0700, 1300, and 1900 hours Eastern Standard Time (EST); these hours correspond 

 to the times that the NWS creates daily synoptic weather maps. During storms, 

 hourly data recordings were made. Since the Modcomp/phoneline and NOVA-4 sys- 

 tems were automated, recording data during nonduty hours and on weekends and 

 holidays created only a minimum of problems. Prior to October, the FM re- 

 corder was run manually, and for most dates only two observations, one in the 

 morning and one in the afternoon during duty hours, were obtained. In gen- 

 eral, the FM recorder was not run on the weekends and holidays unless there 

 was a particularly interesting event in progress, such as a storm or experi- 

 ment. After October, a controller was used to turn the recorder on and off at 

 specified times; this automation permitted FM data collection in the evening 

 and on weekends . 

 Data tapes 



57. The wave data were recorded in digital form with the following 

 basic tape format: two records of header information which include (a) the 

 station identification number, (b) the date and time, and (c) a variable num- 

 ber of records necessary to obtain 20 minutes of data from all sensors at a 

 sample rate of four values per second. Each record contained 384 20-bit 

 integer words (i.e., binary format); each integer word represented the com- 

 puter units corresponding to the instantaneous voltage output of the sensor. 

 The above sequence of records was repeated for each recording interval until 

 the data tape was filled. Seven-track tapes were used for data recorded via 

 the Modcomp computer, while nine-track tapes were used with the NOVA-4. (The 

 20-bit word size is unusual but necessary because CERC processed the data on 



a CDC 6600 machine with a 60-bit word size; when necessary, CERC converted the 

 data tapes to an ASCII format). 

 Analysis/summarization procedures 



58. The CERC procedure for analyzing and summarizing digital wave data 

 was based on a Fast Fourier Transform (FFT) spectral analysis procedure. The 

 final results were also subjected to human editing and quality control before 

 public distribution (Thompson 1977; Harris 1974). The computer analysis 

 routine used 4096 data points (1024 seconds of data sampled four times per 

 second) for each data record processed. The program first edited the digital 

 data record, checking for nonnumeric characters, jumps, and spikes (i.e., 

 deviations greater than 2.5 and 5 standard deviations from the mean, respec- 

 tively). If more than five bad data points were found in a row or more than 



■ 26 ■ ^' 



