The influence of data transformations in simulating Total Suspended Solids using Bayesian inference

Abstract

Water Quality Models (WQMs) can be difficult to calibrate due to the complexity and heteroscedasticity of model errors. With the aim to address error heteroscedasticity and produce more accurate and reliable probabilistic predictions, this research develops a framework for improved uncertainty analysis and Bayesian inference of WQMs in the context of Total Suspended Solids (TSS) models. Along with error formulations applying data transformations in traditional time-space, we introduce an error model based on Flow Corrected Time, in which clock time is expanded during high values and compressed during low values. The results for four case study catchments show that proper data transformation can change the calibration emphasis on the parts of interest, which can modify the simulation reliability and improve the accuracy of critical condition judgments. Overall, this data transformation framework can help to select appropriate residual error models for water resource models and associated water quality predictions.