Substantial progress has been made in the last several decades for quantification and communication of hydrologic uncertainty (uncertainty quantification here is in a broad sense, including parameter estimation, sensitivity analysis, uncertainty propagation, and experimental data and data-worth analysis for uncertainty reduction). For example, due to development of public-domain software (e.g., PEST, UCODE, and DREAM), uncertainty quantification using regression and Bayesian methods has become a common practice not only in academic but also in consulting industry and govern agencies. However, the hydrologic uncertainty community are still facing several grand challenges that have not been fully resolved, and new challenges emerge due to changing hydrology and environmental conditions. Below are three grand challenges that may be addressed in the coming decade:
- Software development: Software for supporting decision-making and for communicating uncertainty with decision-makers and stakeholders is still needed. Given that there have been a number of software in the public domain, it appears to be necessary to launch an effort for community software development such as building libraries of uncertainty quantification, visualization, and communication. An effort is also needed to closely collaborate with software developers of physical models, so that uncertainty quantification can be built as a module of the modeling software for efficient and effective operations.
- Information and knowledge extraction from data: While new technologies of data across multiple scales collection are always needed, it is of tantamount importance to develop methodologies that can extract information and knowledge from data. This includes identification of new and overlooked data needs (e.g., water management data such as water use), revisit of existing data (e.g., those collected by NASA or NOAA but have not been analyzed), and development of machine learning and deep learning methods suitable to hydrologic research. Machine learning is a hot topic in many research fields, and its value to hydrologic uncertainty quantification (especially on reducing model structure uncertainty) has not been intensively explored.
- A particular need: Computationally efficient algorithms: Uncertainty quantification nowadays mainly relies on Monte Carlo approaches, which is computationally expensive particularly for new models that are more complex than models several decades ago. Computationally efficient algorithms (e.g., parallel computing and surrogate modeling) will enable us to conduct more comprehensive and accurate uncertainty quantification. The effort of algorithm development requires close collaboration with scientists in other disciplinaries such as applied mathematics, statistics, and computational science.
- We feel that the research field of hydrologic uncertainty is in its transition stage in two sense. First, substantial progress has been made in the past but we need to finish the last mile. For example, we have developed many methods for uncertainty quantification, but need to work on efficient and effective communication of uncertainty to decision-makers and stakeholders. In addition, we are facing new challenges of developing more advanced methodologies to make a full use of existing data and emerging computational hardware and algorithms.