1. Introduction
The spatiotemporal representativeness of liquid and solid precipitation data is among the most crucial factors in every flow simulation practice. Sporadic meteorological observations, among other data constraints, can result in uncertainties in many hydrological modeling practices performed for flow and inflow forecasting. This is also the case with the “HYDROTEL” system (Bouda et al. 2012, 2014; Fortin et al. 2001a; Turcotte et al. 2003, 2007) set up for the watersheds in Yukon in northwestern Canada, where data constraints due to sparsely distributed precipitation information in major basins of interest have adversely affected the performance of the modeling system. Therefore, it is obvious that augmenting the precipitation observation network could greatly reduce the uncertainty involved with meteorological forcing.
In many forecasting centers around the globe where streamflow simulation is performed in basins with a hydrology dominated by snowpack melt during spring freshet, in the absence of a high-density precipitation observation network, assimilation of in situ and remotely sensed measurements of snowpack state variables has become increasingly important for accurate flow estimation (Helmert et al. 2018). Li et al. (2019) have shown that in snow-dominated basins, where the meteorological uncertainty during the forecast period is significant (which is the case for sparsely gauged networks), reinitializing the model based on observed snow water equivalent (SWE) information can significantly improve streamflow forecasts. Similarly, in the absence of a high-density precipitation observation network, assimilation of snowpack state variables can provide the possibility to handle different sources of uncertainty by merging the value of observed information into the model in order to correct the effects of model errors and improve forecasting capabilities (Turcotte et al. 2010).
SWE reinitialization through various data assimilation (DA) approaches has been proven to be an effective approach to improve the degree of agreement between the simulated and observed discharge values (see, e.g., Clark et al. 2006; De Lannoy et al. 2012; Leisenring and Moradkhani 2011; Nagler et al. 2008; Liu et al. 2013; Saloranta 2016). Several DA techniques are available for updating snow state variables, including direct insertion (Liston et al. 1999), Cressman interpolation (Drusch et al. 2004), optimal interpolation (Brasnett 1999), nudging (Boni et al. 2010), particle filtering (Arulampalam et al. 2002), and various types of Kalman filtering approaches with different levels of complexity (Gelb 1974; Miller et al. 1994; Moradkhani 2008; Evensen 1994). Among these approaches, Kalman filtering, and its Monte Carlo–based implementation, the ensemble Kalman filtering (EnKF) approach, have been widely applied in different hydrological modeling studies (see, e.g., Andreadis and Lettenmaier 2006; Clark et al. 2006; De Lannoy et al. 2012; Durand and Margulis 2008; Huang et al. 2017; Magnusson et al. 2014; Piazzi et al. 2018; Slater and Clark 2006; Su et al. 2008).
Currently, to gain a proper insight into short-term, seasonal, and long-term flow forecasting in northern and midcordilleran alpine, subalpine, and boreal watersheds in Yukon, where the flow regime is dominated by snowpack melt, and also to alleviate the adverse effects of scarce precipitation datasets, two independent DA routines are combined in HYDROTEL. These DA tasks are performed to update (i) flow states, including soil temperature, soil moisture, overland flow routing, and river flow routing, based on in situ discharge measurements, and (ii) snow states, including snow depth, SWE, snowpack thermal deficit, snowpack liquid water content, and surface albedo, based on snow survey data. The first DA routine was implemented by Samuel et al. (2019), where the North American Ensemble Forecasting System (NAEFS) precipitation products are merged into the operational flow forecasting platform in HYDROTEL through EnKF. The snow DA routine, on the other hand, performs a distributed snow correction of the simulated snowpack based on available in situ measurements. When snow surveys are available, the simulated state variables including SWE and snow depth are corrected based on site measurements. The correction is performed by interpolating the three nearest sites, where measurements are taken from, over the entire watershed (Turcotte et al. 2007). Thus, the application of the snow DA routine in HYDROTEL is in line with the same practice followed by a number of other forecasting centers (see, e.g., Brasnett 1999; Barrett 2003; Drusch et al. 2004).
There are other sources of information, such as gridded numerical products, which can reduce the input data uncertainty. For instance, the numerical weather prediction datasets produced by Environment and Climate Change Canada (ECCC), which are adjusted through an assimilation technique known as statistical interpolation (SI), represents a prime example of such atmospheric analysis gridded precipitation products. Currently, these adjusted products are created by the Canadian Precipitation Analysis (CaPA) system (Fortin et al. 2015; Mahfouf et al. 2007), the product of which is known as the Regional Deterministic Precipitation Analysis (RDPA). The CaPA-RDPA products are currently available in grib2 format on a polar-stereographic grid with a 10-km resolution (true at 60°N) at two temporal resolutions (6 hourly and 24 hourly). A high-resolution version of the system, known as High Resolution Deterministic Precipitation Analysis (CaPA-HRDPA) System is also in operation since 2018 and takes the HRDPS 2.5-km-resolution field as the trial.
The CaPA system has gained considerable momentum in recent years, and the suitability of its precipitation products for application in hydrological modeling studies in Nordic watersheds in Canada have been the subject of a number of studies (e.g., Deacu et al. 2012; Eum et al. 2014; Gbambie et al. 2017; Haghnegahdar et al. 2014; Hanes et al. 2016; Wong et al. 2017; Zhao 2013). Boluwade et al. (2018) compared the performance of CaPA-RDPA data with precipitation observations in the Lake Winnipeg basin, which entails many of the hydroclimatological characteristics associated with the northern Great Plains and concluded that CaPA-RDPA data is a reliable precipitation product in sparsely gauged basin. Xu et al. (2019) evaluated daily total precipitation data derived from CaPA-RDPA, ERA-Interim, ERA5, JRA-55, MERRA-2, and NLDAS-2 over the Assiniboine River basin, and concluded that in general, except for convective rainfalls in summer, CaPA-RDPA products demonstrated the best performance among all.
CaPA-RDPA data have also been used for establishing observational priorities in poorly instrumented basins in Canada. For instance, Abbasnezhadi et al. (2019) used the SI technique and simulated the products and by-products of the CaPA system to design a stochastic meteorological network density assessment scheme. In this approach, the network assessment is undertaken with the objective to maximize the accuracy of precipitation products for hydrological modeling applications. This scheme can be used to find the optimal density of a new observation network, only if the RDPA products in the sparsely gauged region, where the observation network is investigated for augmentation, are assumed to represent the truth. Given such a proposition, a controlled assessment approach (one in which observation uncertainty is accounted for), as suggested by Abbasnezhadi et al. (2019), would then be necessary to find the optimal station density. However, the benchmark that the snow assimilation routine in HYDROTEL provides for accurate flow estimation would mean that the accuracy of the CaPA-RDPA products could be first validated prior to undertaking the network assessment. In other words, it is possible to claim or at least expect that the current SWE correction performed in HYDROTEL can result in accurate streamflow estimates against which the simulated streamflow for given CaPA-RDPA forcing can be compared. Such an evaluation would provide us with valuable information (i.e., benchmark) with respect to the accuracy or the intrinsic added value of using the CaPA-RDPA products in sparsely gauged basins for meteorological network assessment. Given this approach, it would then be possible to perform the precipitation observation network assessment through a parsimonious approach. Therefore, this study was designed to provide a framework for performing a proxy validation (i.e., indirect validation of gridded weather products by means of hydrological modeling) of the RDPA products through the application of snow assimilation in physically based hydrologic models. The proxy-validation experiment and the network assessment framework designed in this study can therefore be undertaken to complement the precipitation network assessment approach designed by Abbasnezhadi et al. (2019). The assessment scheme introduced in this study may also be implemented autonomously in sparsely gauged basins, providing that snow survey data would be readily available.
The remainder of the paper is organized as follows. In section 2, the study area is described and specific details with respect to the hydrometeorological data used in the study are provided. Section 3 describes the HYDROTEL model and outlines the approaches carried out to: (i) perform HYDROTEL parameter sensitivity analysis and optimization, (ii) validate the CaPA-RDPA products through the application of the snow data assimilation routine in the model, and (iii) undertake the network assessment. Thereafter, results are presented and discussed in section 4, and conclusions are drawn in section 5.
