Estimation of watershed-level distributed forest structure metrics relevant to hydrologic modeling using LiDAR and Landsat

Abstract

A detailed characterization of vegetation structure is fundamental for physically-based hydrologic models to simulate various processes that determine rates of snow accumulation and ablation, evapotranspiration and water dynamics. However, major efforts focused on developing complex equations to describe hydrologic processes as a function of vegetation structure at the plot level have not been accompanied by corresponding attempts to adequately extrapolate these metrics over the wider landscape in order to parameterize fully-distributed models. Recent advances in remote sensing technologies offer alternatives to overcome these difficulties and therefore improve our capacity to monitor vegetation and hydrologic processes extensively. Airborne Laser Scanning (ALS) stands out as the most promising tool to provide detailed, 3-dimensional representations of vegetation from which a wide array of structural metrics can be estimated. On the other hand, moderate scale optical remote sensing imagery such as Landsat Thematic Mapper (TM) offers the capacity to extrapolate these metrics across the landscape by virtue of its spatial and temporal resolutions. Here we correlate ALS-derived forest cover (FC), tree height (H), leaf area index (LAI) and sky view-factor (SVF) – the four main structural parameters used by hydrologic models – with a suite of spectral indices obtained from six spectral bands of a Landsat 5 TM image. Despite numerous sources of variation that affect the relationships between 2-dimensional spectral indices and three-dimensional structural metrics, models to predict FC, H, LAI and SVF with reasonable accuracy were developed. The extrapolation of these variables across a watershed in British Columbia severely affected by insect disturbance resulted in highly-detailed 30 m spatial resolution maps and frequency distributions consistent with the natural variation ranges of each metric – a major improvement compared to traditional approaches that use coarse, discrete vegetation classes in fully-distributed models. This article fits well with repeated calls from researchers to maximize the use of remote sensing tools in hydrologic studies, especially for larger catchments where satellite-derived data might be the only alternative to properly initialize models.