An enhanced representation of forest cover for distributed hydrologic modeling based on national-scale forest monitoring data, remote sensing, and biophysical variables in a statistical learning algorithm

Accurate representation of the ecohydrologic effects of forest disturbances in distributed hydrologic models requires maps of overstory leaf area index (LAI) at sufficiently fine resolution to indicate loss of tree cover. We present a new method for estimating overstory and understory LAI at 30-m resolution using a combination of ground vegetation measurements, remote sensing data, and biophysical predictors in a statistical learning model (random forests) applied to the 3,000-km² upper South Fork Flathead watershed, Montana. Ground vegetation data were acquired from the US Forest Service's Forest Inventory and Analysis (FIA) program, which collects data from a probabilistic sample of 0.4-ha permanent plots with mean spacing of 5 km throughout the US. Because FIA does not directly measure LAI, we used FIA's measurements of areal vegetation coverage, plant height, and canopy height to estimate the proportion of total canopy volume in the overstory canopy versus understory. We used these proportions to partition total LAI, as estimated from remote sensing, into overstory and understory LAI components at FIA plots. We then developed a statistical model of overstory LAI based on spatially explicit biophysical and remote sensing predictors to generate a gridded overstory LAI layer. The advantage of representing overstory rather than total LAI is that overstory trees exert stronger controls than understory vegetation on hydrologic processes such as interception of rain and snow, sublimation of intercepted snow and ground snowpack, evapotranspiration from both overstory and understory vegetation, and radiation transmission. When we applied the method to two time periods (2003-2009 and 2010-2016) within the South Fork Flathead watershed, we found that the overall frequency distributions of overstory, understory, and total LAI shifted downward at the basin scale, reflecting recent canopy loss due to a combination of drought- insect-caused tree mortality, as well as total canopy decreases caused by fire. The importance of this method is its ability to capitalize on existing distributed models that represent overstory influences on ecohydrologic fluxes, thus enabling enhanced understanding of process-level hydrologic responses to forest disturbance.