1. Introduction
Chilean Patagonia contains some of the largest water reserves in the southern cone of South America. Tied with economic growth, increasing energy demands have turned public attention to the region’s rivers (the largest in Chile) and their yet untapped hydropower potential. Although the statistical properties of flow for the major rivers in the area are known because of a reasonably comprehensive hydrometric network, the processes responsible for the hydrology of Patagonia are poorly understood, and so the reliability of historical information for streamflow prediction cannot be evaluated. In light of ongoing environmental change, hydrological investigations are crucial to achieve a better understanding of environmental systems dynamics and their possible response to human interference across spatial and temporal scales. A particular concern is climate change, which has recently resulted in substantial glacier retreat in the region (Rivera et al. 2007) with unknown impacts on river flow.
In spite of these concerns, very few studies of the hydrology and climatology of the region have been conducted. Aravena (2007) developed a 400-yr precipitation reconstruction using tree ring data and glacier fluctuations in the Austral Chilean Andes, finding important decadal variations for the northwest and central Patagonia and also a strong biannual oscillation for the southernmost region. Garreaud et al. (2009) described the mean annual and decadal patterns of precipitation in South America and how they are influenced both by climatic indexes (Pacific decadal oscillation, Antarctic Oscillation, and El Niño–Southern Oscillation) and orographic effects using a large-scale paleoclimatic approach. Lopez et al. (2008) studied variability in snow-covered areas (SCAs) of the Northern Patagonian Ice Field (NPIF) during 2000–06 and correlated SCA to precipitation and air temperature, obtaining the highest correlations for temperature (r2 = 0.75). Rivera et al. (2007) quantified decreases in the NPIF of up to 4.0 ± 0.97 m yr−1 for ice thickness and up to 3.2% ± 1.5% or 140 ± 61 km2 for area over 1975–2001, based on remote sensing and in situ data.
Although the above studies represent improvements on their respective fields, there remains a major gap in published comprehensive hydrological investigations in the region. Dussaillant et al. (2012) present the first description of the main hydrological patterns (precipitation, temperature, and streamflow) of the Baker River basin using observed data. They also discuss the difficulties associated with undertaking this task given the sparse distribution of gauging stations, together with the significant topographical and climatological gradients existing in the region. Barría (2010) developed a statistical approach for obtaining monthly streamflow forecasts for the Baker and Pascua river basins. However, because this research was entirely data driven, it does not increase knowledge of the physical processes governing water movements in the basin. Given the combination of climate change and increased pressure to use the water resources in the region, it becomes paramount for the scientific and decision-making community to increase their knowledge on the hydrological functioning of this relatively pristine environmental system.
Physically based models offer the opportunity of comprehending physical interactions between processes and variables within the hydrological cycle, an advantage that cannot be achieved with other types of models (empirical, conceptual, or statistical, for example). The Cold Regions Hydrological Model (CRHM; Pomeroy et al. 2007) is a physically based model developed at the Centre for Hydrology, University of Saskatchewan, with the aim of improving the understanding of hydrological processes in cold environments, which are particular in the sense that a host of specific phenomena such as snow and ice accumulation, interception, transport and melt, infiltration through frozen soils, and cold water bodies control the hydrograph timing. CRHM has a limited need for calibration (Pomeroy et al. 2007), and most (but not all) of its parameters can be inferred from intensive field or modeling studies. This, together with its modular nature and open structure, makes it particularly suitable for testing hydrological hypothesis in poorly gauged or ungauged basins. Gonthier (2011) developed the first hydrological study in Chile using CRHM. He analyzed three high mountain basins in the Chilean semiarid Andes (32°S), calibrating parameters regarding soil moisture and routing processes against streamflow records. All results showed Nash–Sutcliffe coefficient (NS) values below 0.6 and overestimation of snow accumulation up to 400% with respect to local snow pillow data. Poor modeling results were attributed in part to the very low density of meteorological measurements within the basin and thus great uncertainty in meteorological driving variables. Fang and Pomeroy (2007) developed a CRHM with the aim of understanding the dynamical processes that govern drought phenomena in the Canadian prairies. A sensitivity analysis to meteorological input data was carried out, showing that even under moderate drought scenarios of 15% reduction in winter precipitation and 2.5°C increase in winter mean air temperature, spring runoff may disappear completely. Ellis et al. (2010) developed a CRHM to assess the differences in snowmelt and snow accumulation in forest and clearing sites, achieving an NS model efficiency value of 0.51 for snow water equivalent (SWE), with slightly better representation on clearing sites; these results show the CRHM predictive potential when no calibration is undertaken. Another study was developed by Fang and Pomeroy (2009), who characterized blowing snow redistribution in prairie wetlands, obtaining good results either with an aggregated or a fully distributed spatial representation. Pomeroy et al. (2012) developed a CRHM in a forested mountainous basin with minimal calibration in order to simulate the impacts of forest disturbance in the basin hydrology. Results show different streamflow volume responses for each scenario, ranging from 2% for small forest reduction impacts to 8% for a complete forest-burning event. A recent study developed a CRHM for an alpine to subalpine Canadian Rockies catchment and evaluated uncalibrated model performance against snow accumulation, soil moisture, groundwater, and subbasin and basin streamflow over several years (Fang et al. 2013) with acceptable predictive performance for all variables except groundwater. Tests of CRHM in alpine and steppe environments of the Qinghai Tibetan Plateau show good performance for snowpack, runoff, and streamflow simulation when blowing snow, energy balance snowmelt, and frozen soil infiltration options are used (Zhou et al. 2014).
Atmospheric reanalyses are a scientific method for developing a comprehensive record of weather and climate change over time and can complement the information provided by scarce meteorological observations in remote regions in order to obtain surface meteorology to force land surface models (Sheffield et al. 2004). Quoting Saha et al. (2010, p. 1015), “[t]he general purpose of conducting reanalyses is to produce multiyear global state-of-the-art gridded representations of atmospheric states, generated by a constant model and a constant data assimilation system.” The current generation of reanalyses assimilates data from satellite observations; in situ surface measurements such as 2-m temperature, relative humidity, and wind speed; and upper-atmosphere variables from radiosondes, wind profilers, and aircraft (Dee et al. 2011; Saha et al. 2010). In this study, we test two of the most recent products available: the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) Climate Forecast System Reanalysis (CFSR) and the European Centre for Medium-Range Weather Forecasts (ECMWF) interim reanalysis (ERA-Interim). ERA-Interim is the latest global atmospheric reanalysis produced by the ECMWF, and it covers the period from 1 January 1989 to the present. The gridded data product includes a large variety of 3-hourly surface parameters and 6-hourly upper-air parameters (Dee et al. 2011). This reanalysis has a spatial resolution of 1.5° and 37 pressure levels [increasing by 14 levels from the preceding version 40-yr ECMWF Re-Analysis (ERA-40)]. On the other hand, CFSR spans the 31-yr period from 1979 to 2009. It was designed to be a high-resolution coupled atmosphere–ocean–land surface–sea ice system and to provide the best estimation of the state of these domains over this period (Saha et al. 2010). Its spatial resolution varies from 0.25° at the equator to 0.5° beyond the tropics, with 40 pressure levels. Ward et al. (2011) evaluate several precipitation products [ERA-40, NCEP–NCAR Global Reanalysis 1 (R-1), Precipitation Estimation from Remotely Sensed Imagery Using Artificial Neural Networks (PERSIANN), and Tropical Rainfall Measuring Mission (TRMM)] over the Andes Cordillera (Baker and Paute basins, Chile–Argentina and Ecuador, respectively), comparing them with observed data interpolations based on Thiessen polygons. For the Baker River basin, precipitation products were always above observed data, which can be explained by the low density and low elevation of the station. Ward et al. (2011) also highlighted the secondary importance of the interpolation scheme used, arguing that errors in interpolation are dominated by the low density of rain gauges throughout the Baker basin. Silva et al. (2011) compared CFSR and other NCEP–NCAR reanalyses, that is, R-1 and the NCEP–U.S. Department of Energy (DOE) Second Atmospheric Model Intercomparison Project (AMIP-II; R-2), over South America. Over the Andes Cordillera, in particular, all three reanalyses (CFSR, R-1, and R-2) seem to overestimate total precipitation, but significant improvements are associated with CFSR, probably because of its higher spatial resolution and better representation of regional topography.
