Joining interdisciplinary modeling and field-based methods to document riparian forests in eastern New York

Abstract

Riparian floodplain forests persist in a small fraction of their historical extent in the United States with estimated cumulative losses as high as 95% for some regions. Many remaining occurrences are also degraded due to changes to local flood dynamics, disturbance pressure from adjacent land use, and exotic species invasions. Yet these communities are disproportionally valuable for the area they occupy as they provide vital ecosystem services such as flood mitigation, erosion control, runoff interception, and wildlife habitat. To strengthen their protection and management, we present a novel approach for identifying riparian forests in eastern New York using low-complexity flood modeling and land cover analysis. We enlisted the Height Above Nearest Drainage method to compute ten-year floodplains for rivers and streams in the Mohawk River Watershed of eastern New York. We then extracted the forested portions of these floodplains using the National Land Cover Dataset Tree Canopy Cover. This process produced approximately 21,500 acres of predicted riparian forest spread across 1,063 occurrences. Our field verification surveys took us to 17 modeled locations where we successfully captured examples of riparian forests at 76% of sites and correctly predicted overbank flood occurrence, though not necessarily extent, at 88%. Our model also outperformed several other publicly available datasets in remotely identifying floodplains illustrating that this method shows promise for identifying community occurrences unrepresented in other datasets. In the field, we documented a diverse set of riparian forests with varied ecological condition and species composition. Our cluster analysis produced three compositional groups adding weight to ongoing efforts to formally recognize distinct riparian forest types in the Northeast. As predicted, our disturbance metrics were negatively correlated with floristic quality and percent native species. Yet contrary to our hypothesis, larger model occurrences typically had lower floristic quality and higher disturbance scores though this was the result of overestimated polygon extent in heavily modified areas rather than a true phenomenon. Our results demonstrate the power of blending remote and field methods while presenting an approach for the rapid and inexpensive identification of some of our most valuable and threatened natural communities.