Space-borne scatterometers have been continuously providing global sea-surface vector winds since the launch of ERS-1 mission in 1991. Although a single scatterometer has a large daily global coverage, it passes most ocean areas only once per day. Thus, for effectively monitoring ocean surface winds, multiple satellite scatterometers are needed. An ocean surface vector wind virtual constellation (OSVW-VC) has been proposed by CEOS (http://ceos.org/), in which, for example, the equator crossing times are being coordinated to make the OSVW-VC effective. The combined wind products from these multiple scatterometers in the OSVW-VC will increase spatial coverage and decrease revisiting period, which will greatly benefit the end users. However, significant systematic differences do exist in winds derived from different scatterometers and may result in detrimental impacts on the applications if unknown. Therefore, describing the error characteristics and inconsistencies of scatterometer wind products are very meaningful for data users, and also can help researchers to solve the reported issues.
The SCATSAT-1 mission scatterometer (OSCAT-2) was launched by the Indian space research organization in September 2016, with its main purpose to measure ocean vector winds. Based on many collocations of wind retrievals from the current operating ASCAT and OSCAT-2 scatterometers by Royal Netherlands Meteorological Institute (KNMI) or Jet Propulsion Laboratory (JPL), the inconsistencies of ASCAT and OSCAT-2 wind products are analyzed, and the potential causes are discussed. The error characteristics and inconsistencies of ASCAT and OSCAT-2 wind products that we will present are good references for data users and will promote further improvements in wind products from scatterometers. Such error knowledge are also very important for the community which would benefit from multiple scatterometer missions.
In addition, some preliminary results of wind products retrieved from the newly launched Chinese HY-2B and CFOSAT missions are also planned to be presented.

Area C - Board 272

Following the success of the QuikSCAT, Oceansat-2, HY-2A, and RapidScat missions, a new Ku-band rotating pencil-beam scatterometer, ScatSat-1 from the Indian Space Research Organization (ISRO) was launched in September 2016. Scatterometer sea surface winds have been used in a wide variety of atmospheric, oceanic, and climate applications. Moreover, thanks to the near-real-time data distribution of most missions, scatterometer wind data have been successfully assimilated into numerical weather prediction models for more than two decades.

In the framework of the EUMETSAT Numerical Weather Prediction Satellite Application Facility (NWP SAF) and Ocean and Sea Ice Satellite Application Facility (OSI SAF), the Royal Netherlands Meteorological Institute (KNMI) has developed the so-called Pencil-beam Wind data Processor (PenWP), which has provided and provides near-real-time Level 2 (swath-based) sea surface wind fields for all past and current rotating pencil-beam scatterometer missions. The main components of PenWP include calibration, inversion, quality control, and ambiguity removal. Research & Development activities within the NWP SAF and OSI SAF over the past 15 years have focused on the improvement of the different algorithms of the scatterometer wind data processors, including PenWP. Recent results show that both the Ku-band forward model or geophysical model function (GMF) and the quality control (QC) can be further improved. In this paper, we focus on these two relevant aspects of the scatterometer wind data processing in order to improve the ScatSat-1 wind retrieval quality.

Recent developments on the wind GMF of Ku-band (~2 cm wavelength) scatterometers include a sea surface temperature (SST) dependent term. Moreover, it has been found that the SST effects on the radar backscatter are wind speed dependent, and are mainly relevant at radar wavelengths smaller than C-band (~5 cm wavelength). A new Ku-band GMF, NSCAT-5, has been developed based on a physical model and RapidScat radar backscatter measurements. The objective of this work is to verify the NSCAT-5 GMF at slightly different incidence angles. First, the ScatSat-1 backscatter sensitivity to sea surface wind and SST is assessed using the best available wind reference at scatterometer scales, i.e., the C-band Advanced Scatterometer (ASCAT) winds, for which plenty of collocations are available (note that during the first year of the ScatSat-1 mission, both ScatSat-1 and ASCAT were flying on a very similar orbit, which resulted in a large amount of nearly coincident measurements in space and time). Second, the approach used to derive the NSCAT-5 GMF for RapidScat is adapted to derive a SST-dependent GMF for ScatSat-1. In particular, the GMF sensitivity to quality control effects is also investigated. The new GMF will be used to consolidate the current NSCAT-5 model, and then evaluated for ScatSat-1 wind retrieval.