The objective of this paper is to describe the first in-depth analysis of the hydrological function of the upper Baker River basin (UBRB), based on a physically based hydrological model (CRHM) driven using meteorological station observations and global meteorological model reanalyses. Hence, the main goals of this research are (i) to describe a plausible combination of hydrological processes giving rise to the observed streamflow record in order to enable future global change impact assessments and (ii) to demonstrate the value of reanalysis data in combination with a physically based hydrologic model in achieving the former goal. In this paper, we also highlight the existing information gaps that preclude a better understanding of the hydrology of the region and suggest future avenues for improving such knowledge.
2. Study site and observations
a. Location
The UBRB is defined by the Bertrand Lake outlet and has an area of 15 904 km2 (see Fig. 1). At this location, the Baker River has a mean annual discharge of 566 m3 s−1 (Fig. 3), being the largest river in Chile when it reaches the ocean. The basin is characterized by very heterogeneous climate, geology, and land cover features, with landscape types that include glaciers and icefields [2787 km2 (17.5%)], rivers and lakes [2109 km2 (13.3%)], dense forest [2777 km2 (17.5%)], grassland and shrubland [5984 km2 (37.6%)], peatlands [66 km2 (0.4%)], and bare rock [2181 km2 (13.7%)] (CONAF/CONAMA 1999; see Fig. 7, described in greater detail below). The regional climate is dominated by the interaction of weather fronts traveling east from the Pacific Ocean with the topographic barrier of the Andes Cordillera. Here, mountains reach from 200 up to 4000 m MSL over a distance of less than 100 km and generate a steep west-to-east precipitation gradient (Warren and Sugden 1993); as a result, precipitation in the mountainous western part of the basin can reach more than 2000 mm annually, whereas the low-lying eastern region has a steppe-like climate (Pampas) with annual precipitation on the order of 400 mm (see Figs. 2, 4). Although precipitation decreases somewhat during the spring–summer season (September–March), rainy conditions persist throughout the year. Air temperature reaches freezing conditions during the winter (June–August) season, whereas maximum summer temperatures usually oscillate around 15°C. The prevalence of cold conditions indicates that snow and ice formation and melt should be a dominating process in the region’s hydrology. Figure 3 illustrates this effect, with UBRB flows peaking during the warm season (February) with a strong seasonal pattern.
Baker River basin at the drainage of Bertrand Lake. Dashed line shows a cross section at 46.5°S.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Baker River basin at the drainage of Bertrand Lake. Dashed line shows a cross section at 46.5°S.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Baker River basin at the drainage of Bertrand Lake. Dashed line shows a cross section at 46.5°S.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Monthly mean precipitation and temperature observed at meteorological stations. Error bars correspond to temperature standard deviations.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Monthly mean precipitation and temperature observed at meteorological stations. Error bars correspond to temperature standard deviations.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Monthly mean precipitation and temperature observed at meteorological stations. Error bars correspond to temperature standard deviations.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
(left) Mean monthly streamflow of the Baker River at the drainage of Bertrand Lake. Error bar shows monthly streamflow standard error. (right) The hypsometric curve with approximate meteorological station location.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
(left) Mean monthly streamflow of the Baker River at the drainage of Bertrand Lake. Error bar shows monthly streamflow standard error. (right) The hypsometric curve with approximate meteorological station location.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
(left) Mean monthly streamflow of the Baker River at the drainage of Bertrand Lake. Error bar shows monthly streamflow standard error. (right) The hypsometric curve with approximate meteorological station location.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
b. Observed data
Standard streamflow and meteorological data are available from stations operated by the Dirección General de Aguas (National Water Directorate) of Chile. Streamflow is more reliably measured than precipitation in the region, mainly because only rainfall but not snowfall is measured and because of the sparseness of the rainfall gauge network and gaps in the rainfall data records. Rainfall observations are only available in valley bottom locations, with no measurements in the mountains where most runoff occurs. Rainfall that is measured is subject to undercatch because of wind and freezing effects, whereas streamflow measurements are obtained in mostly stable river sections with rating curves that are updated periodically (B. Nazarala, Dirección General de Aguas, 2014, personal communication). Tables 1 and 2 show information on the existing stations. Data gaps in weather stations, sometimes representing up to 83% of a station’s records within a modeling period, prevented the use of all station records existing for the region. The density of rain gauge stations over the basin is approximately 0.2 stations per 1000 km2. This value is compared with the recommendation of the World Meteorological Organization (WMO 1994), which set a minimum precipitation station density of 1 station per 250 km2 or 4 stations per 1000 km2 for mountainous regions; this suggests that UBRB instrumentation is below international standards.
Available weather stations within UBRB. Monthly precipitation and temperature are from within the modeling period (source: Dirección General de Agua).
Available stream gauge stations within UBRB. Monthly streamflow is from within modeling period (source: Dirección General de Agua).
c. Reanalysis of precipitation
Atmospheric model data from ERA-Interim and CFSR were considered in an attempt to overcome the lack of snowfall measurements and the limited number and unrepresentative location of meteorological stations for hydrological modeling. These data products represent state-of-the-art global climate characterization and constitute improvements with respect to previous versions. As a first step for using these as input to the hydrological model, they are analyzed in the context of local weather and streamflow data.
In light of the documented effect of topography on the spatial distribution of precipitation, a first step involves comparing how each reanalysis product represents effects of the regional topography; this is closely linked to each product’s spatial resolution. Figure 4 shows a cross section at 46.5°S showing ERA-Interim and CFSR elevation compared with the Advanced Spaceborne Thermal Emission and Reflection Radiometer–Global Digital Elevation Model (ASTER-GDEM; NASA 2014) with a 30-m spatial resolution; annual precipitation for the modeling window is also shown in Fig. 4 (top), with bars appropriately located given the spatial resolution of each reanalysis (three centroids for ERA-Interim, nine centroids for CFSR). Bars representing observed precipitation are located at the approximate longitude of the corresponding meteorological stations. Although both reanalyses include realistic approximations of the regional topography, CFSR’s higher spatial resolution allows for a better representation than ERA-Interim, with steeper and higher topography at the western side of the basin. Although the absence of observations renders it impossible to assess which reanalysis achieves better precipitation estimates at the western edge of the region, for the central section—around 72°W—CFSR matches the observed values more closely. When comparing within a reference centroid (72°S, 46.5°W), differences of about 3 times the mean annual precipitation were detected (682 and 1870 mm yr−1 for CFSR and ERA-Interim, respectively). As mentioned before, CFSR has 3 times the spatial resolution of ERA-Interim in this region, but in terms of density of centroids within the basin, the difference rises to 6 times, with six centroids for CFSR (about 0.4 centroids per 1000 km2) and one centroid for ERA-Interim (0.06 centroids per 1000 km2).
(top) Mean annual precipitation for the modeling period (from 1 Jul 2004 to 30 Jun 2006). (bottom) Altitude cross section at 46.5°S. ASTER-GDEM topography corresponds to average altitude between 46.25° and 46.75°S.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
(top) Mean annual precipitation for the modeling period (from 1 Jul 2004 to 30 Jun 2006). (bottom) Altitude cross section at 46.5°S. ASTER-GDEM topography corresponds to average altitude between 46.25° and 46.75°S.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
(top) Mean annual precipitation for the modeling period (from 1 Jul 2004 to 30 Jun 2006). (bottom) Altitude cross section at 46.5°S. ASTER-GDEM topography corresponds to average altitude between 46.25° and 46.75°S.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
A second analysis focused on the meteorological characteristics of individual storm events as represented by each data source. This analysis is considered to be very relevant to runoff generation as the characteristics of large storms are critical to mountain hydrology regimes. Scatterplots (see Fig. 5) show important differences; for example, the observed record shows almost no precipitation events concurrent with air temperatures below 0°C, likely because of the lack of snowfall measurements, whereas reanalysis data show a significant amount of precipitation occurring at temperatures below 0°C, for instance, 127 (17% of all events) and 93 (13% of all events) events for ERA-Interim and CFSR, respectively, during the modeling period. Given that snowstorm events are routinely noticed in the region, it can be concluded that the current precipitation observational record does not include snowfall, and a low bias in recorded annual precipitation results from this lack of measurement. Another difference between reanalysis and observed data is the fact that most of the observed precipitation events occur at about 10°C, which is 5°C above the temperature at which the majority of precipitation events from the reanalyses occur. In this context, reanalysis data tend to include colder precipitation events, with most precipitation occurring at temperatures below 15°C. On the other hand, observed temperatures show precipitation events with daily temperatures up to 20°C; the lack of snowfall measurements clearly puts a substantial bias on observations that are expected to have a large impact on the ability of any mountain hydrology model to simulate streamflow.
