Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions A FRAMEWORK FOR HYDROLOGICAL MODELLING IN MAGS E.D. Soulis1, N. Kouwen1, A. Pietroniro2, F.R. Seglenieks1, K.R. Snelgrove3, P. Pellerin4, D.W. Shaw5 and L.W. Martz5 1Department of Civil Engineering, University of Waterloo, Waterloo, ON N2L 3G1 Water Research Institute, Environment Canada, Saskatoon, SK S7N 3H5 3Department of Civil Engineering, University of Manitoba, Winnipeg, MB R3T 2N2 4Recherche Prévision Numérique, Environment Canada, Dorval, PQ H9P 1J3 5Department of Geography, University of Saskatchewan, Saskatoon, SK, S7N 5A5 2National ABSTRACT There is a strong global research effort in coupling atmospheric and hydrological models for improved hydrological flow modelling and improved atmospheric simulation. The land surface is an important hydrological control as it is the primary influence in the surface-water budget, and it is almost always a requirement in the implementation of either hydrological or atmospheric models. Sophisticated soil-vegetation atmospheric transfer schemes also known as landsurface schemes (LSS) are currently being implemented in global climate models, regional climate models and day-to-day operational forecasting numerical weather prediction models. Rarely have these been incorporated into hydrological models. Over the last 10 years, there has been a systematic attempt, through collaborative research in Canada and under a variety of research programs, to couple atmospheric and hydrological models using the LSS as the common link. Our approach has been to combine LSS with hydrological streamflow models to provide stand-alone hydrology-land-surface schemes (H–LSS). These stand-alone models are also incorporated as the LSS in the atmospheric models, creating a fully coupled system. The ability and flexibility of this system permits the analysis of sensitivities of H-LSS to parameterization and physical conceptualizations, and the models impact on hydrological and atmospheric prediction. 119 Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions Soulis et al. RÉSUMÉ À l’échelle mondiale, des efforts de recherche importants sont consacrés au couplage des modèles atmosphériques et hydrologiques pour la modélisation améliorée des débits hydrologiques et la simulation atmosphérique améliorée. La surface terrestre est un important déterminant hydrologique en ce sens qu’elle exerce la principale influence sur le bilan hydrique de surface et qu’il est presque toujours exigé d’en tenir compte dans la mise en œuvre des modèles hydrologiques ou atmosphériques. Des mécanismes évolués de transfert entre sol, végétation et atmosphère, également appelés en anglais land-surface schemes (LSS) sont actuellement mis en œuvre dans des modèles de climat du globe, des modèles de climat régional et des modèles numériques opérationnels de prévision météorologique au jour le jour. Ces modèles ont rarement été incorporés aux modèles hydrologiques. Au cours des 10 dernières années, il y a eu une tentative systématique, dans le cadre des travaux de recherche concertée au Canada et d’une variété de programmes de recherche, de coupler les modèles atmosphériques et hydrologiques à l’aide des LSS à titre de lien commun. Notre approche consistait à combiner les LSS aux modèles des débits hydrologiques pour fournir des mécanismes hydrologiques autonomes de transfert vers et depuis la surface (en anglais, hydrology-land-surface schemes (H-LSS)). Ces modèles autonomes sont également incorporés à titre de LSS dans les modèles atmosphériques, créant ainsi un système entièrement couplé. Les capacités et la souplesse de ce système permettent l’analyse des sensibilités des H-LSS pour le paramétrage et les conceptualisations physiques, et les modèles ont des répercussions sur la prévision hydrologique et atmosphérique. INTRODUCTION The recently announced IAHS Decade on Predictions in Ungauged Basins (PUB) (Sivapalan et al., 2003), initiated in the Spring of 2001, is aimed at achieving advances in the capacity within the hydrological community to make predictions in ungauged basins. The overall objective of PUB is to identify the key gaps in our knowledge that limit our capacity to generate reliable predictions in ungauged catchments. This includes improved understanding of the water and energy balance at a number of scales. In data-sparse areas, this poses a particularly difficult challenge, and modelling plays an essential role by providing a physically consistent framework for the integration of observations into descriptions of the balances. The particular challenge of PUB is that not only are the existing hydrological databases extremely limited, but also many processes within the water and energy cycles on the earth’s surface still remain poorly 120 A Framework for Hydrological Modelling in MAGS understood. Therefore, any modelling effort within the context of PUB must be as comprehensive and systematic as possible, involving atmospheric and hydrological models that are tightly integrated and that use all available data. The land surface is an important hydrological control as it is the primary influence in the surface-water budget, and it is almost always a requirement in the implementation of either hydrological or atmospheric models. Sophisticated soil-vegetation atmospheric transfer schemes, also known as land-surface schemes (LSS), are currently being implemented in global climate models, regional climate models and day-to-day operational forecasting numerical weather prediction models. Rarely have these