3.2.1. Humber River

The run-off results for Humber Village for the winter season December 2017–May 2018 show very good agreemen<sup>t</sup> with the measured run-off reference values (Figure 7). Assimilation of the in situ information, mainly adjusting the solid precipitation, provide a perfect match of the volume of water. For the entire region, the amount of water stored as snow was perfectly represented by the model. The volume of total water released in the winter season 2017/18 was modelled by 95% (80% without assimilation). The coefficient of determination (**R**2) reaches 0.90 at Humber Village. The results for the upper part of the catchment down to Reidville also show a very reasonable accordance between measured and modelled run-off behaviour (Figure 8). Due to invalid run-off measurements during the first flood peak in January 2018, no full analyses of the period were performed. However, the volume of total water released in up to Reidville was modeled by 90% (65% without assimilation). The coefficient of determination (**R**2) reaches 0.80 at Reidville.

**Figure 7.** Daily run-off results from PROMET without (grey), with assimilation (blue) of stations and EO against the values (dashed red) measured at Humber Village (catchment: 7600 km2) for the winter season 2017/18.

**Figure 8.** Daily run-off results from PROMET without (grey), with assimilation (blue) of stations and EO against the values (dashed red) measured at Reidville (catchment: 2100 km2) for the winter season 2017/18.

## 3.2.2. Exploits River

The analyzed run-off results for the down-stream gauges "Below Noel Pauls Brook" and "Charlie Edwards Point" at the Exploits River show a very notable agreemen<sup>t</sup> regarding timing and volume with the measured values in the winter season from December 2017–May 2018 (Figures 9 and 10). Due to the strong influence of the controlled water release at Millertown Dam, and the not daily adapted model parameters during the demo, the results are a little below the ones for Humber River. The coefficient of determination (**R**2) reaches 0.80 at Noel Paul Brook. For the downstream section of the Exploits River, after the conjunction with Badger River, the occurrence of river ice impeded the full evaluation of the results. Within the period of mid-December to late April, the measurement point was affected by ice jams, which resulted in an increased water level and therefore a false determination of the run-off. Due to the thawing in April, the flood peak from the snow melt could be well compared (Figure 10).

#### *3.3. Advantages and Potential Limitations*

Regarding the SnowSense in situ station, it is capable of measuring reliably SWE and LWC using freely available GNSS signals and low- cost GNSS sensors. Besides the derivation of SWE and LWC, Koch et al. [8] recently presented an approach to additionally derive snow height, which might even extend the range of application, not only for hydrological targets. A grea<sup>t</sup> advantage of the SnowSense in situ station is its light-weight design making an easy transportation and installation possible, which is highly valuable especially for remote and difficult to access areas. In total, only two people are needed for the set-up and all components can be carried in a big backpack. As the stations have an integrated on-board-processing module and satellite communication capabilities, the results can be transmitted (sub-)daily to the users. This makes the station autonomous and guarantees low maintenance. In general, the in situ station can either be used as a stand-alone component for snow cover property determination or can easily be integrated in the entire service encompassing the EO and modelling components.

**Figure 9.** Daily run-off results from PROMET without (grey), with assimilation (blue) of stations and EO against the values (dashed red) measured at Noel Pauls Brook (catchment: 7400 km2) for the winter season 2017/18.

**Figure 10.** Daily run-off results from PROMET without (grey), with assimilation (blue) of stations and EO against the values (dashed red) measured at Charlie Edwards Point (catchment: 8850 km2) for the winter season 2017/18. Due to river ice for the period mid-December to late April, the measurements have been separated into measurements with ice cover (normal dashed line) and without (bold dashed line).

As GNSS signals are globally available, the application of such in situ stations is potentially possible all over the world. However, a potential restriction of the in situ stations might be the availability of satellite reception in extreme locations, e.g., in narrow alpine valleys or in dense forests, with reduced GNSS signal reception. As presented in Lamm et al. [26], the integration of Galileo satellites besides GPS satellites increases satellite availability markedly, which increases also the

availability in potentially difficult areas. Further studies will focus in more detail on, e.g., tilted terrain like avalanche-prone slopes, different climate and altitudinal ranges as well as more challenges on the station design regarding its protection from wildlife or its mounting on top of bare rocks and ice. Until now, we were able to test the GNSS SWE-derivation for quite huge amounts of snow with up to 1000 mm w.e. in high alpine regions [8]. In this study, we reached SWE values of up to 500 mm w.e. However, further studies on even more extreme amounts of snow and especially wet snow will be conducted in the future, as the limited operational time span of the demonstration project did not provide an opportunity to test the sensor performance for such extreme events performance. This is also true for further tests on extreme temperatures for the entire sensor hardware, which is designed for minimum temperatures of −40 ◦C, though temperatures down to −35 ◦C were reached in December 2017 and January 2018 at the NEIGE testsite at Forêt Montmorency (Figure 6b).