2. Study area and data characteristics
a. Study basins
Figure 1 illustrates the location of the three study basins in Yukon, Canada, including the Mayo River basin, Aishihik (/eyzhak/) River basin, and Upper Yukon River basin. These watersheds are located in northern and midcordilleran alpine, subalpine, and boreal ecoclimatic regions (Strong 2013) of central and southern Yukon. The Mayo basin covers a drainage area of roughly 2670 km2. The mean annual precipitation and mean daily 2-m temperature are 456 mm (257 mm as rain; 199 mm as SWE) and −5.9°C, respectively (true for 1981–2018). The flow volume varies on a seasonal basis, peaking in summer between June and July and dropping during winter in January and December. There are two generating stations in Mayo: Mayo A and Mayo B. The Aishihik basin covers a larger drainage area in the order of 4550 km2 and is housing the Aishihik hydroelectric Facility. The mean annual precipitation is around 302 mm (126 mm as rain; 176 mm as SWE), and the mean daily annual 2-m temperature is approximately −6.6°C (true for 1981–2018). The streamflow peaks in June, and the flow volume is relatively higher between May and October (Brabets and Walvoord 2009). The Upper Yukon River basin is the largest of the three and covers a drainage area of around 19 600 km2. The basin is mountainous and is largely covered by sporadic permafrost. Runoff in the Upper Yukon is derived primarily from snowmelt and rainfall. The mean annual precipitation is around 299 mm (101 mm as rain; 198 mm as SWE), and the mean daily annual 2-m temperature is approximately −3°C (true for 1981–2018). The streamflow peaks in August and is low between November and May. There is a generating station in Whitehorse and one control structure on Marsh Lake. For all three basins, the dominant hydrological processes are governed by snow accumulation and melting that produce high flow volume, which peaks in summer. In addition, the Upper Yukon River summer runoff involves glacier melting from the southwest region of the basin.
The locations of Mayo, Aishihik, and Upper Yukon River basins in central and southern Yukon. The southern half of the Upper Yukon basin is located within northern British Columbia.
Citation: Journal of Hydrometeorology 22, 4; 10.1175/JHM-D-20-0106.1
b. Meteorological data
Table 1 provides a list of the meteorological stations located within and in the vicinity of the boundaries of each basin. The MAYOMET and AISHMET stations are operated by Yukon Energy (YE), and the other stations are operated by the Meteorological Survey of Canada (MSC).
Meteorological Survey of Canada meteorological networks in Mayo, Aishihik, and Upper Yukon (see Fig. 2).
Figure 2 shows the distribution of the meteorological stations within and in the vicinity of the study basins. In Mayo, the precipitation gauge at the Mayo airport (Mayo A), which is located just in the outskirts of the basin, is the only historical active weather station with close to 100 years of available record. The MAYOMET station located near the outlet of Mayo Lake was installed in late 2018 and is the only active station within the basin. In Aishihik, the majority of the stations (17 of 27) have less than 25 years of available data. There are three active MSC stations within a 75-km distance from the basin boundaries, including Carmacks CS (recording since 1999), Haines Junction (recording since 1944), and the one at Burwash airport, which is 50 km east of Aishihik, providing more than 50 years of historical precipitation data in conjunction with its nearby stations (Burwash and Burwash A). Within the basin boundaries, however, there are only two weather stations available (AISHMET and Otter Falls NCPC), of which Otter Falls NCPC has not been recording since 2015, and AISHMET was activated in late 2018. In Upper Yukon, more than 65% of the stations have less than 20 years of record, the majority of which have been installed in the past 10 years. The MSC station at Atlin is the only historical active station with more than 120 years of recorded precipitation amounts. It should be reminded that solid precipitation undercatch is rather an important issue to consider when assimilating snow measurements. Pierre et al. (2019) assessed the undercatch to be as much as 20%–70% of the solid precipitation, which is, to the authors’ knowledge, the most recent assessment available. This can justify and explain why snow assimilation is necessary and beneficial.
The distribution of meteorological (solid black squares), hydrometric (concentric green circles), snow course sites (blue asterisks), and GMON stations (solid red three-dot triangles) within and in the vicinity of the study basins (Mayo, Aishihik, and Upper Yukon). Meteorological stations are graduated on the basis of the number of years of available record. Active meteorological stations (hollow red squares with dashed perimeter) and the synoptic weather stations currently assimilated in CaPA (hollow red circles) are identified.
Citation: Journal of Hydrometeorology 22, 4; 10.1175/JHM-D-20-0106.1
The grib2 CaPA-RDPA v3.0.0 data from 2010 to 2018 at daily time steps were also downloaded from ECCC ftp repository and decoded using the NOAA/National Weather Service wgrib2 program. The decoded datasets were then converted from the polar-stereographic grid onto a rectangular grid covering each basin’s drainage area with a spatial resolution of 0.10° in latitude and 0.15° in longitude (roughly 10 km in both directions at 60°N).
c. Hydrometric data
Table 2 provides a list of available hydrometric stations at which streamflow measurements were taken in each basin (see Fig. 2 for the specific location of the hydrometric stations). The inflows to Aishihik Lake and Mayo Lake do not represent naturally observed discharge values and were reconstructed based on recorded water levels [see Samuel et al. (2019) for a detailed description of the reconstruction method]. For Mayo Lake, water level data obtained from the 09DC005 station and streamflow observed at the YECMAYO station were used for reconstructing inflows. Similarly, water levels recorded at station 08AA005 and streamflow recorded at 08AA008, 08AA009, and 08AA010 stations were used to reconstruct the inflows to Aishihik Lake. All flows and water levels were provided by the Water Survey of Canada (WSC), except those at reconstructed stations #0000003, ##0000003, and YECMAYO, which are recorded by YE.
Water Survey of Canada and Yukon Energy (YE) hydrometric networks in Mayo, Aishihik, and Upper Yukon basins. The gauges used for the HYDROTEL model calibration are in boldface type (see Fig. 2).
d. Snow readings
Table 3 provides the metadata of the snow depth and SWE monitoring networks managed by the Water Resources Branch (WRB) of Environment Yukon as well as the Gamma Monitoring (GMON) automatic snowpack sensor readings provided by YE. The GMON (a.k.a. Campbell Scientific CS725) sensor measures SWE by detecting the attenuation of naturally occurring electromagnetic energy from the ground. This contactless approach can offer highly reliable and accurate local SWE measurements with an uncertainty level that does not exceed ±5% at maximum snow depth. Traditional SWE measurement approaches, such as the application of snow pillows, by which the snowpack weight is directly measured, are prone to higher uncertainty levels since snowpack properties (e.g., radiation characteristics) can be altered during the measurement. The GMON gauge, which monitors snowpack properties in a contactless mode, does not suffer from the same disadvantages. During the past few years, a number of GMON gauges were installed at those locations identified in Table 3 and Fig. 2 (five stations were initially installed in Upper Yukon, but two were removed and relocated: one in Mayo and one in Aishihik). Once monitored, the collected information is transmitted via satellite connection and goes through quality control. Added and relocated GMON gauges intend to complete the existing snow survey site or at least offer specific measurements within the basin limit (in Aishihik, Mayo). In situ snow measurements are relevant and aim to capture snow evolution, but local measurements may not be representative for the entire basin conditions.
