The development of an optimal design for a space-based Doppler lidar wind sounder (DWL) has been an iterative process as the science community provides better data and product generation algorithms to DWL system simulation/impact studies. Without a space-based heritage, the Doppler lidar programs have had to rely on ground-based and airborne experience to provide some sense of the reasonableness of model simulations. As part of this effort, Simpson Weather Associates developed a series of computer tools to process and analyze NASA/MSFC and NOAA ground-based Doppler lidar polypulse-pair and single shot data. The following describes the models and a series of ground-based experiments in system stability, accuracy versus SNR, winds from cirrus decks, single shot pairing and detailed error analyses.
Signal Processing Tools And Data Display Models
Validation of the LSM is a key priority for DWL studies. While GLOBE has improved the basis for modeled backscatter, the ground-based Doppler lidars at MSFC and NOAA have provided some checks on assumed atmospheric variability parameterization and signal processing performance.
Many of the expectations for the performance of a space-based Doppler lidar are based upon the experienced gained from ground-based (and to a lesser extent airborne) systems. NOAA and NASA have ground-based systems with pulse rates of 30-100 Hz and signal processing based upon the pulse-pair algorithm involving 3-50 shots per estimate. A space-based system will probably operate at < 20 Hz. Single laser shots can be accumulated to produce a LOS wind or each shot will be processed separately. SWA's experiments, therefore, have required that the single shot In-phase and Quadrature time series be recorded for post processing. Since the data volumes involved with single shot recording are much greater than the MSFC data system can handle, only 2 shots per record were recorded.
SWA has developed and refined a series of computer tools that are designed to process and analyze NASA/MSFC and NOAA ground-based Doppler lidar polypulse-pair and single shot data. The models are the Pre-processor models, FFT Model, Sine Fit Model, PPI/UV and Range Display Models, Shot Pairing Model and the SNR/Consensus Model.
The DWL ground-based pre-processor reads the raw DWL data files providing data record counts to reformat the data into either VAD files (DWL.VAD) for scanning angles or RANGE files (DWL.RAN) for fixed scan angles. For single shot data, the models correct for the in-phase and quadrature data for offset and gain. For the scanning data sets, the models create line-of-sight wind velocity files (DWL.PPI) for the PPI/UV Display Model and velocity/azimuth/signal (DWL.VAS) files for the sine fit model.
After the polypulse-pair and single shot data sets are reformatted, the in-phase and quadrature single shot data is run through a FFT model to produce line-of-sight wind (LOS) velocities. Throughout this effort, we have evaluated several FFT models (MathLIB, MathCAD, numerical recipes, etc.) for Doppler lidar data processing. It became evident that a FORTRAN or C source code FFT model not only provides more flexibility (such as applying unity amplitude windowing) over a commercial off the shelf product, but is also faster. One VADS' worth of lidar processing is approximately 7136 FFTs and takes ~3 minutes on a Pentium II 400 Mhz class PC.
In an attempt to establish "ground truth" winds using poly-pulse data and in order to compute U/V horizontal wind components for the PPI/UV displays, SWA developed two sine wave fitting models that work with incomplete and extremely noisy data sets. The first model is a Fortran coded "brute force" model that tries to best fit the data by varying amplitude, phase and vertical offset for many combinations. Data filtering based on signal strength is used to remove wild data points. This model goes through many iterations to arrive at a best fit and is computationally expensive. Also, the sine fits have to be visually examined to evaluate the quality of the fits and to assess whether the applied filter was too weak or too strong. The second model was developed in Visual Basic and allows the user to quickly move the best sine fit around the data by changing amplitude, phase and offset. These sine fits are saved in DWL.APO files.
Two Examples of the Visual Basic Sine Fit Model shows a sine fit to some single shot DWL data at range gate 9. The first example is unfiltered and has a very poor goodness of fit - 235.9. The second example applies a signal filter to the data to remove weak pieces of information. The sine fit quickly improved to 10.9.
The PPI/UV Display Model uses the DWL.PPI line-of-sight velocity data file from the pre-processor model to produce a Plan Position Indicator plot. The model also uses the APO file from the Sine Fit Model to compute a vertical distribution of horizontal wind components. The RANGE Display Model uses the DWL.RAN line-of-sight velocity data files from the pre-process or model to produce range verses elapsed time plots.
A space-based wind sounder will produce a pattern of shots that will provide generally bi-perspective samples within 100x100 km target areas. The angle between the two perspectives will vary depending upon the location of the target areas relative to the sub-satellite ground track. The wind estimate error will also vary with the geometry of the bi-perspective samples. The Ground-Based Shot Pairing Model processes the Doppler lidar data to yield shot combinations that are similar to those that will be achieved from space. Wind components are calculated for a range of bi-perspective shot pairs, which are compared to the true components "ground truth" derived from the sine fitted poly-pulsed pair VAD.
Ground-Based Doppler Lidar Laboratory Block Diagram
| This page managed by Sidney A. Wood | Last modified: 3 June 1999 |