6. Discussions

6.1 Representativeness of the 5-y study period

Key basin-average budget parameters computed for the current (1997-2002) and longer-term (1979-1999) periods are given in Table 6.1. Despite the fact that the record-breaking 1997-98 El Nino event occurred within the 5-y study period, the 5-y and 20-y climatologies are very similar. In particular, both the observed and modeled 5-y and 20-y mean precipitation show very little difference. The 5-y and 20-y mean basin evaporation also exhibit little difference for the ERA-40 data and a slight (5%) increase in the NCEP-R2 estimate. Both the WSC discharge measurements and NCEP-R2 reanalysis show little difference in the 5-y and 20-y mean runoff from the basin while the 5-y mean runoff from the ERA-40 reanalysis shows an 8% decrease when compared to the longer-term mean. Both the observed and analyzed surface air temperatures show that the basin during the 5-y study period was 0.6-0.7°C warmer than the previous 20 years on average. This warming in the region could be related to the strong 1997-1998 El Nino event or it could be part of the strong warming trend that has been observed in the region (Zhang et al., 2000). These inter-comparisons show that the 5-y period chosen for the study while exhibited some abnormalities in its mean hydroclimatic state (e.g., it is a relatively warm period for the basin), many of the water and energy fluxes computed for the period should also be representative of longer-term climates. Further comparisons of the 5-y mean budgets with longer-term climatologies can also be found in discussions of the individual budget parameters.


Table 6.1 Intercomparison of 1979-1999 (clim) and 1997-02 (5) climatologies of precipitation (P, mm/d), evaporation (E, mm/d), Runoff (N, mm/d) and T2m(K) for different datasets.

6.2 Budget annual cycles and the MRB climate system

A brief description of the regional water and energy cycle of the MRB as revealed from the budget results will be given here. Like other major high-latitude continental basins, the MRB acts as a sink region for heat and water in the global climate system, as reflected in the budget results. During the boreal cold season (Nov. – Feb.) when the mean N-S global temperature contrast is strong and the atmosphere is dynamically active, a large amount of heat is transported into the MRB from the warm southern and oceanic regions (HC in Fig. 5.2.3a ). As much of the basin receives no or little solar radiation during these months, there is a net radiation deficit at the surface(QRS in Fig. 5.2.4a ) hat cools the surface of the basin to low temperatures during the winter. Some of the heat that is transported into the basin is lost to the cold underlying surface via sensible heat transfer (SH in Fig. 5.2.3a and Fig. 5.2.4a ) and hence cooling the lowest levels of the atmosphere. As a result, surface-based temperature inversions, and thus super-stable conditions, are created over much of the basin’s area and limit evaporation and latent heat transfer at the surface during the cold season (see E and LE during the cold season in Fig. 5.2.1a and Fig. 5.2.4a , respectively). Consequently, the atmospheric heat convergence into the basin is largely balanced by thermal radiation loss to outer-space (QR in Fig. 5.2.3a ) during the cold season.

Although the strong mean westerly flow entering the continent from the North Pacific is moisture-laden, much of the moisture is precipitated out of the atmosphere as the flow converges and rises over the coastal mountainous regions. While the enhanced condensation and associated latent heat release in the forced ascent over the western slopes effectively enhances the transport of dry static energy into the basin on the lee-side of the mountains, net moisture convergence into the basin is reduced and its magnitude is typically small throughout the year (see MC in Fig. 5.2.1a ). During the cold season, precipitation (P, and the associated condensational heating LP) is relatively low and comes solely from synoptic systems that pass through or develop within the basin. As surface evaporation is extremely weak, the winter precipitation is largely balanced by the large-scale moisture convergence into the basin. Because of the low winter temperatures that characterize the region, the winter precipitation falls almost exclusively in the form of snow over much of the basin. Apart from possible wind transport and enhanced sublimation in the blowing snow over tundra-covered regions, sublimation is typically weak over much of the basin and the snowpack grows through the season (as reflected in the negative RESW in Fig. 5.2.2a ) and much of the basin is snow-covered by the end of winter. Expectedly, runoff is extremely low under such conditions ( Fig. 5.2.2a ).

Solar insolation increases and warms the surface during spring (see QRS in Fig. 5.2.4a during Apr. and May). A substantial portion of the solar input is used to melt the snow on the surface. As much of the basin lies within the continuous and discontinuous permafrost zone, and there is abundant snow on the surface, the melt water often recharge the active soil layer in many areas of the basin to saturation and produces huge runoff during spring (see N in Fig. 5.2.2a during Apr.-Jun.). The balance between the runoff and snowmelt during the spring is also reflected in the balance between the N and RESW budgets during these months in ( Fig. 5.2.2a ).