3. Models and method
a. HYDROTEL: Sensitivity analysis and model calibration
The semidistributed physically based HYDROTEL model can simulate a variety of hydrological processes. These processes and the physically based approaches used to simulate each one along with a list of parameters associated with each process used in the version of HYDROTEL utilized in this study are listed in Table 4. In HYDROTEL, the vertical water budget is computed over a computational unit called the Relatively Homogeneous Hydrological Unit (RHHU), which represents either a hillslope or elementary subwatershed and are derived based on a digital elevation model and a digital network of lakes and river sections using PHYSITEL, a specialized GIS for distributed hydrological models (Turcotte et al. 2001; Rousseau et al. 2011; Noël et al. 2014), both of which overlaid by a multilayer soil model. The soil column of a RHHU is stratified into three layers. The first soil layer (Z1) governs infiltration, and the other two layers (Z2 and Z3) control interflow and baseflow. The interpolation of meteorological variables is based on the weighted mean of the nearest three stations to resolve the amount of total precipitation, which is then partitioned into rain and snow according to a threshold temperature and a simple weighted scheme based on daily minimum and maximum temperatures, on each RHHU. For missing station values, HYDROTEL fills the gap by using the values available at the three nearest stations based on the interstation temperature and precipitation altitude variations. The accumulation and melt of snowpack processes are based on a mixed degree-day energy budget approach and determine the timing and peak of the spring freshet. In the glacier module, a mixed degree-day energy budget approach is also used in the exact same fashion used for the snowmelt process. In the soil temperature and soil frost process, the only associated parameter (soil freezing temperature threshold) is not distributed over the entire RHHUs, and therefore, is not recommended to be modified. The next process is designed to identify the potential evapotranspiration that is dominantly going to impact the total annual runoff and baseflow in summer. The flow process at the RHHU scale simulates the water flux toward the river network using a hydrogeomorphological unit hydrograph (a.k.a., HGM).
HYDROTEL parameter sets associated with each hydrological process. Importance-level-0 parameters refer to those that are often physically based constants (noncalibrated parameters), and levels 2 and 3 indicate lower importance levels and were not calibrated. Parameters calibrated in OSTRICH are identified by the importance level of 1. The lower and upper bounds shown in the table are used in parameter optimization.
While other studies have performed different types of sensitivity analyses of HYDROTEL on other basins (e.g., Bouda et al. 2014; Turcotte et al. 2003), a global sensitivity analysis was performed using the Variogram Analysis of Response Surfaces (VARS) toolbox (Razavi et al. 2019). The toolbox allows the user to identify the parameters by the importance level (i.e., model sensitivity to changing parameter conditions) through a multimethod approach that unifies different theories and strategies. With sensitive parameters in hand, the model calibration becomes a less challenging task. However, since calibration of HYDROTEL, in essence, is a multiobjective optimization problem (due to the number of stream gauges reporting flows in the basin for which several error criteria might be assessed), defining what makes the model calibrated is not a straightforward task. Moreover, other factors affecting the quality of the calibration result include error due to lake/reservoir inflow reconstruction and the quality of precipitation or temperature forcing data (elaborating on these concerns is beyond the scope of the current study). To properly respond to these challenges, model calibration was completed in OSTRICH (Optimization Software Toolkit for Research Involving Computational Heuristics), which is a model-independent and multi-algorithm optimization tool (Matott 2017). The toolkit, which supports both single- and multicriteria optimization options, can be used for the weighted nonlinear least squares calibration of the model parameters or for constrained optimization of a set of design variables according to predefined cost functions. OSTRICH can incorporate different algorithms to search for the optimal value of the objective functions and to identify the set of parameter values associated with such optima. There are several optimization algorithms available in the toolkit, which can be classified as either deterministic local or heuristic global search methods incorporating elements of structured randomness. For multicriteria optimization, the Pareto Archive Dynamically Dimensioned Search (PA-DDS; Asadzadeh and Tolson 2009, 2013) and the simple multiobjective optimization heuristic algorithms are available, while for uncertainty-based calibration, several sampling-based algorithms (i.e., generalized likelihood uncertainty estimation and Metropolis–Hastings Markov chain Monte Carlo) are available. In addition, the asynchronous parallel processing architecture provided by OSTRICH, which is based on the industry standard Message Passing Interface (MPI), provided the means to speed up the calibration procedure.
The model was calibrated for the period of 2010–18 using PA-DDS by maximizing the Kling–Gupta efficiency (KGE; Gupta et al. 2009) and minimizing the root-mean-square errors (RMSE). HYDROTEL was forced with CaPA-RDPA and meteorological data, including daily precipitation and maximum and minimum temperatures time series described in Table 1, as well as snow survey observations provided in Table 3. Daily historical discharge data measured at the location of available hydrometric stations described in Table 2 and identified in Fig. 2 were obtained from WSC, while the reconstructed inflows were calculated and used for model calibration.
b. Impact of snow data assimilation and CaPA-RDPA forcing
To investigate the impact of SWE assimilation on model performance, and also to understand how robust the accuracy of CaPA-RDPA products were over the three study basins for hydrologic application purposes, two separate sets of modeling experiments were designed. In the first set (experiment Set 1), the model was trained with forcing CaPA-RDPA, while in the second set (experiment Set 2), MSC meteorological data were used as input. Depending on whether the GMON and snow survey monitoring information were assimilated during the calibration and the “stand-alone” run (i.e., when the model runs once the calibration is completed), two separate runs were considered for each set (see Table 5). In Expt 1.1, the model was calibrated while assimilating SWE measurements. The assimilation was then switched off and the calibrated model was forced with CaPA-RDPA once again for the same time period (2010–18) (Expt 1.2). This experiment was designed to indicate the extent by which the model would be able to preserve the flow estimation accuracy with forcing precipitation analysis products only. The second set of experiments (Expt 2.1 and Expt 2.2) is similar to the first set, except that CaPA-RDPA data were replaced with gauged in situ meteorological forcing. For each experiment, goodness-of-fit metrics can be used to quantitatively measure the representativeness of the experimental flow estimations to the hydrometric observations (the metrics used in this study can be found in the online supplemental material). Such an evaluation helped us perform an intercomparison of the results between the two sets of experiments.
Application of snow assimilation during the experiments. Cross: snow assimilation was not performed during the stand-alone run; check mark: snow assimilation was performed during the calibration/stand-alone run.
c. Network assessment
Depending on whether the former assessment of the CaPA-RDPA forcing in HYDROTEL may suggest if the gridded analysis products can be adequately used for streamflow simulation, a simple network density sensitivity analysis based on CaPA gridded products was proposed for flow simulation in HYDROTEL. Such an assessment was designed to guide future network assessment procedures. Therefore, a network assessment procedure similar to that of Abbasnezhadi et al. (2019) was followed here, except that the assessment did not include artificially generated reference fields. Rather, a subset of grid points was extracted to create network scenarios of different resolutions from the RDPA domain over each basin, while the respective precipitation analysis was directly used during the assessment. Such an uncontrolled framework could be specifically useful for the case of this study as the SWE DA-CaPA coupling could prove to output such streamflow estimation that could closely match flow observations. Sampling grids (Θν), where ν is the resolution of the pseudo network in decimal arc-degrees, pertaining to each study basin are defined in Table 6 (refer to the online supplemental material to see individual scenarios for each basin).
Sampling grid of pseudo-network scenarios Θν of different resolutions in decimal arc-degrees ν for each study basin, extracted from the CaPA grid. See individual scenarios in the online supplemental material.
4. Results and discussion
a. Sensitivity analysis and model calibration
The results of the sensitivity analysis provided by VARS indicated that among the parameters used to regulate the vertical water budget, the second soil layer thickness (Z2), which affects flow peaks, is a sensitive parameter. The third soil layer thickness (Z3), which mostly affects baseflow, was identified to be a less sensitive parameter in this group. Also, the recession coefficient (CR), which affects summer baseflow and works with Z3, was found to be a relatively sensitive parameter. Among the parameters used for calculating the weighted mean of the nearest three stations, VARS indicated that the third parameter in this group (PPN) has more impact on the results, and the first two (GT and GP) are almost equal in sensitivity. Also, for the snow processes, the melting temperature thresholds and rates for all three land classes in this group (SFC, SFF, SFD, TFC, TFF, TFD) were shown to have equal sensitivity levels. Both glacier melting parameters (MR and TT) were found to be sensitive too, and the multiplicative coefficient (FETP) applied to the Penman–Monteith equation was found to be the only sensitive evapotranspiration parameter. None of the parameters related to the flow process at the RHHU scale was found to be sensitive, while any modification to these parameters would force the model to recalculate the HGM file, which would be time consuming. The parameters associated with the channel flow process, computed using the kinematic wave equation, were also not found to be sensitive.
