The absence or failure of hydrometric data with long and reliable series, by factors of physical and / or economic order, is one of the main challenges faced in hydrological studies. In order to work around this problem, in this thesis, the application of regionalization method of flow duration curve was proposed. In this context, the main idea was to divide the region of the Amazon in homogeneous regions defined by the methods of hierarchical cluster analysis of Ward and diffuse Fuzzy C-Means, and for each, formulate regional models of flow duration curves. For both methods of the Euclidean distance cluster analysis was used as a similarity measure, and the explanatory variables the flow (drainage area, mean annual precipitation, length and slope of the river), as input data. We obtained four homogeneous regions through the Ward method and 14 regions by Fuzzy C-Means. Duration curves were constructed for each of the 214 gauged stations distributed in their respective regions, and calibrated according to 6 mathematical models (linear, power, exponential, logarithmic, quadratic and cubic). For each homogeneous region formed by cluster analysis methods, a regional model of flow rates of duration curves using multiple regression analysis was formulated, relating the parameters of the best model calibrated with the physical characteristics (drainage area, length and slope of the river) and climate (average annual precipitation) basins. The obtained regional models were validated by the method Jack-Knife cross validation. The performance indices found – values of NASH ≥ 0,75 in over 62% of cases, standing in the performance range from acceptable to good – showed that the Fuzzy C-Means method was the most suitable for the formation of homogeneous regions of flow. The regional models developed for each of the regions formed, are presented as a good option for modeling of flow duration curve for medium and small basins without flow data in the Amazon region.
|