In the current version of PenWP, a maximum likelihood estimator (MLE-) based QC is used to discern between good- and poor-quality winds. MLE QC is generally effective in flagging rain contamination and increased sub-cell wind variability in the ocean surface wind vectors derived from Ku-band pencil-beam scatterometers. However, the MLE is not an effective quality indicator over the outer swath where the inversion is underdetermined due to the lack of azimuthal diversity (including lack of horizontal polarized measurements). Besides, it is challenging to discriminate rain contamination from “true” high winds. In this paper, recent developments for Ku-band QC are adopted for ScatSat-1 data. In addition to the MLE, two wind quality-sensitive indicators are used: the spatially averaged MLE value (MLEm), and the singularity exponent (SE) derived from an image processing technique called singularity analysis. Their sensitivities to data quality and rain are evaluated using collocated Advanced Scatterometer (ASCAT) wind data, and Global Precipitation Measurement satellite’s Microwave Imager (GMI) rain data respectively. It shows that MLEm and SE are the most effective indicators for filtering the poorest-quality winds over ScatSat-1 inner and outer swath, respectively. A simple combination of SE and MLEm thresholds is proposed to optimize ScatSat-1 wind QC. In comparison to the operational PenWP QC, the proposed method mitigates over-rejection at high winds, and improves the classification of good- and poor-quality winds.

The recently launched Ku-band scatterometer missions CFOSAT (October 2018) and HY-2B (November 2018), currently in commissioning phase, will certainly benefit from efforts like those proposed here to improve and/or validate the Ku-band forward model. Reprocessing of past missions will benefit as well. In particular, the verification of the NSCAT-5 GMF at other incidence angles will be carried out with the CFOSAT scatterometer, a rotating fan-beam system which samples the surface at varying incidence angles.

Area C - Board 273

Multi-temporal imagery from a single geostationary (GEO) satellite such as NOAA’s Geostationary Operational Environmental Satellite (GOES) or Europe’s Meteosat are routinely used to derived Atmospheric Motion Vectors (AMVs) that represent winds in the atmosphere. The AMV method generally assumes that the cloud or moisture feature being tracked is undergoing horizontal motion that is observed by the displacement of the feature in a sequence of images. Observations from a single vantage point provide poor information about the height of an AMV in the atmosphere; therefore, AMV heights are generally assigned using IR temperatures and an a priori model atmosphere. Such height assignments can have large uncertainties, especially lower in the troposphere, and are error prone in the presence of multiple cloud layers. We show that multi-angle, multi-satellite stereo imaging is a powerful tool for observing atmospheric dynamics in three dimensions. When a tracked feature is viewed from multiple vantage points, additional information in the form of geometric parallax enables accurate determination of feature height. The Multi-angle Imaging SpectroRadiometer (MISR) instrument on NASA’s Terra spacecraft provides such a capability from low-earth orbit (LEO) thanks to nadir, forward, and backward-looking cameras. MISR can accurately measure the vertical height of stationary cloud tops and the cross-track component of wind, but the along-track wind and height are coupled and difficult to separate accurately. We have developed an improved retrieval method that allows for the full 3D geometry of an AMV to be observed by simultaneously tracking features in both MISR and geostationary satellite imagery to take advantage of the best characteristics of each system. It does not require synchronization between the observations from each satellite and can resolve vertical structure with an accuracy of ~100m with horizontal sampling every 2.2km. This is a significant improvement over existing single-satellite winds products and the method should be of interest to investigations of vertical structure from the tropopause to the top of the planetary boundary layer. The algorithm developed is applicable to all LEO-GEO combinations, including GOES, Meteosat, Himawari, MODIS, VIIRS, and others. We describe our 3D-Winds retrieval algorithm and present results using various LEO-GEO combinations. We show validations against operational satellite wind products and over clear-sky terrain where stationary ground features can be tracked. Stereo imaging and 3D-Wind retrievals will be important techniques to enable future LEO-GEO, LEO-LEO, or GEO-GEO systems that are proposed to satisfy 2017 Earth Science Decadal Survey objectives to observe 3D atmospheric dynamics.

