2B.1

**Multi-frequency radar estimation of cloud and precipitation properties using an artificial neural network**

Andrew L. Pazmany, University of Massachusetts, Amherst, MA; and J. B. Mead, S. M. Sekelsky, D. J. McLaughlin, and H. B. Bluestein

The University of Massachusetts (UMass) and the University of Oklahoma (OU) have been developing an artificial neural network algorithm for the estimation of liquid water content (LWC) and drop size in liquid clouds and precipitation from multi-frequency profiles of radar reflectivity. The problem of extracting cloud parameters from the measured range profiles of backscattered power is a good example of a problem without well-defined rules for estimation. The forward problem is straightforward: for a given drop-size distribution, the radar observed (attenuated) reflectivity, can easily be calculated using Mie scattering formulas. Also, cloud and precipitation properties, such as LWC, rain rate or drop size can be directly calculated from drop-size distribution. Solving the inverse problem, that is, calculating cloud parameters from measured reflectivity profiles, is very difficult, in part due to the non-linearity of the forward problem. Artificial neural networks are ideal for solving problems where the forward problem is well characterized but the inverse problem is complex.

A neural network was trained using a simulated data set of cloud parameter and multi-frequency radar reflectivity profiles, based on the modified Gamma drop size distribution. The network estimated LWC, mean volume diameter (MeVD) and mean Z diameter (MeZD) in the middle three volume cells from five range gate profiles of radar observed (attenuated) radar equivalent reflectivity factors. The drop size parameters MeVD and MeZd were defined, similarly to the commonly used median volume diameter (MVD), as the diameter that corresponds to the mean volume and that corresponds to the mean radar reflectivity factor respectively.

The algorithm was tested with multi-frequency radar reflectivity data collected during the Mount Washington (New Hampshire) Sensors Project (MWISP) in March and April of 1999. MWISP was a multi-investigator experiment with participants from UMass, Quadrant Engineering, NOAA Environmental Technology Laboratory (NOAA/ETL), and others. Radar systems from UMass and NOAA/ETL were used to measure X, Ka and W-band backscatter data from the base of Mt. Washington, while simultaneous in-situ particle measurements were made using ballone sounds, from aircrafts and from the observatory at the summit. Figure 1 shows the neural net estimated (*) and the in situ ATEK probe derived (solid line) LWC, agreeing in altitude to within a few hundred meters and in magnitude to an error of less than 20%. This paper will describe the measurement concept, the neural net training model and present estimated LWC and drop size images of liquid clouds and precipitation.

Session 2B, Algorithms—Microphysical Retrieval & Particle Typing I (Parallel with Session 2A)

**Thursday, 19 July 2001, 4:00 PM-6:00 PM**** Next paper
**