Quantifying a LiDAR Return Intensity Signature Intrinsic to Water Under Canopy, and Leveraging it to Map Occluded Waterways

Mapping headwater streams and understanding their temporal dynamics can provide insight into watershed ecological connectivity, inform management or hydraulic design decisions, and shed light on long-term trends in the face of a changing climate. Despite making up nearly 75% of streams in the U.S. and being the most abundant streams within a stream network, their geographic extent is poorly understood, primarily due to mapping limitations. Remote sensing of these headwater streams with optical imagery can be challenging, becoming impossible under dense canopy cover. LiDAR (Light Detection and Ranging) presents itself as an ideal tool to delineate hidden wet channels occluded from aerial view by canopy cover. The capacity of LiDAR to map drainage networks under canopy has been well-documented, as has its ability to identify water bodies from a low return-intensity signature. Here, we quantify a decrease in return-intensity intrinsic to a water body under canopy cover with a GIS analysis. Ten study areas were selected: five in central Colorado, and five in Yellowstone National Park. First, drainage networks and areas with canopy cover (defined as any 1 m pixel with vegetation height > 2 m) were delineated from spatial LiDAR data. Normalized values of return-intensity were compared for canopy-occluded areas between wet channels, dry channels, and areas outside of the drainage network, using the National Hydrography Dataset as reference. In all cases, a statistically significant reduction in return-intensity was observed in the wet-channel group relative to both dry channels and non-channels. Furthermore, we identified a key threshold, whereby wet channels under canopy cover never exhibited a normalized return-intensity value greater than 0.80. This threshold was leveraged in a mapping workflow, using an intensity-density raster to identify regions displaying a high frequency of pixels with normalized intensity < 0.8, resulting in a classified map product that accurately categorizes wet and dry channels under canopy cover. Mean accuracy was calculated at 81%, compared to an average false positive rate of just 12%. These results promote optimism towards enhanced mapping and a better understanding of temporal dynamics in forested watersheds.