The Artificial Intelligence Identification and Filtering Methods for Non-meteorological Radar Echo in Ningbo

The products of Doppler weather Surveillance Radar are occasionally contaminated by non-meteorological echo. For example, the accuracy and reliability of the precipitation-estimating products are reduced significantly by the clutter contamination. For the purpose of improving the accuracy of the products, the clutter removalwork should be firstly done before further developing new radar products. Statistical analysis of non-meteorological radar echo in Ningbo from 2003 to 2006 shows that these clutters mainly are normal propagation ground clutters (NPGC) and abnormal propagation ground clutters (APGC) . The radar-derived products are contaminated by the two types mostly. The following characteristics are shown: they are mostly common in low elevations, discrete distribution of reflectivity products, radial velocity values and spectrum widths values in clutter around zero occasionally mixing with high values and significant horizontal and vertical gradient. Based on echo characteristics and the traditional experience on clutter identification, a stationary safe multidimensional linear approximation algorithm simulating manual eyes' fuzzy logic is designed, which takes full account of the relationship between computation efficiency and data quality. The following three rules should be obeyed by clutter removal. The first is conservation, which is to avoid removing precipitation echoes improperly. The second is independency, that is different clutter is identified and filtered respectively. The third is diversity, that is to utilize three moments of R, V, W of more than two elevations (now use 0.5°and 1.5°) as many as possible and other real time data. The flow chart of clutter removal quality control is given and key points are as below: to build echo background to classify the echoes such as clear or light rain, moderate rain, heavy rain, the edge of strong echoes, strong echoes and strong clutters and so on, identification of isolated clutters, clutter filtering based on the horizontal and vertical reflectivity gradient, identification of small region where the reflectivity values have a sharp change. 190 cases have been tested with the techniques under weather conditions including clear sky, showers and deep convections. Results show that it works well, and has good performance to remove the nonmeteorological radar echoes completely and accurately within the range of 150 km. But out of the range, it doesn't operate well while the radial velocity is fussy. Because of the great horizontal and vertical gradient in small scale deep convective systems, for the sake of conservation, not only a very strict threshold is set but also the removal work is given up to protect the real precipitation echo where fussy radial velocity exists. The cases in the article prove that it doesn't cause wrong removal in small-scale deep convections. If satellite data, real time surface and sounding data can be combined together with radar base data, and time continuity of radar data is taken into account as well, the technique may probably be applied up to a valid range of 200 km even further away in the future.