Much of the basin experiences long hours of solar insolation during the summer (see QRS during Jun.-Aug. in Fig. 5.2.4a ). As a large portion of the basin’s surface is covered with vegetation, wet soil or surface water bodies, much of the solar insolation is consumed in evapotranspiration processes and inducing large evaporation and latent heat flux at the surface (see E and LE during Jun.-Aug. in Fig. 5.2.1a and Fig. 5.2.4a ). A relatively smaller portion of the solar radiation is used to warm the surface which in turn warms the lower atmosphere via sensible heat transfer and turbulent heat fluxes (SH in Fig. 5.2.3a and Fig. 5.2.4a ). The surface sensible and latent heat fluxes destabilize the atmosphere over the basin and consequently, despite the northern location of the basin, a considerable portion (typically between a third to one half of total precipitation in the ERA-40 data) of its warm-season precipitation comes from moist convection. The considerable evapotranspiration and precipitation (and their strong phase coherence) that characterize the basin during the warm season ( Fig. 5.2.1a ) suggest that moisture recycling might play an important role in governing the warm-season water cycle of the region, an inference that was supported by results in a previous study (Szeto, 2002) which showed that up to half of the summer precipitation in the basin could be derived from local evaporation. Although summer precipitation contributes to warm-season runoff in the basin, runoff tapers off steadily from the spring snowmelt freshet. Due to the strong surface heat flux and condensation heating in the atmosphere (SH and LP in Fig. 5.2.3a ), the basin transformed into a heat source region (i.e. HC is negative) for the large-scale airflow during the summer months. However, despite the strong evapotranspiration that occurs in the basin, the basin on the whole remains as moisture sink during the summer ( Fig. 5.2.1a ).

Solar input decreases rapidly as the basin progresses into the autumn months of September and October and the net basin surface radiation heating becomes negative again during October. As the basin surface water and energy processes enter their dormant cold-season states, the atmospheric moisture and energy convergence into the basin increases as the large-scale thermal and moisture gradients intensify during the fall. In particular, the moisture convergence into the basin maximizes during October when the moisture contrast between the basin and the upstream region enhances and the synoptic processes become active again over the North Pacific.

6.3 Discussions of Budget Parameters

Because the evaluations of many budget terms are based on analysis data, it is convenient to discuss the results with reference to the degree by which the source analysis variables are affected by observations. In particular, we will adopt the convention used in describing the NCEP analysis variables (e.g. Kistler et al., 2001 and others) in the following discussion. In this convention, type-A variables are those that are strongly influenced by observations (e.g., air temperatures), type-B variables are those affected by both the model performance and observations (e.g., specific humidity, screen temperatures) and type-C variables are purely forecast variables with no correction from observations (e.g., surface evaporation or sensible heat fluxes). In the following discussion, we will focus on assessing the variability of basin-average budgets from the different datasets, the self-consistency of the budget components within each dataset, and intercomparisons of the current results with previous estimates where available. Discussions of the interannual and spatial variability of the budgets as well as their applications to better understand the MRB hydroclimate system can be found in Szeto (2006a and 2006b).


The vertically-integrated atmospheric moisture content or precipitable water (Q) gives a measure of the storage of water in the atmosphere. Despite the fact that Q is typically a type-B analysis variable, i.e., both observations and model performance could exert significant effects on their analyzed values, the variability of Q among the different estimates is relatively small. In particular, the annual basin-average Q agrees well among the different datasets including the estimates from the global NVAP (1988-1999 climatology) and regional rawinsonde datasets (the global and regional observations of Q in Table 5.1). The high bias of annual Q from the rawinsonde measurement can be related to the bias of the site locations in the warmer southern regions. The low bias of Q in the CMC analysis is diagnostically related to the relatively low moisture convergence and high precipitation bias in CMC as compared to other datasets.


The vertically-integrated soil moisture (M) accounts for the sub-surface storage of water in the basin. Although top-soil wetness is becoming available from satellite measurements, regular in situ measurements of soil moisture are relatively scarce, particularly for remotely located regions like the MRB. As such, M is typically a type-C variable in determining its analyzed values. Since the depth of the soil layer varies substantially among the different models, it is very difficult to compare their total soil moisture content. As such, the depth-to-bedrock information for the region from Soil Landscape of Canada (http://sis.agr.gc.ca/cansis/nsdb/slc/intro.html) was used to normalize the physical depth of the model soil layers in the calculation of their total soil moisture content. With this normalization, the range of annual basin-average M still varies from about 230 mm for the CMC estimate to about 325 mm for the NCEP-R2 estimate. Due to the uncertainties introduced in normalizing the soil water in these results, it is more meaningful and physically more interesting to inter-compare the differences in the spatial and temporal variability of M exhibited in the different models rather than their annual averages. Results from these inter-comparisons are presented in Szeto (2006a).


