Abstract:
Flooding is increasing in tropical regions, where millions of people are at risk, and
challenges exist in providing reliable predictions and warnings. This research
responds to this challenge by identifying and applying physics-based and data-based hydrological modelling approaches for large-scale flood modelling in lowland tropical regions. First, a distributed hydrological model was developed to accurately
represent catchment conditions and processes in the model. Second, empirical data
from nested catchments were analysed using statistical scaling relationships to
complement the accuracy of peak discharge estimates. Finally, the effects of
uncertainty propagation and interactions were quantified to increase the reliability of
model results.
The research was conducted in the Grijalva catchment area (57 958 km²)
southeast of Mexico. A large-scale model with a 2 x 2 km grid cell resolution was
developed using the SHETRAN hydrological model and run enforced with 3-hour
input rainfall data. Geostatistical techniques were used to quantify and reduce errors
in input data, and all diverted flows were accounted for to optimise simulations. For
the first time, the application of the Scaling theory of floods was applied in the study
area to improve the estimation of peak discharge. A Monte Carlo technique was
used to propagate and quantify rainfall and parameter uncertainties through a
coupled hydrologic and hydraulic model and into model results.
Although the model under-predicted the magnitude of peak discharge,
calibration results showed satisfactory model performance (NSCE = 0.72, CC =
0.74, Bias = –0.44% and RMSE 139.56 mm) and validation results were good
(NSCE = 0.56, CC = 0.60, Bias = –6.3% and RMSE 62.59 mm). A statistical log-log
relationship between intercepts (α) and peak discharge, from the smallest nested
catchment, was used to complement the simulation of peak discharge magnitudes.
It was observed that given rainfall uncertainties of ±71%, ranging from 63 to 73%;
the model generates discharge with uncertainties of ± 46%, ranging from 45 to 49%
and errors of ±46% ranging from 45 to 46%. The propagated uncertainties resulted
in flood inundation extents of ±4.34 km² varying from 1.66 to 7.02 km²
Thus, flood modelling in large tropical regions can be achieved by optimally
integrating several datasets with the best combination of the model parameter, input
and output datasets based on uncertainty and error quantification and removal
approaches.