Observed precipitation and temperature from Puerto Ibáñez meteorological station compared to ERA-Interim and CFSR precipitation for centroid (46°S, 72.5°W) and centroid (46.5°S, 72°W), respectively. Dashed line at 0°C is presented to approximate between snowfall and rainfall events.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Observed precipitation and temperature from Puerto Ibáñez meteorological station compared to ERA-Interim and CFSR precipitation for centroid (46°S, 72.5°W) and centroid (46.5°S, 72°W), respectively. Dashed line at 0°C is presented to approximate between snowfall and rainfall events.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Observed precipitation and temperature from Puerto Ibáñez meteorological station compared to ERA-Interim and CFSR precipitation for centroid (46°S, 72.5°W) and centroid (46.5°S, 72°W), respectively. Dashed line at 0°C is presented to approximate between snowfall and rainfall events.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Also, the total number of precipitation events in the modeling period differs significantly between reanalysis and observational data. For example, the weather stations recorded 216 precipitation events between 1 July 2004 and 30 June 2006, while CFSR and ERA-Interim produced 664 and 616 events, respectively. Many of these excess events occurred during the winter season, which was particularly problematic for weather station observations. When inspecting observed daily streamflow (Río Ibáñez) and precipitation (Puerto Ibáñez) data (Fig. 6), several high-flow events are concurrent with zero measured rainfall. Two examples of this are displayed and highlighted with dashed lines in Fig. 6 (bottom), this time with reanalysis data shown in columns. While high-flow events can be caused by snowmelt without precipitation, the streamflow peaks show a good agreement with the reanalyses storm precipitation timing, in contrast with observed precipitation, where several gaps can be seen. These observation gaps strongly affect the potential of the observed data series to inform the modeling exercise.
(top) A comparison between observed precipitation and streamflow at Puerto Ibáñez station. (bottom) Two precipitation gap examples are shown, along with the precipitation data from both reanalyses.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
(top) A comparison between observed precipitation and streamflow at Puerto Ibáñez station. (bottom) Two precipitation gap examples are shown, along with the precipitation data from both reanalyses.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
(top) A comparison between observed precipitation and streamflow at Puerto Ibáñez station. (bottom) Two precipitation gap examples are shown, along with the precipitation data from both reanalyses.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
3. Methodology
a. Cold Regions Hydrological Model
The CRHM is a modular, physically based model developed by the University of Saskatchewan for simulating the hydrological cycle across temporal and spatial scales, with specific characteristics suited for representing hydrological processes in both cold region and temperate environments (Pomeroy et al. 2007). One of the main features of this model is its flexible modular structure, where different modules represent each hydrological process, such as snow redistribution by wind, snow interception, sublimation, evapotranspiration, and infiltration. These modules may be selected and activated in order to achieve a better representation of the hydrological system, according to the modeler’s understanding of the local processes relevant for runoff generation. Each module is interconnected in order to simulate the hydrological cycle. The spatial distribution of hydrological processes over the basin is represented through hydrologic response units (HRUs), which are used to discretize the basin. The HRU is the spatial unit for mass and energy balance calculations and is defined as being able to be represented by one set of biophysical, pedologic, geomorphic, hydrometeorological, and hydraulic parameters. HRUs may be disjointed spatially, but they can always be linked to observable features that define hydrological behavior, such as shallow or deep soils, steep or shallow slopes, orientation, and elevation range.
CRHM is primarily a physically based model, which means that most hydrological processes are represented following realistic equations, with observable parameters. However, a minority of processes relevant at the watershed scale, such as runoff routing and soil moisture dynamics, can also be represented using empirical/conceptual formulations with parameters that require calibration. One of the advantages of physically based modules is that parameterization can be accomplished through direct observations in the basin or transfer of values observed in similar basins.
b. Model setup
The UBRB catchment area is 15 904 km2 and was divided into subbasins (SBs), defined by common drainage and similar hydrometeorological or terrestrial ecological features. The criteria for each SB delineation included, hierarchically, (i) the existence of a stream gauge defining an SB of the larger basin and (ii) the existence of climatic differences that would suggest different hydrological behavior. The second criterion is supported by the strong climatic gradients across the region, which result in variations in precipitation, temperature, humidity, and exposure to wind. The combination of these criteria resulted in the definition of eight SBs. The first five SBs are Puerto Ibáñez (SB1, 2403 km2) and Murta River (SB2, 906 km2), where stream gauges exist, and three ungauged SBs associated with significant streams: (>500 km2 contributing area) Jeinimeni y los Antiguos River1 (SB3, 1985 km2), León River (SB4, 823 km2), and Delta River (SB5, 640 km2). The remaining basin area is characterized by different meteorological conditions (most notably defined by the west–east precipitation gradient), resulting in the windward SB7 (3385 km2) and lee SB6 (3857 km2). General Carrera Lake is a large water body that defines SB8 (1905 km2). Figure 7 shows SB delimitation, including land cover over the Baker basin area, and also SB1 and SB2 HRU designation. HRU designation criteria are further discussed in section 3c(1).
Land cover map, with each SB labeled: SB1, Puerto Ibáñez; SB2, Bahía Murta; SB3, Jeinimeni y Los Antiguos; SB4, Río León; SB5, Río Delta; SB6, lee side; SB7, windward side; and SB8, General Carrera Lake.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Land cover map, with each SB labeled: SB1, Puerto Ibáñez; SB2, Bahía Murta; SB3, Jeinimeni y Los Antiguos; SB4, Río León; SB5, Río Delta; SB6, lee side; SB7, windward side; and SB8, General Carrera Lake.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Land cover map, with each SB labeled: SB1, Puerto Ibáñez; SB2, Bahía Murta; SB3, Jeinimeni y Los Antiguos; SB4, Río León; SB5, Río Delta; SB6, lee side; SB7, windward side; and SB8, General Carrera Lake.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Observed rainfall was first assigned to each subbasin by (i) adopting the value of the measurement station or reanalysis grid centroid closest to the subbasin, when no stations or centroids within the subbasin were present, or (ii) adopting the average of the values observed at the stations or centroids within the subbasin. As this first approximation provided very poor modeling results, these data were then adjusted based on Chile’s countrywide isohyetal maps of precipitation variation with topography, as described in detail in section 3b(3). Because of the lack of any regular observations of precipitation variation over mountain topography in the region to inform or confirm interpolation methods, no further interpolation methods were deployed. This is consistent with the study objective to assess the quality of the forcing datasets in the context of a hydrologic modeling exercise and not to find the most accurate representation of the real precipitation fields. Additionally, Dussaillant et al. (2012) have shown that interpolation methods such as Thiessen polygon, kriging, and cokriging do not necessarily perform well for hydrological calculations in this region, in part because of the very low spatial coverage and support of the existing stations and in part because of the very large gradients that can be observed to depend both on elevation and on longitude.