been incorporated into hydrological models. Over the last 10 years, there has been a systematic attempt within the Mackenzie GEWEX Study (MAGS) (Stewart et al., 1998) in Canada as part of a strong global research effort to couple atmospheric and hydrological models using the LSS as the common link. This paper describes the evolution of such a system for the Mackenzie GEWEX Study (MAGS). Our approach has been to combine LSS models with hydrological streamflow models to provide stand-alone hydrology-land-surface schemes (H-LSS). These stand-alone models are also incorporated as the LSS in atmospheric models, creating fully coupled systems. The ability and flexibility of these systems permits the analysis of sensitivities of H-LSS to parameterization and physical conceptualizations, and the models impact on hydrological and atmospheric prediction. For these reasons, we consider this as an appropriate conceptual framework for PUB modelling activities in Canada. THE MACKENZIE GEWEX MODELLING FRAMEWORK MAGS is a major Canadian scientific initiative with the goal of closing the water and energy budget within the Mackenzie River basin. A major component of MAGS was an integrated modelling initiative that involved coupling atmospheric and hydrological models. One of the challenges (Pietroniro and Soulis, 2003) was to resolve the significant difference in spatial scales, domains, and landsurface representations that the two modelling systems typically use. Atmospheric models typically are applied continentally or globally, use grid elements ranging in size from 25 to 250 km, and use simple terrain models; hydrological models are more regional in nature and use computational elements that range from 10’s of metres to 10’s of kilometres. Physical aspects, such as hydraulic cross-sectional information, land-surface slopes and major waterstorage elements such as small lakes or glaciers, are rarely considered in atmospheric model formulations. 121 Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions Soulis et al. Another challenge in integrating hydrological and atmospheric models has been to resolve the differences between the two modelling traditions. As noted by Pietroniro and Soulis (2003), heterogeneity in the landscape has forced hydrologists to conceptualize the physics and to seek effective parameter values. Hydrologists traditionally have used hydrographs to permit the development of simple objective functions to describe a complex terrain. The relatively simple hydrological comparison between observed and modelled streamflow, however, can often lead to reasonable simulations yet poor physical representation and/or parameterization. The atmospheric scientist is dealing with a simpler medium (air) but requires more complex objective functions. Hence, hydrologists tend to focus on effective parameters and precise objectives, while the atmospheric community has precise parameters and fuzzy objectives. The result has been completely different approaches to parameter identification and diagnostics. The MAGS modelling system (Figure 1) was built in stages by combining two well-established Canadian codes; the first is CLASS, the Canadian Land Surface Scheme (Vershegey, 1991) and the second is WATFLOOD, a distributed hydrological model (Kouwen et al, 1993) developed at the University of Waterloo. CLASS pays great attention to the vertical and waterenergy budget and has an appropriate interface for atmospheric models. WATFLOOD has a well-developed routing scheme and provides an effective connection between runoff and the streamflow records. WATLOOD uses the grouped response unit (GRU) approach for basin segmentation (Kouwen et al., 1993). The GRU was adapted from the early hydrology storm-water management models and is very similar to the “tile” or mosaic approach adopted for atmospheric-land surface model interaction (Bartlett et al., 2002). The resulting code, WATCLASS, combines these strengths and can be run in coupled mode with any atmospheric model. The linking of models was initially based on uncoupled (linked) tests through simple forcing of WATFLOOD and CLASS using the operational numerical weather prediction model (NWP), the Canadian Regional Climate Model (CRCM), and the Canadian General Circulation Model (CGCM) (Snelgrove et al., 2005). The results are a good first step but can be slightly misleading since the land surface is treated independently in each model, resulting in inconsistent basin state variables. Nevertheless, this exercise was important in order to test the integrity of the modelling database and to identify major shortcomings in the models. We call this Level 1 modelling. For the next step, Level 2, the two models share the same land surface scheme but still run independently. Thus, the physics are common but there is still no feedback. 122 A Framework for Hydrological Modelling in MAGS Process Studies Level 0 Level 1 Level 2 Level 3 GEM/RCM GCM GEM/RCM GCM GEM/RCM GCM GEM/RCM GCM CLASS with hydrology WATFLOOD WATFLOOD WatCLASS CLASS WATFLOOD CLASS WatROUTE Process Studies Figure 1. Atmospheric-hydrological coupled modelling strategy for MAGS. Finally, for Level 3, the models are truly combined in the expectation that there is improvement in both the estimates of streamflow and of fluxes to the atmosphere. This three-level framework for coupling, described by Soulis et al. (2000) and Pietroniro and Soulis (2003), is part of an overall strategy to provide a consistent and rational approach to coupling atmospheric climate models with hydrological models. The Mackenzie River basin is a data-sparse environment. There are approximately 82 operational stream gauges in the 1.6 million square kilometre basin and a similar number of synoptic weather stations. Although not typical of the developed world, this low-network density is characteristic of much of the globe, hence the approach presented may be very appropriate for PUB. This gives special importance to work in the Mackenzie River basin, as the techniques developed will be more appropriate globally than models developed in data-rich environments. 