Snowcover plays an important role in governing the water and energy cycle for this northern basin. Snow depth is measured routinely at various locations in the basin and there are also remote-sensing snowcover estimates from satellite (e.g. SSM/I). The analyzed snow water equivalent (SWE) is typically a type-B variable that is derived from the observed snow depth, model-generated precipitation, and the snow densities that are assumed in the analysis procedures. The procedure used by the forecast centers to do the SWE analysis can however vary substantially between each other, and hence the large variability exhibited in their annual estimates for the region. The annual basin SWE varies with values less than 40 mm in the NCEP-R2 and CMC estimates to high values exceeding almost 50 mm in ERA-40 and exceeding 70 mm in the CRCM (Table 5.1).

The high bias of SWE in the CRCM can be related to the cold bias in the lower troposphere and the low sublimation rates (Table 5.1) in the model during the cold season. Consequently, snow accumulation starts early while snowmelt is delayed in the CRCM which in turn affected its water budgets in the spring ( Fig. 5.2.2c ). Further discussions of the model snowcover in the CRCM can be found in MacKay et al. (2003).

Since snow mass is a type-B variable while precipitation is a type-C variable in the analysis datasets considered here, the bias of SWE in each dataset usually does not correlate well with their corresponding cold season precipitation bias. For example, although the ERA-40 cold season precipitation is biased low, its SWE is biased high while the reverse is true for SWE and precipitation from the CMC analysis.


The screen temperature (T2m) is presented here instead of the surface skin temperature (Ts) because the two variables are very closely correlated with each other. However, there are extensive observations for surface air temperatures while there is very little observations available for Ts in the region. Consequently, T2m is a type-B variable while Ts is a type-C variable in the assimilated datasets. It just needs to note that the spatial-temporal variability of both variables are very similar. There are small systematic differences between the two that vary on the diurnal and seasonal time-scales, i.e., T2m exceeds Ts during the winter and at night time while the reverse is true during summer and the day time. The differences are typically small with basin-average values less than 1 K and 0.5 K during the winter and summer, respectively.

The fact that T2m is a type-B analyzed variable in the assimilation datasets partially accounts for the relatively low variability of this variable across the different estimates from the analysis datasets (Table 5.1). In terms of annual basin-average T2m, both the CMC and NCEP-R2 analyses are closest to the CANGRID regional “observations” while the ERA-40 estimates showed a moderate warm bias (< 1 K). It should however be noted that the good agreement of the annual average of the estimates could be misleading because of the compensating biases of different signs in different seasons. For example, NCEP-R2 showed relatively large warm (>3 K) bias during winter and weak (< 1K) cold bias during the rest of the year. Strong cold bias in low-level atmospheric temperature is found in the CRCM data with the strongest bias occurred during the fall and winter (>4K for SON) and less so during the summer (<1K). Examination of the vertical structure of the temperature bias showed that the cold bias extends from the surface up to ~ 700 hPa with the strongest bias occurring near the surface and over the plains region of the basin. An attempt to explain the cold temperature bias in the CRCM can be found in Szeto (2006c).

The spatial distributions of the seasonal average T2m are presented in Fig. 5.3.4. Sub-freezing temperatures characterize the whole basin during the winter and most of the basin (roughly N. of 60°) during spring and fall. The horizontal variation of T2m is controlled by three main factors, namely, the latitude, altitude and continentality effects. The latitude and continentality effects dominate to create strong SW-NE gradients in T2m during the cold season, particularly during DJF. In addition, T2m increases weakly towards the mountainous region during DJF. This latter result reflects the strong surface-based temperature inversion that typically occurs in the basin during the winter. The N-S temperature gradients weaken while the E-S gradient reverses and increases in magnitude during spring and summer. During JJA, surface air temperatures are relatively uniform over the whole basin interior and surface temperatures decrease rapidly towards the higher altitudes.


As atmospheric temperature is a type-A variable in the analyses, its vertical integral (or equivalently the atmospheric enthalpy or heat content, H) exhibits little variability among the different estimates(Table 5.1). The basin average H from the CRCM is very close to others despite of the strong cold temperature bias that exists at its lower model troposphere, suggesting that the model atmosphere has adjusted dynamically to the low-level cold bias to restore dynamic and thermal balance of the basin with its environments. Similar to the high bias in the annual Q estimates from the rawinsonde measurement, the high bias in H estimated from the rawinsondes (the "regional observations” of H in Table 5.1) is related to the bias of the site locations in the warmer southern regions.


