A NEW THREE-DIMENSIONAL VISUALIZATION SYSTEM FOR COMBINING AIRCRAFT AND RADAR DATA AND ITS APPLICATION TO RICO OBSERVATIONS

Dan K. Arthur1, Sonia Lasher-Trapp1, Ayman Abdel-Haleem2, Nicholas Klosterman2, and David S. Ebert2

1Department of Earth and Atmospheric Sciences, Purdue University, West Lafayette, IN,

USA 2Purdue University Rendering and Perceptualization Laboratory, West Lafayette, IN, USA

1. INTRODUCTION AND MOTIVATION

Meteorological field studies often provide researchers with diverse data sets gathered from different sources. The Rain In Cumulus over the Ocean (RICO) field campaign (Rauber et al. 2007) was such a study; along with the use of satellite observations, three research aircraft, a research vessel, surface observations, and the NCAR SPolKa radar were deployed to the Caribbean on and around the islands of Antigua and Barbuda for several weeks in December 2004 – January 2005. The observational strategy of RICO was to gather data on shallow maritime convection and trade wind cumuli at a wide range of scales. One goal of the project was to gain a better understanding of the warm rain process, as traditional theory has been unable to explain how the growth of cloud droplets by condensation alone proceeds to large enough drop sizes to begin the coalescence process and produce precipitation as rapidly as has been observed (Beard and Ochs 1993).

Examination of such data has generally involved using separate applications, often specific to the type of data being analyzed, making determination of correlations and synthesis of information between data sets extremely challenging. For example, during RICO, the aircraft conducted statistical sampling of cloud dynamic, thermodynamic, and microphysical properties along one-dimensional transects through clouds, while the radar gathered data on larger scales of cloud and hydrometeor motion and evolution. C130 aircraft measurements ranged from roughly 4 m to 100 m in spatial resolution, depending on sampling rate of the instrumentation, at constant altitude for a particular field of clouds, on a GPS-based

lat-lon coordinate system. One-dimensional C130 data in netCDF files traditionally have been used with an application to produce time series plots such as that shown in Fig. 1. Radar measurements are based in a polar coordinate system with resolution decreasing with distance from the radar site. Radar scans during RICO were conducted primarily at successive constant elevation angles over an azimuthal sector of at least 180°, with a set of increasing elevation scans comprising one volume scan. The half-power beam width of the radar was 0.91° (Keeler et al. 1991), with 150 m a range resolution between samples. The S-Pol data, in its own format, can be viewed in successive two-dimensional slices for example with SOLO II (Oye et al. 1995). Figure 2 depicts a portion of an elevation scan within the same time period as that in Fig. 1.

Although useful in their own right, such traditional analysis tools force the investigator to spend significant amounts of time trying to collocate the data, and synthesize it into some conceptual picture, before being able to use it to test hypotheses. Here we present a new tool for combining multi-source, multi-scale data in three dimensions, allowing users to make queries from the combined data sets, applied toward the effort of attaining a better understanding of precipitation development in trade wind cumuli. In this work we examine microphysical probe data collected by aircraft simultaneously with radar data viewed in three dimensions. The new tool ___________________________________ Corresponding author’s address: Dan K. Arthur, Dept. of Earth & Atmos. Sciences, Purdue University, West Lafayette, IN, 47907; E-Mail: dkarthur@purdue.edu.

facilitates data organization and synthesis and may be of use in both individual case

studies and in the statistical analysis of the properties of the entire cloud field.

Fig. 1. A segment of aircraft time series data from four cloud droplet and precipitation probes during the 23 Jan 2005 RICO flight, using ncplot (Webster 2007). Time is indicated along the abscissa, and total drop concentrations are plotted on the ordinates.

Fig. 2. A portion of one radar elevation scan during 23 Jan 2005, displaying reflectivity factor (Ze). Color scale along bottom indicates increasing ranges of Ze in colors toward the right (in dBZ, scale labels duplicated below colors for clarity). Ground clutter from the northern end of the island of Barbuda is visible near the bottom center of the figure. A field of small cumuli is indicated by the green clusters.