123 Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions Soulis et al. A land cover-based approach is essential for modelling a region such as the Mackenzie River basin. It minimizes the amount of ground-based data required for model calibration and allows detailed physics in the vertical water budget on a landscape basis. This is entirely consistent with the mosaic approach currently being implemented in CLASS, and the GRU approach used in WATFLOOD. An extensive database has been developed to support this approach. This includes topography, land-cover and drainage data, as well as a long-term land-surface water balance for the Mackenzie basin that consists of approximately 10 km gridded, mean monthly precipitation, evapotranspiration and runoff fields for the Mackenzie basin. Another challenge is the northern climate of the watershed. Although many existing hydrological models have winter components, the parameterizations are usually not developed well enough to accommodate the extreme northern conditions in the Mackenzie basin. Also, physiographic considerations such as permafrost, frozen soils, snow sublimation and snow redistribution are rarely incorporated in most hydrological models. As noted by Pietroniro et al. (1996), there are major problems when models are transposed from temperate climates and are applied to northern environments. The modelling efforts in MAGS required incorporation of results from hydrological research basins in the modelling framework. This involved reviewing the research basin results (Quinton and Hayashi, 2005), and developing parameterizations such that they are included in the integrated modelling scheme. The WATFLOOD Grouped Response Unit Treatment of land covers by WATFLOOD is a balance between the need for detail associated with physically based modelling and the inevitable complexity that comes with the detail. Physically based simulation models often require breaking the watershed down into smaller units to more closely represent the observed hydrological and hydraulic phenomena. Distributed models are defined by their ability to incorporate the distributed nature of watershed parameters and inputs into a modelling framework. Fully distributed models apply detailed physics in differential form but are too complex and data intensive to solve for large basins. Lumped hydrological models often lack the detail of physics and distributed inputs. Two widely accepted distributed model procedures are the Hydrological Response Units (HRU) approach (Leavsley and Stannard, 1990) and the GRU approach. An HRU is a real element within a basin where the hydrological properties are definable and would not be significantly different if a smaller scale of discretization was used. This approach is appropriate for small basins and grids since it is data and CPU intensive. An approach more suitable 124 A Framework for Hydrological Modelling in MAGS Figure 2. The GRU approach to basin discretization used in WATFLOOD and WATCLASS. The shading with each circle represents the land-cover based GRU within a grid square. Water is routed along the stream in the down-slope direction. for large basins is the GRU, which is a grouping of all areas with a similar land cover (or other attribute) such that a grid square will contain a limited number of distinct GRUs. Runoff generated from the different groups of GRUs are then summed together and routed to the stream and river systems (Figure 2). Two GRUs with the same percentages of land-cover types, rainfall, and initial conditions will produce the same amount of runoff regardless of how these landcover classes are distributed. The major advantage of the GRU approach is that it can incorporate the necessary physics while retaining simplicity of operation. The surface-water budget in WATFLOOD is computed for each GRU within a grid square. Infiltration schemes use the well-known Green-Ampt approach. When the infiltration capacity is exceeded by the water supply, and the depression storage has been satisfied, the model then computes overland flow from the 125 Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions Soulis et al. Manning equation. Infiltrated water is stored in a soil reservoir referred to as the Upper Zone Storage (UZS). Water within this layer percolates downward or is exfiltrated to nearby water courses using simple storage-discharge relationships. All GRUs within a grid contribute to the shared lower zone storage (LZS). An initial LZS is determined through trial and error by matching stream baseflow at the outlet to the observed hydrograph. Ground water, or LZS, is replenished by recharge from the UZS according to a power function. A ground-water depletion function is used to gradually diminish the base flow. There is only one LZS for each grid. Flow rates through soil depend upon the hydraulic conductivity that is optimized on the basis of land cover. The total inflow to the river system is found by adding the surface runoff and interflow