| Abstract |
Overflow in the lower part of the Mekong River, which happens regularly during rainy
season, is beneficial to its riparian areas in terms of agriculture and fishery. Catastrophic
floods, however, can occur once in 5 to 6 years and cause damage to life and property.
Therefore, flood prevention is not quite a suitable measure but forecasting with early warning
is a must to mitigate inundated losses. We compared 2 flood forecasting methods,
Muskingum routing technique (MR) and Artificial Neural Networks (ANN), at Chiang Khan
as upstream monitoring site and the city of Nong Khai, Thailand, the downstream target site
which is 184 km apart. Daily flow data of the 2 stations from 2001 to 2010 were used in this
study, the data of the years 2001 to 2003 were used for calibrations and 4 pairs of
correspondent flood hydrographs each from 2004, 2006, 2008, and 2010 were used for
comparisons. Three indices, i.e. the coefficient of determination (R2), the Nash-Sutcliffe
efficiency index (EI), and the relative error (RE), were used for comparisons. The R2 for MR
and ANN are in the ranges of 0.88-0.96 and 0.85-0.96, respectively. The EI for MR and ANN
are -0.04-0.95 and 0.87- 0.96, respectively. The RE are 0.06-0.11 and 0.02-0.08 for MR and
ANN, respectively. The high values of R2 show that both methods are accurate in the sense of
random error. The negative and small EI values of MR method indicate its systematic error is
high especially when the flood discharge is low. The RE values for ANN are less than those
of MR method showing that the ANN is superior to the MR. |
Publication
Journal Publication
ไทย