Regarding the different SWE measurement techniques applied at the NEIGE study site at Forêt Montmorency, the GMON sensor is based on passive gamma rays, whereas the GNSS based measurements are based on electromagnetic waves. Both techniques are capable of deriving SWE in good accordance with standard measurement techniques like manual measurements or snow pillows [6,8,46]. Of course, slight differences in the derivation of SWE might occur considering these two relatively new sensors, however, as the sensors are not installed at the exact same place and are located up to 25 m from each other and have a distance (up to 150 m) to the manual measurements. The main differences might originate in different amounts of snow at each location, e.g., due to different wind effects and the different physical principles of the measurements.

Regarding the spatially distributed components EO and the hydrological model, it is another big advantage that both can also be potentially used worldwide and are often free of charge as, e.g., the Sentinel data. Of course, remote sensing products might be restricted in temporal and spatial resolution and face, depending on the wavelength and if the systems are active or passive, different limitations as, e.g., cloud cover or foreshortening effects. Therefore, it is often more difficult to apply EO in mountainous terrain. The applied hydrological model PROMET was already tested and validated for various applications for small and large catchments (e.g., [22,32,33,47]) and also globally (e.g., [48]) in different temporal and spatial resolutions. Although there are a few limitations in the model setup like, for example, the difficulty of implementing small-scale features regarding snow variability or the run-off generation in extreme alpine surroundings, the modelled output provides very good results for different scales, also in case of sparse input data as it is the case in Newfoundland. Especially in those remote areas with sparse data, real-time information and forecasts of run-off, fresh water availability, SWE, snow extent and the snowmelt onset can be significantly improved. Up to now, a limitation in our hydrological model setup is the lake ice formation. In general, SWE is calculated from the snow cover on the ground. However, until now, snow accumulation on top of frozen lakes is not implemented, although this SWE also contributes to the run-off after the onset of snow- and ice-melt. Moreover, we aim to describe river ice formation as an additional feature in the model since it can build up to ice dams with a subsequent banking up of water masses. These two model improvements will be implemented in the future, e.g. as suggested in [49,50].

The big advantage of the entire combined SnowSense service is that it picks up the advantages of all three components to deliver a reliable, assimilated product of snow and hydrology. In case data of one component is missing, the service can still rely on two other pillars and is therefore less vulnerable to data losses or other failures. The service can be used as entire combined system relying on the three pillars in situ measurements, EO, and hydrological modelling, but each of these pillars can also be applied as a modular stand-alone solution if desired. It therefore enhances and combines existing solutions and is, due to its modularity, a customer friendly approach.

#### *3.4. Demo User Feedback*

The feedback from the two demo users, Nalcor/NL Hydro and the Water Resources Management Division in the Department of Municipal Affairs and Environment (WRMD) of the Government of Newfoundland and Labrador, confirmed the significant importance of snow and run-off monitoring, which could be improved by using such a combined approach like the SnowSense service. They underlined that this is especially important for remote locations, as many regions in Newfoundland are only accessible by helicopter or snowmobiles and face limited access possibilities due to poor weather conditions. From their experience, standard in situ measurements sometimes fail, as the snowpack is often icy with a varying snow density and snow hardness, which makes it difficult to measure SWE with manual snow core equipment. Normally, manual snow surveys are performed three times per winter, which is expensive and is difficult to be completed under bad weather conditions, and causes labour safety risks. Existing automatic instrumentation (GMON sensors) is not providing enough information. The demo users state a need for accurate, reliable SWE information for each hydrological watershed and runoff forecasting that affects the real-time scheduling of hydrological assets and minimizes the use of thermal heat generation. From their point of view, a network of stations across the island of Newfoundland and Labrador would be the ideal scenario to give a province wide estimate of SWE for all users. In addition, they confirmed that SnowSense has the potential to fill a gap regarding the provision of spatial and temporal data at a high resolution by applying EO and modelling for spatial SWE maps of the entire region and run-off information at specific points of interest for real-time and forecasts. The received products matched with other sources of information, which they had for comparison and could even provide insights in hydrological processes. Both users stated that the in situ stations are competitive in their operation compared to other SWE monitoring technologies and therefore they have the potential to replace existing SWE monitoring stations like snow pillows or manual field observations.

In general, the feedback provided by the demo users was very positive, which encourages us in our further developments and improvements.
