Large sample hydrology and the value of data

A key aim of large sample hydrology is to gain generalizable insights by comparing the functioning of hydrological systems across many locations and, by doing so, across many spatial gradients. To be informative, large sample datasets must therefore contain locations that cover the gradients of interest (e.g. in climate, topography, geology) and they must quantify these gradients in a meaningful way (e.g. as catchment attributes). Despite the increasing availability of open datasets with growing coverage and diversity, recent research has highlighted several limitations of the datasets currently used, which may compromise the insights gained in large sample studies. Here we will discuss three problems associated with large-sample hydrological data, as well as some possible consequences and solutions.

(1) Data uncertainty, which arises because we cannot exactly measure the variables of interest at the relevant scales.

(2) Data representativeness, i.e., the issue that our data may not represent the actual variables of interest because they are either measured indirectly, must be processed in some way, or contain subjective choices.

(3) Data imbalance, i.e., the issue that certain regions are omitted or disproportionately represented in our datasets, which may lead to biased results.

While identifying and addressing these issues is challenging, it will not only increase the value of large sample datasets, but ultimately also improve our understanding of hydrological processes and our predictive modeling capabilities.