TITAN 

One of the greatest shortcomings in evaluations has been the accurate measurement of the final product, the precipitation, or more precisely, the change in precipitation due to seeding from individual clouds and over an area. Research in the substantial scatter and standard errors of the seemingly inexhaustible correlation between radar reflectivity factor Z and rainfall rate R, has come to a point where the relation Z-R is not very useful, because we are looking for a good temporal and spatial resolution for research and hydrological problems. It follows that cloud seeding evaluations based on Z-R relations are subject to large errors. Yet, unlike other options, only radar offers continuous, wide-area surveillance. It is now becoming well established that R can be estimated much more accurately using K_dp (^o/km) which is the specific propagation differential phase.

Reinking (et al. 1999) on their paper, they are focusing on the delivery of the seeding agent, its dispersion, and the response of microphysical transitions within the seeded volume, resulting precipitation as it falls, and the consequent quantity of rain, and they mention that rapid PPI or low-angle RHI, narrow-sector scanning of the cloud volume with a short-wavelength dual polarization radar (or radars) will measure the elapsed time, transport and dispersion of the chaff, and follow the hydrometer evolution in the treated/untreated regions between the chaff tags. Profiles of Depolarization ratio and Z are shown in Fig. 1, thus treated and untreated volumes of one cloud or separate clouds can be compared to look for natural vs. seeded patterns and timing of transitions. Transitions A to C in Fig. 1, are expected to be detectable from the corresponding evolution in depolarization (DR). Depolarization increases with drop size. Cloud drops and most drizzle drops are spherical and do not depolarize a radar’s signal. A cloud in early stages of growth (low reflectivity; Z_e £ -20 dBZ, droplet D £ 30 mm) will likely be below the threshold of detection in DR, but still be observable in reflectivity. As the hygroscopic material increase drop concentration and size, Z will increase and DR will become measurable at the cross-talk limit of the radar, indicating minimum depolarizations (droplets and drizzle). Growth through D » 0.3 mm leads to sine nonsphericity, so large drizzle drops cause slight depolarizations (transitions A, Fig 1). If the ice process becomes involved, greater changes in DR are expected. A transition to pristine crystals will result in the largest depolarization (B, Fig 1); which will be moderated by subsequent transformations to aggregates and graupel, which tend more toward sphericity (lesser DR). Rain drops of D ³ 3mm are oblate and cause considerable depolarization relative to drizzle, but less than ice.

Weather radars have proved to be excellent tools for the study of spatial structure of rain. Attempts have been made to use radar for the quantitative measurement of precipitation, especially for the use in the evaluation of weather modification schemes; Crane (1977) used radar calibration and radar-rain gauge comparisons, to get an estimated of augmentation of rain fall in 1976 form a cloud seeding program in Goodland, Kansas, but he had some problems due to the small number of seeded and non seeded clouds that he sampled. While many studies have shown gage-calibrated radar rainfall measurements to be a very promising technique, several questions have remained unanswered. Most importantly, the relative accuracy of the gage only and gage-radar rainfall analyses needs to be investigated for various gage densities and analysis intervals.
 

Fig. 1 Schematic combined with actual data, illustrating expected transitions in depolarization as hygroscopic seeding broadens the cloud droplet spectrum to produce drizzle (A), the drops freeze and grow first as ice crystals (B), which then rime and/or aggregate as the fall to produce rain (C).

Hildebrand et.al. (1977), used the same technique (radar rainfall measurements in precipitation modification experiments) and they found out that using gage-radar techniques can help in the measurements of the effects of seeding single raincells. This is due to the lack of gage information during cloud seeding programs.

There has been different cloud seeding programs working with radars over the years, trying to find a more suitable way of estimate the rainfall that can be produced in a seeding experiment.

Now we will be focused on the use of a radar for the quantitative evaluation of the seeding experiments that took place in Mexico (PARC, Program for the Augmentation of Rainfall in Coahuila). And how this radar and its software TITAN: Thunderstorm Identification, Tracking, Analysis, and Nowcasting developed at NCAR (Dixon and Weiner, 1993) can be used to calculate the different properties of the seeded clouds.

The success of the South African hygroscopic seeding experiment (Mather et al., 1997; Bigg, 1997) was sufficiently encouraging to investigate its applicability to other regions of the world; and Monclova, province of Coahuila on the north of Mexico was not the exception.  The goal was to develop a cloud seeding strategy that was optimal for the region and could be tested. Following the 1996 field season, an experimental design was written, based largely on the work and results in South Africa (Mather et al., 1997), and a randomized seeding experiment, utilizing hygroscopic seeding material released at cloud base, was begun in 1997 (July-September).

