Improving flood estimation in ungauged catchments
Visualitza/Obre
Estadístiques de LA Referencia / Recolecta
Inclou dades d'ús des de 2022
Cita com:
hdl:2117/363263
Correu electrònic de l'autorihtshamzafar4gmail.com
Tutor / directorRico-Ramirez, Miguel Angel
Realitzat a/ambUniversity of Bristol
Tipus de documentTreball Final de Grau
Data2021-05-28
Condicions d'accésAccés obert
Llevat que s'hi indiqui el contrari, els
continguts d'aquesta obra estan subjectes a la llicència de Creative Commons
:
Reconeixement 3.0 Espanya
Abstract
For the ungauged sites, in order to link the index flood to catchment descriptors, statistical models such as multiple regression are most widely used. In this paper, the Flood Estimation Handbook (FEH) statistical methodologies are analysed and hence recalibrated using all data available as well as cross validation process, the availability of more flow records produced better results. Similarly, using correlation analysis, appropriate catchment descriptors are selected, and a new simple non-linear regression model is proposed. The use of Artificial Neural Networks (ANNs) is investigated to estimate the flood index from the catchment descriptors and are compared to traditional non-linear regression models. The National River Flow Archive (NRFA) data has been utilized to estimate the index flood (QMED) for 337 ungauged catchments in the UK. The results showed that i) there are several catchment descriptors (e.g. catchment area, mean annual rainfall) that are directly correlated to QMED; ii) the recalibrated FEH models produce better results than the original models and so it is important to recalibrate this model as new data becomes available; iii) the proposed non-linear power-law model produces slightly better results than the FEH model; iv) the QMED estimates obtained from ANNs have shown improved performance as compared to the traditional non-linear regression models. In addition to that, given the fact that the number of catchments is not large enough to separate the data set in calibration and validation, we used cross validation (where each model is trained for n-1 data points, and then tested against the remaining 1 unseen data point demonstrating the real performance of model), to test the true performance of the models. The cross-validation results showed that the performance of the models decreases, but the ANNs still produces slightly better results compared to the nonlinear regression models.
TitulacióGRAU EN ENGINYERIA CIVIL (Pla 2017)
Fitxers | Descripció | Mida | Format | Visualitza |
---|---|---|---|---|
2021RP080_paper.pdf | 975,6Kb | Visualitza/Obre |