Modelling urban floods in megacities: a comparative bibliometric review of traditional physically based and artificial intelligence models

Traditionally, flood models are based on physical and hydrodynamic processes, which require massive data. However, this information is not always easily accessible. As an alternative, the development and application of data-driven models, such as Artificial Intelligence (AI) techniques, have grown. These have the advantage of requiring less parameterization than traditional models. However, some of those models demand long time series data for model training. We present comparison of traditional and AI approaches based on a bibliometric review.  We used the Scopus database, selecting papers from 1990 to 2020, divided into two groups: physical models, with thekeywords: GIS-based model, Hydrodynamic model, HEC-RAS and PCSWMM, and AI models, with the keywords: Artificial Intelligence, Machine learning, and Neural. To restrict the search, the following keywords were also inserted: Urban flood, Urban rainstorm flood, Flood mapping, Inundation mapping, Flood hazard, Flood, Pluvial, and Fluvial. Table 1 shows the results for each search string. The main metrics evaluated were the number of papers per year (Figure 1) and co-citation networks (Figure 2). For physical models, the main keyword found was “HEC-RAS”, with twice as many occurrences compared to the second most found “Hydrodynamic model”, demonstrating its wide application. The United States has the highest production of papers, followed by China and the United Kingdom. However, the United Kingdom has a more central role when evaluating the co-citation network. P. D. Bates (Bristol University, UK) stands out regarding the authors with the highest citation. For the AI models, the results for the keyword "Machine Learning" and the keywords "Artificial Neural Network" and "Neural Network" together have an equal count. China and the United States also presented the highest production. However, in the co-citation network, Iran has greater centrality and citations. The main authors include B. Pradhan (University of Sydney, Australia) and K. Khosravi (Sari Agricultural Science and Natural Resources University, Iran). In Figure 3, we present word clouds for each category. This review identifies the most relevant methods in flood modeling, which will be used to select solutions for comparative case studies in megacities.