Effect of Geological Feature on Regional Flood Frequency of Watersheds of North-East of Iran

Finding out homogeneous watersheds based on their flood potential mechanisms, is needed for conducting regional flood frequency analysis. Similarity of watersheds based on flood potential severity depends on many factors such as physiographic and meteorological features of the watershed, geographical location and geological features. These criteria although are sound ones, they suffer from this concept that there is no attention to hydrological losses of runoff into the soil. As a result, current literature lacks for considering geological features into delineating homogeneous regions. The primary contribution of this paper is to include one geological criterion on flood regionalization. In a previous study we made a homogeneous classification for Khorasan Province of Iran without taking into consideration of infiltration features of the region. So, by taking geological features there may provide a sound comparison to regionalization issue.
To find out the effect of geological feature on delineation of homogeneous regions, 73 hydrometric stations at North-East of Iran with arid and semi-arid climate covering an average of 29 years of record length were considered. Initially, all data were normalized. Watersheds were clustered in homogeneous regions adopting Fuzzy c-mean algorithm and two different scenarios, considering and not considering a criterion for geological feature. Three validation criteria for fuzzy clustering, Kwon, Xie-Beni, and Fukuyama-Sugeno, were used to learn the optimum cluster numbers. Homogeneity approval was done based on linear moment’s algorithm for both methods. We adopted 4 common distributions of three parameter log-Normal, generalized Pareto, generalized extreme value, and generalized logistic. Index flood was correlated to physiographic and geographic data for all regions separately. To model index flood, we considered different parameters of geographical and physiological features of all watersheds. These features should be easily-determined, as far as practical issues are concerned. Cumulative distribution functions for all regions were chosen through goodness of fit tests of Z and Kolmogorov-Smirnov.
Watersheds were clustered to 6 homogenous regions adopting Fuzzy c-mean algorithm, in which fuzziness parameter was 1.9, under the two different scenarios, considering and not considering a criterion for geological feature. Homogeneity was approved based on linear moment’s algorithm for both methods, although one discordant station with the lowest data was found. For the case with inclusion of genealogic feature, 3-parameter lognormal distribution was selected for all regions, which is a highly practical result. On the other hand, for not considering this feature there were no unique distribution for all regions, which fails for practical usages. As far as index flood estimation is concerned, a logarithmic model with 4 variables of average watershed slope, average altitude, watershed area, and the longest river of the watershed was found the best predicting equation to model average flood discharge. Determination coefficient for one of the regions was low. For this region, however, we merged this region to other regions so that reasonable determination coefficient was found; the resulting equation was used only for that specific region, however. By comparing the distributions of stations and also two evaluation statistics of median relative error and predicted discharge to estimated discharge ration corresponding to 5 different return periods (5, 10, 20, 50, and 100 years). Both perspectives showed acceptable results, and including geological feature was effective for flood frequency studies. With considering the geological feature for regionalization, Besides, Log normal 3 parameters distribution was found appropriate for all of the regions. From this point of view, geological feature was useful. Median of relative error was lower for small return periods and gradually increased as return period was increased. Median of relative error was between 0.21 to 00.45 percentages for the first method, while for the second method it varied between 0.21 to 0.49 percentages. These errors are quite smaller than those reported in literature under the same climatic region of arid and semi-arid. The probable reason may due to the fact that we made a satisfactory regionalization via fuzzy logic algorithm., We considered another mathematical criterion of “predicted discharge to the observed discharge”. The optimum range for this criterion is between 0.5 and 2. While under-estimation and over-estimation are found if this criterion is lower than 0.5 and higher than 2, respectively. Based on this premise, 75 to 95 percentages of stations were categorized as good estimation under the first method of analysis. On the other hand, 78 to 97 percentages of stations were considered good for the second approach.