Previous VARS applications performed by Foulon et al. (2019) in two basins in southern Quebec yielded different results for the vertical water budget parameters. Z1 was shown to be the least sensitive soil layer thickness, while Z2 and Z3 were the second most and the most sensitive parameters, respectively. Also, the recession coefficient (CR) was indicated to be one of the most sensitive parameters in the model. This signifies that HYDROTEL is rather sensitive to basin location and governing hydrological processes. In fact, Yukon and southern Quebec are both governed by snow accumulation and melt, yet summer baseflow plays a more prominent role in southern Quebec.
With sensitive parameters in hand, comprising of a set of 16 parameters indicated in Table 4 by those with the importance level of 1, the model was calibrated in OSTRICH. The standard upper and lower bound values used for each parameter in OSTRICH are provided in Table 4, which are based on the physical meaning of each parameter and the works of Fortin et al. (2001b) and Turcotte et al. (2003). Also, the initial estimates for each parameter were based on those derived in previous calibration efforts, in which each parameter was manually adjusted in order to achieve the desired hydrological performance. The toolbox utilized eight computational cores for asynchronous parallel processing at the budget of 2–18 h (depending on the basin’s drainage area) for 1000 iterations.
In Mayo, the model calibration was completed in OSTRICH based on the inflow time series into Mayo Lake associated to YE gauge ##0000003 (see Fig. 2). In Aishihik, the model calibration was completed in two stages. In the first stage, the model was calibrated for Sekulmun River streamflow time series at the outlet of Sekulmun Lake observed at WSC Gauge 08AA008 (see Fig. 2). The Sekulmun portion of the Aishihik model was isolated and separated in HYDROTEL GUI (graphical user interface) to decrease the model run time. In the second stage, the model was setup to simulate the reconstructed inflow time series to Aishihik Lake associated with YE gauge #0000003. The original reconstructed inflow data display high-intensity fluctuations and were not deemed suitable for the calibration. Instead, they were first smoothed by using a 7-day moving average window (windows of longer durations were also tested and did not show to enhance the calibration results). In Upper Yukon, the model calibration was also performed in two stages. In the first stage, the model was calibrated separately for three gauged subbasins, including Atlin River (WSC gauge 09AA006), Tutshi River (WSC gauge 09AA013), and Wheaton River (WSC gauge 09AA012) (see Fig. 2). In the second stage, the model was then setup to simulate the flow time series in Yukon River at Whitehorse observed at WSC gauge 09AB001.
Figure 3 shows the flow duration curves for Mayo, Aishihik (including the Sekulmun subbasin), and Upper Yukon (including the Atlin, Tutshi, and Wheaton subbasins) (refer to the online supplemental material to see discharge time series). In Mayo, the simulation has fully preserved the exceedance probability of observed flows. In Aishihik and Sekulmun, other than some overestimation of winter low flows, the remainder has been well captured by the model. In Upper Yukon, in general, the exceedance probabilities of the simulated flows closely resemble the observed ones although the low flows are underestimated in subbasins with small drainage areas (Tutshi and Wheaton), which has similarly impacted the low flows in Yukon too. In Atlin, the exceedance probability of the observed high flows (corresponding to the flow peaks) is marginally underestimated.
Calibration flow duration curves for different hydrometric stations. Observations are shown as solid lines, and simulations are dashed (refer to the online supplemental material to see flow hydrographs).
Citation: Journal of Hydrometeorology 22, 4; 10.1175/JHM-D-20-0106.1
b. Proxy validation of CaPA-RDPA
The impact of the snow DA routine in HYDROTEL and CaPA-RDPA forcing data on modeling results were assessed based on the set of experiments discussed in section 3b. Figure 4 compares the metrics in Mayo for the first and the second sets of experiments (for the full description of the metrics used in the figures of this section, see the online supplemental material). The metrics reported by the experiments indicate that the calibration results for the case when CaPA-RDPA are used as input (Expt 1.1 and Expt 1.2) surpass, in both cases, those derived by MSC meteorological data (Expt 2.1 and Expt 2.2). In addition, the best outcome is obtained with Expt 1.1 when the model calibration is performed with CaPA-RDPA forcing and the snow DA routine in active mode. Expt 1.2 (CaPA-RDPA forcing and no snow DA), on the other hand, indicates that the model’s performance is not undermined if the snow DA routine is turned off in HYDROTEL (when the model has already been calibrated with the snow DA routine in active mode). In other words, for this experiment, the assimilation of snow monitoring data has relatively no impact on the flow estimation accuracy if CaPA-RDPA data are used as input. In contrast, the metrics obtained from the second set of experiments indicate that when the model is calibrated using MSC meteorological data as input and with the snow DA routine in active mode (Expt 2.1), the metrics are on the ballpark of an acceptable level, while still falling short of those obtained with CaPA-RDPA. However, as Expt 2.2 indicates, if the snow DA routine is turned off, the flow estimation accuracy declines significantly. This illustrates that for the second set of experiments with sparsely gauged meteorological input data, the snow DA routine has a compensating impact on the flow estimation accuracy.
Radial diagram for the performance of the model in response to the set of experiments completed in Mayo (Station ##0000003). NSE, VE, bR2, md, mNSE, and KGE stand for Nash–Sutcliffe efficiency, volumetric efficiency, modified coefficient of determination, modified index of agreement, modified Nash–Sutcliffe efficiency, and Kling–Gupta efficiency, respectively.
Citation: Journal of Hydrometeorology 22, 4; 10.1175/JHM-D-20-0106.1
While the new GMON stations do not provide a long record of measurements yet, the snow course sites in all three basins provide long-enough and continuous records of snow depth and SWE measurements. Results from the second set of experiments shown in Fig. 4 indicate that, in Mayo, these snow course measurements provide valuable information by which the SWE data simulated using the meteorological network can be corrected through the snow assimilation routine in HYDROTEL. In other words, the flow estimation accuracy in Mayo is highly dependent on the external information from the snow survey sites. Although this outcome does not indicate the representativeness of the snow survey sites, it hints at their value. The same debate is found in the literature where hydrological models, for example, are run by interpolating snow depth measurements from a few selected sites to larger areas despite their limited spatial representativeness (Grünewald and Lehning 2015; López-Moreno et al. 2013). Other studies have quantified the issue of snow sites representativeness. For example, Winstral and Marks (2014) proved that an index site representative of the basin conditions can be valid for a basin wide SWE in most years.
On the other hand, the proxy validation of the CaPA-RDPA in Mayo based on the reconstructed inflow associated with gauge ##0000003 shows that the analysis is accurate enough to the extent that would not call for any correction through snow measurements. To this point, these results indicate that in Mayo (i) CaPA-RDPA products can be used for flow estimation, (ii) given the fact that very few precipitation stations are currently assimilated in CaPA, if the current network is extended, the modeling accuracy will improve, and (iii) in the absence of a precipitation observation network with an optimal density, the snow assimilation routine plays a significant role to compensate for proper precipitation information.