Air temperature was spatially distributed to the HRUs, assuming that elevation is the only control over this variable within an SB. The environmental temperature lapse rate corresponds to that for average adiabatic conditions (i.e., −0.0065°C m−1) for all of the HRU. Nevertheless, for the upper snow/ice HRU for the closest SBs to the Northern Icefield (SB2, SB4, and SB5), different environmental temperature lapse rates (−0.005° and −0.0055°C m−1) were set. We found that the latter values were necessary in order to avoid an overly pronounced positive trend in snow/ice mass balance, which would be unrealistic with respect to recent work published for the area. Such work has documented widespread retreat for major glacier bodies. As shown by Marshall et al. (2007), lapse rates for areas near icefield sites are commonly lower than off-glacier sites because of katabatic drainage flows. They also found a mean annual temperature lapse rate of −0.005°C m−1 for the Prince of Wales Icefield (Canada).
1) HRU delineation and module selection
HRUs were defined using geomorphological and land use criteria, such as elevation, aspect, and land cover (see Fig. 7). HRUs defined here are similar to the catena concept described in Arnold et al. (2010), where water is routed from one unit to another rather than being routed directly from one unit to the outlet of the basin. Hierarchically, the first classification used to obtain HRUs is land cover (CONAF/CONAMA 1999), where six dominant types were identified: (i) rock, (ii) grass and shrubs, (iii) peat, (iv) snow and ice, (v) forest, and (vi) open water (rivers and lakes). Subsequently, each land cover was divided into north- and south-facing slopes, in order to capture potential differences in melt timing due to radiation and wind exposure, which have been found to be important in other mountain environments (DeBeer and Pomeroy 2010). A third classification hierarchical level was defined based on terrain elevation, in order to capture the effect of air temperature spatial variability on the occurrence of solid/liquid precipitation. The appropriate level of vertical discretization was assessed based on the statistical properties of HRU elevation prior to this last hierarchical step. Any HRU with an elevation range greater than 500 m and with an elevation standard deviation larger than 200 m was split at the mean elevation into upper and lower elevation bands. This discretization is needed to distinguish climate regimes; an environmental lapse rate of −0.0065°C km−1 over a range of 500 m results in a temperature range of 3.25°C between the lower and upper boundaries of the elevation band, smaller than the standard deviation for average daily temperature (5°C). The number of HRUs for each SB varies from 1 to 16 and is controlled by (i) the altitude range and (ii) the variety of land cover found at each SB. Once all HRUs are set for each basin, a drainage sequence needs to be specified by the modeler. We set the drainage sequence with upper north and south faces draining to lower faces of the same exposure (see Fig. 8). A set of physically based modules was assembled for each SB to simulate the hydrological processes relevant to the UBRB. These modules are described in the appendix. Also, Fig. 8 shows a flowchart of CRHM module interactions, denoting inputs required for each module and the associated output for the following calculations.
Flowchart showing CRHM modules configuration. This setup is repeated for all SBs except SB8. The SB1 drainage sequence is also shown, where NN (SS) refers to north-facing (south facing) orientation of the HRU.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Flowchart showing CRHM modules configuration. This setup is repeated for all SBs except SB8. The SB1 drainage sequence is also shown, where NN (SS) refers to north-facing (south facing) orientation of the HRU.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Flowchart showing CRHM modules configuration. This setup is repeated for all SBs except SB8. The SB1 drainage sequence is also shown, where NN (SS) refers to north-facing (south facing) orientation of the HRU.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
2) Parameterization
(i) Parameters that were not calibrated
Most of the model parameters lie in this category, for which parameters were set either based on measurable physiographic features (through remote sensing or fieldwork observations) or transferred from similar climatic/landscape conditions. Physiographic characteristics are required in many modules. For example, the longwave radiation module uses the terrain view factor to adjust incoming longwave radiation and the routing module uses the mean HRU slope for adjusting the parameters of the Muskingum method. As such, model parameterization was undertaken using, in the first place, terrain characteristics such as mean elevation, aspect, slope, view, and area obtained from ASTER-GDEM analyses. Also, a set of parameters was assigned depending on the land cover classification. These parameters are presented in Table 3. For routing module parameters, the routing lengths for each HRU were estimated using Hack’s law (Rigon et al. 1996); channel shapes adopted were supposed as parabolic and the Manning coefficient was set referentially from values presented by Barnes (1967) with support from fieldwork observations. For the soil module, parameters representing maximum soil moisture capacity and maximum soil recharge for the “grass and shrubs” land use classification were transferred from western Canada, where a cool subhumid climate in a postglacial landscape has resulted in a similar steppe soil development as in the glacio-fluvio-volcanic landscape of the UBRB that has an overall sandy loam soil texture, moderately deep to deep soil profile, and good root penetration (Casanova et al. 2013, 75–77). For the “forest” land cover classification, parameters were transferred from aspen forests in Canada (Fang et al. 2010); for “peat,” values measured in northern Canadian subalpine tundra by Quinton et al. (2005) were used. Depressional storage (DS) parameters for grass and shrubs and forest were transferred from Fang et al. (2010) in postglacial landscapes where high-resolution lidar DEMs were available to estimate storage capacity. Albedo values for peat, grass, and shrubs were adopted from Armstrong (2011) and for forest and rock from Pomeroy et al. (1997). For ice albedo, although ponding and debris conditions can induce large variability, a value of 0.5 was adopted as an average representative of white ice (Perovich et al. 2002). Vegetation height, which is used for both canopy clearing and evaporation modules, was obtained from fieldwork measurements, estimating a value of 0.5 and 1 m for bottom and upper grass and shrubs and 6–12 m for bottom and upper forest. For snow transport and blowing snow sublimation, given that no usable in situ wind information was available for this study, we adopted wind direction and velocity values from ERA-Interim for the model forced with otherwise measured data. For wind direction, both reanalyses confirm that the principal direction during the modeling period was northwest. Precipitation phase classification is based on a mean day temperature index. As the model is set, when temperature rises above 2°C, precipitation is set as liquid, whereas below 0°C it is set as snowfall; for intermediate temperatures, the precipitation phase was linearly interpolated.
CRHM parameterization.
(ii) Calibrated parameters and calibration procedure
Only two soil moisture parameters and the storage–discharge rating curve relationship for General Carrera Lake were calibrated manually, with the objective of maximizing the NS coefficient criterion applied to daily flows. DS parameters for rock, glaciers, and peat and daily subsurface drainage (SSD) from recharge and soil column for grass and shrubs, forest, and peat were calibrated, assuming the same value for each HRU type. Calibration was carried out by analyzing the results of the SB1 submodel, forced with CFSR data, with the objective of maximizing the NS coefficient value computed against observed daily streamflow data. The calibration procedure is as follows. First, for several arbitrary (within a realistic range) values of SSD, the DS value was tested over the 5–300-mm range. This exercise showed that the model has very little sensitivity to DS, with NS changes smaller than 0.01 for each SSD value. With this, an arbitrary but realistic value of DS = 5 mm was set for rock, glaciers, and peat. Second, with the DS parameter value set, the SSD parameter was tested again in a more selective probable range of 1–20 mm day−1. These runs showed that the model has a somewhat stronger sensitivity to this parameter; for small values (1–4 mm day−1), the NS coefficient varied around 0.3–0.5. For higher SSD values (5–20 mm day−1), the model showed a smaller sensitivity, but the best NS values between 0.5 and 0.53 were also attained. Therefore, an arbitrary 10 mm day−1 midpoint value within the “insensitive” range was selected. Third, the insensitivity of the DS parameter was verified for the selected SSD value, thus finalizing the calibration stage.
(iii) Sensitivity analysis and insights
Calibration was also conducted using observed and ERA-Interim forcing data; however, using “optimal” parameters from each forcing data source revealed that CFSR always resulted in the best performance against streamflow observations; also, the parameter selection for these forcing data are very similar to those found for CFRS (the model is insensitive to DS, and optimal SSD values of 3 and 6 mm day−1 were obtained for ERA-Interim and observed data, respectively). The calibration and parameter transfer method used here corresponds to the deduction–induction–abduction approach recommended by Pomeroy et al. (2013) for designing and parameterizing models for prediction in ungauged basins. Physically based model parameter transference across >1000 km between similar ecozones was demonstrated by Dornes et al. (2008) in northern Canada and by Gelfan et al. (2004) between Canada and Russia. Because of the lack of any kind of detailed (i.e., finescale) hydrological information for this basin, we chose to transfer previously published parameter values whenever plausible in order to minimize calibration. We find that, after this process, only subsurface drainage and depression storage parameters require calibration, which is seen here as a promising result in terms of preserving a parsimonious model in a relatively large, complex hydrological system.
