Developing trustworthy hydrologic models
Abstract
Numerical models of watershed hydrology are used to assess impacts of climate change and land use, to manage Great Lakes water levels, and to inform weather, carbon cycling, and nutrient transport models. Unfortunately, the challenge of building operational models that get the right answers for the right reasons is a considerable one: we need to address issues of uncertainty, upscaling, information loss, parameterization, soft data, faulty conceptual models, and a legacy of poorly constructed computational codes pushed beyond their original design. Here, I will discuss some of these challenges, their potential solutions, and suggest some steps for moving forward. I will also discuss how these challenges have helped to motivate the development of Raven, a flexible hydrological modelling framework which can be deployed to investigate many of these issues and used to build models which are only as complex as can be justified by the available data.