02971nas a2200253 a 450000500170000000800410001703500150005810000210007324500810009426000670017530000220024249000270026450001420029150200580043352018240049165000500231570000380236570000490240370000430245270000530249571000440254881000590259285600660265120260817162834.0110314s2009    th uu|m rtt   0|  a1eng d  a.b120922651 aBadgujar, Girish10aApplication of artificial neural networks for rainfall forecasting in Mumbai  aPathum Thani, Thailand :bAsian Institute of Technology,c2009  a66 leaves :bill.1 aThesis ;vno. WM-08-05  aA thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Water Engineering and Management  aThesis (M.Eng.) - Asian Institute of Technology, 2009  aAccurate forecast of both spatial and temporal distribution of rainfall is necessary for flood management in urban areas. Mumbai is one of the metropolitan cities in India. It receives high  intensity  rainfall  during  four  monsoon  months.  High  intensity  rainfall  and  poor drainage  system causes  flooding  in  Mumbai almost  every  year.  In  the  present  study  three different  short  term  rainfall  forecasting  models  were  developed,  using  ANN,  to  forecast rainfall from 1 to 4-hr ahead at Santa Cruz station in Mumbai. The inputs for the first model, Model A, were rainfall time series at Santa Cruz station, The inputs   for   Model   B   were   rainfall   and   other   meteorological   namely,   temperature, atmospheric  pressure,  relative  humidity  and  wind  speed  of  Santa  Cruz  station.  The  inputs for   Model   C   were   selected   using   Mutual   Information technique   from   rainfall   and meteorological   variables   of   Santa   Cruz   station   and surrounding   Colaba   station.  Continuous  time  series  data  from  June  to  September was  used  for  model  development.  2 and  1  year  data  was  used  for  training  and  cross  validation  while  1  year  data  used  for testing. Normalized Mean Square Error (NMSE) was used to evaluate the performance of the developed models. For  1-h  forecast,  the  least  NMSE  (0.65)  was  observed  in  Model  C.  For  2,  3  and  4-h forecasts  model  B  gives  lower  NMSE.  Model  A  always produced  high  NMSE.  It  was found that MI technique successfully identified input variables for 1-h forecasts. Moreover, the rainfall at Santa Cruz station cannot be forecasted by only using historical time series of rainfall data. Higher lead period forecasts were found to be dependent on target station input variables. 0aRain and rainfallzIndiazMumbaixForecasting1 aBabel, Mukand Singh,eChairperson1 aClemente, Roberto S.,eExamination Committee0 aSutat Weesakul,eExamination committee0 aTawatchai Tingsanchali, eExamination committee 2 aThailand (HM King), eScholarship donor2 aAsian Institute of Technology.tThesis ;vno. WM-08-05  3Full-Textuhttp://203.159.5.9/ait-thesis/detail.php?q=B04300