Search for: All records

Abstract A variety of classification approaches are used to facilitate understanding, prediction, monitoring, and the management of lakes. However, broad‐scale applicability of current approaches is limited by either the need for in situ lake data, incompatibilities among approaches, or a lack of empirical testing of approaches based on ex situ data. We developed a new geographic classification approach for 476,697 lakes ≥ 1 ha in the conterminous U.S. based on lake archetypes representing end members along gradients of multiple geographic features. We identified seven lake archetypes with distinct combinations of climate, hydrologic, geologic, topographic, and morphometric properties. Individual lakes were assigned weights for each of the seven archetypes such that groups of lakes with similar combinations of archetype weights tended to cluster spatially (although not strictly contiguous) and to have similar limnological properties (e.g., concentrations of nutrients, chlorophylla(Chla), and dissolved organic carbon). Further, archetype lake classification improved commonly measured limnological relationships (e.g., between nutrients and Chla) compared to a global model; a discrete archetype classification slightly outperformed an ecoregion classification; and considering lakes as continuous mixtures of archetypes in a more complex model further improved fit. Overall, archetype classification of US lakes as continuous mixtures of geographic features improved understanding and prediction of lake responses to limnological drivers and should help researchers and managers better characterize and forecast lake states and responses to environmental change. 

Abstract The LAGOS‐US RESERVOIR data module classifies all 137,465 lakes ≥ 4 ha in the conterminous U.S. into three categories using a machine learning predictive model based on visual interpretation of lake outlines and a lake shape classification rule. Natural Lakes (NLs) are defined as naturally formed, lacking large, flow‐altering structures; Reservoir Class A's (RSVR_A) are defined as lakes likely human‐made or human‐altered by a large water control structure; and Reservoir Class B's (RSVR_Bs) are lakes likely human‐made but are not connected to streams and have a shape rare in NLs. We trained machine learning models on 12,162 manually classified lakes to predict assignment as an NL or RSVR, then further classified RSVRs based on NHD Fcodes, isolation, and angularity. Our classification indicates that > 46% of lakes ≥ 4 ha in the conterminous U.S. are reservoir lakes. These data can be easily combined with other LAGOS‐US modules and U.S. national databases for the broad‐scale study of reservoir lakes and NLs.