3) Precipitation data correction
Initial test model runs revealed important discrepancies (bias) between observed and simulated runoff when forcing the model with either observed or reanalysis data. This bias problem has been found in other studies, such as that developed by Pan et al. (2003), where they compare SWE from four land surface models [Noah, Mosaic, Sacramento Soil Moisture Accounting (SAC-SMA), and Variable Infiltration Capacity (VIC)] against 3 years of Snow Telemetry (SNOTEL)-measured data. Experiments with the VIC model indicated that most of the bias in SWE is removed by scaling the precipitation by a regional factor based on the regression of the North American Land Data Assimilation System (NLDAS) and SNOTEL-measured precipitation.
Correction factor used for each precipitation data source simulation.
4. Results and discussion
a. Model testing
Large differences were found for every data source and SB (see Table 5). In general, model performance was enhanced when forcing CRHM with reanalyses instead of observed data, especially for the entire Baker basin, where the differences are more significant. The well-established calibrated relationship between lake level and outlet runoff can explain, in part, the satisfactory results associated with Baker basin model performance. This relationship constrains the range of possible discharge values, attenuating peak flows due to the substantial storage in the lake. The best performance in SB1 and SB2 discharge was when CHRM was forced with CFSR precipitation. This can be explained through the CFSR spatial resolution, which, as shown in Fig. 4, is 3 times higher than that of the ERA-Interim and many times the meteorological station density. In particular, the representation of spatially variable meteorological conditions within SB2 is crucial for model performance, whereas for SB1, basin average weather conditions are well correlated with SB1 observed outflow. This can be explained by higher meteorological variability in SB2 due to the effect of the Northern Icefield in dampening temperature lapse rates in its vicinity (Marshall et al. 2007). Such high variability in local hydrometeorology suggests that a higher spatial resolution reanalysis like CFSR should provide better results. ERA-Interim has only one centroid within the Baker basin, and so its representativeness is lower than that of CFSR.
Model performance for different input source.
Figure 9 shows the simulated streamflow time series for each SB in the model. Observed flows are included for SB1, SB2, and SB8, and area averaged precipitation input is shown for every SB as well. From Fig. 9 it can be seen that SB1, SB2, SB4, and SB7 show a somewhat synchronized response in the sense that distinct peak flows occur on similar dates and in that daily flows are well correlated. These SBs are located almost entirely west of 72°W, and share traits such as a steep relief, widespread glaciation, and a wetter climate. In contrast, SB5 has a milder hydrograph, with attenuated daily peak flows when compared to the highest observed/modeled events in other basins. SB3 and SB6 do exhibit a different behavior; these are located at the lee side of the basin (semiarid conditions) and also have shallow soil profiles. Consequently, base flows have a relatively small influence over the hydrograph. Although it is not possible to corroborate the model results for the ungauged basins, Fig. 9 illustrates the potential of the CFSR, in that it can actually predict a different climatic regime, which in turn can be associated with a different predicted hydrological response.
Model streamflow simulations for each SB and UBRB using CFSR forcing data.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Model streamflow simulations for each SB and UBRB using CFSR forcing data.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Model streamflow simulations for each SB and UBRB using CFSR forcing data.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Figure 10 shows monthly average streamflow simulations using CFSR input data. It can be seen that the overall features of model performance vary across space, with overestimation of monthly flows during fall and winter months occurring at SB2 and UBRB. The model overestimates winter flows (and underestimates summer flows) for all three SBs, which, together with a small overall bias, suggest that liquid-phase precipitation is being overestimated as a whole over the river basin. This effect is relatively more important in SB2, with water year 2005/06 showing a greater bias than year 2004/05. Figure 10 (bottom) shows scatterplots for observed versus simulated daily flows. For the Baker River (Fig. 10, bottom left), we see a very clear trend of underestimation (overestimation) of flows below (above) the median value of about 600 m3 s−1. The buffering effect of the lake is well represented, and as a consequence, a systematic trend in model fit is preserved (r2 = 0.84). For the smaller, unregulated basins SB1 and 2 (Fig. 10, bottom middle and right), the scatter is much higher. The model shows better (although noisy) fit for flow rates below 300 m3 s−1, but for higher values a low bias is clearly perceptible. From Fig. 9, it can be seen that these high daily values occur mostly between the months of October and March, that is, during spring and summer, when direct runoff from liquid precipitation combines with melt from snow and glaciers to give daily streamflow of up to 1000 m3 s−1 during the modeling period. The fact that at the same time the model underestimates spring–summer monthly flows confirms that a specific runoff-generating process or high-elevation precipitation is misrepresented under our current modeling framework. We hypothesize that rain-on-snow events may be more important that currently modeled, and this area will be the focus of further research.
(top) Monthly streamflow simulations and observations. (bottom) Observed vs simulated daily streamflow scatter. Simulation results shown using CFSR forcing data.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
(top) Monthly streamflow simulations and observations. (bottom) Observed vs simulated daily streamflow scatter. Simulation results shown using CFSR forcing data.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
(top) Monthly streamflow simulations and observations. (bottom) Observed vs simulated daily streamflow scatter. Simulation results shown using CFSR forcing data.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
b. Snow component
No long-term snow accumulation data exist for this region, but modeled SWE time series at four SBs are shown to illustrate the differences resulting from selection of model parameters and the regionalization of forcing data. Maximum SWE accumulation goes up to 2000 mm for high-mountain SB4 and SB2 and down to 200 mm for SB3 (semiarid conditions), revealing very heterogeneous snow conditions in the basin (see Fig. 11). Ablation curves are also very different, as well as the duration of the snow-free period. The ablation curve slope is steeper for semiarid SBs, whereas gradual ablation curves were found for snow-dominated SBs. The semiarid SB has longer snow-free periods, about 6–7 months long, beginning in late October. On the other hand, snow-dominated SBs have shorter snow-free periods, about 1–2 months long, beginning in February. Figure 11 can also be related to SB3 simulated streamflow (see Fig. 9) in the sense that the rapid decrease in snow cover relates well with the quick rise in streamflow during the snowmelt period (November).
Simulated SWE for upper-snow/ice north-facing HRU, using CFSR forcing data.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Simulated SWE for upper-snow/ice north-facing HRU, using CFSR forcing data.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Simulated SWE for upper-snow/ice north-facing HRU, using CFSR forcing data.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
As shown in Table 6, simulated blowing snow transport and sublimation have a small impact in the entire water balance, representing 0.6 mm yr−1 when forcing with CFSR. These small values, lower than expected given the local meteorology, are attributed to our use of daily wind speeds from reanalysis data instead of the hourly or shorter wind speeds required to correctly calculate these fluxes (Pomeroy and Li 2000). Table 7 shows disaggregated values for blowing snow transport and sublimation (and other water balance components) for each SB when forcing CRHM with CFSR data. As expected, blowing snow transport is consistently higher for SBs with higher wind speeds, which correspond to SB1 with a mean annual rate of 4.5 versus 3.8 for SB2 and 3.2 m s−1 for SB4 and SB6. Although wind speed is a relevant factor for blowing snow, so is vegetation height (associated with land cover classification), as it controls shear stress at the snow surface. The relative contribution of blowing snow at SB6 (lee side of the basin) is the highest (0.4 over 73 mm yr−1 of snowfall, about 0.7%), which we attribute to negligible forest cover and predominance of grassland and shrubs land cover (height of 0.5–1 m). Other studies, such as that presented by MacDonald et al. (2009), show that in mountainous regions like the Canadian Rockies, blowing snow transport can reach up to 23% and down to 9% over the total snowfall. This discrepancy can be explained by the still poor understanding of the blowing snow controls over the basin, compounded with the lack of appropriate wind speed data, which cannot be compared or adjusted with any measurement within the basin. A site visit in October 2012 suggested substantial blowing snow transport in alpine areas after snowfall events.
