|dc.description.abstract||Decades of paired catchment studies have provided insights on how forests regulate water redistribution following vegetation disturbances such as logging and wildfire. In these settings, changes in runoff characteristics are detected by comparing streamflow responses to those of undisturbed catchments nearby. While this is generally considered the best approach for assessing disturbance impacts, the method is prone to confounding factors. Challenges can often include issues relating to spatial scale, limited resources for monitoring, and climate trends that can confound subsequent analyses.
Process-based computer models can be used to address the shortcomings of empirically based paired catchment approaches. One way this is achieved is through simulating impacted catchments under no-disturbance conditions to provide a virtually identical control for comparison. In this thesis, we address two objectives, which are to 1) assess the capability of hydrologic models and hydrologic-vegetation growth models in simulating altered streamflow patterns of forested watersheds following disturbances, and 2) compare the utility of different change detection methods in describing the hydrologic impacts of forest disturbances using hydrologic signatures.
We applied and then evaluated two process-based models (Raven and Raven Robin) using their simulated weekly runoff ratios and calibrated parameter distributions to better understand the individual effects of climate variability and harvesting on streamflow at the Turkey Lakes Watershed in Ontario, Canada. Calibrated models achieved Kling Gupta Efficiency (KGE) scores higher than the climatological reference benchmark specific to their catchment-period scenario. However, most models became less effective outside of their calibration periods due largely to the effect of climate trends (the post-harvest period exhibited significantly warmer and drier climate conditions). Pre-harvest Raven models were an exception to this, likely owing to the canopy losses that initially buffered higher evapotranspiration demands driven by a warmer climate after harvesting. Despite maintaining their predictive skill, such models are subject to equifinality concerns as they do not incorporate changes to the vegetation cover induced by harvesting.
In assessing weekly runoff ratio distributions derived from both observed and simulated data, we found that all three catchments produced low magnitudes of runoff. For this reason, we separated our data into low-flow and high-flow time steps. Through comparisons of weekly runoff distributions describing high flows, we show that pre-harvest and post-harvest periods exhibited distinct hydrologic responses across all three catchments. This resulted in models that simulated functionally different rainfall runoff responses when calibrated to different periods. However, these responses could only be assessed during high-flow weeks; catchments exhibited dry conditions for as much as 16 weeks during the three summer months of each water year – thus suggesting that the weekly runoff ratio is not an ideal change metric for use in the non-lake headwater catchments of the Turkey Lakes watershed. This guided our subsequent assessment of model ability in simulating low-flow conditions that were excluded from our initial analysis: models calibrated to drier post-harvest conditions were generally more successful at simulating dry weeks during the summertime. This was not the case for post-harvest Raven Robin models in catchment C31, which were trained to represent a regrowth forest with lower canopy densities after treatment. This decreased interception and evaporative losses in the model, making them less successful at simulating dry conditions.
We then compared parameter distributions between model ensembles calibrated to different periods in each catchment. Results from this analysis suggest that climate trends altered soil moisture dynamics in control catchments C32 and C35, promoting more infiltration, soil evaporation, altered baseflow regimes, and an increase in plant water uptake rates. Catchment C31 models experienced parameter distribution shifts that created a new evapotranspiration regime for its vegetation in addition to enhanced soil moisture storage. Overall, the significant differences detected between parameter distributions in this analysis suggest that both climate trends and clearcut harvesting have altered hydrologic processes in the watershed.
While our results demonstrate how models can be used for hydrologic change detection, improvements can be made to the employed methods. We recommend considering phenological timescales when designing model calibration experiments to address the non-stationarity of model parameters across time. This will better represent emergent processes throughout the various stages of hydrologic recovery, allowing for improved quantification of forest disturbance impacts on watershed hydrology.||en