Data Assimilation Research For Geosynchronous Interferometric Infrared Sounder Onboard FengYun-4 Satellite

Currently,various data processing techniques have been developed for hyperspectral sounders before the satellite data assimilation.The techniques are specific for those sounders onboard the polar orbiting satellites.However,for the hyperspectral sounder operating on a geostationary satellite,for example the GIIRS?Geosynchronous Interferometric Infrared Sounder?which has an unprecedented advantage to observe the fast-changing water vapor and temperature structures related to severe weather events,the early data processing techniques are not optimal for applying to this new instrument.In order to maximize the benefits of GIIRS observations,some methodologies and technical approaches need to be improved.Thus,this study focuses on the following three research areas:?1?RTTOV?the Radiative Transfer for TOVS?model is an observation operator used in GRAPES?Global/Region Assimilation and Prediction Enhanced system?data assimilation system.For GIIRS data assimilation in GRAPES model,the GIIRS coefficients for RTTOV transmittance model are required.Two innovations are made in this research.Since generation of coefficients is a linear regression problem with small samples for training,the effect of abnormal data on regression accuracy will be intensified.The weighted least squared?WLS?method?a robust regression method?is considered to eliminate the overlarge influence of abnormal data.In addition,considering the fact that GIIRS,as a geostationary advanced IR sounder,only observes a specific region so the local training profiles are developed for RTTOV GIIRS coefficients.Evaluations of brightness temperature?BT?simulations using RTTOV GIIRS coefficients generated from the global training profiles with original least square?OLS?method,threshold value method,and WLS method,respectively,indicate the following findings:Both the threshold value method and WLS have better performances on BT simulation accuracy than OLS method.The improvements are significant in the middlewave band,which is associated with strong water vapor absorption lines.The WLS method has higher level of robustness than the threshold value method.The WLS+threshold method is recommended for generating RTTOV coefficients for IR instruments.Evaluations of BT simulation accuracy from the global and local coefficients indicate the following findings:RTTOV forward BT simulation errors for the GIIRS are decreased in all spectral regions using the local coefficients with an 11.7%accuracy improvement.BT simulation accuracy from local coefficients has a 11.9%,14.10%,14.39%,and 5.06%improvement over the global coefficients for four different latitude regions,respectively.This implies that the local training profiles can enhance the RTTOV GIIRS accuracy especially in the non-extreme atmospheric conditions.More channels meet the 0.05 K simulation accuracy standards from local coefficients,comparing using global coefficients by independent testing profiles.Major improvements from local coefficients are found in the longwave band.The largest ones occur in the regions around 800 cm^-11 associated with CO₂ absorption lines and 1,050 cm^-11 associated with O₃ absorption lines.Since local training profiles can better represent the atmospheric conditions within the GIIRS observation coverage,especially for variables such as atmospheric temperature and ozone,the BT simulation accuracy from local coefficients is obviously higher than that from the global coefficients,especially for CO₂ and O₃ absorption spectral regions.?2?In general,hyperspectral infrared?IR?sounder has thousands of channels.It is impractical to use full observation data volume in real-time or near real-time operations in terms of data transmission,storage,processing and computation resources.Various methodologies are developed to select the subset channels from all the channels for data assimilation.Among those the information entropy iterative method is the most nearly optimal.However,this method is not comparatively best for a hyperspectral IR sounders onboard the geosynchronous satellites which mainly aims to observe the changing weather events over a region.One major shortfall of this method is that it merely considers the decrease in uncertainty of atmospheric state parameters caused by measurements without taking into count the dynamic effect of measurements in time window.The information entropy iterative method is improved with the new index called M index which reflects the Jacobian variance over time The new channel selection method combining the information entropy iterative method and the M index are named as M method here.Comparisons of both methods with single profile and local training profiles show that there are minor differences between two methods in choosing temperature or water vapor sensitive channels.It means that the information entropy iterative method has some ability of choosing the channels which are sensitive to variability of atmospheric status.However,the M index could improve the ability of information entropy iterative method in selecting the sensitive channels on atmospheric variability without decreasing its original advantage.The M method has been proven to be more suitable for a hyperspectral sounder operating on the geostationary satellite.?3?The synergistic use of satellite-based hyperspectral IR sounders and high spatial resolution imagers data are very useful for data assimilation?DA?,for example,combining sounder/imager can help bias correction and cloud detection for DA..Converting the hyperspectral infrared sounder radiance spectrum to the broad bands using spectral response functions?SRFs?is a common approach for applications.Usually the convolution is performed in the wavenumber space.However,numerical experiments in our study indicate that there are brightness temperature differences?BTDs?between wavelength and wavenumber spaces in convolving hyperspectral IR sounder spectrum to broad IR bands,those BTDs should be taken into account for inter-comparison between IR sounders and high spatial resolution imagers.The spectral convolution difference?SCD?has a direct relationship with spectral response function?SRF?width of broadband,and for a given spectral region,with the wider SRF yielding a larger SCD.The spectral resolution of the hyperspectral radiance spectrum has little influence on SCD.The SCDs are relatively larger for the short-wave IR bands than the other IR bands.It may attribute to the fact that the corresponding radiance spectrum of short-wave IR band is relatively weak and thus more sensitive to the radiance variance.In the spectral convolution,the spectral resolution matching procedure is also required when the spectral resolution of SRF is inconsistent with the hyperspectral IR radiance.Interpolation scheme should be selected when the spectral resolution of radiance spectrum taken as a calibration reference is higher than that of IASI apodizated measurements.The scheme of interpolating SRF to hyperspectral IR sounder resolution is recommended for most IR bands.The central wavelength and central wavenumber should be determined separately using the same formula in wavelength and wavenumber spaces,respectively;and there are BTDs between these two approaches for broad IR bands;the differences are related to the spectral region and band width.