Characterizing real-time forest disturbance in a dynamic land surface model with implications for operational streamflow forecasting using WRF-Hydro

Abstract:

Forest cover changes have potential to induce large shifts in water fluxes, including short-term responses like changing downstream flood risk and longer-term feedbacks between land surface water and the atmosphere that can influence local and regional weather patterns. Two main challenges in capturing vegetation changes within hydrological forecasting systems, however, are finding real-time data streams to characterize these changes and the ability to accurately translate readily observable qualities such as “greenness” to modelled biophysical and biochemical properties.

Here we investigate the feasibility of using near real-time remote sensing products paired with a continuous-running atmosphere - land surface - hydrology model to characterize vegetation disturbance and regrowth dynamics, with the ultimate goal of improving short-range (hourly to daily) and mid-range (weekly to monthly) operational streamflow forecasts. We use WRF-Hydro, a framework for coupling components of land surface models, hydrologic models, and the Weather Research and Forecasting (WRF) atmospheric model. We take advantage of newly implemented data assimilation techniques in WRF-Hydro to assimilate MODIS near real-time observations into the NoahMP dynamic vegetation model to simulate disturbance and regrowth following (1) wildfire in the Southwest U.S. and (2) plantation forestry management in the Southeast U.S. Ranges of modeled vegetation structural (leaf and stem area, height, canopy organization) and biochemical (chlorophyll content) characteristics consistent with MODIS observations are used to estimate uncertainty ranges in downstream flow predictions.

The WRF-Hydro framework allows us to characterize changes in streamflow behavior expected after these modeled forest disturbances, evaluate when and where assimilation of vegetation characteristics from remote sensing is most important to streamflow prediction, and quantify the role of recovering vegetation biophysical/biochemical state uncertainty in predicted streamflow probability distributions. These hindcast examples, by relying solely on near real-time data streams, provide a basis for improving representation of post-disturbance vegetation dynamics in future operational hydrological forecasting systems.