2. SYSTEM DESCRIPTION

Previous radar visualization packages capable of 3D output have been based on generation of isosurfaces (e.g., Johnson and Edwards 2001). For microphysical research applications, however, isosurface-based 3D products can present serious drawbacks. Specific values of multiple parameters in the 3D domain are often difficult to obtain from isosurface representations, and navigation of the 3D space, particularly when viewing multiple values in translucent surfaces, can be disorienting and confusing without clear reference coordinates.

Overcoming the challenge of collocating and synthesizing microphysical data such as that in Fig. 1 with larger-scale radar data such as in Fig. 2 is one goal of the application described herein. Collocation of the 3D polar coordinate radar volume with the lat-lon coordinate aircraft data is achieved via calculation of geodesics between the S-Pol site latitude, longitude, and altitude and the same values from GPS data for each aircraft data point (Vincenty 1975).

Once combined, the data sets can be queried to look for correlations between aircraft microphysical probe variables and radar data where they are spatially and temporally collocated. Such correlations can be used to make inferences across the field of trade wind cumuli where radar histories are well-documented but microphysical probe data are lacking due to the statistical sampling nature of the aircraft flights.

Our system uses netCDF data, in order to maintain standardization and transferability, so files from the RICO C130 flights needed no advance preparation. In order to use the radar data, a file translator that is part of the SOLO II software package (Oye et al. 1995) was used on the S-Pol files to convert them to netCDF. Any single radar variable and multiple 1-D aircraft variables can then be visualized simultaneously. The system uses a transfer function scheme to plot data quantitatively by color and/or opacity (Fig. 3). The user can set limits on what range(s) of values can be

viewed by controlling the limits of this function.

Fig. 3. Transfer function for converting numerical data into color and opacity values for 3D plotting. Function shown displays Ze > -1 dBZ at 75% opacity, decreasing from 50% to 10% opacity for Ze between -1 dBZ to -8 dBZ.

A portion of one set of S-Pol elevation scans (one radar volume scan) is shown in Figs. 4, 5, and 6, from three different views. Using the PC mouse and keys, the investigator can rotate, pan, and zoom within the 3D view in order to examine any particular region more clearly. Coordinates can be input to define the data that is displayed in 3D to be within desired absolute geographic (lat-lon), relative Cartesian (in kilometers, centered on the radar site), or polar (also centered on the radar site) coordinate extents. User-defined Cartesian and polar grids can be overlaid onto the data in the 3D view if desired.

After setting color scales via transfer functions for at least one aircraft probe data variable, these data can be displayed in the system’s 3D view along with the radar data using either of two schemes: aircraft-centric time or radar-centric time. The former displays the portions of the path of the aircraft that occurred within the space defined in the 3D view for the entire flight/data file, while the latter limits display of the aircraft variable(s) to those which

correspond to the time of the currently-displayed radar volume scan (Fig. 7).

Fig. 4. Sample 3D view looking down obliquely onto portion of a 23 Jan 2005 volume scan, showing Ze based on the transfer function described in Fig. 3. A Cartesian grid is overlaid onto the boundaries of the 3D space. Longitude scale is visible along the edge of the grid (lower left). Bluish clusters are likely fields of small cumulus, with some brighter colored regions more dense cloud regions, and the bright red to the right of center is a “skin paint” (direct detection of the aircraft by the radar, e.g. Bringi et al. 1991).

Fig. 5. A different view of the same volume as in Fig. 4, also demonstrating

polar grid overlay and ability to alter background and coordinate scale value colors. Latitudes are smaller numbers along bottom, azimuthal scale accompanying polar grid are larger numbers above the latitude scale.

Fig. 6. Looking down onto the same volume as shown in Figs. 4 and 5, with a transfer function set for higher minimum Ze. North is toward top of figure. Clouds are visible as the blue regions, and more dense/more developed clouds or skin paints of the aircraft are the brighter colors.

Fig. 7. Close-up of radar volume showing the aircraft track in radar-centered time mode. The aircraft track corresponding to

the entire time of the displayed radar volume scan is indicated by the ribbon near the center of the figure. Each color band on the ribbon represents a different aircraft data variable, and color changes along a band indicate greater values based on that variable’s transfer function (i.e., color scale). The bright colored cluster of radar data penetrated by the left end of the aircraft track is a skin paint.