from all GRUs in a grid and the base flow. These flows are added to the channel flow from upstream grids and routed through the grid to the next downstream grid using a surrogate channel system with a storage routing technique. The CLASS Mosaic The development of CLASS began in 1987 in response to the perceived need for second-generation land-surface modelling within the Canadian GCM. The following synopsis is based on Verseghy (1991) and Verseghy et al. (1993). Land cover is currently dealt with using a patched land-cover, or mosaic, approach. Each modelled grid cell can have up to four sub-areas representing bare soil, vegetation-covered, snow-covered and snow-and-vegetation covered “patches” of the landscape. Inputs of meteorological variables at the bottom of the atmosphere are used to drive the energy and moisture balances for each of the sub-areas, and the resulting fluxes to the atmosphere are passed back to the host atmospheric model. The surface-energy balance is solved for each sub-area by expressing the various fluxes as functions of a single unknown, the surface temperature (of the vegetation, snow or soil as appropriate), and then solving iteratively. The soil column is divided into three layers of 0.10 m, 0.25 m and 3.75 m. The layer temperatures and liquid and frozen moisture contents are carried as prognostic variables, which are stepped forward in time on the basis of the energy and moisture fluxes at the top and bottom of each layer. The energy fluxes are obtained in the course of the solution of the surface energy balance. The moisture fluxes are calculated using Darcy’s equation (in the case of gravitational drainage) and the Green-Ampt method (in the case of infiltration). If the infiltration capacity is exceeded, water is allowed to pond on the surface up to a maximum surface detention capacity that varies according to land cover. Ponded water is retained on the surface until it either infiltrates or evaporates. 126 A Framework for Hydrological Modelling in MAGS The soil albedo and the soil thermal and hydrological properties vary according to texture and moisture content. When snow is modelled, it is analogous to a fourth, variable-depth “soil” layer. As in the case of the soil, the energy fluxes at the top and bottom of the snowpack are obtained in the course of the solution of the surface-energy balance. The snow albedo and density vary with time according to simple exponential decay functions. Snowmelt is modelled as occurring if either the surface temperature or the average layer temperature is projected to rise above 0°C. In this case, the excess energy is used to melt part of the snowpack and the temperature is set back to 0°C. Meltwater percolates into the pack and refreezes until the temperature of the snowpack reaches the freezing point; any further melt is then allowed to reach the soil. The snow cover is assumed to be complete as long as the modelled snow depth does not fall below 0.10 m. When this occurs, the depth is reset to 0.10 and a fractional snow coverage is calculated through the conservation of snow mass. The vegetation properties are determined based on the vegetation types present over the grid cell in question, each of which is assigned to one of four major groupings: coniferous trees, deciduous trees, crops and grass. The leaf-area index, roughness length, areal mass and rooting depth of each of these groups are considered as varying in a distinctive way over seasonal timescales. The canopy gap fraction, the ratio of the roughness lengths for heat and momentum, and the diurnal variations of albedo and transmissivity are also different for the four groups. The canopy interception capacity is calculated as a function of the leafarea index. Rain or snow that does not fall though gaps in the canopy fills the interception store until the capacity is exceeded. At this point any excess is allowed to reach the ground. Transpiration is controlled by the bulk canopy stomatal resistance that is, in turn, a function of leaf-area index, incoming solar radiation, atmospheric vapour pressure deficit, temperature and soil moisture tension. Model Integration: WATCLASS Coupling WATFLOOD and CLASS was not straightforward. Both models are land covered-based, and have similar land-cover categories, but the treatments of mixed covers are significantly different. For example, each blends different state variables within an element, and many of the parametizations are slightly different. The combined model was constructed using CLASS for the vertical processes and WATFLOOD for the horizontal process. In essence, the generation of runoff was assigned to CLASS and the routing was carried out by WATFLOOD. CLASS was called multiple times to simulate the GRU approach. 