear sky.The dynamic cloud identification algorithm constructed above is applied to DPC data for cloud identification,and the cloud identification results are compared with the cloud mask results of MODIS and CALIPSO.It is found that the accuracy of cloud identification results of a track data is above 90%,and the accuracy of global cloud identification results is above 80%,which also verifies the accuracy and credibility of DPC cloud identification results.Based on the accurate DPC cloud identification results,the retrieval of cloud phase and cloud optical thickness is carried out.In the cloud phase identification of DPC data,due to the different scattering and absorption characteristics of water cloud droplets and ice crystal particles,and the polarized radiation intensity is not sensitive to multiple scattering compared with the total radiation intensity,but sensitive to single scattering of water cloud droplets,which forms a representative phenomenon of water cloud rainbow.According to the simulation of atmospheric radiation transmission,there is an obvious peak of 865 nm polarization reflectance in the water cloud near the 140°scattering angle.Based on this characteristic,four cloud phase identification critera are constructed,namely,60~140°polarization reflectance curve slope identification,100°and 140°polarization reflectance ratio identification,75~120°polarization reflectance neutral point identification and 140~180°polarization reflectance standard deviation identification.Based on above critera,the cloud phase identification algorithm based on DPC data is constructed.Based on water cloud and ice cloud results obtained by the DPC cloud phase identification algorithm,the optical thickness retrieval of DPC water cloud and ice cloud is carried out respectively.In the DPC cloud optical thickness retrieval,the UNL-VRTM vector radiation transmission model is firstly used to simulate and analyze the sensitivity of the reflectance of 670 nm and 865 nm bands to cloud optical thickness under cloud conditions.It is found that polarization is not sensitive to cloud with large optical thickness,and it is proved that the reflectance(scalar)of both the 670 nm and 865nm bands can be used to retrieve the cloud optical thickness.Secondly,the effects of water droplet effective particle radius,ice cloud bulk scattering model,ice cloud effective particle radius,surface reflectance and aerosol on cloud optical thickness retrieval are analyzed.It is found that the sensitivity of cloud particles to reflectance gradually decreases with the increase of cloud effective particle radius.Moreover,with the increase of cloud optical thickness,the contribution of surface reflectance to radiation observation data gradually weakens,but the high surface reflectance still has great influence.For the selection of the bulk scattering model of ice cloud particles,the SC,ASC and GHM scattering models provided by Baum et al.are selected.It is found that ASC is less affected by the effective particle radius than SC and GHM scattering models,and is relatively more suitable for the retrieval of the ice cloud optical thickness.In the retrieval of cloud optical thickness based on DPC data,670 nm band is selected over the land and 865 nm band is selected over the ocean.Standard water drop model and ASC model are respectively selected for cloud optical thickness retrieval on water cloud and ice cloud models.In the retrieval of DPC cloud parameters,the different cloud phases are identified based on the constructed multi-angle polarization cloud phase identification algorithm model,and the results are compared with MODIS and CALIPSO results.It is found that the accuracy of water cloud is 93.70%,and the accuracy of ice cloud is 85%.Based on the constructed cloud optical thickness retrieval algorithm model,the DPC cloud optical thickness results are compared with MODIS cloud optical thickness results,and R~2 of water cloud optical thickness result is 0.86 and R~2 of ice cloud is 0.81.From the verification results of DPC cloud phase and optical thickness,the cloud retrieval results of DPC imagery data have higher accuracy and credibility.The DPC cloud identification,cloud phase and cloud optical thickness retrieval algorithm model constructed in this paper fills in the gaps in cloud identification and cloud parameter retrieval of GF-5 DPC data just been launched in China,and provides the premise and guarantee for the next step of atmospheric pollutant retrieval and weather forecast application. |
PDF Full Text Request