Figure 5a compares the metrics in Aishihik for the first and the second sets of experiments, while the performance of the model in response to the set of experiments completed in Sekulmun are shown in Fig. 5b. The results reported for both Aishihik and Sekulmun are not identical to those of Mayo and the experiments rather exhibit a contrasting outcome. While in Mayo, deactivating the snow assimilation routine in HYDROTEL when forcing the model with CaPA-RDPA (Expt 1.2) would marginally impact the metrics relative to the case when the snow assimilation routine was active (Expt 1.1), in Aishihik (including the Sekulmun subbasin), deactivating the snow assimilation routine led the model performance to decay significantly. This suggests that the RDPA gridded products do not encompass the required accuracy over Aishihik, rendering the assimilation of snow readings an essential component for accurate flow estimation. The inadequacy of the RDPA estimates over Aishihik is an indication of the detrimental impact of the sparse precipitation network in Aishihik, which encompass a relatively larger drainage area, on CaPA products over the basin. In Sekulmun, Expt 2.2 provides marginally better results than Expt 2.1, demonstrating that the precipitation measurements taken at the MSC meteorological stations better represent the ground SWE accumulation than those recorded at the snow course sites. Nevertheless, in Sekulmun, when using CaPA-RDPA data as the input, the combined effect of incorporating the value of information from both the external assimilation of precipitation data in CaPA and the internal assimilation of snow readings in HYDROTEL has obviously improved the flow estimation accuracy (see Fig. 5b). In Aishihik, however, Expt 2.1 displays a declined performance relative to Expt 1.1, while Expt 2.1 and Expt 2.2 are relatively identical. These results, in total, revealed that in Aishihik and Sekulmun the snow data are essential for accurate flow estimation if the model is forced with CaPA-RDPA, while the MSC precipitation input data seem to deliver sufficient accuracy (indicating the accuracy of the precipitation measurements taken as MSC stations, which necessitates minimal correction by the data taken at the snow course sites). This, once again, indicates that the value of precipitation information from the MSC precipitation gauges is superior to those of CaPA-RDPA, which illustrates the low accuracy of CaPA data over the basin.
Radial diagrams for the performance of the model in response to the set of experiments completed in Aishihik at (a) Aishihk (Station #0000003) and (b) Sekulmun (Station 08AA008).
Citation: Journal of Hydrometeorology 22, 4; 10.1175/JHM-D-20-0106.1
Figure 6 compares the metrics in Upper Yukon, including those for Atlin, Tutshi, and Wheaton for the first and the second sets of experiments. In Atlin (Fig. 6a), there are marginal differences between the results derived from all four experiments. This agreement could be the outcome of several factors, including (i) collocation of the snow course site and the MSC gauge in Atlin, (ii) existence of an MSC gauge that is assimilated in CaPA (see Fig. 2), forcing the respective RDPA over the basin to become more or less identical to that of gauge reading, and (iii) the impact of the nearby MSC gauges on the northeast side of the basin (just beyond the basin boundary) on the accuracy of precipitation estimate over the basin. In Tutshi and Wheaton, however, a different outcome is evident. The impact of drainage area on the flow estimation accuracy for the given activity state of the snow assimilation routine seems to be a factor of importance. For instance, for a subbasin such as Tutshi (Fig. 6b) with a small drainage area, the impact of the only snow course site in the basin (site 09AA-SC3) on the flow accuracy can be comprehended by the fact that deactivating the snow assimilation in Expt 2.2 has significantly decayed the flow accuracy by almost half. On the other hand, in Wheaton (Fig. 6c), a subbasin with a comparable drainage area to that of Tutshi, in the absence of any snow course site, Expt 2.2 has apparently yielded about the same metrics obtained from Expt 2.1. In general, the results of the experiments performed in Upper Yukon indicate that since the basin generally enjoys a higher number of weather stations (including those assimilated in CaPA and snow course sites), the results demonstrate better metric values.
Radial diagrams for the performance of the model in response to the set of experiments completed in Upper Yukon at (a) Yukon (Station 09AB001), (b) Tutshi (Station 09AA013), (c) Wheaton (Station 09AA012), and (d) Atlin (Station 09AA006).
Citation: Journal of Hydrometeorology 22, 4; 10.1175/JHM-D-20-0106.1
Table 7 summarizes the significance of the snow assimilation routine for each basin for the given meteorological forcing. In short, activating the snow assimilation routine would have a significant impact on the flow estimation only in Mayo when forcing HYDROTEL with the MSC meteorology and in Aishihik when forcing the model with CaPA-RDPA data. Hence, it appears that snow survey sites are more representative of the watershed snow conditions than the meteorological conditions recorded at the MSC stations or embedded into CaPA-RDPA.
Significance of the snow assimilation routine in HYDROTEL given the meteorological forcing for each study basin. Basin denominations are in boldface type, and subbasins are not. A check mark indicates that performing snow assimilation for the selected basin has a significant impact; a cross shows a nonsignificant outcome.
In Upper Yukon, subbasins did not yield consistent results. It was shown that the model does not necessarily need the assimilation of snow products when the model is forced with either gauged or analysis precipitation products (for three of four subregions). While medium-size watersheds (as Tutshi) could benefit from snow survey measurements, the others could not. For larger watershed with denser meteorological networks, snow assimilation may prove to be superfluous. Overall, where snow assimilation significantly improves the results, it can be concluded that the corresponding meteorological forcing does not have the expected accuracy for hydrologic modeling purposes, including the assessment of the meteorological network density, which is the subject of the next analysis in this study.
c. Network sensitivity analysis
The information gained from the validation stage was used to decide whether the assessments should be undertaken with/without the assimilation of snow course data. The proxy validations indicated that at least in Aishihik, CaPA data do not have the required accuracy, while the validations in the other two basins (Mayo and Upper Yukon) were promising. Therefore, in Aishihik, the network assessment was carried out while assimilating the snow course measurements. In Mayo and Upper Yukon, no snow assimilation was performed when evaluating the impact of different network scenarios. Even though any proposed additional station would probably be equipped with various measuring apparatus for different meteorological variables, the network augmentation assessment was carried out with the assumption that the network would be mainly measuring precipitation. This is mainly due to the fact that precipitation demonstrates a lot more spatial variability than other meteorological variables (e.g., temperature, wind).
Figure 7 shows the variation of the NSE, KGE, and absolute PBias scores in Mayo, Aishihik, and Upper Yukon with the changing resolution of the pseudo-network scenarios (for descriptions of the scores, see the online supplemental material). In Mayo (thick lines in all figures), as the network resolution decreases (and so does the network density) from 0.10° to 0.35°, the scores go through two distinct areas of variation. First, decreasing the network resolution from 0.10° to 0.30° results only in marginal drops in all three performance scores. In comparison, the performance of the CaPA precipitation products for a network with a given resolution of 0.30° or higher is better than that of the current meteorological precipitation network (shown by horizontal lines). The fluctuations and the unexpected drops in performance scores in this range are an artifact of the spatial variability of precipitation that has not been fully resolved by certain grid points. This phenomenon, which is known as singularity, has been reported previously by Abbasnezhadi et al. (2019) and Dong et al. (2005). Decreasing the network density below 0.30°, results in substantial performance deterioration to an extent well below the current sparse MSC network. This indicates that the limit at which the CaPA gridded data can outperform the existing network in Mayo is limited to a network with a density of at least 0.30°.
Variation of the (left) NSE, (center) KGE, and (right) absolute PBias in Mayo (thick solid lines), Aishihik (dashed lines), and Upper Yukon (thin solid lines) based on pseudo-networks (PN) resolutions defined in Table 6. The revenue of the current network (CN) in each basin is also shown (horizontal lines).
Citation: Journal of Hydrometeorology 22, 4; 10.1175/JHM-D-20-0106.1
The variation of the NSE, KGE, and absolute PBias in Aishihik with changing network resolution is shown by dashed lines and compares the performance of the pseudo-network scenarios constructed based on the CaPA grid definition with the current MSC network in the basin. The same overall trend of variation previously observed in Mayo is evident here too where the scores drop (although less abruptly) after negligible changes before the threshold network density. The less sudden drop is an expected attenuation consequence of a larger drainage area that is more evidently manifested by the NSE score, which is known to be a sensitive parameter to peak discharge values [see Abbasnezhadi et al. (2019) for the same performance outcome]. In Aishihik, the network resolution threshold cannot be explicitly inferred. The variation of the NSE indicates that for every decrease in resolution there is a decrease in performance that is rather of the same order of magnitude for all resolutions, whereas those of KGE and PBias assert the 0.4° pseudo-network to entail the optimal resolution below which the accuracy of the ensued flow simulations degrades significantly. Any higher-density network would cause the scores to level off and little would be gained by further increasing the network density. The asserted network density threshold of 0.4° derived for Aishihik resembles the performance established by the current MSC meteorological network in the basin. Moreover, this threshold value is also slightly higher than the one determined for Mayo. This was an anticipated outcome as in basins with a larger drainage area, representativeness errors are averaged out, which makes missing a storm event less impactful on the overall network precision. In contrast, in smaller basins (as in Mayo), mesoscale precipitation systems are essentially significant for capturing proper flow statistics. Accordingly, a higher network threshold value can already be anticipated for Upper Yukon, which has an even larger drainage area than that of Aishihik.