UBRB water balance components. Percentage is over total precipitation.
Water balance component for each SB (mm yr−1). Results associated with CFSR forcing data.
Notwithstanding the high ice depletion rates found for the NPIF, up to 4.0 ± 0.97 m yr−1 (Rivera et al. 2007) for the period between 1975 and 2001, no streamflow trend can be seen for the period 1991–2008 at the UBRB. This fact suggests that this depletion rate over the small portion of the NPIF within this basin has limited influence for the runoff generation process.
c. Water balance
Mean values for modeled components over the modeling period of the water balance were obtained and are shown schematically in Fig. 12. Additionally, water balance components for each data source are displayed in Table 6. Results with CFSR forcing data show that about 28% of the total precipitation input corresponds to snowfall, which is mainly controlled by the elevation discretization and the environmental temperature lapse rate used.
UBRB mass balance using CFSR forcing data over the modeling period.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
UBRB mass balance using CFSR forcing data over the modeling period.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
UBRB mass balance using CFSR forcing data over the modeling period.
Citation: Journal of Hydrometeorology 16, 1; 10.1175/JHM-D-13-0178.1
Model performance in terms of monthly flows (see Fig. 10) is characterized by a moderate overestimation of winter (low) flows and underestimation of high (summer) flows, which in turn could be related by an overestimation of the rainfall fraction of precipitation throughout the basin during the winter months. Four reasons could explain this behavior: (i) a high bias in the index temperature station/reanalysis data used for spatial interpolation, (ii) a too-shallow lapse rate or misrepresentation of its temporal variability, (iii) overestimation of precipitation amounts at lower elevations due to spatial averaging at the subbasin scale, and (iv) inadequate selection of parameter values. No reliable field data exist in order to test the above hypothesis for the modeling period, so this topic constitutes a relevant direction for future research. A recent improvement to CRHM for phase determination using a psychrometric energy balance method (Harder and Pomeroy 2013) holds promise for future evaluations of precipitation phase if reliable humidity measurements become available in the region.
In terms of runoff generation, we found that subsurface flow is significantly more important (68.5%) that overland runoff (21.8%). This can be explained by the high infiltration rates and soil moisture storage capacity, especially for forested and peat land cover. Even though modeled liquid-phase precipitation almost triples snowfall, little difference exists in terms of each precipitation overland contribution; rainfall- and snowmelt-generated overland flows contribute 13.5% and 8.3% to total flow, respectively. Evapotranspiration, sublimation, and lake evaporation take up to 8.2% of total precipitation, with negligible contributions from canopy interception losses (<1%). These evapotranspiration losses account for about 100 mm yr−1, significantly different from the 351 mm yr−1 estimation contained in the Chilean National Water Budget (Dirección General de Aguas 1987), which in turn amount to 20% of the then-estimated long-term precipitation mean for the basin of 1686 mm yr−1. Because our modeling period is drier than the reference period in the Chilean Water Budget (approximately 1400 mm yr−1), it is expected that ET as a percentage of annual precipitation would be different from the climatological value. However, in theory, this percentage should be larger, not smaller, than the reference one. The lack of intermediate state variable information precludes us from formulating a hypothesis with respect to the nature of this discrepancy, and future research should aim at developing point-scale water balances through lysimeter experiments in order to obtain a better understanding of water fluxes along the atmosphere–soil column in various climatic regimes across this region.
Low interception sublimation losses are attributed to the largely deciduous nature of the forest canopies, and low evaporation and sublimation capacity are derived from existing moist and cold conditions. Simulated blowing snow has a negligible influence on the water balance (as discussed in section 4b), and only when the model is forced with CFSR input data is a small amount of drifted snow transported from each HRU (0.7 mm yr−1 on average). Storage change in the basin results from SWE, soil moisture, depression storage, and soil recharge changes. This component has little impact on the modeling, only 1.6%; soil moisture changes represent the most relatively important change (1.4%). Both SWE and depression storage changes are null.
Table 7 shows spatially disaggregated SB water balance components when CRHM is forced with CFSR. Results show that the model is capable of simulating the west–east precipitation gradient within the basin, with high precipitation patterns over SB1, SB2, SB4, and SB5, which total about 72% of the entire precipitation over the basin, to the dry eastern patterns of SB6 and SB3, which only total about 9% of total precipitation input over the basin. Other differences can be highlighted, like evapotranspiration, where SB7 contributes 42% of the total evapotranspiration, with significant forested land (about 1029 km2) and grassland and shrubland (699 km2).
5. Conclusions
This paper presents the first insight into the hydrological cycle and water balance of a Patagonian mountain and lowland basin through physically based modeling. Like most remote and sparsely populated regions in South America, Patagonia has a low density of meteorological stations, many times with incomplete and/or unreliable records. To circumvent this problem and evaluate the impact of data scarcity, data from observed local meteorological stations and reanalyses were analyzed and then used as forcing data to the hydrologic model. Actual meteorological stations have poor representativeness over the Baker basin because (i) only rainfall gauges are available, whereas snowfall events are frequent and significant, and (ii) meteorological stations are all located at low altitudes (<500 m MSL), neglecting higher precipitation magnitudes at higher elevation due to orographic effects.
The performance of a CRHM for the Baker basin was shown to be more satisfactory when forced with reanalyses data, obtaining values of NS > 0.7 for daily flows, than when forced with weather station observations. Because bias was mostly removed from all precipitation forcings by a precipitation correction, the higher performance of the model when forced with reanalysis data can be in attributed in part to a better temporal representation of precipitation events, as the observation record does not include snowfall, and also to better spatial averaging of precipitation when compared to point observations in valley bottoms. When evaluating the performance of the hydrological model in individual subbasins, CFSR proved to be a better estimator of local meteorological conditions, which is reasonable since CFSR has 3 times the spatial resolution of ERA-Interim in this region. These results strongly indicate that reanalysis data have great potential as an effective source of information for hydrological understanding in ungauged or poorly monitored basins, such as those located in Patagonia. Transferring some parameters from hydrological studies in other cold regions ecosystems—in this case, by abductive inference of certain parameters from those developed in Canadian research basins—proved to be a viable approach for this remote and poorly gauged basin and allowed us to avoid or minimize complex calibration schemes, obtaining satisfactory results when no other local source of information was available.
The model results suggest that, although snowfall was only 28% of total precipitation, modeled snowmelt contributed to streamflow for up to 6 months per year and was the major source of runoff. This is not dissimilar from the hydrology of many regions in western Canada (Pomeroy et al. 2007). Infiltration was the principal component in the water balance, capturing about 73% of the total precipitation and showing the influence of subsurface flow generation mechanisms. Some of the infiltrated water remained in the basin until the next hydrological year but most formed runoff. Evapotranspiration from soils and evaporation from lakes represented losses of only 8% of total precipitation, which is half of the mean annual value estimated through simple temperature-based equations in the Chilean Water Budget. The differences are due to consideration of snow cover suppression of evaporation and the deployment of physically based combination method evapotranspiration and aerodynamic lake evaporation schemes in CRHM. Evapotranspiration depends strongly on season, soil moisture capacity, and vegetation, which are based on land cover classification, and lake evaporation depends on wind speed, relative humidity, and mean temperature. Hence, the development of more accurate land cover maps, as well as meteorological stations with wind speed and relative humidity sensors, are crucial in order to obtain more precise estimations for evaporative losses, which should be validated against, for instance, lysimeter experiments in representative locations.
Future research must seek finer-scale spatial representations of mountain meteorology in the region that better approach the HRU spatial discretization, starting from the improved representation that the reanalyses—CFSR, in particular—provide. Also, future research must include refining the estimation of blowing snow phenomena by acquiring finescale, subdaily wind speed data, validating rain-on-snow energy exchanges through point-based field experiments, and further refining the contribution of the Northern Icefield to the hydrology of the region, in view of current climate change projections, which indicate warming for the region and a continued decrease in glacier ice storage.
