Predicting rainfall runoff in Southern Nigeria using a fused hybrid deep learning ensemble

Abstract

Rainfall  as  an  environmental  feat  can  change  fast  and  yield  significant influence  in  downstream  hydrology  known  as  runoff  with  a  variety  of implications  such  as  erosion,  water  quality,  and  infrastructures.  These,  in turn impact the quality of life, sewage systems, agriculture, and tourism of a nation  to  mention  a  few.  It  chaotic,  complex,  and  dynamic  nature  has necessitated  studies  in  the  quest  for  future  direction  of  such  runoff  via prediction models. With little successes in use ofknowledge driven models, many  studies  have  now  turned  to  data-driven  models.  Dataset  is  retrieved from Metrological Center in Lagos, Nigeria for the period 1999-2019 for the Benin-Owena River Basin. Data is split: 70% for train and 30% for test. Our studyadapts  a  spatial-temporal  profile  hidden  Markov  trained  deep  neural network. Result yields a sensitivity of 0.9, specificity 0.19, accuracy of 0.74, and    improvement    rate    of    classification    of    0.12.    Other    ensembles underperformed  when  compared  to  proposed  model.  The  study  reveals annual  rainfall  is  an  effect  of  variation  cycle.  Models  will  help  simulate future floods and provide lead time warnings in flood management.