In Upper Yukon (thin lines), the same features previously observed in Mayo and Aishihik are apparent, while a higher network threshold value is resolved. Similar to what was indicated for Aishihik, a network resolution threshold cannot be explicitly inferred in Upper Yukon. Arguably, if Pbias changes are ignored (which asserts the 0.7° pseudo network to entail the optimal resolution), it can be claimed that the 0.5° pseudo network would be optimal. A pseudo network with a density threshold value between 0.5° and 0.7° would as such provide an optimal resolution range. Very interesting is that the current MSC network maintains an accuracy that is comparable in performance to the highest network density of the original CaPA network.
5. Summary and conclusions
This study is at the crossroads between meteorological data assimilation (in which precipitation observations are merged into numerically modeled precipitation data), and hydrological data assimilation (in which snow survey data are merged into streamflow forecast). Before applying assimilated precipitation products in meteorological network assessment, first it is required to validate the accuracy of these products. In this study, it is indicated that since assimilation of snow survey data could provide the benchmark for accurate flow estimation, it would then be possible to evaluate the accuracy of precipitation assimilation products through the proxy validation of precipitation analysis in such a hydrologic system. The HYDROTEL model snow data assimilation (DA) routine is one such example that provides the opportunity to investigate the added value of using the CaPA-RDPA data for application in meteorological network assessment in sparsely gauged Nordic basins.
The hydrologic footprint of CaPA-RDPA data and MSC ground observations were validated against hydrometric observations. This validation was performed to examine whether assimilating snow monitoring information in HYDROTEL can offset the adverse effects of precipitation data scarcity in Yukon. When snow assimilation could significantly improve the flow simulation outcomes, it was concluded that the corresponding meteorological forcing (either CaPA-RDPA data or ground observations; in this instance, MSC stations) could not exclusively provide the required accuracy for hydrologic modeling purposes. The proxy validation of the CaPA-RDPA data indicated that the gridded analysis products enjoy the level of accuracy required for accurate flow simulation in Mayo and Upper Yukon, which does not entail the application of snow assimilation in HYDROTEL. In Aishihik, however, the validations demonstrated that the regional precipitation analysis does not have the required accuracy, and therefore, assimilation of observed snow course information had a significant impact on the flow estimation accuracy. Based on the results of these experiments, it can be concluded that, although these basins are all located within similar ecoclimatic zones in southern Yukon and in the proximity of each other, the distribution of snow course sites and precipitation gauges has left a substantial impact on the accuracy of precipitation and snow assimilation procedures, which directly affects the accuracy of flow simulations. These results indicate the importance of the snow assimilation routine in HYDROTEL to embed crucial information not readily available from precipitation forcing data. This approach and the lessons learned may also benefit watersheds in other parts of the world facing similar challenges related to incorporating accurate data when such information is not embedded within the forcing data.
With the experiments in hand, a network augmentation assessment was carried out subsequently by incorporating the value of data and products available from the CaPA assimilation system with the assumption that the network would be mainly measuring precipitation. The assessment indicated that a number of additional stations can be installed in each basin to increase the accuracy for streamflow simulation. It is worth reiterating that the analysis was performed based on CaPA-RDPA data and having real measurements on the ground could prove to require fewer stations, especially for Aishihik and Mayo. In addition, the network was assessed in an uncontrolled mode where no observation error was added during the analysis to simulate the impact of such errors (including those related to solid precipitation in winter and convective storms during summer). Instead, CaPA-RDPA data were used directly into the assessment since the assumption of accuracy was validated prior to undertaking the assessment. Given that in the CaPA system, precipitation measurements are subjected to various quality control (QC) procedures before being assimilated, the RDPA products can, therefore, be assumed to be of relatively proper quality. However, the implication of such an assumption is that, the optimal number of stations derived for each basin is valid when those stations satisfy CaPA QC procedures too. In other words, if the quality of measurements available from the proposed extended network can satisfy CaPA QC, they could equally benefit the CaPA system. Moreover, it is ultimately beneficial if any additional precipitation station that can be directly used for flow forecasting in HYDROTEL may also be used for the similar purpose indirectly when embedded into the products of the CaPA assimilation system. Also, if existing snow survey sites could provide the required SWE data for hydrologic snow assimilation, the framework introduced in this study could be easily implemented. Otherwise, in case a network assessment is to be undertaken in a basin where such data are not readily available, proper arrangements should be made to first conduct snow surveys.
Acknowledgments
The authors gratefully acknowledge the contributions of the anonymous reviewers to the final form of the paper. The authors are also grateful for the financial support from Natural Sciences and Engineering Research Council of Canada (NSERC) and Yukon Energy (YE) through Collaborative (CRDPJ 499954-16) and Applied (CARD2 500263-16) Research and Development grants. This project would not have been possible without substantial contributions from staffs at Yukon Research Centre, namely, Brian Horton and Maciej Stetkiewicz; at INRS, Sébastien Tremblay; and at YE, Shannon Mallory, Kevin Maxwell, and Andrew Hall. We also acknowledge the following organizations for their readily available online data used in this study: Meteorological Survey of Canada, Water Survey of Canada, Environment and Climate Change Canada, Water Resources Branch at Environment Yukon, Yukon Energy, and Natural Resources Conservation Service at U.S. Department of Agriculture. The authors declare no conflict of interests in this work.
REFERENCES
Abbasnezhadi, K., A. N. Rousseau, K. A. Koenig, Z. Zahmatkesh, and A. M. Wruth, 2019: Hydrological assessment of meteorological network density through data assimilation simulation. J. Hydrol., 569, 844–858, https://doi.org/10.1016/j.jhydrol.2018.12.027.
Andreadis, K. M., and D. P. Lettenmaier, 2006: Assimilating remotely sensed snow observations into a macroscale hydrology model. Adv. Water Resour., 29, 872–886, https://doi.org/10.1016/j.advwatres.2005.08.004.
Arulampalam, M. S., S. Maskell, N. Gordon, and T. Clapp, 2002: A tutorial on particle filters for on-line nonlinear/non-Gaussian Bayesian tracking. IEEE Trans. Sig. Proc., 50, 174–188, https://doi.org/10.1109/78.978374.
Asadzadeh, M., and B. Tolson, 2009: A new multi-objective algorithm, Pareto archived DDS. 11th Annual Conf. on Genetic and Evolutionary Computation Conf. (GECCO 2009), New York, NY, Association for Computing Machinery, 1963–1966, https://doi.org/10.1145/1570256.1570259.
Asadzadeh, M., and B. Tolson, 2013: Pareto archived dynamically dimensioned search with hypervolume-based selection for multi-objective optimization. Eng. Optim., 45, 1489–1509, https://doi.org/10.1080/0305215X.2012.748046.
Barrett, A. P., 2003: National Operational Hydrologic Remote Sensing Center Snow Data Assimilation System (SNODAS) products at NSIDC. NSIDC Special Rep. 11, 19 pp., https://nsidc.org/sites/nsidc.org/files/technical-references/nsidc_special_report_11.pdf.
Boluwade, A., K.-Y. Zhao, T. Stadnyk, and P. Rasmussen, 2018: Towards validation of the Canadian Precipitation Analysis (CaPA) for hydrologic modeling applications in the Canadian Prairies. J. Hydrol., 556, 1244–1255, https://doi.org/10.1016/j.jhydrol.2017.05.059.
Boni, G., F. Castelli, S. Gabellani, G. Machiavello, and R. Rudari, 2010: Assimilation of MODIS snow cover and real time snow depth point data in a snow dynamic model. 2010 IEEE Int. Geoscience and Remote Sensing Symp., Honolulu, HI, IEEE, 1788–1791, https://doi.org/10.1109/IGARSS.2010.5648989.