127 Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions Soulis et al. Most of the difference between the models is in the treatment of the soil moisture budget, and readers are directed to Soulis et al. (2000) for the details of the approach developed for combining the models. The principles are based on the topographic index approach that focuses on the near-surface water balance of valley slopes (Beven and Kirkby, 1979). Version 2.6 of CLASS has no horizontal drainage (Figure 3a) resulting in the top soil layer remaining wet too long after a rainfall event. This causes evapotranspiration to be overestimated between runoff events and infiltration to be under-estimated during these events. To address this, physically based transfer functions were included in the WATCLASS model to allow movement of water from the soil column into the micro-drainage system in a physically realistic manner. This is depicted in Figure 3b. A fundamental drainage element is conceptualized within the WATFLOOD and WATCLASS framework as a grid element. The element can be viewed as an assembly of sloped blocks, each with three soil layers and each with a connection to the drainage system (Figure 4). One of the relevant morphological properties is LS, the typical length of a block within the element that supplies, or has the potential to supply, a segment of a receiving stream of length LV. This can be determined from the drainage density, DD, of the element, defined as ∑LV/A where A is element area (Soulis et al., 2000). Drainage density is landform dependent and typically ranges from 2/km to 100/km (Dingman, 1994). LS is the average distance from the divides in the micro-drainage system to the stream channels and is equal to 1/2DD (Dingman, 1978). This distance represents the ‘slow’ portion of the water-flow system, via overland routes or through the soil matrix. One limitation is that WATCLASS requires that a stream element be present within each grid unit. Also relevant is ΛI, the typical valley slope, called internal slope in order to distinguish it from the overall slope of the element. ΛI provides the topographic gradient for flow from the soil blocks. Thus an element has a mosaic of sloped tiles, with average dimensions LS and LV and average slope ΛI, drained by a system of micro channels. Elements are large enough that the majority of the inter-element flow is channel flow. As in WATFLOOD, the excess surface water will drain to the micro-drainage system as overland flow, qover, (Figure 3). WATCLASS overland flow is represented by Manning’s equation, which is the momentum equation applied to open-channel flow. The depth of flow at the stream edge will depend on how much of the slope is contributing to overland flow and how much concentration is occurring. Since the two factors offset each other, we assume the best estimate of depth of flow at the stream bank is the average effective depth. 128 A Framework for Hydrological Modelling in MAGS dp q over 0 dp l 0 q over 1 10 cm z1 z1 2 1 q int 10 cm 2 35 cm z2 z2 35 cm 3 3 410 cm z3 q drain (a) z3 410 cm q drain (b) Figure 3. Soil moisture and land-surface drainage representation in a) CLASS and b) WATCLASS. Horizontal near-surface flow, qint, called interflow, will occur through the soil matrix and the macropore structure, leaving the block through the seepage face (Figure 3). Interflow is less straightforward but can be considered a major flow mechanism (Dingman, 1994). The primary interflow mechanism is saturated and unsaturated matrix flow, enhanced by macropore flow near the surface. Moisture content typically varies from saturation immediately after a heavy rainfall event, to the wilting point after a sustained dry period. We can focus our attention on moisture conditions near or above field capacity. Next, because we expect high 129 Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions Soulis et al. A LV n KH, KV LS I Figure 4. Conceptualization of micro-drainage within a GRU incorporated in WATFLOOD. The square grid depicted above represents a WATFLOOD grid square. The slope facets represent the different GRUs within the grid. values for KH/KV (where KH is horizontal hydraulic conductivity and KV vertical) and because naturally occurring slopes generally do not exceed 10%, the DupuitForschheimer (Bear, 1972) approximation is valid, and horizontal flow velocity, VH, can be calculated using a one-dimensional form of Richard’s equation. We represent the variation in hydraulic conductivity with the depth exponential relationship used in TOPMODEL. Furthermore, we extend the relationship to unsaturated conditions by letting hydraulic conductivity vary with soil moisture using Clapp-Hornberger soils physics. 130 A Framework for Hydrological Modelling in MAGS Percolation, qdrain, the downward movement of water from the unsaturated zone to the groundwater system and ultimately to the river system as base flow, is well represented by Darcy’s law and is a function of the integrated volumetric soil moisture in the third soil layer, θ3, and KV (Figure 3). Physiographic Data Use of a distributed hydrological model requires a detailed description of topography, land cover, and sub-basin boundaries of a watershed. WATCLASS requires that the topography of the watershed be outlined and that internal physiographic features such as contour density, drainage direction, channel elevations and densities, in addition to river classifications be parameterized for each grid. These physiographic parameters are key to describing the horizontal transfer (routing) of water in the model. These features are derived using land- cover information, typically from satellitebased imagery and digital elevation information. In the early versions of WATFLOOD, these parameters were derived manually for small basins from topographic map sheets. However, it is clear that repeatable and consistent methods using current GIS and image analysis technologies are required to automate the process of drainage feature extraction. Aggregating land cover and digital elevation information at the gridsegmentation level is not sufficient to deal with the sub-grid representation outlined in the GRU approach. As noted by Shaw et al. (2005), assessing drainage direction and contributing area in square grid-type models is not easily achieved using simple