Characterizing annual flood patterns variation using information and complexity indices

Abstract

In recent decades, the frequency of torrential rain coupled with various complex flood patterns increased. Since floods are changeable and unpredictable, deeper understanding of flood incidents is necessary for better watershed management. The primary objective of this research is to investigate and characterize annual flood patterns embedded in a river catchment located in west of Japan. To fulfill this aim, we proposed a method based on information and complexity principles to examine the annual variation of flood patterns embedded in a riverine system. The key strength of the proposed approach is being established on two fundamental pillars: (1) word pattern; gives an image about the detected flood patterns, i.e., simple flood occurs within one day or severe flood persists for two days or more, and (2) information-complexity indices that report the frequency and randomness of the detected patterns. Information content was quantified using Mean Information Gain (MIG), whereas Effective Measure Complexity (EMC) and Fluctuation Complexity (FC) were indices used to define the complexity in the studied records. The results show that the proposed method is very powerful in detecting hidden patterns. Furthermore, we succeed in capturing the stations that exhibited the same flood patterns. The main finding of the first pillar of this approach revealed that flood events were fundamentally triggered by precipitation occurred during East Asian monsoon and tropical cyclones. Alternatively, the information-complexity pillar was a powerful tool in capturing different internal structures of flood patterns. Hence, higher MIG values indicated higher degree of randomness. On the other hand, higher EMC values reflected the length of flood events, while higher FC values showed higher number of separated flood events. Overall, this study discusses the competence of a new approach capable to capture hidden patterns in dataset and can be extended to numerous applications that investigate system behavior.