Bouda, M., A. N. Rousseau, B. Konan, P. Gagnon, and S. J. Gumiere, 2012: Bayesian uncertainty analysis of the distributed hydrological model HYDROTEL. J. Hydrol. Eng., 17, 1021–1032, https://doi.org/10.1061/(ASCE)HE.1943-5584.0000550.
Bouda, M., A. N. Rousseau, S. J. Gumiere, P. Gagnon, B. Konan, and R. Moussa, 2014: Implementation of an automatic calibration procedure for HYDROTEL based on prior OAT sensitivity and complementary identifiability analysis. Hydrol. Processes, 28, 3947–3961, https://doi.org/10.1002/hyp.9882.
Brabets, T. P., and M. A. Walvoord, 2009: Trends in streamflow in the Yukon River Basin from 1944 to 2005 and the influence of the pacific decadal oscillation. J. Hydrol., 371, 108–119, https://doi.org/10.1016/j.jhydrol.2009.03.018.
Brasnett, B., 1999: A global analysis of snow depth for numerical weather prediction. J. Appl. Meteor., 38, 726–740, https://doi.org/10.1175/1520-0450(1999)038<0726:AGAOSD>2.0.CO;2.
Clark, M. P., A. G. Slater, A. P. Barrett, L. E. Hay, G. J. Mccabe, B. Rajagopalan, and G. H. Leavesley, 2006: Assimilation of snow covered area information into hydrologic and land-surface models. Adv. Water Resour., 29, 1209–1221, https://doi.org/10.1016/j.advwatres.2005.10.001.
Deacu, D., V. Fortin, E. Klyszejko, C. Spence, and P. Blanken, 2012: Predicting the net basin supply to the Great Lakes with a hydrometeorological model. J. Hydrometeor., 13, 1739–1759, https://doi.org/10.1175/JHM-D-11-0151.1.
De Lannoy, G. J. M., R. H. Reichle, K. R. Arsenault, P. R. Houser, S. Kumar, N. E. C. Verhoest, and V. R. N. Pauwels, 2012: Multiscale assimilation of Advanced Microwave Scanning Radiometer-EOS snow water equivalent and Moderate Resolution Imaging Spectroradiometer snow cover fraction observations in northern Colorado. Water Resour. Res., 48, W01522, https://doi.org/10.1029/2011WR010588.
Dong, X., C. M. Dohmen-Janssen, and M. J. Booij, 2005: Appropriate spatial sampling of rainfall or flow simulation. Hydrol. Sci. J., 50, 279–298, https://doi.org/10.1623/hysj.50.2.279.61801.
Drusch, M., D. Vasiljevic, and P. Viterbo, 2004: ECMWF’s global snow analysis: Assessment and revision based on satellite observations. J. Appl. Meteor., 43, 1282–1294, https://doi.org/10.1175/1520-0450(2004)043<1282:EGSAAA>2.0.CO;2.
Durand, M., and S. A. Margulis, 2008: Effects of uncertainty magnitude and accuracy on assimilation of multi-scale measurements for snowpack characterization. J. Geophys. Res., 113, D02105, https://doi.org/10.1029/2007JD008662.
Eum, H., D. Yonas, and T. Prowse, 2014: Uncertainty in modelling the hydrologic responses of a large watershed: A case study of the Athabasca River Basin, Canada. Hydrol. Processes, 28, 4272–4293, https://doi.org/10.1002/hyp.10230.
Evensen, G., 1994: Sequential data assimilation with a nonlinear quasi-geostrophic model using Monte Carlo methods to forecast error statistics. J. Geophys. Res., 99, 10 143–10 162, https://doi.org/10.1029/94JC00572.
Fortin, J. P., R. Turcotte, S. Massicotte, R. Moussa, J. Fitzback, and J. P. Villeneuve, 2001a: Distributed watershed model compatible with remote sensing and GIS data, I: Description of model. J. Hydrol. Eng., 6, 91–99, https://doi.org/10.1061/(ASCE)1084-0699(2001)6:2(91).
Fortin, J. P., R. Turcotte, S. Massicotte, R. Moussa, J. Fitzback, and J. P. Villeneuve, 2001b: A distributed watershed model compatible with remote sensing and GIS data, II: Application to Chaudière Watershed. J. Hydrol. Eng., 6, 100–108, https://doi.org/10.1061/(ASCE)1084-0699(2001)6:2(100).
Fortin, V., G. Roy, N. Donaldson, and A. Mahidjiba, 2015: Assimilation of radar quantitative precipitation estimations in the Canadian Precipitation Analysis (CaPA). J. Hydrol., 531, 296–307, https://doi.org/10.1016/j.jhydrol.2015.08.003.
Foulon, E., M. Blanchette, and A. N. Roussea, 2019: Sensitivity and uncertainty analysis of a semi-distributed model used for the quantitative assessment of watershed-scale hydrological services of wetlands. 2019 Fall Meeting, San Francisco, CA, Amer. Geophys. Union, Abstract H53L-1951, https://ui.adsabs.harvard.edu/abs/2019AGUFM.H53L1951F/abstract.
Gbambie, A. S. B., A. Poulin, M. A. Boucher, and R. Arsenault, 2017: Added value of alternative information in interpolated precipitation datasets for hydrology. J. Hydrometeor., 18, 247–264, https://doi.org/10.1175/JHM-D-16-0032.1.
Gelb, A., 1974: Optimal linear filtering. Applied Optimal Estimation, MIT Press, 102–155.
Grünewald, T., and M. Lehning, 2015: Are flat-field snow depth measurements representative? A comparison of selected index sites with areal snow depth measurements at the small catchment scale. Hydrol. Processes, 29, 1717–1728, https://doi.org/10.1002/hyp.10295.
Gupta, H. V., H. Kling, K. K. Yilmaz, and G. F. Martinez, 2009: Decomposition of the mean squared error and NSE performance criteria: Implications for improving hydrological modelling. J. Hydrol., 377, 80–91, https://doi.org/10.1016/j.jhydrol.2009.08.003.
Haghnegahdar, A., B. A. Tolson, B. Davison, F. R. Seglenieks, E. Klyszejko, and E. D. Soulis, 2014: Calibrating Environment Canada’s MESH modelling system over the Great Lakes basin. Atmos.–Ocean, 52, 281–293, https://doi.org/10.1080/07055900.2014.939131.
Hanes, C. C., P. Jain, M. Flannigan, and G. Roy, 2016: Evaluation of the Canadian Precipitation Analysis (CaPA) to improve forest fire danger rating. Int. J. Wildland Fire, 26, 509–522, https://doi.org/10.1071/WF16170.
Helmert, J., and Coauthors, 2018: Review of snow data assimilation methods for hydrological, land surface, meteorological and climate models: Results from a COST HarmoSnow survey. Geosciences, 8, 489, https://doi.org/10.3390/geosciences8120489.
Huang, C., A. J. Newman, M. P. Clark, A. W. Wood, and X. Zheng, 2017: Evaluation of snow data assimilation using the ensemble Kalman filter for seasonal streamflow prediction in the western United States. Hydrol. Earth Syst. Sci., 21, 635–650, https://doi.org/10.5194/hess-21-635-2017.
Leisenring, M., and H. Moradkhani, 2011: Snow water equivalent prediction using Bayesian data assimilation methods. Stochastic Environ. Res. Risk Assess., 25, 253–270, https://doi.org/10.1007/s00477-010-0445-5.
Li, D., D. P. Lettenmaier, S. A. Margulis, and K. Andreadis, 2019: The value of accurate high-resolution and spatially continuous snow information to streamflow forecasts. J. Hydrometeor., 20, 731–749, https://doi.org/10.1175/JHM-D-18-0210.1.
Liston, G. E., R. A. Sr. Pielke, and E. M. Greene, 1999: Improving first-order snow-related deficiencies in a regional climate model. J. Geophys. Res., 104, 19 559–19 567, https://doi.org/10.1029/1999JD900055.