averaging techniques. They note that by incorporating sub-grid elevation information and respecting the true hydraulic pathways, drainage directions and other physiographic features can be preserved. A series of algorithms (WATMAP) were developed to provide the required internal routing parameters required for implementation of the hydrological models. WATMap An integral component of the WATCLASS modelling framework is the segmentation of the basin into gridded computational units. Unlike the atmosphere, the land surface and drainage network that are subdivided in the gridding process require location to be explicitly considered in order to maintain the proper drainage network and sub-grid properties inherent to a specific location. The segmentation process in these models condenses the data to a format that preserves all of the input information, but greatly reduces the memory requirements of programs (Kouwen et al., 1993). 131 Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions Soulis et al. WATCLASS requires six topographic parameters to be input into the model. Grid elevation is determined at the mid-point of the main channel within the grid. The elevation at mid-point is usually interpolated from elevations observed where contours cross the channel. This elevation is used to estimate channel slope between grid elements for hydraulic routing. The drainage area estimate for each grid is required in order to match the drainage area of the grid to what is physically observed on the land surface. Typically, grid segments that fall entirely within a gauged catchment will contribute 100% of their area to the downstream gauge. Grid segments that border two catchments require that a dominant drainage direction be chosen and assigned to the grid segment, with the remainder of the catchment being assigned to a neighbouring grid segment. Both grid area and dominant drainage direction are used in determining the runoff volumes and the grid segment routing path. The concept is best explained by Figure 5. Another important physiographic parameter is the river classification assigned to the grid. Because the grid segmentation approach assumes at least one dominant stream must be within the segment, hydraulic properties of river roughness and a meandering factor are assigned to that grid segment. The values are chosen by the user, or they can be calibrated using standard calibration techniques outlined by Kouwen et al. (1993). As part of the river routing algorithm, channel density is also used and there can be up to five channels of equal size within a grid segment. This is used because large channels are more efficient at conveying water than smaller channels for the same runoff production. The final parameter for the grid segment is the contour density. This parameter represents the land slope rather than the channel slope for each segment. For each grid square, WATMAP scans every outside DEM element to determine if flow is in or out of the grid square. WATMAP determines the main drainage direction out of the grid square, the midpoint elevation of the main channel and the size of the main channel. The primary and secondary outflows from each sub-grid watershed are also recorded; this allows WATMAP to determine the fraction of the grid that is drained in the main drainage direction. The fraction of the grid not drained in the main drainage direction is added to the appropriate neighbouring square. Also derived from the outflows is the number of equalsized channels flowing out of the grid square, another parameter required by WATCLASS. Past experience with WATFLOOD indicates that surface flow is calculated better using the maximum internal slope rather than the average internal slope. As a result, WATMAP calculates the land-surface slope as the cumulative difference in 132 A Framework for Hydrological Modelling in MAGS Figure 5. Drainage direction and basin contribution for a grid using the WATCLASS segmentation method. elevation between cells for each row and column of the grid square individually, and the row or column with the highest cumulative difference is identified. This value is multiplied by a contour factor determined by comparing internal slopes using the manual method and WATMAP on various sample watersheds. Meteorological Forcing WATCLASS requires gridded surface meteorological data to drive its hydrological calculations. Because WATCLASS resolves the water and energy budget, it requires hourly precipitation, air temperature, wind speed, relative humidity, incoming short-wave radiation and net long-wave radiation for each grid element in the modelling domain. Meteorological forcing data for hydrological models have been traditionally derived from interpolated measured station data and measured weather radar; however, more recently Global Circulation Model (GCM) and Numerical Weather Prediction (NWP) archive data have been applied (Snelgrove et al., 2005). This relatively new source of meteorological data has positive implications for both the hydrological and the atmospheric modeller. For the atmospheric modeller, the climate simulation or weather forecast is evaluated against streamflow data using a hydrological model. In these cases the watershed effectively acts as “large rain gauge” even though the caveats and uncertainties of hydrological modelling must be considered. For the hydrological modeller, a new source of data becomes available to drive their model from the atmospheric archive. This offers the opportunity to model remote watersheds for