The impact of initial conditions on convection-permitting simulations
of flood events
Abstract. Western Norway suffered major flooding after 4 days of intense rainfall during the last week of October 2014, resulting in damages totalling hundreds of millions Norwegian kroner. These types of events are expected to become more frequent and severe as Norwegian (and the earth's) climate continues to warm. However, due to the strong effects that local features and conditions can have on these kinds of events, coarse-grained global or regional models are unable to capture their characteristics. Very high resolution models run at so-called convection-permitting scales have shown some promise for reliably capturing such events. Doing so in a robust manner is, as a matter of course, of high interest to both scientists and stakeholders in both climate prediction and projection contexts. Despite this promise, the impacts of initial conditions on convection-permitting simulations, i.e., precipitation pattern and discharge, are uncertain, especially over complex, mountainous terrain. Complicating matters, these areas also usually lack dense measurement networks. In this paper, we apply a distributed dynamic regional atmosphere-hydrological modelling system (WRF-Hydro) at convection-permitting scale and assess its performance over four catchments in western Norway for the aforementioned flood event. The model is calibrated and then evaluated using observations and benchmarks obtained from the HBV light model. Interestingly, the calibrated WRF-Hydro model with NSE value of 0.86 exceeds the upper benchmark obtained from the conceptual HBV model with NSE value of 0.80 suggesting that the former performs as well or better than the simpler conceptual model, especially in areas with complex terrain and poor observational coverage. Confident in the capabilities of the modelling system we then examined the sensitivity of precipitation pattern and discharge, especially peak flow, to poorly constrained elements such as spinup time and snow conditions. The results show that: (1) overall the convection-permitting WRF-Hydro simulation captures the precipitation pattern/amount, the peak flow volume and the timing of the flood event; (2) precipitation is not overly sensitive to spinup time, while discharge is slightly more sensitive due to the influence of soil moisture, especially during the pre-peak phase; (3) the idealized snow depth experiments show that a maximum of 0.5 m of snow is converted to runoff irrespective of the initial snow depth and that this snowmelt contributes to discharge mostly during the rainy and the peak flow periods. This suggests that snow-cover, in these experiment at least, intensifies the extreme discharge instead of acting as sponge, which further implies that future rain-on-snow events may contribute to higher flood risk. While targeted experiments on the changing characteristics of projected future rain-on-snow events are needed to confirm this study suggests that WRF-Hydro is an ideally formulated tool to investigate these questions.