Precipitation (P) is the best observed water budget variable and, along with evapotranspiration, it is also one of the two processes that couples the atmospheric and surface branches of the water cycle. Despite the relatively abundant observations for precipitation, it is a type-C variable, i.e., it is a purely forecasted field in the analysis datasets that are used in this study. Hence, variability among the different estimates are expected. However, as shown in Table 5.1 and Fig. 5.2.2b , the precipitation from the various datasets, with the exception of NCEP-R2 and the global blended datasets, agrees relatively well on both their monthly and annual means. The NCEP-R2 precipitation is substantially higher than others while both the CMAP and GPCP global blended precipitation datasets give similar annual P estimates for the region that are substantially lower than those from other datasets. P exhibits strong seasonal variability for the MRB ( Fig. 5.2.2b ). Depending on the datasets, the ratio of basin summer precipitation to winter precipitation varies from 2.5 for the CRCM to over 6 in NCEP-R2. In fact, the high bias in the NCEP basin-average precipitation occurs mainly during the summer ( Fig. 5.2.2b ). Neglecting the global satellite and NCEP P estimates, the mean annual precipitation estimated for the basin ranges from 467 mm (CANGRID) to 507 (CRCM) mm which are substantially higher than the values assessed previous by using older datasets (~410 mm in Stewart et al., 1998 and ~421 mm in Louie et al., 2002).

The spatial distributions of seasonal precipitation from the various datasets are given in Fig. 5.3.6. The CANGRID observations show that maximum precipitation is found over the mountainous western basin during all seasons. With the exception of NCEP-R2 during the spring and summer, the observed spatial characteristics of seasonal precipitation are reproduced fairly well in all models, suggesting that the principal precipitation-producing mechanisms in the basin are reasonably well-represented in the models.


Although evapotranspiration (E) plays a critical role in the coupling between the atmosphere and surface branches of the water cycle, it is in general much more poorly observed than precipitation, especially for remote regions like the MRB. Consequently, E is also a type-C variable in the analysis datasets. As surface solar radiation exerts a strong control on E, E exhibits strong seasonality over the basin ( Fig. 5.2.2b ). Evaporation is weak over the basin during DJF in general, with basin-average E varies from ~ 0.1 mm d-1 in ERA-40 and CMC to about 0.5 mm d-1 for NCEP-R2. The CRCM differs from others in that it is actually characterized by general weak surface condensation during the winter which only occurs over the extreme northern basin in the other datasets. This enhanced deposition in the CRCM during the cold season might have partially contributed to the high SWE bias in the CRCM that we have discussed earlier. With warmer temperatures, surface evaporation increases markedly with snowmelt during the spring and maximizes during summer where the basin-average E approaches or even exceeds P in all datasets. The relative low (high) bias that characterizes the CRCM (NCEP-R2) E is consistent throughout all seasons and they consequently give the lowest and highest values for the annual E estimate, respectively. It is of interest to note that although vastly different treatments of surface and boundary layer models are used in the CMC and ERA-40 analysis, their annual E estimates agrees to be within 0.1 mm d-1. Previous MAGS estimate of E for the basin was conducted by Louie (2002) by using the empirical model of Morton (1983). Their estimated annual basin-average E value of about 0.76 mm d-1 is comparable to the CRCM value but substantially lower than others. If we neglect changes in surface water storage over the study period then E ~ N – P; using the observed discharge and CANGRID precipitation data suggest that annual basin-average E is about 0.8 mm d-1. Hence, it is believed that, at least on the annual and basin-scales, the E estimate from the CRCM might be closer to the “truth” than the estimates from the analysis data.


As a high-latitude continental basin, the MRB is naturally a moisture sink region in the global water balance. As discussed earlier, moisture convergence into the region is strongly affected by the synoptic systems over the North Pacific. However, the coastal mountains shield off much of the moisture transported into the continent from the ocean through precipitation over the windward slope and thus effectively reducing the moisture convergence into the basin. Nevertheless, large-scale moisture convergence into the basin plays an important role in producing the precipitation over the region. Based on the 45 year archive, the correlation coefficient between daily ERA-40 MC and precipitation is about 0.56 during July and about 0.26 during January over the central basin (Slave region). The correlation is even stronger for high-elevation regions such as the Liard sub-basin where r = 0.7 and 0.5 for the same months.

The seasonal variability of MC over the basin is presented in Fig. 5.2.1c . Except for the CRCM, seasonal variability of MC is relatively low in all datasets and typically with maximum moisture convergence occurring in October and minimum in August. The NCEP-R2 and CRCM exhibited a high bias of net moisture convergence during the cold season which is probably a consequence of the under-prediction of topographic precipitation over the coastal regions in the two datasets when compared to CMC and ERA-40. It is of interest to note that despite the strong evapotranspiration that occurs in the basin, the basin on the whole remains as moisture sink during the summer in most datasets. The only exception is found in the CRCM budget which suggests that the basin becomes a source of moisture for the atmosphere during August when the its MC estimate is lower than others ( Fig. 5.2.1c ).