Along with the visual presentation of

collocated multi-source data for qualitative analysis, the system contains quantitative tools useful in determining correlations not easily done when using separate applications for each data set. Visible in Figs. 4 and 6 is a distinct 3D cube shape; this cube is a “data probe” that can be sized and moved throughout the 3D view space by the user. It displays values for selected variables collocated at its center point, including the sample time. Also, as shown in Fig. 8, it is accompanied by a display window of basic statistical information on values contained within it. If desired, the probe can be locked onto the aircraft track contained in the displayed timeframe, and scrolled along the track in order to display the values of collocated radar and aircraft data points easily.

Fig. 8. Data probe control window, displaying its center location, size, minimum, maximum, and average values for radar data within, as well as values from both data sets collocated at the center. Control of data probe size, sampling resolution, and/or scroll speed (the latter when not locked to an aircraft track) are possible.

In addition to the information available

from the data probe, 2D time series plots of the user’s choice can be generated along the aircraft flight track, either within the application as a quick-look (Fig. 9), or, alternatively, variable values from both data sets along the flight track are exported to a text file and can be subjected to further analysis with other tools.

Fig. 9. Sample 2D plot of one aircraft variable (drizzle number concentration per liter – blue curve) and one radar variable (Ze – green curve) along the aircraft track for one radar volume scan. Abscissa is time in seconds relative to start of volume scan. Peak toward the left on the green curve is the same skin paint visible in Fig. 7, Peak in blue curve corresponds to drizzle values visible along the aircraft track ribbon (yellow, red pixels) shown in Fig. 7.

Time is another important

consideration in collocating radar and aircraft probe data. The aircraft is not always in the same place at the same time as the radar beam, and if the two do exactly coincide temporally as well as spatially, the radar values must be discarded as a “skin paint” (direct detection of the aircraft by the radar, e.g. Bringi et al. 1991). In order to attempt to account for the movement of radar echoes between the time the associated clouds were sampled by the aircraft and scanned by the radar, the system has a flexible advection adjustment mechanism (Fig. 10). The user can input an advection time, around which the radar is advected based on a user-input wind speed and direction. Radar samples after this time are advected along the direction of the wind, the distance depending on the wind speed and the difference between the sample time and the advection time. Radar samples occurring before the advection time are advected opposite the wind direction in a similar manner. This is admittedly a simplistic approach, and may introduce some errors in resulting

collocations, but because the radar echoes evolve in time as well as in space, no advection scheme will achieve perfectly collocated results.

Fig. 10. Advection control window. To reduce calculation time, the user can limit the extent of advected radar data to be within a region of interest.

The application also includes the

ability to perform multi-parameter database queries (Fig. 11). These can be executed to define a data set to locate echo regions of interest in any radar volume scan accessible by the database. For example, specific collocated Ze and ZDR pairs within user-defined limits accompanying radial velocity data (also within defined limits) could be located. Once found, these echo regions could be visually tracked across multiple radar volume scans to observe their evolution.

Fig. 11. Sample screenshot of the database query interface. Future expansion will add the ability to query several aircraft variables in addition to the radar variables shown.

Fig. 12. A drastic example of a 3D radar volume scan viewed before selection of data by a database query. Values shown are -20 < Ze < 55 dBZ, so much of the echo filling the view is outside of cloud (cf. the 2D slice in Fig. 2).

Fig. 13. 3D radar plot of a volume scan after selection of data by query: -1 < Ze < 10 dBZ and -0.5 < ZDR < 0.5 dB. Aircraft track (dark stripe) is now visible. Background color was changed from Fig. 12 to emphasize differences between selected and unselected data.

In order to display query results in 3D, the user can switch the 3D rendering mode from using transfer functions exclusively for color scales and opacity to “context mode”. In this mode, the currently defined transfer function controls color and opacity of the selected data, and non-selected data can still be viewed using the same colors of the transfer function, at a different opacity controlled by the user. For example, as shown in Figs. 12 and 13, non-selected data is still visible after execution of the query. In this example, the opacity of non-selected data was set to 0.3, which is multiplied by the opacity of the current transfer function (90%), resulting in an opacity of 2.7%. The ability to view non-selected data along with selected data can aid keeping that data in context, to see possible cloud edges, areas of fractocumulus, Bragg scattering, and other weak echoes. Although Figs. 12 and 13 show a drastic example of the use of this feature, the “context mode” can easily help the investigator focus on a particular set of clouds within the entire field, based on the strength of their radar echoes, for example.

