| Abstract: | The distributed hydrology soil–vegetation model (DHSVM) was applied to the small watersheds WS1, 2, 3 in H.J. Andrews Experimental Forest, Oregon, and tested for skill in simulating observed forest treatment effects on streamflow. These watersheds, located in the rain–snow transition zone, underwent road and clearcut treatments during 1959–66 and subsequent natural regeneration. DHSVM was applied with 10 m and 1 h resolution to 1958–98, most of the period of record. Water balance for old‐growth WS2 indicated that evapotranspiration and streamflow were unlikely to be the only loss terms, and groundwater recharge was included to account for about 12% of precipitation; this term was assumed zero in previous studies. Overall efficiency in simulating hourly streamflow exceeded 0·7, and mean annual error was less than 10%. Model skill decreased at the margins, with overprediction of low flows and underprediction of high flows. However, statistical analyses of simulated and observed peakflows yielded similar characterizations of treatment effects. Primary simulation weaknesses were snowpack accumulation, snowmelt under rain‐on‐snow conditions, and production of quickflow. This was the first test of DHSVM against observations of both control and treated watersheds in a classic paired‐basin study involving a long time period of forest regrowth and hydrologic recovery. Copyright © 2005 John Wiley & Sons, Ltd. |
