Optimum quantization of remotely sensed rainfall data

Quantization is an approximation of a signal value by a whole multiple of an elementary quantity. Unlike the sampling process (providing the Nyquist frequency is above the maximum process frequency) this results in an irretrievable loss of information since it is impossible to reconstitute the original analogue signal from its quantized version. Hence there is a tendency to use higher resolution digitizing cards to convert data from analogue to digital signals. However, it is important to notice that despite extensive research in remote sensing technology, many factors exist that influence the accuracy of the measured data. This paper illustrates that although longer Quantization length could reduce the overall errors caused by Quantization, this is only true when the data to be converted are error-free. If there are uncertainties in the data, a longer Quantization length may increase the overall error and reduce the data quality. A simple case study shows that 3-bit data are more accurate than 8-bit data. A demonstration of random Gaussian data series is used to illustrate that high resolution representation of the data series may produce less desirable results.