Study on spatial-temporal feature extraction of rainstorm in urban areas

Under the influence of climate change and the continuous expansion of urban scale and other factors, the spatialtemporal dynamics of frequent extreme rainstorms in urban areas is becoming more and more obvious, but the commonly used expression methods such as point/area rainfall are difficult to reflect this spatial-temporal dynamics.With the advancement of rainfall observation technology, most cities have accumulated long-term rainfall observation data, which contains rich information on rainfall spatiotemporal processes, providing possibilities for predicting rainfall or designing the spatiotemporal distribution of rainfall.Based on the comprehensive use of machine learning, spatial analysis and other methods, this paper designs a set of technical processes for the extraction of spatial-temporal characteristics of rainstorms in urban areas, including data collection, standardized processing, rainstorm field division, spatial-temporal feature extraction, achievement expression methods and other links, and analyzes the processing steps and key problems of each link, Taking the rainstorm in Beijing as an example, the technical process is instantiated and verified.The temporal and spatial characteristics of the rainstorm in Beijing during 12 h, 24 h and 72 h were successfully extracted.The extraction results showed the temporal and spatial dynamic process of the rainstorm, and good matching with the actual rainfall processes.This indicates that the method can be used for extracting typical rainfall patterns in urban areas and for spatiotemporal distribution of rainfall processes.This method uses historical rainfall data to extract rainfall distribution templates, inputs the total rainfall (or forecast rainfall) of a rainfall event, and obtains the spatiotemporal distribution process of the rainfall event.The distribution results can provide more accurate rainfall input conditions for flood forecasting.