Acknowledgments
The authors acknowledge the financial support from Fondecyt Project 1090479, Graduate Division of the Mathematical and Physical Sciences Faculty, University of Chile, with their Grant “Pasantías Cortas de Investigación”; from NSERC Discovery Grants, Canada Research Chairs; and from the Inter-American Institute for Global Change Research (IAI) under Grant SGP-CRA2047. The authors also thank the Dirección General de Aguas (DGA), which provided all the observed data used in this study. Sebastian Krogh also thanks Xing Fang and Tom Brown for their important advice on modeling strategies with CRHM.
APPENDIX
CRHM Modules Description
Global calculates the theoretical global radiation and direct and diffuse solar radiation, as well as maximum sunshine hours based on latitude, elevation, ground slope, and azimuth (Garnier and Ohmura 1970).
Annandale estimates incoming shortwave radiation from daily minimum and maximum temperatures (Annandale et al. 2001) and theoretical global radiation from the “global” module.
The longwave radiation module estimates incoming longwave radiation using temperature, humidity, and shortwave transmittance (Sicart et al. 2006).
Albedo estimates snow albedo throughout the winter and into the melt period. Albedo is estimated following a linear decay rate for each time period based on snow depth, new snow, and melting (temperature and radiation criteria) occurrence (Gray and Landine 1987).
Netall models net all-wave radiation to snow-free surfaces from the Brunt equation (Brunt 1932), using inputs from the “global” and “Annandale” radiation modules.
Prairie Blowing Snow Model (PBSM) calculates SWE from snowfall and blowing snow transport, redistribution, and sublimation (Pomeroy and Li 2000).
Energy-Budget Snowmelt Model (EBSM) estimates snowmelt by calculating the energy balance of radiation, sensible heat, latent heat, advection from rainfall, and change in internal snowpack energy (Gray and Landine 1988).
Ayers is an empirical relationship that estimates rainfall infiltration into unfrozen soils based on soil texture and ground cover (Ayers 1959).
The evaporation module has two types: (i) Granger’s evaporation expression (Granger and Gray 1989; Granger and Pomeroy 1997) estimates actual evapotranspiration from unsaturated surfaces (canopy, crops, soils) using a complementary solution to the Penman equation, and (ii) the Priestley and Taylor evaporation expression (Priestley and Taylor 1972) estimates evaporation from saturated surfaces, wetlands, or small water bodies including advection effects.
The canopy module estimates the snowfall and rainfall intercepted, sublimated, and evaporated by the forest canopy, subcanopy snowfall, rainfall, and shortwave and longwave radiation (Ellis et al. 2010).
The soil moisture module computes soil moisture balance for frozen and unfrozen periods (Pomeroy et al. 2007, 2012); moisture content exceeding field capacity is routed away from the HRU using “netroute.”
The Muskingum routing module is based on a variable discharge–storage relationship (Chow 1964) and is used to route runoff between HRUs.
The lake evaporation module is an empirical relationship that estimates monthly large lake actual evaporation using monthly wind speed, relative humidity, and temperature, following the Meyer formula with coefficients as determined by the Prairie Provinces Water Board in western Canada (Martin 2002).
REFERENCES
Annandale, J. G., Jovanovic N. Z. , Benadé N. , and Allen R. G. , 2001: Software for missing data error analysis of Penman–Monteith reference evapotranspiration. Irrig. Sci., 21, 57–67, doi:10.1007/s002710100047.
Aravena, J. C., 2007: Reconstructing climate variability using tree rings and glaciers fluctuations in southern Chilean Andes. Ph.D. thesis, Graduate Program in Geography, University of Western Ontario, 220 pp.
Armstrong, R. N., 2011: Spatial variability of actual evapotranspiration in a prairie landscape. Ph.D. thesis, Dept. of Geography and Planning, University of Saskatchewan, 194 pp.
Arnold, J. G., Allen P. M. , Volk M. , Williams J. R. , and Bosch D. D. , 2010: Assessment of different representations of spatial variability on SWAT model performance. Trans. ASABE, 53, 1433–1443.
Ayers, H. D., 1959: Influence of soil profile and vegetation characteristics on net rainfall supply to runoff. Spillway Design Floods: Proceedings of Hydrology Symposium No. 1, National Research Council of Canada, 198–205. [Available online at www.usask.ca/hydrology/papers/Ayers_1959.pdf.]
Barnes, H. H., 1967: Roughness characteristics of natural channels. USGS Water Supply Paper 1849, 213 pp. [Available online at http://pubs.usgs.gov/wsp/wsp_1849/.]
Barría, P. L., 2010: Pronóstico de Caudales Medios Mensuales en las Cuencas de los Ríos Baker y Pascua. Civil engineering thesis, Civil Engineering Dept., University of Chile, 167 pp.
Brunt, D., 1932: Notes on radiation in the atmosphere. Quart. J. Roy. Meteor. Soc., 58, 389–420, doi:10.1002/qj.49705824704.
Casanova, M., Salazar O. , Seguel O. , and Luzio W. , 2013: The Soils of Chile. Springer, 185 pp.
Chow, V. T., 1964: Handbook of Applied Hydrology. McGraw-Hill, 1468 pp.
CONAF/CONAMA, 1999: Catastro y evaluación de los recursos vegetacionales nativos de Chile. Project Rep., 89 pp. [Available online at www.bcn.cl/carpeta_temas_profundidad/ley-bosque-nativo/archivos-pdf/Catastro.pdf.]
DeBeer, C. M., and Pomeroy J. W. , 2010: Simulation of the snowmelt runoff contributing area in a small alpine basin. Hydrol. Earth Syst. Sci., 14, 1205–1219, doi:10.5194/hess-14-1205-2010.
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, doi:10.1002/qj.828.
Dirección General de Aguas, 1987: Balance hidrológico nacional. Tech. Rep. 70115, Dirección General de Aguas, 62 pp. [Available online at http://documentos.dga.cl/SUP1540.pdf.]
Dornes, P. F., Bryan A. T. , Davison B. , Pietroniro A. , Pomeroy J. W. , and Marsh P. , 2008: Regionalization of land surface hydrological model parameters in subarctic environments. Phys. Chem. Earth, 33, 1081–1089, doi:10.1016/j.pce.2008.07.007.
Dussaillant, J. A., Buytaert W. , Meier C. , and Espinoza F. , 2012: Hydrological regime of remote catchments with extreme gradients under accelerated change: The Baker basin in Patagonia. Hydrol. Sci. J., 57, 1530–1542, doi:10.1080/02626667.2012.726993.
Ellis, C. R., Pomeroy J. W. , Brown T. , and MacDonald J. , 2010: Simulation of snow accumulation and melt in needleleaf forest environments. Hydrol. Earth Syst. Sci., 14, 925–940, doi:10.5194/hess-14-925-2010.
Fang, X., and Pomeroy J. W. , 2007: Snowmelt runoff sensitivity analysis to drought on the Canadian prairies. Hydrol. Processes, 21, 2594–2609, doi:10.1002/hyp.6796.
Fang, X., and Pomeroy J. W. , 2009: Modelling blowing snow redistribution to prairie wetlands. Hydrol. Processes, 23, 2557–2569, doi:10.1002/hyp.7348.
Fang, X., Pomeroy J. W. , Westbrook C. J. , Guo X. , Minke A. G. , and Brown T. , 2010: Prediction of snowmelt derived streamflow in a wetland dominated prairie basin. Hydrol. Earth Syst. Sci., 14, 991–1006, doi:10.5194/hess-14-991-2010.
Fang, X., Pomeroy J. W. , Ellis C. R. , MacDonald M. K. , DeBeer C. M. , and Brown T. , 2013: Multi-variable evaluation of hydrological model predictions for a headwater basin in the Canadian Rocky Mountains. Hydrol. Earth Syst. Sci., 17, 1635–1659, doi:10.5194/hess-17-1635-2013.
Garnier, B. J., and Ohmura A. , 1970: The evaluation of surface variations in solar radiation income. Sol. Energy, 13, 21–34, doi:10.1016/0038-092X(70)90004-6.