The annual-average MC is highest in NCEP-R2 (0.69 mm d-1) and lowest in ERA-40 (0.46 mm d-1). Previous estimates of annual MC by Walsh et al. (1994; 0.67 mm d-1), Roads et al. (2002; 0.67 mm d-1 estimated by using the same R2 data but for 1988-1999, a value that we have verified by re-computing the R2 MC for the same period) and Strong et al. (2002; 0.73 mm d-1) are all closer to the NCEP estimate. It is arguable that MC estimates from the CMC and ERA-40 analyses might be closest to the “truth” because of the higher resolutions and generally more sophisticated model physics that are employed in these models, as well as the smallest difference between these MC estimates and the observed discharge from the basin. Although the annual estimates of MC in the CRCM is close to those from the CMC and ERA-40, it is actually a result of compensating bias during different seasons (relative high bias during the cold season and low bias during the summer).

6.3.9 RUNOFF

Although runoff is an important component in the surface water budgets, it is only poorly simulated at best in the model datasets that are used in this study. In the CMC analysis, runoff is simply computed as P-E. As such, we will not further discuss the CMC runoff. Although quite sophisticated surface modules are included in the CRCM, NCEP and the ERA-40 models, runoff processes, especially cold region runoff processes are not well represented in these surface models. In addition, there is no river routing scheme in the versions of the models used to generate the data used in this study.

The annual cycle of basin-average runoff is presented in Fig. 5.2.2c . For comparison, the mean discharge from the Mackenzie River at Arctic Red for the same period is also shown. Most river flow in the basin follows a nival regime in which spring melt generates high flows that are orders of magnitude larger than the winter discharge. The flow declines after the spring freshet but it might increase occasionally by summer rainstorms. With the exception of runoff in the CMC analysis, all datasets exhibit to some degrees these runoff characteristics over the MRB. However, none of the model runoff replicates faithfully the phase and magnitude of the observed discharge. Runoff in ERA-40 maximizes in spring during the snowmelt period and its basin-average runoff is quite comparable to the observed streamflow data although its maximum runoff occurs two months earlier than the peak observed discharge and its summer values are lower than those of the observed discharge. These discrepancies between the model runoff and measured discharge are of course expected because of the large size of the basin and the lack of river routing implemented in the IFS model used for the analysis.

For the CRCM, the model runoff peaks a month later, in May; this runoff is all drainage from the model’s deepest layer, which is parameterized to increase rapidly once a threshold soil water is reached in the model base layer (100–289 cm). Both the peak and annual model runoff is higher than the observed discharge as one might expect from the high bias exhibited in its model snowcover (Table 5.1).

As NCEP-R2 exhibits low bias in its snowcover, its runoff from the snowmelt is slightly less than those from ERA-40 and the CRCM. The NCEP model, however, significantly over-predicts its runoff from precipitation during the warm season when compared to other modelled or observed data.


As discussed earlier, the MRB, as a high-latitude continental basin is an important heat sink region in the global energy budget. While the coastal mountains shield off much of the moisture transported into the continent from the ocean through precipitation over the windward slope and effectively reducing the moisture convergence into the basin, the condensation heating effectively enhances the convergence of dry static convergence (HC) into the basin. In fact, HC for the MRB is the highest among the mid- and high-latitude GEWEX CSE basins (see Table 1 and Fig. 11 of Roads et al., 2002). Contrary to the MC for the basin, the basin-average HC exhibits strong seasonality ( Fig. 5.2.3b ). As expected, HC into the basin is strongest during the winter when the pole-to-equator and continental-oceanic temperature contrasts are the greatest. Estimates of HC from all datasets suggest that the basin is a heat source for the circulation (i.e., HC <0) during the summer. The annual basin-average HC for the basin agrees reasonably well among the analysis datasets (within 0.03 K d-1 or 10% of each other, Table 5.1) while the annual HC from the CRCM is biased high among the estimates. Fig. 5.2.3b shows that the high bias of HC in the CRCM is largely found between the months of May and November (i.e., the basin is a much weaker heat source region during the summer in the CRCM when compared to others). The current estimates of HC values for the MRB are lower than the HC value of 0.51 K d-1 estimated by Roads et al. (2002) for the period 1988-1999 by using the same NCEP-R2 dataset. Re-computing the HC values for the same 1988-1999 period with the NCEP-R2 data confirms the value given in Roads et al., suggesting a recent decrease in the mean convergence of dry static energy into the basin.


