Mountain disaster-causing storm identification method based on Doppler radar and deep learning

Abstract

The invention relates to a mountain disaster-causing storm identification method based on Doppler radar and deep learning. The method comprises the following steps: extracting heavy rainfall related features of Doppler radar data, establishing the mountain disaster-causing storm model feature database by combining multisource meteorological factors and the nationwide mountain torrents disaster research and appraisal result, and acquiring all disaster-causing storm features in a radar strong echo area by adopting a fuzzy matching method; and constructing a mountain disaster-causing storm identification model with the regional property based on the deep learning method, optimizing the model, evaluating a model identification result, thereby realizing full-automatic and efficient mountain disaster-causing storm identification.

Description

technical field [0001] The invention relates to the technical field of early warning and mountain torrent defense of disaster-causing rainstorms in mountainous areas, specifically a method for identifying disaster-causing torrential rains in mountainous areas based on Doppler radar and deep learning, which is mainly used for identifying short-term heavy rainfall in mountainous areas and torrential rain disasters in mountainous areas. Early warning and forecasting work. Background technique [0002] It is jointly affected by various factors such as special natural and geographical environment, extreme disastrous weather, and economic and social activities. Mountain torrent disasters have the characteristics of many points and wide areas, obvious seasonality and regionality, frequent bursts, rapid disaster formation, and strong destructive power. However, short-duration mountain torrential rain is the main factor causing mountain torrent disasters. Therefore, for mountain torr...

AI Technical Summary

Problems solved by technology

Using the physical parameter model to identify short-term heavy precipitation mainly relies on manual judgment, and fails to consider radar data and other meteorological elements

The short-term heavy precipitation identification method based on Doppler data is also limited to extracting a single echo feature from the radar basic reflectivity product, and the accuracy of the echo shape is not high, and it is all aimed at a certain feature ( Such as headwind area, monomer) to identify, and not to extract all the radar features related to heavy precipitation

[0006] However, there are many meteorological features related to short-term heavy precipitation, and the relationship between the physical quantity parameters used in weather forecasting is not clear. Using a single feature to judge is easy to cause misjudgment. It needs to be judged based on multi-source and multi-element features. Comprehensive use of radar features, physical features, vector features, and risk features requires regional matching of these features to ensure that all relevant feature parameters in the area can be obtained in strong echo areas. It needs to adopt the method of automatic identification and interpretation

[0007] Therefore, the current short-term heavy precipitation identification methods are not very applicable to the forecast of disastrous rainstorm in mountainous areas, and the application effect is poor. Therefore, in practical applications, it is difficult to provide assistance for early warning and decision-making of mountain torrent disasters

Images