Garreaud, R., Vuille M. , Compagnucci R. , and Marengo J. , 2009: Present-day South American climate. Palaeogeogr. Palaeoclimatol. Palaeoecol., 281, 180–195, doi:10.1016/j.palaeo.2007.10.032.
Gelfan, A. N., Pomeroy J. W. , and Kuchment L. S. , 2004: Modeling forest cover influences on snow accumulation, sublimation and melt. J. Hydrometeor., 5, 785–803, doi:10.1175/1525-7541(2004)005<0785:MFCIOS>2.0.CO;2.
Gonthier, C., 2011: Influencia de la escala espacial y representación hidrológica de una Cuenca de montaña sobre la capacidad predictiva de un modelo hidrológico. M.S. thesis, Civil Engineering Dept., University of Chile, 165 pp.
Granger, R. J., and Gray D. M. , 1989: Evaporation from natural non-saturated surfaces. J. Hydrol., 111, 21–29, doi:10.1016/0022-1694(89)90249-7.
Granger, R. J., and Pomeroy J. W. , 1997: Sustainability of the western Canadian boreal forest under changing hydrological conditions. II. Summer energy and water use. IAHS Publ.,240, 243–249. [Available online at http://iahs.info/uploads/dms/iahs_240_0243.pdf.]
Gray, D. M., and Landine P. G. , 1987: Albedo model for shallow prairie snow covers. Can. J. Earth Sci., 24, 1760–1768, doi:10.1139/e87-168.
Gray, D. M., and Landine P. G. , 1988: An energy-budget snowmelt model for the Canadian Prairies. Can. J. Earth Sci., 25, 1292–1303, doi:10.1139/e88-124.
Harder, P., and Pomeroy J. , 2013: Estimating precipitation phase using a phycrometric energy balance method. Hydrol. Processes, 27, 1901–1914, doi:10.1002/hyp.9799.
Lopez, P., Sirguey P. , Arnaud Y. , Pouyaud B. , and Chevalier P. , 2008: Snow cover monitoring in the Northern Patagonia Icefield using MODIS satellite images (2000–2006). Global Planet. Change, 61, 103–116, doi:10.1016/j.gloplacha.2007.07.005.
MacDonald, M., Pomeroy J. , and Pietroniro A. , 2009: Parameterizing redistribution and sublimation of blowing snow for hydrological models: Test in a mountainous subartic catchment. Hydrol. Processes, 23, 2570–2583, doi:10.1002/hyp.7356.
Marshall, S. J., Sharp M. , Burgess D. , and Anslow F. , 2007: Near-surface-temperature lapse rates on the Prince of Wales Icefield, Ellesmere Island, Canada: Implications for regional downscaling of temperature. Int. J. Climatol., 27, 385–398, doi:10.1002/joc.1396.
Martin, F. R., 2002: Gross evaporation for the 30-year period 1971–2000 in the Canadian Prairies. Hydrology Rep. 143, Agriculture and Agri-Food Canada Prairie Farm Rehabilitation Administration Technical Service, 35 pp.
NASA, cited 2014: ASTER Global Digital Elevation Model. [Available online at http://gdem.ersdac.jspacesystems.or.jp/.]
Nash, J. E., and Sutcliffe J. V. , 1970: River flow forecasting through conceptual models. Part I—A discussion of principles. J. Hydrol., 10, 282–290, doi:10.1016/0022-1694(70)90255-6.
Pan, M., and Coauthors, 2003: Snow process modeling in the North American Land Data Assimilation System (NLDAS): 2. Evaluation of model simulated snow water equivalent. J. Geophys. Res., 108, 8850, doi:10.1029/2003JD003994.
Perovich, D. K., Grenfell T. C. , Light B. , and Hobbs P. V. , 2002: Seasonal evolution of the albedo of multiyear Arctic sea ice. J. Geophys. Res., 107, 8044, doi:10.1029/2000JC000438.
Pomeroy, J. W., and Li L. , 2000: Prairie and Arctic areal snow cover mass balance using a blowing snow model. J. Geophys. Res., 105, 26 619–26 634, doi:10.1029/2000JD900149.
Pomeroy, J. W., Granger R. J. , Pietroniro A. , Elliott J. E. , Toth B. , and Hedstrom N. , 1997: Hydrological pathways in the Prince Albert Model Forest. Final Rep. to Prince Albert Model Forest Association, 220 pp. [Available online at www.pamodelforest.sk.ca/pubs/PAMF3400.pdf.]
Pomeroy, J. W., Gray D. M. , Brown T. , Hedstrom N. R. , Quinton W. L. , Granger R. J. , and Carey S. K. , 2007: The cold regions hydrological model: A platform for basing process representation and model structure on physical evidence. Hydrol. Processes, 21, 2650–2667, doi:10.1002/hyp.6787.
Pomeroy, J. W., Fang X. , and Ellis C. , 2012: Sensitivity of snowmelt hydrology in Marmot Creek, Alberta, to forest cover disturbance. Hydrol. Processes, 26, 1891–1904, doi:10.1002/hyp.9248.
Pomeroy, J. W., Whitfield P. H. , and Spence C. , Eds., 2013: Putting Prediction in Ungauged Basins into Practice. IAHS, 362 pp.
Priestley, C. H. B., and Taylor R. J. , 1972: On the assessment of surface heat flux and evaporation using large-scale parameters. Mon. Wea. Rev., 100, 81–92, doi:10.1175/1520-0493(1972)100<0081:OTAOSH>2.3.CO;2.
Quinton, W. L., Shirazi T. , Carey S. K. , and Pomeroy J. W. , 2005: Soil water storage and active-layer development in a sub-alpine tundra hillslope, southern Yukon Territory, Canada. Permafrost Periglacial Processes, 16, 369–382, doi:10.1002/ppp.543.
Rigon, R., Rodriguez-Iturbe I. , and Maritan A. , 1996: On Hack’s law. Water Resour. Res., 32, 3367–3374, doi:10.1029/96WR02397.
Rivera, A., Benham T. , Casassa G. , Mabmer J. , and Dowdeswell J. , 2007: Ice elevation and areal changes of glaciers from the Northern Patagonia Icefield, Chile. Global Planet. Change, 59, 126–137, doi:10.1016/j.gloplacha.2006.11.037.
Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System Reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1057, doi:10.1175/2010BAMS3001.1.
Sheffield, J., Ziegler A. , and Wood E. , 2004: Correction of the high-latitude rain day anomaly in the NCEP–NCAR reanalysis for land surface hydrological modeling. J. Climate, 17, 3814–3828, doi:10.1175/1520-0442(2004)017<3814:COTHRD>2.0.CO;2.
Sicart, J. E., Pomeroy J. W. , Essery R. L. , and Bewley D. , 2006: Incoming longwave radiation to melting snow: Observations, sensitivity and estimations in northern environments. Hydrol. Processes, 20, 3697–3708, doi:10.1002/hyp.6383.
Silva, V., Kousky V. , and Higgins R. , 2011: Daily precipitation statistics for South America: An intercomparison between NCEP reanalyses and observations. J. Hydrometeor., 12, 101–117, doi:10.1175/2010JHM1303.1.
Ward, E., Buytaert W. , Peaver L. , and Wheather H. , 2011: Evaluation of precipitation products over complex mountainous terrain: A water resources perspective. Adv. Water Resour., 34, 1222–1231, doi:10.1016/j.advwatres.2011.05.007.
Warren, R., and Sugden D. , 1993: The Patagonia Icefield: A glaciological review. Arct. Alp. Res., 25, 316–331, doi:10.2307/1551915.
WMO, 1994: Guide to hydrological practices. 5th ed. WMO Publ. 164, 735 pp.
Zhou, J., Pomeroy J. W. , Zhang W. , Cheng G. , Wang G. , and Chen C. , 2014: Simulating cold regions hydrological processes using a modular model in the west of China. J. Hydrol., 509, 13–24, doi:10.1016/j.jhydrol.2013.11.013.
Río Jeinimeni y los Antiguos corresponds to two close SBs, which were combined into one because of the close distance between both outlets and similarity in land cover characteristics.
