Application of the HBV model for the future projections of water levels using dynamically downscaled global climate model data

The Hydrologiska Byråns Vattenbalansavdelning (HBV) model was used to project the future water levels of the Mackenzie River at selected stations. The Weather Research and Forecasting (WRF) model was utilized to dynamically downscale the Global Climate Model data. The calibrated and validated HBV model was run with the WRF downscaled CanESM2 data and with the PCIC data for the historical (1979–2005) period, and then compared with the observed flow data at the Fort Simpson station and the Arctic Red River station. The simulated streamflow showed a good correlation with the observed streamflow (R value was around 0.85). The HBV model was then forced with the bias-corrected WRF downscaled daily rainfall and temperature data driven by the CanESM2 RCP 4.5 and RCP 8.5 climate scenarios to simulate the future streamflow for the 2041– 2070 period. Rating curves were used to convert streamflow to water levels. At the Fort Simpson station, mean flow was projected to decrease by about 5% under both RCP 4.5 and RCP 8.5 scenarios, whereas the peak flow was likely to reduce by about 12 and 9% for RCP 4.5 and RCP 8.5 scenarios, respectively, in the 2050s. The projected lower water levels could affect the navigability and the northern ferry operations of the Mackenzie River.


INTRODUCTION
The long-term increase in air temperature globally has modified the energy and water fluxes, and thus the hydrological cycle.
Rising air temperature may accelerate the atmospheric moisture transport which could alter the climate system (Newton et al. ). Changes in the hydrologic cycle would occur as temperature increases, and observations along these lines have already been documented (Groisman et al. ; Pryor et al. ). Changes to seasonal air temperature and precipitation can have a significant long-term impact on the hydrology of river basins across the world, including that of Canada. The Mackenzie River Basin (MRB), the largest River Basin in Canada, is expected to be subject to the significant impact of climate change. Several studies have pointed out that the MRB is experiencing some of the greatest rises in temperature anywhere in the world (Wayland ), and the studies also have concluded that over the last few decades, the warming trend is going up sharply in Canada including the MRB (Shabbar et al. ; Zhang et al. ; Aziz & Burn ). Conversely, we expect that climatic change will modify the future water resources of MRB, especially during summer when there will be more demand for water to adequately maintain aquatic and terrestrial ecosystems, agricultural production, northern ferry operations, and hydroelectricity generation (Stewart et al. ; Trenberth et al. ).
Hydrological simulations to predict the future streamflow at the basin scale require future climate data. This future rainfall and temperature data can be obtained from the Global Climate Models (GCMs) outputs. However, the GCMs climate data resolutions are too coarse for direct use in catchment-scale hydrological modeling and need to be downscaled before the simulation process (Fowler et al. ). The downscaled GCM data could be a reliable way to simulate the streamflow of a river using any hydrological In this study, the Weather Research and Forecasting (WRF) model was selected to dynamically downscale the GCM data into a regional scale. The downscaled temperature and rainfall data will be the input to the hydrology model (HBV) to simulate the streamflow. The WRF model is a next-generation mesoscale numerical weather prediction system that is being used by many researchers in different parts of the world. Several studies have reported that the WRF model could reliably downscale the future temperature and rainfall data into much finer scale. It has been utilized to simulate rainfall over various regions (Chawla et  Considering this fact, we incorporated the WRF model data into the HBV model to simulate the streamflow and to project the future water levels of the Mackenzie River. The Mackenzie River possesses a vital role in the northern transport system. In an open water season, the ferry transport system keeps the northwest part connected to the rest of Canada. Here, the water level in the summer season is important for planning and managing the ferry operation system, which could be affected by the recent trend of climate change. The key objective of this study was to simulate the streamflow and to predict the water levels of the Mackenzie River for the 2050s based on the downscaled data of an RCM (WRF model) driven by boundary conditions of a selected GCM. The outcome of this study will provide some quantitative estimate of possible hydrologic changes, especially the water levels of the Mackenzie River in future.

Study area
MRB is a high latitude continental basin with a total area of about 1.8 million square kilometers, extending from 52 to 69 N and 140 to 102 W, making it the largest river basin in Canada. The western part of MRB is mountainous and has an average elevation of about 1,000 m, whereas the interior part is a plain and the eastern part is the Canadian Shield. The map of the MRB with terrain height is shown in Figure 1. The northern part of the basin experiences harsh winter, with subzero temperature for the whole winter, with an average winter temperature ranging from À25 to À35 C, and À50 C is not uncommon; whereas, in the summer, the average monthly temperature ranges from 8 C in the north to 20 C in the south, which could be as high as 35 C. So, the inter-annual temperature variation is substantial within this region. From 1979 to 1998, the average annual precipitation over the MRB was about 410 mm. In this study, climate data such as precipitation and temperature were calculated as a weighted mean of climate stations in and around the basin. A Thiessen polygon method was used to determine the relative weight of different stations for mean daily temperature and precipitation.
When gridded climate data are used, the areal average is obtained by this method. A similar method (Thiessen poly- Appendix 2) that essentially minimizes differences between the simulated runoff of the HBV with observed streamflow.
The calibrated HBV parameters are then validated using data independent of the calibration period.

ANUSPLIN data
The Australian National University SPLINe (ANUSPLIN) data are daily observational climate data produced by Natural Resources Canada and were available at 300 arc-second (10 km) spatial resolution over Canada from 1950 to 2015.
The ANUSPLIN likely is the best-gridded dataset available for Canada and has been used as the source data to compare climate products (Eum et al. ; Wong et al. ) and to evaluate the accuracy of regional climate models (Eum Model is necessary to downscale the GCM model data into a regional scale before using it in any hydrological models. Figure 2 illustrates the typical GCM, Regional Climate Model (RCM), and local hydrology parameters.
The calibrated and validated WRF model setup (Pervin & Gan ) was used to simulate temperature, rainfall, and other climate data for May, June, July, August, September, and October (MJJASO) over the MRB for the historical period (1979 to 2005), and the WRF simulated historical data were compared with the reference ANUSPLIN data.     temperature, rainfall, and PET data will be used as the input to the HBV hydrological model to simulate the streamflow for the future period at the selected stations of the Mackenzie River to assess the future water levels.     Red River will be experiencing reduced peak flow which is also projected to shift about 2 weeks earlier than the historical peak flow.

Calibration and validation of the HBV model
Since the Fort Simpson station shows a significant change in the future water levels, we use the boxplots to statistically compare the results. In Figure 12, the boxplots represent the average weekly water levels of MJJASO for each time period (historical and 2050s for RCP 4.5 and RCP 8.5). The observed historical mean water level at the Fort Simpson station matches with the WRF downscaled CanESM2 data-derived HBV simulated mean historical water levels. In the 2050s, the mean water level will be lowered by 300 cm with respect to the historical period; on the other hand, the peak flow and the high water events will be lowered significantly for both RCP 4.5 and RCP 8.5 scenarios.
At the Fort Simpson Station, mean flow is projected to decrease by about 5% under both RCP 4.5 and RCP 8.5

CONCLUSIONS
To project the future streamflow and to assess the changes in water levels of the Mackenzie River, a conceptual hydrologi-    Again, we also applied the PCIC data for the same period