Troubles in the Paradise: Hydrology Does not Respond to Newtonian Mechanics and the Rise of Machines

There are two broad approaches to doing science: deduction and induction. While the primary objective of a deductive approach is the application of a theory to solve a problem, an inductive approach strives to develop a theory from data through generalization. Banking on Newtonian mechanics, scientific studies often prefer deductive approaches due to their perceived superiority, even though Newton himself highlighted the importance of observations. Here, we provide an overview of how hydrological science has largely progressed through inductive or observation-driven research. We also discuss recent advances in machine learning modeling and why superior performance achieved with these models cannot undermine the importance of understanding hydrological processes. In our opinion, the progress of hydrological science has been significantly hindered by the popular notion that each catchment is unique, necessitating a hydrological model to employ multiple free parameters representing its characteristics, particularly those related to soil. It is argued that efforts should be directed towards developing climate-centric models that require little or no calibration, focusing on similarities among catchments. In this regard, machine learning models have the potential to further improve our understanding of hydrological processes.