Cloud Height Estimation Based On Geostationary Satellites And Its Application In Numerical Model Evaluatio

Cloud structure affects the radiative energy budget of the earth-atmosphere system.In this paper,machine learning algorithms are applied to estimate the cloud top heights(CTHs)and cloud base heights(CBHs)using geostationary imager measurements and reanalysis data.Taking the influence of radiative transfer algorithms on the simulation of numerical weather prediction(NWP)model as an example,the application of the estimated CTHs and CBHs in model evaluation is investigated.The major contents are as follows:(1)A machine learning algorithm(Gradient Boosting Regression Tree)is applied to estimate the CBHs using measurements from the ABI aboard GOES-16 and data from ERA5.The CBHs from the CALIOP aboard CALIPSO satellite covering 2018 and 2019 are used as the labels(which are the true values/classes of the model output for regression/classification problem in machine learning terminology).By evaluating the errors of estimated CBHs and analyzing the feature importance,it is found that the errors of estimated CBHs are affected by cloud optical depth,cloud phase information and CBHs.The machine learning algorithm for classification(Gradient Boosting Decision Tree)is used to classify the qualities of estimated CBHs into high and low confidence.The root mean square error(RMSE)of the CBH estimates with high confidence is 1.14 km,and 96% of the CBH errors are within ± 2 km.Although there is no requirement of the CBH estimates for ABI at present,the presented accuracy has met the CBH uncertainty requirement(less than 2 km)for VIIRS aboard SNPP satellite.According to the case study of Hurricane Dorian in 2018,the errors of CBH estimates based on ABI are smaller than those based on VIIRS.(2)The algorithm for CBH estimation based on ABI is revised and applied to the AGRI aboard FY4 A satellite.Using the CTHs and CBHs of CALIOP observation in 2020 as labels,the CTH and CBH estimates based on AGRI are obtained.The CTH and CBH estimates with high confidence take up 80% and 85% of total sample and have 96% estimation error less than2 km.In the case of Meiyu in 2020,by comparing with the CTHs from CALIOP observation,it is found that the AGRI L2 product using traditional retrieval methods underestimates the CTHs,while the CTH estimates using machine learning algorithms are generally unbiased.The RMSE of CBH estimates based on AGRI are smaller than those of VIIRS.(3)Taking the influence of different radiative transfer algorithms on the three-dimensional structure of cloud simulated by WRF model as an example,the application of cloud height estimates based on AGRI in numerical model evaluation is discussed.Firstly,by comparing with the benchmark model(128-stream discrete ordinate method),the accuracy of different radiative transfer algorithms is analyzed.The shortwave radiative transfer algorithm includes four-stream adding algorithm for the discrete ordinate method and two-stream practical improved flux method,and the longwave radiative transfer algorithm includes four-stream adding algorithm for the discrete ordinate method and non-scattering Simplification.The results show that the adding algorithm for four-stream discrete ordinate method are more accurate than the two-stream practical improved flux method or the non-scattering simplification.Secondly,to evaluate the influence of radiative transfer algorithms on the cloud-radiation interaction with synoptic scale in the model simulation,the WRF model is used to simulate the Meiyu case and the tropical cyclone Haishen case in 2020.Two experiments are designed: 1)the radiative schemes adopt the adding algorithm for four-stream discrete ordinate method(Revised RRTMG experiment,the adding algorithm for four-stream discrete ordinate method is used for both longwave and shortwave radiative schemes);2)the original RRTMG radiative schemes are used(RRTMG experiment,the two-stream practical improved flux method is used for shortwave radiative scheme,and the non-scattering simplification is used for longwave radiative scheme).In the Meiyu case,comparing the cloud simulations between the two experiments,the Revised RRTMG experiment produces positive radiative heating rate forcing and upward velocity forcing,and generates clouds with a larger spatial extent(2%-4%)and a larger average cloud amount(5%-7%)in the middle and higher troposphere(about 8-16 km height).Taking the cloud amount data of ERA5 and the cloud structure estimated by AGRI as the reference value,the overall result of the Revised RRTMG experiment is closer to the reference value,and its Equitable Threat Score(ETS)is about 6% higher than that of the RRTMG experiment on the 12 km resolution grid.The accuracy advantage of the Revised RRTMG experiment is further varified in the simulation of tropical cyclone Haishen in 2020.In general,the CTH and CBH estimation based on geostationary satellite imager are developed in this paper.The CTH and CBH estimates are applied to evaluate the NWP model.It is found that in the simulation of convection process,the four-stream adding algorithm for the discrete ordinate method can improve part of the WRF model simulation.