Sensible heat is heat energy transferred between the surface and air when there is a difference in temperature between them. Since sensible heat flux (SH) is not part of the routine observations, it is a purely forecasted variable in all the assimilated datasets and it is thus strongly dependent on the model physics. All datasets agree that there is downward transfer of sensible heat into the surface in general during the cold season when the basin surface is substantially colder than the overlying air ( Fig. 5.2.3c ). While the magnitude of the basin-average cold-season SH is similar among the CMC, ERA40 and CRCM results, the NCEP-R2 SH is substantially stronger than the others( Fig. 5.2.3c ). For the basin as a whole, upward sensible heat fluxes occurs between March and September for all datasets except for NCEP-R2 which exhibits negative basin-average SH except for the months of June and July. The weaker upward SH in the CRCM during the spring is consistent with the low bias in Ts and high bias in SWE that are exhibited in its results. The lower than average SH in the NCEP-R2 analysis is reflected in its negative annual SH while all three other annual estimates are positive (Table 5.1).

6.3.12 CLOUDS

Although cloud amount is not one of the explicit variables that enter into the water and energy budget equations, it is included here for the discussion because of the critical roles it plays in affecting both the water and energy cycles in many different locations of the earth, including the MRB. Cloud amount is typically a type-C variable in the analysis datasets and it is also generally agreed that it is one of the more poorly simulated variables in current models. Most datasets show a weak seasonality for the mean cloud coverage (not shown), and both the models and manual observations suggest that maximum cloud coverage over the basin occurs during the autumn months between September and November and relatively low cloud coverage occurs during the spring between March and May. The high mean cloud coverage over the basin during the autumn is likely a result of the combined effects of high frequency of synoptic systems that visit the west coast of Canada as well as lee cyclones that commonly develop within the basin during the autumn. Annual basin-average cloud cover is lowest in NCEP-R2 with mean basin average below 50% and followed by CMC with mean cloud coverage of 56% (Table 5.1). Annual estimates from the ERA-40, ISCCP or the CRCM are all within 10% of the observed values of 65% with the CRCM values biased low, especially during the cool and cold seasons.


Although the MRB receives almost no incoming solar radiation for an extended period of the year, radiative transfer nevertheless plays a critical role in affecting its water and energy balance due to the strong coupling between radiative transfer and other hydrometeorological processes in the region. There are very little ground-based radiative flux measurements for the region and subsequently most information on the radiative fluxes in the region come from either satellite measurements or model data. Although radiative fluxes are type-C variables in the analyses, the variability among their estimates (e.g., measured by the spread among the estimates using the percentage error) are typically smaller than those for other flux variables in the budget estimates (Table 5.1) and much of the revealed variability can be related to the variability of cloud cover in the different datasets.

Although all models use fairly standard top-of-the-atmosphere (TOA) incoming shortwave (TOA_SWD) values in their specifications, the shortwave radiation that reaches the surface (BOA_SWD) exhibit substantial variability among the estimates depending critically on the model cloud cover (Table 5.1). In particular, the high bias of BOA_SWD in the NCEP-R2 and CMC data can be related to the lower model cloud cover in these datasets. The high BOA_SWU in NCEP-R2 can be related directly to its high BOA_SWD despite its lower than average SWE (and hence lower than average mean surface albedo during the spring). Top-of-the-atmosphere SWU, on the other hand, is highest in the CRCM, likely a result of the enhanced mean TOA albedo from the above average cloud cover and more extensive and longer duration of snowcover in the model.

Relatively low variability are found in the estimates for both the BOA and TOA longwave fluxes. Both the downward and upward longwave fluxes at the surface, as well as the TOA_LWU are slightly lower in the CRCM, presumably a result of the cold bias that is found in the model surface and lower-atmospheric temperatures as discussed earlier. On the other hand, the slightly above average BOA longwave fluxes in both the ERA-40 and ISCCP data can be related to the above average cloud cover in these dataset. In fact, the ERA-40 and ISCCP FD radiative flux estimates are very close to each other which could be partially related to their close estimates of cloud coverage for the basin. Both of them exhibit a 10% higher than average atmospheric radiative cooling (QR) for the basin while the CRCM is characterized by the lowest net surface radiative heating (Table 5.1).

Although all the datasets exhibit very similar seasonal variability of QR and QRS ( Fig. 5.2.3b and Fig. 5.2.4b ), there is substantial seasonal dependence shown in the relative biases of the flux estimates from the different datasets. In particular, the net atmospheric radiative cooling QR is typically strongest in ERA-40 data during the autumn, weakest in the CMC during the summer, and weaker in the RCM and NCEP-R2 during the cold season. For basin-average QRS, surface cooling (heating) is strongest in NCEP-R2 during the winter (summer). The results in Fig. 5.2.4b also show that the low bias in the annual QRS from the CRCM is mainly a results of the lower than average QRS in the model during the warm season, particularly between May and September.