which no gauge atmospheric data are available, and the benefit of adding spatial structure to atmospheric forcing data that is lost in the interpolation of gauge data. 133 Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions Soulis et al. River Routing with WATROUTE Routing between grid elements is required to convey the water to the downstream node or gauge location. The routing of water through the channel system is accomplished using a storage routing technique, WATROUTE. More sophisticated routing models are available but the application of such models does not appear to be warranted. In fact, for large watersheds, differences between the routing methods may well be smaller than the noise in the data (Kouwen et al., 1993). The method involves a straightforward application of the continuity equation, where the flow is related to the storage through the Manning formula. Large lake and reservoir (bigger than a grid element) routing can be explicitly taken into account in this routing scheme by assigning contiguous lake-grid segments a reach number. Once the lakes have been located, the outlets are located at the outlet grid. Water is then routed through the lakes using a user-specified power function. With the development of the previously described topographic analysis tools and appropriate meteorological data sources, the data input framework for WATCLASS is in place. The framework used to create H-LSS is not limited to WATFLOOD and CLASS. It is our view that this framework will allow for a number of H-LSS to be developed and tested at a variety of scales, and to be directly incorporated into atmospheric models. We expand on this in the next section. A PROPOSED MODELLING PROGRAM FOR PUB - THE MEC SYSTEM Future modelling efforts will focus on better modelling structure, physics and parameterization via improved physical understanding. This includes better schemes for the interaction of soil moisture and transpiration from vegetation, optimization of land-surface parameters using hydrographs, and the parameterization of cold-region processes including permafrost, frozen soil, snow sublimation and snow redistribution. In order to assist in H-LSS model testing, a number of research projects are in place. The first of these is a coupler that allows for off-line testing of H-LSS models, yet provides a consistent framework for coupled atmospheric/hydrological applications, lending itself well to the goals of PUB. The Modelling the Environment Community modelling system (MEC) was developed by the Meteorological Service of Canada (MSC) in collaboration with many Canadian groups including universities and the private sector. The main objective of this project is to optimize research and development (R&D) in environmental modelling by sharing a unified modelling environment. With advances in numerical modelling and computer power, we now have the 134 A Framework for Hydrological Modelling in MAGS Meteorological Data - Models - Observations - Analyses - Radar - Satellite HLSS in MEC 2 or 1 way Coupling Land-surface Models MEC system Hydrological Models Figure 6. A strategy adopted for the Environmental Modelling. The MEC system (Modelling Environmental Community) includes different environmental models (ocean-ice, landsurface, hydrology that are coupled to an atmospheric model using a systematic coupling allowing the HLSS to run a different time and space scales than the hydrological model). scientific and technical capabilities to build comprehensive environmental prediction systems, integrating expertise from a wide range of disciplines and addressing important R&D and operational issues. For the moment, the MEC system includes one atmospheric model, one ocean-ice model, and two landsurface schemes. The system includes also a coupler (RPN-Coupler (Recherche en Prévision Numérique)) (Pellerin et al., 2004) which allows it to efficiently link different MEC configurations or environmental models together. Figure 6 presents this strategy adopted by MSC for the environmental modelling. MEC allows the hydrology-land-surface model to run independently of the atmospheric models, enabling hydrologists to test the H-LSS at the basin scale using observed forcing data within experimental basins. This coupling system also allows the H-LSS to run at those time and space scales that are not the same as the atmospheric model, but more appropriate for hydrological simulation, while still providing two-way water and energy feedback between the atmosphere and the land surface. This allows for a number of H-LSS models to be tested in a coupled mode with the atmospheric model. More importantly, it also allows for off-line testing of the H-LSS at the smallbasin scale where the energy and water budget can be closed experimentally, thus reducing both parameter and model uncertainty. Findings at the experimental basin scale (i.e grid square scale in WATCLASS) can then be translated to larger domains, allowing hydrologists to scale from the small basin to the larger basin. 