Interpretable Classification of the Contiguous United States River Catchments using Network Science Methods
Abstract
The classification of river catchments has been an active field of study for a few decades, and the recent surge in availability of hydrological and environmental data has enabled new approaches to be applied to this problem. Catchment classification can be performed with several methodologies, such as grouping by spatial proximity, physical similarity and hydrological signatures. In the context of the physical similarity approach we present a novel method for catchment classification based on network science applied to data on watershed traits, where the similarity among catchments is reflected by the edges of a network. Under this framework we can leverage the network topology to find clusters of catchments with similar physical traits. This choice is dictated by the capability of networks to capture non trivial patterns and collective behaviors, such as the ones that we show being present among the catchment traits. The method has been tested on over 9000 river catchments across the contiguous United States, each one accompanied by environmental attributes such as climate, geology, vegetation, and anthropogenic features such as land use, proximity to dams and developed areas. The resulting classification shows a big heterogeneity in terms of cluster size, which spans three orders of magnitude. In order to provide interpretability of the identified clusters we determine how distinctive each trait is in that specific group of catchments. We find that, along with classes defined by environmental traits, there are many others for which the dominant traits are related to anthropogenic factors. Additionally, we aggregated various hydrological signatures (e.g., streamflow and temperature statistics) of catchments in each cluster and found that different groups show different hydrological characteristics. This, along with the information on dominant traits, allows us to establish a connection between hydrological behavior and the physical traits of the classes of catchments. Finally, the current study can enable the prediction of hydrological variables of interest in unmonitored basins by providing interpretable sets of donor catchments at different granularities.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H11J..03C