MAPPING THE URBAN AND OROGRAPHIC ENHANCEMENT OF PRECIPITATION IN BELGIUM WITH MACHINE LEARNING

Abstract

Precipitation plays a vital role in the global water cycle, impacting meteorology, climatology, and hydrology. The potential for precipitation to trigger water-related hazards gains significance in urbanised or steep terrain areas. In extreme weather scenarios, accurate precipitation nowcasts are essential for decision-making regarding processes such as transportation strategies and evacuation. Therefore, real-time nowcasting systems are crucial in mitigating risks during extreme weather events. However, challenges persist in accurately measuring and understanding precipitation distribution.
Current extrapolation techniques assume constant precipitation intensity, leading to inaccuracies because of the evolving precipitation patterns. Incorporating growth and decay dynamics is vital, as they can help overcome this assumption. Growth and decay, influenced by factors such as orography and urbanisation, have an impact on precipitation systems.
An artificial neural network (ANN) model, namely multi-layer perceptron (MLP), is trained on radar QPE data of five years from 2018 to 2022 in Belgium to investigate the
predictability of growth and decay patterns. The ANN model is used to learn the dependence of growth and decay on flow speed, flow direction, and time of the day. This thesis study aims to overcome the constant precipitation intensity assumption and identify the effects of topography and cities on the growth and decay of precipitation.
Results show that the ANN model generally underestimates the growth and decay patterns. Thus, even though the overestimation of precipitation intensity is avoided, because of the underestimation, it balances out the improvements and results in a similar error obtained by extrapolation-based nowcasts when compared to observed growth and decay. Therefore, only using the model proposed in this study is proven to be not enough for predicting growth and decay. However, this approach has potential and can be improved to achieve better results.

Date of Award	8 Sept 2023
Original language	English
Supervisor	Simon De Kock (Advisor), Lesley De Cruz (Promotor) & Boud Verbeiren (Co-promotor)