The primary tool for nowcasting and for quantitative evaluation of the seeding experiment is an Enterprise EEC/WR100/77 5-cm wavelength weather radar with a beam width of 1.6° (2.43 m parabolic-dish antenna). The beam width limits the useful range for data collection to 90 km in the randomized seeding experiment. The radar is operated in an automated volume-scan mode - 16 tilts from 1.5° to 48.4° resulting in a volume every 4 min. The TITAN software is used for the display of radar data and aircraft position in real-time for the purpose of directing operations. TITAN is also used as the automated evaluation software, defining the experimental unit and producing time-series of storm properties for use in analysis.

In convective situations, the forecasting problem encompasses storm initiation, evaluation, and movement. For storm initiation, progress has been made in the use of data from sensitive Doppler radars to detect those boundary-layer features that are important for the forecast. One of the principals concerns in cloud radar remote sensing is the need for an objective procedure for the detection of existing storms and extrapolating their evolution and movement. TITAN deals with the development of such procedure.

The experimental unit is defined here as a contiguous region, all of which exhibits reflectivities above a given threshold (T_z), and the volume of which exceeds a threshold (T_v). Clearly, the value of T_z determines the type of storm that will be identified. Some possibilities are:

Individual convective cells, T_z = 40-50 dBZ,
Convective storms, T_z = 30-40 dBZ,
Mesoscale convective complexes, T_z = 25-30 dBZ,
Snow bands, T_z = 15-25 dBZ.

For the study of the seeding experiments and its quantification, the experimental unit is the summertime convective clouds. Based on the results from Dixon et al. (1993) T_z was set to 30 dBZ for the study of the seeded and non-seeded cases. The use of the volume threshold Tv is necessary to prevent the tracking of noise or small regions of ground clutter and to keep the number of identified storms within reasonable limits. For this study, T_v was set to 50 km³.

For the purposes of the nowcasting experiment and storm analysis, various storm properties were computed with TITAN:

Centroid for whole storm and for each plane;
Reflectivity-weighted centroid for whole storm and for each plane;
Top, Base and Volume of the seeded cloud;
Area for each plane, and mean area;
Mass of precipitation for whole storm and for each plane (based on a Z-R relationship);
Precipitation flux (based on a Z-R relationship);
Height of maximum reflectivity.
Position, size and shape of precipitation region;
Histogram of reflectivity as a function of volume and as a function of area.

Many of this data where used as forecast parameters.

The results from the South African experiment point to increases in certain response variables in the period from 20 to 50 minutes after decision time. Therefore, time series of the following quantities were considered as response variables:
 

Radar-estimated precipitation flux.
Total storm mass, where storm is measured in kilotons.
Storm mass (computed as for item 2) above 6 km MSL.
Storm area.
Height of maximum reflectivity - Z-weighted vertical centroid.
 
 

Quite more frequently two or more convective storms merged to form a single storm, and some what less frequently a single storm will split into two or more storms, then a new track is created combining the properties of both storms; Fig. 2 show with real data how TITAN handle a complex matrix of storms that merge together, (6 storms) to form a large rain area. Is necessary to enhance the tracking scheme to handle these situations correctly.
 

Fig. 2 Storm merger

The spliting situation is treated similarly in Fig. 3, The storm at the beginning shows one area of heavy rainfall (»50 dBZ) and after the next volume scan, there are present, two areas of heavy rainfall, which TITAN consider as two separate storms, and following with the development of the storm, the two rain areas are completely separated after a few volume scans.

Figure 4 through Fig. 10 present some examples of TITAN display and how are they used in the evaluation of seeded and unseeded cases. Fig 4 shows the forecast at 30 minutes for a storm developed close to the field project site. Fig. 5 shows the past of the storm at 30 minutes. All this parameters are used to predict the evolution of the storm in a seeded experiment, to keep a track of how the storm is developing over time, and also to decide if the convective cell is a good candidate for an experiment. This way the lifetime of the storm is covered.
 

Fig. 3 Storm split.

Figure 6 is an example of time-height plot, which are in common use in meteorology to depict the history of a height-distributed variable (such as wind speed) over time. In this case the graphic is providing information on the vertical structure of the storm throughout its lifetime.
 
 

Fig. 4 Storm track in the vicinity of Monclova 22:22:43 UTC July 20 1998. The current storm position is represented by the heavy solid blue polygon, the forecast to 30 minutes is shown as the red polygon.	Fig. 5 Same storm from fig 4, the thin yellow polygon represents the past 30 minutes of life of the storm. The heavy red lines represent the orography of the area.

