This cycle will include continuous data assimilation, a 50-member EDA, additional weak coupling of the atmospheric and ocean data assimilation systems. In data assimilation, a stronger coupling will be introduced between the atmosphere and the surface, with Jacobians computed from the EDA used in the surface analysis. Observation usage will be improved thanks to upgrades in the infrared and the microwave all-sky packages.
The Continuous Data Assimilation contributions allow use of later arriving observations, and crucially decouples the start time of the assimilation from the observational cut-off. This permits the beneficial introduction of an additional outer loop without affecting delivery time. Finally as part of the continuous data assimilation package, the early delivery assimilation window length has been increased from 6 hours to 8 hours, thus ensuring that \emph{all} observations that have arrived can be assimilated.
The Ensemble of Data Assimilations (EDA) has increased from 25 to 50 members (with benefits to both the HRES and ENS) which was enabled by significant work on the efficiency of the system. In a progressive step in ocean--atmosphere weakly coupled assimilation, the atmospheric analysis sea-surface temperature in the tropics is now taken from the ECMWF OCEAN5-NRT analysis rather than using the OSTIA product directly.
The spatial interpolation of the model to observation locations in trajectories and minimisations has been made consistent. Interpolation in the nonlinear trajectories has been changed from bicubic to bilinear interpolation, resulting in better fits to observations as well as a slight improvement to computational efficiency.
The radiative transfer model RTTOV has been upgraded from v12.1 to v12.2. This is a broad scientific and technical upgrade to the microwave observation operators RTTOV and RTTOV-SCATT. The upgrade allows RTTOV to use the most accurate science possible, it prepares the way for future sensors like Ice Cloud Imager (ICI). Band corrections are implemented for all microwave sensors, improving accuracy of simulated microwave radiances. RTTOV-SCATT now does its radiative transfer in terms of radiance, not brightness temperature, improving accuracy by several tenths of a Kelvin in some channels (this also prepares the way for future mm-wave sensors like ICI). All-sky dynamic emissivity retrievals have been externalised and use a newly-provided RTTOV framework. This involves minor changes and improvements in the scientific basis of the retrieval, such as the use of the correct clear-sky emissivity in the clear-sky column. New Mie-tables are used that have been generated using the new permittivity model of Rozenkranz (2015).. There has been an update to the infrared aerosol detection channels, and re-specification of related tuning parameters to make them consistent for AIRS, IASI, and CrIS.
For the surface analysis of soil moisture, the Simplified Extended Kalman Filter (SEKF) has been significantly upgraded to improve the computational efficiency by computing its Jacobians directly from the EDA rather than with perturbed nonlinear trajectories. The SMOS neural network soil moisture product is now an additional observational input into the SEKF.
Cycle 46R1 has introduced a package of changes to microwave all-sky assimilation. This includes assimilation of SSMIS-F17 150h GHz and GMI 166 v/h GHz channels as well as improving the use of the land sea mask in the field of view for microwave imagers. Each Microwave observation has a footprint depending on its frequency. We use the 10GHz footprint for AMSR2 and GMI and the 19GHz footprint for SSMIS-FOV to compute how the land-sea mask is affected by this footprint. We name this land-sea mask ``lsm_fov'', stored in odb space. lsm_fov is more accurate than the lsm used in CY45R1, which depends on the resolution of each loop. lsm_fov is then later used (if available) to decide if a superob is contaminated by land (not yet assimilated) or not (assimilated). 1% land contamination is allowed as it has been identified to cause a change in brightness temperature of 1K, which is in the noise level of first guess departures.
Interchannel observation error correlations have been introduced for ATMS. The error covariance matrix is diagnosed using the Desroziers method from an experiment using a previously diagnosed matrix from the Hollingsworth-Lonnberg method (the same method was used to produce the IASI error covariance matrix implemented from cycle 43R1). The diagnosed error standard deviations are inflated by an empirically tuned factor of 1.75.
Similarly interchannel observation error correlations have been introduced for geostationary water vapour channels. The current instruments are: GOES IMAGER (1 WV channel), SEVIRI (Meteosat Second Generation) (2 WV channels), and AHI (Himawari) (3 WV channels). The Desroziers diagnostic R matrix has been computed for each of these instruments, and the resulting errors have been scaled by a factor of 6 in order to provide the best first guess fit to water vapour channels on other instruments (as well as impact at longer lead times). Further upgrades to the use of geostationary radiances is to extend their use to higher zenith angles (i.e. at the edge of the disk). At these high angles, line of sight of the satellite passes through a number of model columns. In order to improve the representation of these observations, the ``slant path'' method is used to perform the radiative transfer calculation along the line of sight. The configuration for CY45R1 used only vertical information from a single location. Currently, geostationary radiances are rejected in the blacklist if the satellite zenith angle is greater than 60 degrees. This change allows data to be used up to zenith angles of 74 degrees, thus improving coverage at the edges of the geostationary disks. This is particularly significant in the North Atlantic, where a significant amount of Meteosat-10 data is currently not being used.
This contribution updates the use of DBNet data with better timeliness (for AMSU-A and MHS). Addition of DBNet data from South American and New Zealand stations. Thinning to give preference to global rather than DBNet data for AMSU-A. Correction of azimuth angle provided in the data from some DBNet stations.