Area C - Board 274

Reliable estimation of cloud parameters, and particularly of cloud-top-height, play an important role in many meteorological applications such as: weather forecasting, climate modeling and cloud classification schemes. Assimilation of cloud drift winds into numerical weather prediction (NWP) models improves forecast performance and requires assigning a height level to cloud motion vectors. Matching two stereoscopic satellite images and calculating depth information from the resulting disparity map, provides a means to determine cloud-top height over large spatial areas at high temporal resolution. Our contribution is an automated method for clouds-top height estimation that can also be used for atmospheric motion vector (AMV) calculation if applied to a sequence of temporally consecutive satellite images. By the joint estimation of cloud heights and motion vectors with a related image matching technique, we can contribute to improving NWP models into which the motion vectors are assimilated.
In this work we compare two stereo matching methods with respect to cloud-top-height estimation accuracy: a local block-based, and a global optical-flow-based method.
The local block-based method uses a window around each pixel and the cost function is calculated as a sum of squared differences of gray values within that window. To limit the computational cost, the disparity range for each pixel is determined as a function of latitude given that the cloud-top-height is a function of it.
In contrast to the block-based local method, the optical flow-based method uses a global energy function that needs to be minimized using iterative reweighted least squares algorithm. The standard gray level constancy assumption in stereo matching is extended with the gradient constancy assumption in order to avoid errors introduced by changes of gray level values between the matching pair of images (Edata term in eq. 1). Furthermore, in order to avoid the aperture problem, and the influence of outliers, a smoothness term is introduced that penalizes the total variation of the flow field (Esmooth term in eq. 1). The final energy function that needs to be minimized takes the following form:
E(u,v)=Edata+αEsmooth, (1)
where α > 0 is the smoothness weight that serves as a regularization parameter: larger values result in stronger penalization of large flow gradients and lead to smoother flow fields. Finally, in order to avoid the minimization algorithm to be trapped in a local minimum, the method is implemented in a multiscale coarse-to-fine manner using a Gaussian pyramid.
We used rectified pairs of high resolution visible (HRV) channel images collected from the Meteosat-10 (MSG-3) and Meteosat-8 (MSG-1) satellites. MSG-3 observes the earth from the standard position at 0° longitude, while from October 2016 onwards the MSG-1 makes observations from a new position at 41.5° East over the Indian Ocean. Stereo matching can be done in the subimages that correspond to the region viewed by both satellites (which includes Europe and Africa). For this study, we limit ourselves to the subimages containing Europe, that is included in the HRV product from both satellites.
The measured stereo cloud heights are validated by comparison with available cloud-top-height space lidar measurements from the CALIPSO environmental satellite. The data collected every day, over a period of one month is used for validation and the results show good agreement of the automatically determined stereo heights with the lidar reference data. Resulting cloud-top-height estimation for one pair of stereoscopic images is shown in figure 1.

Area C - Board 275

Improving our understanding of clouds and convection at a process level and their representation in everything from cloud-resolving to large-scale models is amongst the most important issues in atmospheric science today. Two separate teams were funded to perform preliminary design studies for a future field campaign by a NASA Severe Storms Research proposal call in 2015. These teams marked the first time that ground-based mobile radar scientists, satellite remote sensing experts, and modelers were brought together in the same location to address the big question: How can we improve detection of tornadoes from space? Both teams convened two separate sessions each over the 2015-2016 time frame with many attendees from government laboratories, universities and other experts from the severe storms field. Each team wrote separate white papers to summarize findings from their meetings and now both teams have combined to write one white paper. As a result, the primary purpose of the Tornado Warning Improvement including Satellite-enhanced Tools (TWIST) field campaign design is to understand if tornadogenesis is related, if at all, to features that can be detected with spaceborne remote sensors. This talk will summarize the two studies and their impact on the path forward for the science with renewed focus from the 2017 Decadal Survey, which re-emphasized the importance of understanding dynamics within severe storms and applications for societal impacts.

Area C - Board 276

The CYGNSS constellation of eight satellites was successfully launched in December 2016 into a low inclination (tropical) Earth orbit. Each satellite carries a four-channel bi-static radar receiver which measures signals transmitted by Global Positioning System (GPS) satellites and scattered back into space by the Earth surface. Over the ocean, surface roughness, near-surface wind speed and air-sea latent heat flux are estimated from the direct measurements of surface scattering cross section. The measurements are able to penetrate through heavy precipitation due to the low microwave frequency at which GPS operates. The spatial density and revisit time of sampling resulting from the number of satellites in the constellation and from their continuous data-taking operation makes possible the reliable detection of tropical cyclone intensification and the resolving of diurnal cycles of tropical winds. Engineering commissioning of the constellation was completed in March 2017 and the mission is currently in the science operations phase.

Level 2 science data products have been developed for near surface (10 m referenced) ocean wind speed, ocean surface roughness (via the mean square slope statistic) and latent heat flux. Level 3 gridded versions of the L2 products are also available. A set of Level 4 products have also been developed specifically for direct tropical cyclone overpasses. These include the storm intensity (peak sustained winds) and size (radius of maximum winds), its extent (34, 50 and 64 knot wind radii), and its integrated kinetic energy. Assimilation of CYGNSS L2 wind speed data into the HWRF hurricane weather prediction model has also been studied, and their impact on forecast skill demonstrated. Level 2 science data products over land related to near-surface volumetric soil moisture content and flood inundation extent are also in development.

The current status of the mission will be presented, followed by summaries of recent results related to the assimilation of CYGNSS data into hurricane prediction models and their impact on resolving the structure and dynamics of surface winds and winds aloft in the inner core of the storm and on forecast skill related to storm track, intensity and structure.

Area C - Board 277