A New Lumped Descriptor of Basin-Wide Hydrological Connectivity

In their efforts to study the rainfall-runoff conversion process, hydrologists have deployed a variety of approaches. Despite the huge range of methodologies, a general theme can be identified: there is a trade-off between how generalizable a model can be across different basins and the degree of detail in basin characterization. On one hand, regionalization approaches and deep-learning models use lumped information, typically covering some combination of average geometric, topographic, land-cover, and climatic characteristics of a basin. Based on these descriptors, some general, typically empirical relationship is derived to explain the hydrological response of any watershed within a homogeneous region, e.g., by fitting a regional equation to predict the 10-yr flood at ungauged locations, or by developing regional statistical models on the pooled, standardized data from all the hydrologically similar basins. On the other hand, distributed, physically-based models attempt to simulate the water exchanges occurring within a catchment at different spatial and time scales, at the cost of a detailed, spatially-explicit basin characterization, with the resulting lack of transferability to other watersheds.

While a lumped characterization of basins is crucial for a variety of approaches aimed at model transferability, such as regionalization techniques for flood prediction or deep learning models for flood forecasting, most procedures only consider basin-averaged properties or at most their distribution. Thus, they are unable to account for hydrological connectivity, even though it is well known that it has strong effects on a watershed’s response. For example, the percentage of impervious area is often used as a proxy for the level of urbanization in catchments, but it cannot provide any information about how urbanized areas are located with respect to each other and the watershed outlet, although different spatial configurations of these may result in different hydrological behaviors, for the same precipitation input.

We propose a new, lumped hydrological connectivity index that can incorporate information on how different parts of a basin, with their various topographic and land-use characteristics, are connected to each other and the stream network. In this way, we incorporate their relative contributions to the hydrologic response of the watershed, depending on their location. This index can be regarded as a condensed measure of the potential that each location has for generating runoff at the watershed outlet, given spatially-explicit characterizations of its properties. It can be used in synergy with other lumped descriptors to provide a more detailed basin characterization that reflects hydrological connectivity.

We test the predictive power of the proposed index in the framework of regional flood frequency analysis finding that it benefits well-established approaches for hydrological prediction in ungauged basins.