3. APPLICATION

Although it is difficult to convey the utility of the new software in a 2D medium, because its greatest advantages are in displaying the data in 3D and manipulating the view interactively, an initial application is presented here with a subset of the RICO data.

The initial task in examining the S-Pol and C130 aircraft data as a combined set in 3D involved finding peaks in drizzle/raindrop number concentration from the 260X, 2DC, and 2DP optical array probes1 mounted on the C130 aircraft, and studying the radar echoes closest to the cloud penetration times where the peaks occurred. Because the clouds evolve and travel through the radar volume in time, the maximum Ze and ZDR values detected in a particular cloud may not be collocated with the peaks in drizzle/raindrop number concentrations detected by the aircraft at the same spatial location. This offset is easily detectible with the new software described here. From 42 volume scans across three different days during RICO, only 10% of the clouds sampled by the aircraft had peaks in drizzle/raindrop number concentration collocated with a maximum in Ze, and only 7% with a maximum in ZDR. In addition, only 57% of the maxima in Ze were collocated with the maxima in ZDR within the clouds. Although this latter estimate requires analysis of additional cases to generalize these percentages as an overall trend in the trade wind cumuli observed during RICO, it is in accord with analysis of the RICO radar data by Knight et al. (2008). In an earlier study by Knight et al. (2002), the authors examined early development of Ze and ZDR in Florida cumuli, and speculated that separation of the greatest Ze and ZDR signals in the cloud may result from larger drops appearing in weaker areas of clouds, before coalescence begins in earnest in regions often indistinguishable from Bragg scattering. 1 Manufactured by Particle Measurement Systems, Inc., Boulder, CO, USA.

The new software is also useful for studying individual clouds and gaining perspective on the time a cloud was sampled by the aircraft versus its stage of evolution on the radar. Figures 14 through 17 show the aircraft track (or a portion thereof) within a given radar volume. It can be seen in these figures that the relative peak radar values are not collocated with the peak drizzle/raindrop concentrations measured by the aircraft. Due to the time differences between drop detection by the 260X and/or 2DC and scanning by the S-Pol, the maximum radar echoes are located below the aircraft tracks in Figs. 14 through 17. The aircraft wind measurements indicated that downdrafts were present at the time of cloud penetration in Figs. 14 through 16, and thus the maximum radar echoes were detected at lower levels by the radar than the flight level at which the aircraft sampled the clouds. In Fig. 17, the radar scanned the cloud well before aircraft penetration, and the entirety of the cloud echo appears below the sampling altitude of the aircraft. Aircraft measurements in this case indicated a cloud was penetrated at this location and an updraft was present at this time, so the drizzle sampled by the aircraft must have developed from lower altitudes that were scanned earlier by the radar. These examples demonstrate the ease with which the aircraft data and radar data can be combined, thus allowing the investigator to focus upon the precipitation evolution occurring within the cloud. Simple calculations constrained by the aircraft and radar observations, with the understanding of their spatial and temporal relationships, can then be used to provide insight into the possible mechanisms influencing precipitation development in these cases. Such analysis is planned, over the entire RICO data set.

Fig. 14. Example of aircraft penetration before radar scanning of a cloud, from radar volume scan beginning at 17:56:20 UTC on 20 Dec 2004, viewed from above. Arrow highlights position along track of 260X peak value and cloud echo below track. Peak 260X concentration is 10.7 L-1, with collocated radar values of Ze = -12.9 dBZ and ZDR = 0.8 dB. Below the aircraft track, Ze = 1.51 dBZ and ZDR = 0.01 dB. Z and ZDR calculated from all aircraft droplet probes are 1.47 dBZ and 0.03 dB, respectively.

Fig. 15. As in Fig. 14, side view, arrow highlighting maximum 260X concentration point along aircraft track.

