World Meteorological Organization Symposium on Nowcasting - WMO

58
can be solved for position and amplitude decision variables using stochastic methods, thus
corresponding with ensemble data assimilation. We then show that if an Euler-Lagrange
approximation is made, can solve the objective nearly as well in two steps. This approach is
entirely consistent with contemporary data assimilation practice. In the two-step approach,
the first step is field alignment, where the current model state is aligned with observations by
adjusting a continuous field of local displacements, subject to certain constraints. The second
step is amplitude adjustment, where contemporary assimilation approaches are used. We will
then demonstrate several choices of constraints on the displacement field, first starting with
fluid-like viscous constraints and then proceeding to a multiscale wavelet representation that
allows better balance in the factorization of error into position and amplitude subspaces. Our
new data assimilation by field alignment approach does not rely on the detection of storm
features, and can be used with sparse station observations, just as easily as with detected
features. The two-step approach can be used with any assimilation method in practice;
3DVAR, 4DVAR, and EnKF (and variants). The new method has been implemented with
multivariate fields, and extensions to 3D is straightforward. Ancillary benefits to velocimetry
for rainfall modeling and wind-from-satellites have also been realized. A new data
assimilation system, FAVAR, has been developed to co-exist with using the WRF-VAR
system. Demonstrations on storms using OSSEs and identical twins will be shown as a runup
to a live Hurricane DA & Forecasting System at the Atmospheric Sciences Group at MIT.
SESSION 4: Applications or Op Systems
4.1
INCA - A new operational nowcasting system for mountainous areas
6.12
Data mining for thunderstorm nowcast system development
John K. Williams, David A. Ahijevych, Matthias Steiner and Susan Dettling
National Center for Atmospheric Research, Research Applications Laboratory
This paper describes a data mining statistical analysis approach to developing real-time
thunderstorm nowcasts for aviation users. While fuzzy logic expert systems are often used to
combine observation and NWP model data to form short-range predictions of thunderstorm
intensity evolution, evaluating and incorporating new data sources is often a time-intensive
manual process, and it is difficult to know whether available information is being used
efficiently. A technique called random forests (RFs) provides a means of objectively
identifying the potential contribution of candidate predictor variables, along with a method for
creating a nowcast logic using a minimal skillful set of predictors. The RF methodology was
used to evaluate radar, satellite, lightning, and RUC model data and derived features
collected during the summer of 2007 and 2008 over the eastern U.S., including data fields
produced by MIT Lincoln Laboratory’s Corridor Integrated Weather System (CIWS). The
resulting data fusion system produces real-time probabilistic and deterministic nowcasts of
thunderstorm intensity (VIP level). Statistical evaluations of the RF-based nowcast’s
performance are shown and several case studies are analyzed, demonstrating the value of
this approach. This research has been funded by the U.S. Federal Aviation Administration to
support the development of the Consolidated Storm Prediction for Aviation (CoSPA), which is
intended to provide the thunderstorm nowcast capability for the Next Generation Air
Transportation System (NextGen).
SESSION 7: Verification and Impacts
Thomas Haiden, Alexander Kann
Central Institute for Meteorology and Geodynamics, Vienna, Austria 7.1
Verification methods for spatial forecasts
The high-resolution analysis and nowcasting system INCA (Integrated Nowcasting through
Comprehensive Analysis) provides 3-D hourly fields of temperature, humidity, and wind, and
2-D fields of cloudiness, precipitation rate, and precipitation type at an update frequency of 15
min. The system operates on a horizontal resolution of 1 km and a vertical resolution of 100-
200 m. It combines surface station data, remote sensing data (radar, satellite), forecast fields
of numerical weather prediction models, and high-resolution topographic data. In the alpine
area, the system provides meteorological input for operational high-resolution flood
forecasting and is used for winter road maintenance. INCA employs a new radar/raingauge
combination algorithm and includes elevation effects on precipitation using an intensitydependent
parameterization. In temperature analysis and nowcasting the pooling of cold air is
parameterized as a function of terrain parameters. Verification results showing the skill of the
nowcast compared to a high-resolution NWP model (AROME) are presented and the
potential for application of INCA to the 2010 Winter Olympics in the Vancouver/Whistler area
is discussed.
Barbara Brown [1] Eric Gilleland [1] David Ahijevych [1] Barbara Casati [2] Beth Ebert [3]
[1] NCAR, USA [2] Ouranos, Canada [3] Bureau of Meteorology, Australia
In recent years, many new methods have been developed to evaluate forecasts that have
coherent spatial structures. Several different categories of approaches have been developed,
including object- or features-based, scale separation, neighborhood, and field deformation
methods. Because the majority of nowcasts (e.g., convective nowcasts based on radar
reflectivity) are characterized by identifiable spatial features and structures, the spatial
verification methods are appropriate for the evaluation of these products and can provide
meaningful information on their quality. Over the last several years, the developers of many of
the spatial verification methods have been involved in an intercomparison project that was
designed to compare the capabilities of the methods and provide information about how the
aspects of performance are measured by each approach. The intercomparison was based
on the evaluation of the same set of high-resolution spatial forecasts by each method. In
addition, the methods were applied to a set of artificial geometric cases with simple, known
forecast errors. This presentation will summarize the results of the intercomparison and
83