Noise filtering helps to eliminate erroneous data. Errors may originate from various different phenomena, for example, from electrical spikes.
The filtering is made by calculating the median of a given number of consecutive samples. The median is the middlemost sample in the order of magnitude. This algorithm cuts all remarkably higher and lower measurement results compared with other measurement results near the measurement (in a given time window). The filtering window (the number of consecutive samples, where the median is made) is given to the ground equipment during ozone sounding preparation.
The median calculation algorithm is well-defined for odd values only: number of filtering window length (1, 3, 5, 7, ...) (number of measurement samples in the window). Therefore, the filtering window is defined in the ground equipment as the window radius; from the middle to the last sample, or, equally, from the middle to the first sample. In other words, the window radius indicates how many samples before and after the corresponding sample will be taken with in the filtering window. The number of samples where the median is calculated is as follows (sample radius = window radius):
Filter Window Length = 2 × Window Radius + 1
The sample radius is the value given for the ground equipment. If the windows radius is defined as an Integer, the filter window is always an odd integer and the median algorithm is well-defined.
The ozone sensor response time is typically about 20 seconds, as indicated in the GAW report 201 (see Performance Review Literature for details). To avoid cutting real ozone values of the measurement results, the median filtering window length must be clearly shorter than the sensor’s response time. Note that the length is defined as the amount of samples. Therefore, different values for radiosondes with different sample rates must be used. A typical value for the window radius for RS41 is 4 (filtering window = 9 seconds).
The median filtering algorithm is disabled by setting the window radius to 0.