| Retrieving cloud vertical and three-dimensional structures using satellite remote sensing data is highly valuable for research,but also posesses technical challenges.This paper extends a recently developed conditional adversarial neural network(CGAN)model for retrieving the Cloud Sat cloud profile radar(CPR)reflectivity from MODIS cloud products,and evaluates its performance for eight cloud types and latitude zones.Thereafter,a three-dimensional cloud radar reflectivity inversion technique is developed out,and the reliability of the retrievals is analyzed.The CGAN-based inversion model is extended with additional training datasets and improved with a new normalization method.The model was trained with Cloud Sat/CPR reflectivity data labels with inputs of MODIS cloud top pressure,cloud water path,cloud optical thickness,and effective radius data.The training data are collected from 2010 to 2017 over the global oceans.A test dataset containing 24,427 samples was statistically analyzed to evaluate the performance of the model for eight cloud types and three latitude zones using multiple verification metrics.The results show that the CGAN model has reliable performance for retrieving clouds with reflectivity greater than-25 d BZ.The model performs best for deep convective systems,followed by nimbostratus,altostratus,and cumulus,but has limited ability for stratus,cirrus,and altocumulus.The model performs better in the low and middle latitudes than in the high latitudes.Overall,the CGAN model can reliably retrieve the vertical structures of deep convective clouds and nimbostratus in the mid-and lower latitude regions,laying a foundation for constructing three-dimensional cloud structures of deep convective systems including convective storms and hurricanes from MODIS cloud products.A scheme for constructing three-dimensional cloud radar reflectivity is developed by effectively integrating the line segment retrivals of the CGAN model.The MODIS cross-track resolution is approximately 1.72 km on average,the along-track resolution is approximately 1km on average,and the vertical resolution is 250 m,covering a height range of 750 m to 16750 m.The scheme includes an overlapping stacking calculation(OVERLAP)for averaging,postOVERLAP processing,and a refinement with cloud masking to achieve a continuous 3D cloud field.The scheme can produce 3D reflectivity data for a region of 2330 km(cross-track)* 2000km(along-track)* 16 km(total height)in one run.Typhoon In-Fa is selected for a case study and sensitivity analysis comparisons.To demonstrate the reliability of the method,two sets of observational comparison experiments were designed and multiple cases were selected for test and evaluation.Firstly,the inverted 3D radar reflectivity of Typhoon Chaba was qualitatively compared with ground-based radar observations at different heights,and the inverted Typhoon In-Fa and Typhoon Chaba was quantitatively compared with ground-based radar observations by taking the intersection.Secondly,20 cases were slected to retrieve and construct the 3D radar reflectivity and compared with the corresponding Cloud Sat along-track observations.The advantages and differences between the retrieved 3D cloud reflectvity and the weather radar observations were analyzed for Typhoon Chaba,squall lines,and ordinary convection.The results show that the 3D retrievals is generally consistent with the weather radar observations for different meteorological scenarios,locations,and times.This study provides a new and feasible way to enhance the applications of satellite cloud remote sensing,including convection nowcasting,cloud data assimilation,solar energy forecast and climate modeling. |
PDF Full Text Request