Modeling six years of stream discharge using the distributed hydrology soil vegetation model (DHSVM) in a coastal, timber harvest catchment, Humboldt County, California

Accurate modeling of forest hydrology regimes (i.e. hourly, daily and annual discharges, peak flows) requires detailed datasets to define the physiography, land cover, and meteorological variables. In this study, I used the Distributed Hydrology Soil Vegetation Model to simulate hourly stream discharge for 6 hydrologic years in a small, coastal watershed (463 ha), using meteorological data sourced from a nearby National Weather Service station, and a combination of remote sensing and timber harvest records to prepare model inputs. During the model simulation period, approximately 20 percent (~100 ha) of the basin was harvested. I calibrated and validated the model with continuous discharge measurements made in the study area, and used the Nash-Sutcliffe Model Efficiency, the coefficient of determination, and percent bias metrics to evaluate model success at simulating hourly, mean daily, and total annual discharges. Simulated annual discharges over the six year period display a 2 percent to -23 percent bias compared to observed values. Nash / Sutcliffe and coefficient of determination metrics iv were evaluated for hourly and daily discharges, with the daily metrics improving two percent to 10 percent over hourly metrics. Daily Nash / Sutcliffe values ranged from 0.58 to 0.85 and daily coefficient of determination values show the greatest success ranging from 0.82 to 0.94. The mean absolute error for hourly discharges ranged from 0.05 to 0.1 cubic meters per second and showed a slightly better performance when evaluated at the daily level. Daily mean peak flows display varying degrees of scatter year to year, with Root Mean Square Errors ranging from 0.35 to 0.68 cms. Poor model performance can be somewhat explained by the uncertainty in the meteorological and solar record (out-ofbasin observations) and limited soil data. The model's performance could be improved with either the inclusion of preferential flow pathways (model coding), detailed soil data (spatial variability of soil properties), or in-basin meteorological observations.