6.4 Budget Closure and Error Analysis

One of the applications of WEBS results is the assessment of the completeness and correctness of our knowledge for the water and energy cycle of a region through examining the accuracies of the budget estimates and the degree by which the budgets can be closed on various spatial-temporal scales. In addition, such assessments often give useful information on the areas that we should focus our effort towards improving our model prediction of water and energy cycling in the region. Unfortunately, apart from routine precipitation, temperature and discharge measurements, there is little observation that we can used to compare with the model budget estimates for this remote basin.

The inter-comparisons of budget estimates with available observations was discussed in the previous section. Not surprisingly, WEBS parameters that are derived from strongly “corrected” variables in the analysis datasets (e.g., atmospheric enthalpy, screen temperatures and precipitable water) compare the best to observed values. For the purely forecasted fluxes that also have observations, precipitation estimates compare much better with observations than annual runoff values. Since precipitation in the MRB is the end result of many strongly-coupled hydrometeorological processes that occur either outside or within the basin, the fact that precipitation, including its spatial and temporal (at least on monthly and longer time scales) variability, can be simulated quite successfully in many models for the region (see Table 5.1, Fig. 5.2.2b , and Fig. 5.3.6) should give us a lot of confidence in the representations of northern water and energy processes in current models. On the other hand, the wide discrepancies between modeled and observed snowcover and runoff in the basin suggest that we might have to improve the cold region surface and runoff processes in the models before substantial improvements in runoff predictions for the MRB can be achieved.

When no measurement is available to validate the budgets, the spread of the budget estimates among the different datasets will give a measure of the uncertainties in their evaluations. While large variability in the budget estimates would indicate large uncertainties in their evaluations, it should however be noted that good agreements between the estimates might not guarantee that the estimates are accurate. Similar to the results from the intercomparison of budget estimates with observations, smaller budget estimate variability are found in parameters that are derived from strongly-corrected analysis variables and wider spreads are found in the purely forecasted flux variables (e.g., in evapotranspiration, runoff and sensible heat flux, see Table 5.1). Exceptions are found in the estimates of precipitation and radiative fluxes. In particular, when neglecting the NCEP-R2 and global blended precipitation, annual precipitation estimates from the various sources, and many of the radiative flux estimates agree with each other to be within 10% of the corresponding ensemble mean values.

As mentioned earlier, an accurate and complete quantitative characterization of the water and energy cycle for a region requires both accurate evaluations of the budget components and adequate closure of the budget balance. The closure of the budgets from the various datasets is given conveniently by the residues in balancing their corresponding budgets. Theoretically, there should be complete balance in the water and energy budgets from a single model. However, budget imbalance might occur as a result of non-conservative numerical schemes that are employed in the model or from errors that might have occurred in off-line budget computations with archived model outputs. Nevertheless, the residues in balancing the surface and atmospheric water and energy budgets for the CRCM are in general smaller than those for the analysis datasets. Residues in balancing the budgets from the analysis datasets are generally expected because of the nudging of the forecast variables with observations during an analysis cycle. For example, as discussed in Betts et al. (2003), the liquid hydrological budget is not balanced in the ERA-40 for two reasons, the spinup of the model hydrologic cycle and the imbalance in the model analysis cycle. The model analysis cycle is not closed, because soil water is added through a nudging term calculated from short-term forecast errors in the humidity at the lowest model level. In addition, non-closure of water and energy budgets for a region in the analysis datasets could be a result of neglecting important processes in modeling the water and energy cycle for the region. In the present case, the neglect of lake effects or snow ablation effects such as sublimation in blowing snow in the models could have significant effects in affecting the water and energy budget estimates from the analysis data. In addition, results from Szeto (2006a) and Strong et al. (2002) suggest that the diurnal cycle plays an import role in governing the warm-season water cycling in the region. Inaccurate representations of these processes in the models or the inappropriate sampling frequency used in archiving the data could thus strongly affect the water budget assessments from these datasets.

Results in Table 5.1 show that the residues in balancing annual atmospheric water budgets (RESQ) can ranges from ~25% of observed runoff (for CRCM), to >50% for CMC and over 100% for NCEP-R2. Similarly, residues in closing the atmospheric energy budgets (REST) are in general comparable in magnitudes to the budget terms themselves. Residues are generally smaller in the surface budget balances, suggesting that a large part of the inaccuracy in closing the atmospheric budgets might come from numerical errors that are introduced during the procedure of interpolating data from model grids to the archiving grids, and subsequently propagated into the computation of the convergence terms. It is anticipated that improved closure might be able to be achieved if we can get access to budgets that are computed in-line with the model simulations. It is also of interest to note that there are characteristic seasonal dependencies exhibited in the residue that could vary from dataset to dataset (Fig. 5.2.1c, Fig. 5.2.2c, Fig. 5.2.3d, and Fig. 5.2.4d).

