Regional flood frequency analysis for the island of Newfoundland, Canada using L-Moments

The regional flood frequency analysis (RFFA) for the Island of Newfoundland carried out by the provincial government of Newfoundland and Labrador in 1989 was revisited using the index-flood method based on L-moments. L-moment-based homogeneity tests showed that two of the 1989 regions were possibly redundant. The Water Survey of Canada (WSC) sub regions Y and Z were found to be statistically as well as operationally homogeneous. The conventional goodness-of-fit tests, including the L-moment-based tests were not particularly powerful in discriminating between the fits of generalized extreme value (GEV) and the three-parameter log-normal (LN3) distributions to the regional data from the 1989 regions as well as from the WSC sub regions. However, the robustness evaluation based on Monte Carlo simulation revealed that LN3 was comparatively more robust than the competing GEV distribution. A comparison between the return period flows estimated based on the 1989 regions and those based on the WSC sub regions showed that the estimates based on the WSC sub regions had, in general, equal or better accuracy than the estimates obtained with the 1989 regions. Likewise, the estimates based on the index-flood method using L-moments were found to be more accurate than those based on regression-on-quantile approach of the 1989 study. A similar comparison with the quantile estimates obtained from the more recent RFFA study for the Island of Newfoundland carried out by the provincial government also showed that the index-flood based on L-moment approach produced more accurate estimates than its regression-on-quantile counterpart. The L-moments based LN3 growth factors for the WSC sub regions and nonlinear regional models for index-flood estimation are recommended for carrying out the RFFA on the Island of Newfoundland