On Interactive Adaptability of Hydrological Model and Input Data

Abstract

Hydrological models play a vital role in diverse applications like flood and drought prediction,
and water resources management. However, the accuracy of the two critical inputs,
precipitation (P) and potential evapotranspiration (PET) is often compromised by multiple
sources of errors. While these errors can be compensated for within the modelling process,
such compensation effects can lead to misinterpretations regarding the reliability of both the
input data and the resulting model performance. A thorough understanding of these
compensation effects remains a critical knowledge gap. This thesis focuses on three key aspects:
(1) the compensation of input errors from model calibration (error adaptability), (2) the mutual
compensation between biases in P and PET inputs (interacting effects), and (3) efficient
methods for identifying bias compensation between P and PET inputs. A large number of
datasets, catchments, and machine learning techniques are employed to provide a quantitative
understanding of the mechanism underlying these compensation effects.
Specifically, the key findings of this thesis are: (1) Hydrological model can often adapt to some
inaccurate P inputs showing good streamflow simulations. This adaptive ability is controlled
by an adaptable threshold of the overall P bias and how the event-based P biases shape the
overall P bias. (2) Biased P and PET inputs could compensate for each other to some extent in
reproducing satisfactory streamflow simulations. This mutual compensation is quantified as a
Compensational Interaction Angle (CIA) that corresponds to the bias ratio of P to PET. Drier
catchments have larger CIAs. (3) Empirical models are proposed to efficiently estimate the
CIA, requiring only the input of aridity index and runoff ratio values. These models eliminate
the need for extensive hydrological modelling experiments under numerous biased scenarios.
Furthermore, the thesis discusses overarching remarks and identifies future research needs in
hydrological modelling, particularly in light of the big data era characterised by the increased
availability of hydroclimatic datasets.
Overall, this thesis significantly contributes to our understanding of the three above-mentioned
aspects, highlighting the importance of recognising these compensation effects within the
hydrological modelling research community, in order to get the right answers for the right
reasons. Leveraging the quantitative compensation effects has the potential to improve the
reliability of hydrological modelling and robust quality control of P and PET data products.
Ultimately, this advancement can benefit stakeholders involved in decision-making processes
concerning flood forecasting and water resource management.

Date of Award	7 May 2024
Original language	English
Awarding Institution	University of Bristol
Supervisor	Dawei Han (Supervisor) & Miguel A Rico-Ramirez (Supervisor)