135 Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions Soulis et al. Clearly, modelling and data collection cannot be independent activities. The need for close integration and continuous feedback between models, observations and process studies are essential elements to making significant progress in identifying and reducing uncertainty in prediction of water-cycle variability. To accomplish this, a number of mechanisms should be designed and built into any research program. The most important of these is to sustain the close interaction between modellers and experimentalists -- to develop performance criteria for observation and process studies, including identifying the most important measurements to make, specification of the appropriate spatial and temporal scale for measurements, and recommendations about the optimal resolution and accuracy of the measurements. Success of the Canadian MAGS program and the development of the MAGS modelling strategy were initial attempts at such integration. FUTURE EFFORTS Future modelling efforts will focus on the interaction of soil moisture and transpiration from vegetation as well as optimization of land-surface parameters using hydrographs. It is evident that applications of LSS models incorporated into traditional hydrological models are still exploratory and have many of the same pitfalls of hydrological models developed over the last twenty years (Snelgrove et al., 2005). The framework described and recent advances with MEC allow for systematic coupled atmospheric-hydrological model development. Expertise from both experimental and modelling hydrologists is essential to the success of this modelling framework. Land-surface schemes belong in hydrological models, and the framework described allows for advances in both hydrological and atmospheric modelling, based on sound experimental data and approaches. Through continual feedback between research groups, PUB will be able to make steady and systematic progress toward quantifying and reducing the uncertainty in basin-scale water-cycle models. 136 A Framework for Hydrological Modelling in MAGS REFERENCES Bartlett, P.A., J.K. McCaughey, P.M. Lafleur and D.L. Verseghy. 2002. A Comparison of the Mosaic and Aggregated Canopy Frameworks for Representing Surface Heterogeneity in the Canadian Boreal Forest using CLASS: A Soil Perspective. Journal of Hydrology, 266: 15-39. Bear, J., 1972. Dynamics of Fluids in Porous Media. Dover Publications, Inc., New York. Beven, K.J., and M.J. Kirkby. 1979. A Physically Based, Variable Contributing Area Model of Basin Hydrology. Hydrological Sciences Bulletin, 24: 43-69. Dingman, S.L. 1978 Drainage Density and Streamflow: A Closer Look. Water Resources Research, 14: 1183-1187. Dingman, S.L. 1994 Physical Hydrology. Macmillan Publishing Company: New York. Kouwen, N., E.D. Soulis, A. Pietroniro, J. Donald and R.A. Harrington. 1993. Grouped Response Units for Distributed Hydrologic Modelling. Journal of Water Resources Planning & Management. ASCE, 119(3): 289-305. Leavesley, G. and L. Stannard. 1990. Application of Remotely Sensed Data in a Distributed Parameter Watershed. In: Proceedings of Workshop on Applications of Remote Sensing in Hydrology, NHRI Symposium #5, Kite, G. and A. Wankiewicz (eds.), Saskatoon, 47-68. Pellerin P., H. Ritchie, F. Saucier, F. Roy, S. Desjardins, M. Valin and V. Lee. 2004. Impact of a Two-way Coupling between an Atmospheric and an Ocean-Ice Model over the Gulf of St. Lawrence. Monthly Weather Review, 132: 1379-1398. Pietroniro, A., T.D. Prowse, L. Hamlin, N. Kouwen and E.D. Soulis. 1996. Application of a Grouped Response Unit Hydrologic Model to a Northern Wetland region. Hydrological Processes, 10: 1245-1261. Pietroniro, A. and E.D. Soulis. 2003. A Hydrology Modelling Framework for the Mackenzie GEWEX Program. Hydrological Processes, 17: 673-676. Quinton, W.M. and M. Hayashi. 2005. The Flow and Storage of Water in the WetlandDominated Central Mackenzie River Basin: Recent Advances and Future Directions. In: Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions. Spence, C., J.W. Pomeroy and A. Pietroniro (eds.), Canadian Water Resources Association, 45-66. Shaw, D.A., L.W. Martz and A. Pietroniro. 2005, “A Methodology for Preserving Channel Flow Networks and Connectivity Patterns in Large-scale Distributed Hydrological Models”. Hydrological Processes, 19: 149-168 137 Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions Soulis et al. Sivapalan, M., K. Takeuchi, S.W. Franks, V.K. Gupta, H. Karambiri, V. Lakshmi, X. Liang, J.J. McDonnell, E.M. Mendiondo, P.E. O’Connell, T. Oki, J.W. Pomeroy, D. Schertzer, S. Uhlenbrook and E. Zehe. 2003. IAHS Decade on Predictions in Ungauged Basins (PUB), 2003-2012: Shaping an exciting future for the hydrological sciences. Hydrological–Sciences-Journal-des–Sciences Hydrologiques, 48: 857-880. Snelgrove, K.R., E.D. Soulis, F.R. Seglenieks and N. Kouwen. 2005. The Application of Hydrological Models in MAGS: Lessons Learned for PUB. In: Prediction in Ungauged Basins: Approaches for Canada’s Cold Regions. Spence, C., J.W. Pomeroy and A. Pietroniro (eds.), Canadian Water Resources Association, 139-164. Soulis, E.D., K. Snelgrove, N. Kouwen, F. Seglenieks and D.L. Verseghy. 2000. Towards Closing the Vertical Water Balance in Canadian Atmospheric Models: Coupling of the Land- surface Scheme CLASS with the Distributed Hydrological Model WATFLOOD. Atmosphere-Ocean, 30: 251-269. Stewart, R.E., H.G. Leighton, P. Marsh, G.W.K. Moore, H. Ritchie, W.R. Rouse, E.D. Soulis, G.S. Strong, R.W. Crawford and B. Kochtubajda. 1998. The Mackenzie GEWEX Study: The Water and Energy Cycles of a Major North American River Basin. Bulletin of the American Meteorological Society, 79: 2665-2683 Verseghy, D.L. 1991. CLASS - A Canadian Land Surface Scheme for GCMs, I. Soil Model. International Journal of Climatology, 11: 111-133. Verseghy, D., N.A. McFarlane and M. Lazare. 1993. CLASS - A Canadian Land Surface Scheme for GCMs, II. Vegetation Model and Coupled Runs. International Journal of Climatology, 13: 347-370. 138