Since changes in the atmospheric and surface water storage can be neglected in the long term, (i.e., MC ~ P-E ~ N), the water budget closure is traditionally assessed by the balance between the long-term average atmospheric moisture convergence and observed runoff. With this definition, the regional water budget for the MRB is closed to be 6, 8 and 10 % of the observed runoff using the moisture convergence from ERA-40, CMC, and CRCM, respectively. These are substantial improvements over the closure of ~26% assessed by using the previous generation CMC analysis dataset (Strong et al., 2002); and these improvements possibly reflect the recent advances in the modeling of atmospheric water cycling processes for the region. Nevertheless, more useful measures of the how well we can close the water and energy budgets are given by the residues in balancing the budget equations with the different datasets along with the inter-comparisons of the individual budget parameters estimates with available observations as discussed in the above. In that regard, much improvements in the current models are still needed before we can make use of their results to improve substantially the water and energy budget assessments for the MRB.

6.5 Summary and Conclusions

This study represents the first attempt at developing a comprehensive climatology of water and energy budgets for the Mackenzie River Basin. Different observed, remotely-sensed, (re-)analyzed and modeled data were used to obtain independent estimates of the budgets. Apart from the development of state-of-the-art budget estimates for the MRB, the capability of current models and data assimilation systems in capturing the water and energy cycle of this northern and data-spare region were also assessed.

Although the CRCM simulation was performed in "climate mode", the model simulated a very respectable climate for the MRB when compared with observations and analysis data. Noteworthy points for its basin budgets include its substantially weaker than average HC, MC and E during the warm season and its delayed snowmelt (and subsequently weak SH and over-estimated peak runoff) during spring. All of these budget biases can be partially attributed to the strong cold bias in its low-level tropospheric temperatures within the MRB in the model, especially during the cold season.

Despite the big differences between the model resolutions (24-35 km for GEM and T159 or ~75 km in the region of interest for ERA-40) and model physics (for instance, there is no real surface module in the GEM model during most part of the study period) that are employed in the CMC and ERA data assimilation systems, the water and energy budgets derived from these two datasets for the MRB are very similar, and in general compared the best to available observations.

Although the NCEP-R2 re-analysis have been used in numerous previous hydrometeorological studies related to the MRB, the water and energy budgets evaluated for the basin from the datasets exhibit the strongest deviation from the ensemble mean budgets and are compared most unfavorably to available observations in general. In particular, the R2 dataset represents a significantly more intense warm-season water cycle for the MRB than the ones assessed from other datasets. The NCEP model produced consistently higher surface evaporation than others throughout the year. The surface sensible heat fluxes from the R2 datasets also differ from others in that its sensible heat flux into (from) the surface during the cold (warm) season is substantially stronger (weaker) than those from other datasets. These results suggest that NCEP-R2 dataset should be used with cautions for hydrometeorological studies for the basin.

Although most global satellite or blended datasets are known or expected to perform poorly in northern regions for obvious reasons, it is still of interest to quantify their merits and deficiency in representing the water and energy budgets in northern regions through intercomparisons with other datasets. In brief, the results show that (i) the annual basin average precipitable water estimates from the NVAP dataset compare extremely well with those estimated from analysis datasets; (ii) both the CMAP and GPCP precipitation are lower than others with the low biases particularly worse during the summer; (iii) the ISCCP FD radiative fluxes compare closely with estimates from others (particularly the ERA-40 fluxes).

The spread of budget estimates from the different datasets are large. However, it is noteworthy that the different estimated budgets (although to a lesser extent for the NCEP and some of the global satellite datasets) have exhibited very similar interannual variability to available observed budgets, suggesting the feasibility of using these synthetic or analysis datasets in studying the inter-annual variability of water and energy cycling for the region.

The regional water budget for the MRB is closed to be 6, 8 and 10 % of the observed runoff using the moisture convergence from ERA-40, CMC, and CRCM, respectively. While these are noted improvements over the previous water closure assessments for the region, magnitudes of the residues in balancing the budgets are often comparable to the budget terms themselves in all the model and analysis datasets, suggesting that substantial improvements to the models and observations are needed before we can vastly improve the assessments of the water and energy budgets for this northern region.

The climate of the MRB is governed by complex interactions between the atmosphere and surface features and processes that occur on a wide range of spatial-temporal scales. Some of these processes are common among cold regions while some of them are specific to the MRB. A number of these processes are generally not (e.g., organic soil and evaporation from the northern lakes), or are only crudely (e.g. ground frost processes, topographically-induced precipitation, sublimation from canopy-top) represented in current climate models. These limitations will certainly affect the representation of the region’s water and energy cycle in the models and thus the water and energy budgets that are estimated from the model results. The improved understanding of these processes in MAGS and the incorporation of the knowledge into the numerical models will certainly enhance our predictive capability for the basin, and the results from this study will provide a reference climatology to gauge the progress in future budget estimates from these improved models and newly available remotely sensed data.