Fig. 16. As in Figs. 14 and 15, for radar volume scan beginning at 14:05:48 UTC 23 on Jan 2005. Peak 260X concentration (red pixel) is 189 L-1, collocated Ze = 6.61 dBZ. Z calculated from aircraft probes is 18.51 dBZ. High values on the right side foreground are part of a skin paint of the C130.

Fig. 17. Example of aircraft penetration after the radar scanned the cloud, from radar volume scan beginning at 14:33:48 UTC 23 on Jan 2005. Arrow highlights peak 2DC concentration of 242 L-1, with collocated Ze = -6.4 dBZ. Below the aircraft track, Ze = 11 dBZ. Z calculated from aircraft probes is 23.68.

4. FUTURE WORK

As discussed previously, approximately half of the examined maximum Ze echoes were not collocated with the max ZDR echoes. Past studies have suggested that size-sorting of raindrops within the clouds may cause such a spatial disparity in these echoes [e.g., Knight et al. (2002, 2008)]. Such size sorting also suggests that the earliest raindrops falling from such clouds may result from a different microphysical path (for example, giant aerosol particles, or earlier thermals) than the bulk of the later rainfall. It is our intent to examine this dislocation of Ze and ZDR more closely with this 3D visualization system and the combined RICO datasets.

Analysis of cloud evolution across entire cloud fields, as mentioned above, will be conducted to determine if precipitation development occurs differently across the cumulus field on different days. The new visualization system can aid in easily detecting differences in cloud number, depth, width, spatial separation, organization, cloud lifetime, etc., across different days within the RICO field campaign that can be related to precipitation development. 5. ACKNOWLEDEGMENTS The authors wish to acknowledge the National Center for Atmospheric Research (NCAR) for the use of the ncplot and SOLO II software, and Robert Rilling and Scott Ellis of the NCAR Earth Observing Laboratory for assistance with the NCAR S-Polka radar data. NCAR is sponsored by the National Science Foundation. We would also like to thank the participants of RICO for their data collection efforts, Amanda Sheffield for general statistical analysis of C130 data from each RICO flight, and the Purdue Cloud Microphysics Group for numerous helpful discussions and suggestions. Support for this project was provided by NSF grant IIS-0513464.

6. REFERENCES Beard, K., V., and H. T. Ochs, 1993:

Warm-rain initiation: an overview of microphysical mechanisms. J. Appl. Meteor., 32, 608-625.

Bringi, V. N., D. A. Burrows, and S. M. Menon, 1991: Multiparameter radar and aircraft study of raindrop spectral evolution in warm-based clouds. J. Appl. Meteor., 30, 853-880.

Johnson, S. G., and J. Edwards. “Vis5d+ Home Page.” Vis5d+. 2 April 2001. SourceForge. 1 May 2008 http://vis5d.sourceforge.net/.

Keeler, R. J., W. Lutz, and J. Vivekanandan, 2000: S-Pol: NCAR’s polarimetric Doppler research radar. IGAARS-2000, Honolulu, HI, 24-28 July, 2000, 4 pp.

Knight, C. A., J. Vivekanandan, and R. A. Rilling, 2008: Structural and evolutionary aspects of trade wind cumulus determined from dual-polarization measurements. J. Atmos. Sci., in press.

_____, _____, and S. G. Lasher-Trapp, 2002: First radar echoes and the early ZDR history of Florida cumulus. J. Atmos. Sci., 59, 1454-1472.

Oye, R., C. Mueller, and S. Smith, 1995: Software for radar translation, visualization, editing, and interpolation. Preprints, 27th Conference on Radar Meteorology, Vail, CO, Amer. Meteor. Soc., 359-361.

Rauber, R. M., et al., 2007: Rain in shallow cumulus over the ocean: the RICO campaign. Bull. Amer. Meterol. Soc., 88, 1912-1928.

Vincenty, T., 1975: Direct and inverse solutions of geodesics on the ellipsoid with application of nested equations. Survey Review XXII, 176, 88-93.

Webster, C. “EOL/CDS Aircraft Platform Software Product Guide.” Research Aviation Facility. 27 Feb. 2007. NCAR Earth Observing Laboratory. 23 Apr. 2008 http://www.eol.ucar.edu/raf/Software/.