Liu, Y., C. D. Peters-Lidard, S. Kumar, J. L. Foster, M. Shaw, Y. Tian, and G. M. Fall, 2013: Assimilating satellite-based snow depth and snow cover products for improving snow predictions in Alaska. Adv. Water Resour., 54, 208–227, https://doi.org/10.1016/j.advwatres.2013.02.005.
López-Moreno, J. I., S. R. Fassnacht, J. T. Heath, K. N. Musselman, J. Revuelto, J. Latron, E. Morán-Tejeda, and T. Jonas, 2013: Small scale spatial variability of snow density and depth over complex alpine terrain: Implications for estimating snow water equivalent. Adv. Water Resour., 55, 40–52, https://doi.org/10.1016/j.advwatres.2012.08.010.
Magnusson, J., D. Gustafsson, F. Hüsler, and T. Jonas, 2014: Assimilation of point SWE data into a distributed snow cover model comparing two contrasting methods. Water Resour. Res., 50, 7816–7835, https://doi.org/10.1002/2014WR015302.
Mahfouf, J. F., B. Brasnett, and S. Gagnon, 2007: A Canadian precipitation analysis (CaPA) project: Description and preliminary results. Atmos.–Ocean, 45, 1–17, https://doi.org/10.3137/ao.v450101.
Matott, L. S., 2017: OSTRICH: An optimization software tool. Documentation and user’s guide, version 17.12.19. University at Buffalo Center for Computational Research Doc., 79 pp., www.eng.buffalo.edu/~lsmatott/Ostrich/OstrichMain.html.
Miller, R. N., M. Ghil, and F. Gauthiez, 1994: Advanced data assimilation in strongly nonlinear dynamical systems. J. Atmos. Sci., 51, 1037–1056, https://doi.org/10.1175/1520-0469(1994)051<1037:ADAISN>2.0.CO;2.
Moradkhani, H., 2008: Hydrologic remote sensing and land surface data assimilation. Sensors, 8, 2986–3004, https://doi.org/10.3390/s8052986.
Nagler, T., H. Rott, P. Malcher, and F. Müller, 2008: Assimilation of meteorological and remote sensing data for snowmelt runoff forecasting. Remote Sens. Environ., 112, 1408–1420, https://doi.org/10.1016/j.rse.2007.07.006.
Noël, P., A. N. Rousseau, C. Paniconi, and D. F. Nadeau, 2014: An algorithm for delineating and extracting hillslopes and hillslope width functions from gridded elevation data. J. Hydrol. Eng., 19, 366–374, https://doi.org/10.1061/(ASCE)HE.1943-5584.0000783.
Piazzi, G., G. Thirel, L. Campo, and S. Gabellani, 2018: A particle filter scheme for multivariate data assimilation into a point-scale snowpack model in an Alpine environment. Cryosphere, 12, 2287–2306, https://doi.org/10.5194/tc-12-2287-2018.
Pierre, A., S. Jutras, S. Craig, J. Kochendorfer, V. Fortin, and F. Anctil, 2019: Evaluation of catch efficiency transfer functions for unshielded and single-alter-shielded solid precipitation measurements. J. Atmos. Oceanic Technol., 36, 865–881, https://doi.org/10.1175/JTECH-D-18-0112.1.
Razavi, S., R. Sheikholeslami, H. V. Gupta, and A. Haghnegahdar, 2019: VARS-TOOL: A toolbox for comprehensive, efficient, and robust sensitivity and uncertainty analysis. Environ. Modell. Software, 112, 95–107, https://doi.org/10.1016/j.envsoft.2018.10.005.
Rousseau, A. N., J.-P. Fortin, R. Turcotte, A. Royer, S. Savary, F. Quévy, P. Noël, and C. Paniconi, 2011: PHYSITEL, A specialized GIS for supporting the implementation of distributed hydrological models. Water News, Vol. 31, No. 1, Canadian Water Resources Association, Calgary, AB, Canada, 18–20.
Saloranta, T. M., 2016: Operational snow mapping with simplified data assimilation using the seNorge snow model. J. Hydrol., 538, 314–325, https://doi.org/10.1016/j.jhydrol.2016.03.061.
Samuel, J., A. N. Rousseau, K. Abbasnezhadi, and S. Savary, 2019: Development and evaluation of a hydrologic data-assimilation scheme for short-range flow and inflow forecasts in a data-sparse high-latitude region using a distributed model and ensemble Kalman filtering. Adv. Water Resour., 130, 198–220, https://doi.org/10.1016/j.advwatres.2019.06.004.
Slater, A. G., and M. P. Clark, 2006: Snow data assimilation via an ensemble Kalman filter. J. Hydrometeor., 7, 478–493, https://doi.org/10.1175/JHM505.1.
Strong, W. L., 2013: Ecoclimatic zonation of Yukon (Canada) and ecoclinal variation in vegetation. Arctic, 66, 52–67, https://doi.org/10.14430/arctic4266.
Su, H., Z. L. Yang, G. Y. Niu, and R. E. Dickinson, 2008: Enhancing the estimation of continental-scale snow water equivalent by assimilating MODIS snow cover with the ensemble Kalman filter. J. Geophys. Res., 113, D08120, https://doi.org/10.1029/2007JD009232.
Tolson, B. A., and C. A. Shoemaker, 2007: Dynamically dimensioned search algorithm for computationally efficient watershed model calibration. Water Resour. Res., 43, W01413, https://doi.org/10.1029/2005WR004723.
Turcotte, R., J.-P. Fortin, A. N. Rousseau, S. Massicotte, and J.-P. Villeneuve, 2001: Determination of the drainage structure of a watershed using a digital elevation model and a digital river and lake network. J. Hydrol., 240, 225–242, https://doi.org/10.1016/S0022-1694(00)00342-5.
Turcotte, R., A. N. Rousseau, J.-P. Fortin, and J.-P. Villeneuve, 2003: A process-oriented, multiple-objective calibration strategy accounting for model structure. Advances in Calibration of Watershed Models, Q. Duan et al., Eds., Water Science and Applications Series, Vol. 6, Amer. Geophys. Union, 153–163, https://agupubs.onlinelibrary.wiley.com/doi/book/10.1029/WS006.
Turcotte, R., L. G. Fortin, V. Fortin, J.-P. Fortin, and J.-P. Villeneuve, 2007: Operational analysis of the spatial distribution and the temporal evolution of the snowpack water equivalent in southern Québec, Canada. Nord. Hydrol., 38, 211–234, https://doi.org/10.2166/nh.2007.009.
Turcotte, R., T.-C. F. Filion, P. Lacombe, V. Fortin, A. Roy, and A. Royer, 2010: Simulation hydrologique des derniers jours de la crue de printemps: Le problème de la neige manquante. Hydrol. Sci. J., 55, 872–882, https://doi.org/10.1080/02626667.2010.503933.
Winstral, A., and D. Marks, 2014: Long-term snow distribution observations in a mountain catchment: Assessing variability, time stability, and the representativeness of an index site. Water Resour. Res., 50, 293–305, https://doi.org/10.1002/2012WR013038.
Wong, J. S., S. Razavi, B. R. Bonsal, H. S. Wheater, and Z. E. Asong, 2017: Inter-comparison of daily precipitation products for large-scale hydro-climatic applications over Canada. Hydrol. Earth Syst. Sci., 21, 2163–2185, https://doi.org/10.5194/hess-21-2163-2017.
Xu, X., S. K. Frey, A. Boluwade, A. R. Erler, O. Khader, D. R. Lapen, and E. Sudicky, 2019: Evaluation of variability among different precipitation products in the Northern Great Plains. J. Hydrol. Reg. Stud., 24, 100608, https://doi.org/10.1016/j.ejrh.2019.100608.
Zhao, K.-Y., 2013: Validation of the Canadian Precipitation Analysis (CaPA) for hydrological modelling in the Canadian Prairies. M.S. thesis, Dept. of Civil Engineering, University of Manitoba, 173 pp.
