DISCUSSION

Through linking hydrological fluxes at local scales to regional climatic teleconnection patterns beyond seasonal variations, we can better understand local hydrological processes across multiple years (Prein et al., 2015) and the nature of regional climate and hydrological extremes (Langendijk et al., 2019). Regional climatic patterns can explain nonlinear hydrological behaviors. For example, the hydroclimatic phases one and three in this study had similar annual air temperature and precipitation (<10% difference (Table 3)). Snow and ROS runoff, however, showed substantially different responses to variations of air temperature and precipitation. The observed peak SWE was 63% above normal in phase one and only 24% above normal in phase three. The modeled ROS runoff was 7% below normal in phase one, while it was 45% above normal in phase three (Table 3). Phase one was influenced by negative SST with La Niño events over 1984–1986, while positive AAO influenced phase three. Similar local hydrology and different climate teleconnection pattern (negative SST in phase one and negative AAO in phase three) explained a 52% difference in the modeled ROS runoff (Table 3). The same but opposite-sign teleconnections in phases five and six resulted in different snow and runoff conditions, wet in phase five and dry in phase six. This suggests that there is a strong linkage between teleconnection patterns and local hydrological regimes despite a scale mismatch. The different ROS runoff in phases one and three and the snow and runoff regimes in phases five and six highlighted the role of climate teleconnection patterns in dictating hydrological conditions at local scales. This is consistent with the findings of Whitfield (2001), which showed that small variations in climatic patterns can lead to large hydrological responses. Bonsal and Shabbar (2008) reported that low flow events in western Canada are associated with positive phases of the PNA pattern. The positive NAO, however, is more influential than the positive PNA pattern in decreasing the annual runoff and ROS runoff in Reynolds Mountain (phases two and six, Table 3).
The modeled ROS runoff had large temporal and spatial variations (Figure 9). High variability of ROS runoff implied that not only the regional climate teleconnections affected the local scale hydrology, but also vegetation heterogeneity played an important role. The sheltered forest landscapes with minimal blowing wind generated the highest ROS runoff during phase three (Figure 9a), while the blowing snow source HRUs showed the lowest ROS runoff in phase two. The above-normal precipitation (206 mm, 21%) and the below-normal winter air temperature (0.5 °C) during phase three prolonged the snow cover period by 8 days above normal (Table 3) and increased the frequency of the ROS events in the sheltered forest landscapes. Annual precipitation was 163 mm (17%) below normal and the winter air temperature was 0.3 °C warmer than the normal values (Table 3) in phase two, which led to the lowest ROS runoff in blowing snow sources HRUs among the six phases (Table 3 and Figure 9). This is because of a short period of snow cover in the warm phase and snow transport out of the source HRUs by blowing wind. The frequent ROS events are correlated with the positive phase of SST and El-Niño events (McCabe et al., 2007). Consistently, the hydroclimatic conditions in phase one with a negative SST (Figure 9) and La Niño events showed less frequent events by decreasing ROS runoff by 7% below normal (Table 3). A strong positive NAO in the hydroclimatic phases two and six (Figure 8) affected the interannual variability of snow cover (Derksen et al. 2008; Ge & Gong 2009; Bao et al. 2011) and locked the polar cold air in the Arctic region (Francis & Vavrus, 2012; Cohen et al., 2014), leading to a warmer, drier, and shorter than normal winter in the study area. Warm and dry conditions restricted the generation of ROS runoff, especially in HRUs with short vegetation (blowing snow source HRUs in Figure 9b). Therefore, positive NAO and AAO have more pronounced effects on ROS runoff than negative SST or positive PNA pattern. The positive NAO decreased the ROS runoff by 26% below normal, while the positive AAO increased the ROS runoff by 45% above normal (Table 3). A strong negative correlation between AAO and rainfall ratio (0.70, Figure 6a) indirectly showed the effect of AAO on the magnitude of ROS runoff by changing precipitation phase from snowfall to rainfall.
Despite a small effect of the negative SST on ROS runoff, it had the largest effect on annual peak SWE and runoff increase in Reynolds Mountain and it elevated the observed peak SWE and annual runoff by 63% and 57% above normal, respectively (phase one, Table 3). On the other hand, the positive NAO had the largest impact on snow and runoff drop and it decreased the observed peak SWE and annual runoff by 55% and 43%, respectively (phase six, Table 3), and the modeled ROS runoff by 31% (phase two). A high correlation between ROS runoff and NAO (Figure 6d) explained their potential relation.
The runoff generations in drift and north facing HRUs are sensitive to a wet climate with warm air temperatures (Figure 10). The modeled peak SWE is almost similar to a modeled runoff in these HRUs during dry phases (e.g., phase two) while under wet conditions (e.g., phase one) annual runoff is greater than peak SWE. The spatial variability of runoff under low flow conditions is very similar to that in the low ROS conditions (phase two and six in Figure 10). The highest annual runoff in Reynolds Mountain occurred when rainfall ratio was near the long term average and mean annual air temperature was 0.6 ºC cooler than the average (phase one, Table 3). Under cold air temperatures, snowpack is usually deep and sustains the baseflows in summer, which may increase the annual runoff. Under warm air temperatures, however, we may not expect an increase in annual runoff (i.e., phase six) as both rainfall proportion of precipitation and ET will increase, which may cancel each other out (Rasouli, 2017; Rasouli et al., 2019a).
As the detected hydroclimatic phases are temporally evolving, they can be applied in short and medium range hydrological predictions. For instance, based on the distance of the present-year precipitation from the long term average (Figure 7) and because of a climatological persistence, it is likely that annual precipitation will be below normal for the following year after the sixth phase in Reynolds Mountain. Dutta and Maity (2018) found that a temporally evolving hydroclimatic teleconnection can improve the predictability of the monsoon rainfall at local scales. The application of climate variability patterns as an input to machine-learning methods has been shown to be skillful in improving short-term (Rasouli, Hsieh, & Cannon, 2012) and long-term (Rasouli, Nasri, Soleymani, Mahmood, Hori, & Torabi Haghighi, 2020) streamflow forecasts. A numerical representation of governing hydrological processes by a physically based model can help simulate and forecast the hydrological processes in the present and future climates under different phases of climate variability patterns such as AO (Thompson & Wallace, 1998), NAO (Walker & Bliss, 1932), AAO (Thompson & Wallace, 2000), SST in the Niño 3.4 region (Trenberth, 1997), and PNA, (Blackmon, Lee, Wallace, & Hsu, 1984).
The uncertainty in modeling the high flows using the CRHM model and accurate delineation of the hydroclimatic phases due to occasional overlapping of the positive and negative phases of different teleconnection patterns are the main limitations of this study and should be taken into account when interpreting the results.