Fig. 6 Time-height series of maximum reflectivity for the storm in figure 4 and 5. The vertical and reflectivity-weighted centroid (Z centr) is shown, as well as the height of the maximum reflectivity at each time.	Fig. 7 Time series of the reflectivity distribution in volume for the storm in figure 4 and 5. This details the intensity of the storm. It is a cumulative histogram in which the color represents the percent of the volume which for which the reflectivity exceeds the respective value.

Figure 7 shows the time series of the reflectivity distribution in volume.
 

Figure 8, shows how TITAN displays the aircraft position. This very important when working on a seeding experiment, this way is easy to identify the aircraft position, and the cell that is been treated. Aircraft position is shown here as a white line, and at the end of the line, TITAN displays an arrow showing the direction of the aircraft at that period of time.
 

Fig. 8. Aircraft Position.

Figure 9, shows a vertical and horizontal cross-sections, this is one of the features of TITAN that is very important for this seeding experiment. Once a case is call, an analysis of the reflectivity in both vertical and horizontal coordinates is done.
 

Fig. 9. Vertical and Horizontal cross-sections of a seeded storm.

Using this cross-sections we are able to detect the updraft region of the cloud that is been treated, this way the manager or the person working with the radar can serve as a guide for the pilot of the seeding aircraft, he can guide the pilot to the zone of updraft so the seeding material will be placed in the correct place. The presence of a downdraft in the cloud can be taken as the updraft region so this analysis has to be confirmed by both the pilot and the radar operator. On figure 9 the white line represents the area of the horizontal cross-section and the reflectivity graph shows the possible presence of the updraft area, in this experiment the seeding material was placed close to the area of highest horizontal reflectivity; the track of the plane, show the exact position where the seeded material was placed.
 
 

Fig. 10 Time series for the storm in figure 4 and 5. (Time series of LWC (kg/m2), storm precipitation flux (m3/s), storm volume (km3), storm precipitation area (km2), storm-estimated storm mass (kilotons)).

Figure 10 shows the parameters that TITAN retrieves for the quantification of the seeded experiments, variables like storm precipitation area, are used to minimize the statistical error, when comparing two experiments, this is due to the variability of storm sizes and storm precipitation fluxes. The natural variability is very big, so is necessary to account for this differences when the final precipitation flux is reported.

The Z-R relationship of Z = 200 R^1.6 (Marshall-Palmer) was used because this allows for the significant evaporation which occurs between the cloud base and the ground, because Monclova’s is a region of very light precipitation and very high temperatures.

The data ingested in radar coordinates from the radar site using rdata_to_shmem. The data are transformed into Cartesian coordinates by dobson_from-shmem, which uses a lookup table and optionally the clutter table created during the system preparation phase.

TITAN make the precipitation forecast with the program precip_map, which uses the Cartesian volumes for the reflectivity data and the storm track files for storm position and trend information. Precip_map produces data in dobson format Cartesian files, which are sent to rview via a data server (dobson_server).

The experimental unit as defined before here, is the storm measured by the radar and tracked by TITAN, using a 30dBZ threshold, for a time period of 20 min prior to decision time to 60 min after decision time. If TITAN does not track the storm during this entire time period, then the period is defined as the time at first detection by TITAN to the time when TITAN no longer detects it. The TITAN tracking and analysis is fully automated to avoid the possibility of bias based on knowledge of the seed/no-seed decision. TITAN as described above produces time series of storm properties for use in the analysis. As suggested from the South African experiment, the time-series response variable include precipitation flux, total storm mass, storm mass above 6 km MSL, and the height of maximum reflectivity minus the Z-weighted vertical centroid. Total response variable include total precipitation using the Z-R relationship described before which is applied to a composite of maximum reflectivities at any height in the storm, which minimizes any range bias.
 
 

Fig. 11 Real time operations

The hypothesis to be tested for each time-series response variable is that the value of the variable is larger for seeded cases than for non-seeded cases during the time period 10 to 60 minutes after decision time. For variables that are total quantities (i.e. measured over the lifetime of the storm), the value of the response variable is hypothesized to be larger overall for seeded cases than for no-seeded cases.

The null hypotheses are for no differences in response variables for the seeded and non-seeded cases. The time-series variables will be tested using the first three quartiles of the distribution of values for seeded and non-seeded cases.
 
 

Fig. 12. Life time of a case (July, 20th 1998) The lifetime of the case, following the development of the cloud to a thunderstorm that suffers both mergers and splits. Is clear the augmentation of the reflectivity over time due to natural variability or the response of the storm to the seeding technique. The seeding experiment started on the 7th picture and ended on the 10th picture, the rest is just the following of the lifetime of the storm.

Preliminary results are shown in the statistical analysis section.