The Spatial Extent of Hydrological and Landscape Changes across the Mountains and Prairies of the Saskatchewan and Mackenzie Basins

Abstract. East of the Continental Divide, the cold interior of Western Canada has one of the world's most extreme and variable climates and is experiencing rapid environmental change. In the large Saskatchewan and Mackenzie River basins, the warming climate is changing the landscape, vegetation, cryosphere, and water. This study of a large number (395) of gauged basins in these large river basins provides the basis for a large-scale analysis of observed hydrological and landscape changes. In this region, the existing data sets are complex; observed streamflow records are available for differing series of years and streamflow measurements consist of both continuous and seasonal records. This diversity has been compensated for using novel analytical approaches: [1] a Streamflow Regime classification using dynamic time-warping that covers only the common period of the calendar year amongst all stations, and which is restricted to a time window of seasonal observations, [2] a classification of seasonal Streamflow Regime change using k-means clustering of the year divided into five-day bins. An assessment of landscape and hydrological storage change for each gauged basin was conducted using Landsat 5 TM imagery of Normalized Difference Vegetation, Water, and Snow Indices (NDVI, NDWI, and NDSI) for 1985 to 2010. Therefore, this analysis is for a different time period than the streamflow regime and trend patterns. Twelve Streamflow Regime Types were identified using dynamic time-warping to overcome the problem of timing differences producing flow clusters due to latitude or elevation, rather than from the shape of the hydrograph resulting from differing processes. The success of this approach suggests that there is sufficient information in the time window to adequately resolve regions; Streamflow Regime Types exhibit a strong connection to location; the spatial distribution follows ecoregions and shows a strong distinction between mountains and plains. Clustering of seasonal trends resulted in six Trend Patterns. The Trend Patterns also have a strong and distinct spatial organization. The Trend Patterns include one with decreasing streamflow, four with different seasonal increasing streamflows, and one without any apparent trend structure. Trends in the mean, minimum, and maximum of three satellite indices were determined; the spatial patterns of NDWI and NDSI were similar to each other, but NDVI patterns were generally dissimilar. Streamflow Regime Types, the Trend Patterns, and satellite indices showed spatial coherence. The overlap between hydrological and landscape change was not perfectly coherent, suggesting that landscape changes may have a different domain from the existing hydrological regimes and from the observed trend patterns. Three particular areas of change are identified: [i] north of 60° where streamflow and greenness are increasing while wetness and snowcover are decreasing, [ii] in the western Boreal Plains where streamflows and greenness are decreasing but wetness and snowcover are not changing, and [iii] across the Prairies where there are three patterns of increasing streamflow and an increase in the wetness index, the largest changes occur in the eastern portion of the Canadian Prairies, with only few increases in greenness and snow indices. The results demonstrate the spatial extent of these changes.


7 time series with step changes. Trends of these types should be based only on years with complete record and are not addressed further. In the supplementary information, Figures S1-S12 show up to four example hydrometric stations for each Streamflow Regime cluster as 160 described below. Many of these plots show that the operation of stations have alternated between being seasonal and continuous (e.g. Figure 2). These also demonstrate the variability of period of record between stations.

Streamflow Regime Types
Statistical methods, such as k-means (Likas et al 2003;Steinley 2006) or self-organized maps (Kohonen and Somervuo 1998;Hewitson and Crane 2002;Kalteh et al 2008;Céréghino and Park 2009;van Hulle 2012), are unable to group hydrographs when they are not aligned in time (Halverson and Fleming 2015). Across this study domain this is a difficulty as timing of snow 170 accumulation and melt are strongly affected by both latitude and elevation.
To avoid the effects of gauged basin area and location, the five-day streamflow records were converted to Z scores, by subtracting the mean value and dividing by the standard deviation, of the series five-day. Early snowmelt at low latitudes and elevations resulted in some stations having flow events preceding the sampled date range (Figure 3). Only the data in the periods 175 between the two vertical dashed lines in Figure 3 were used in the clustering (and trend analysis) reported here. https://doi.org/10.5194/hess-2019-671 Preprint. Discussion started: 28 January 2020 c Author(s) 2020. CC BY 4.0 License.

Streamflow Regime Types
The Streamflow Regime Types from the twelve-cluster solution are shown in Figure 5.  Figure S1). Streamflow Regime Type 2 basins are 285 reflective of Prairie streams with spring snowmelt and long periods with low or zero flow ( Figure S2). Streamflow Regime Type 3 basins are in the Athabasca River Basin ( Figure S3).
Streamflow Regime Type 4 ( Figure S4) have both strong snowmelt and late summer streamflow. Streamflow Regime Type 5 basins are predominately Boreal Plains ( Figure S5).
Basins from Streamflow Regime Types 6-8 and 10 are unique (or nearly so) but are similar to 290 adjacent types ( Figures S6, S7, S8, and S10). An interesting feature of these four types is that peak events occur at different times in different years. Streamflow Regime Type 9 basins peak later in the summer, and are generally smoother that other basins ( Figure S9). Streamflow Regime Type 11 basins have an early snowmelt peak and higher flows extending through into the fall ( Figure S11). Streamflow Regime Type 12 basins have an early snowmelt peak and 295 persistent high flows during summer and extends into the fall ( Figure S12).
Figures S1-S12 The spatial locations of the twelve Streamflow Regime Types are mapped over ecozones in Figure 6; there is a clear spatial organization and not a random pattern. This association is also https://doi.org/10.5194/hess-2019-671 Preprint. Discussion started: 28 January 2020 c Author(s) 2020. CC BY 4.0 License. evident in Table 2. Two large-scale features are evident; similar types tend to be from the same 300 spatial areas, and some similar types follow along major rivers. Streamflow Regime Types 3 and 11 follow along rivers and Types 2 and 5 overlap. Streamflow Regime Types 1 (104 members) and 5 (148) occur in the greatest numbers of ecozones (8 and 6 respectively; Table   2). demonstrates the persistence of a mountain runoff signal along the Athabasca River as this hydrograph contains the late melt signal from glaciers ( Figure 5). Streamflow Regime Type 2 basins (n=85) are associated with the Prairie ecozone (n=57) and Boreal Plains (n=28), but this patterns also occurs in the Southern Arctic (n=1) and Taiga Plain (n=3); this pattern has the 320 earliest snowmelt and the records often start with snowmelt already in progress ( Figure 5). there is overlapping of types in space (particularly Streamflow Regime Types 2 and 5), and cases where individual basins of a Type occur quite separately from each other (Types 9 and 12) as is evident in Figure 6). Types, both similarities and differences exist between the basins.

340
Streamflow Regime trend patterns are based upon the statistical trends in five day flows discussed above. Figure   shown in blue and red, no trend in gray, and no available data in white. These data were ordered by the six Trend Patterns determined using only the data for the five-day periods from 23 to 61; Figure S14 shows the data order by stationID. The periods 1-22 and 62-73 were not included in the clustering, but were plotted as they may also be of interest.   (Table 3). Trend Pattern 5, which lacks a pattern, is very prominent in most of the Streamflow Regime types having many members. Again, the caution regarding rivers sourced in the mountains and 365 continuing downstream propagating the upstream signal also applies to their trend pattern as well. The six Trend Patterns of the 395 hydrometric stations are mapped to ecozones in Figure 9. At the ecosystem scale, 51% of basins in the Prairies exhibit a definite Trend Pattern with 45% showing one of the increasing patterns (Trend Patterns 1, 2, 4, and 6) and 6% the decreasing Trend Pattern 3 ( Pattern (5%) ( Table 4). Trend Pattern 6 only occurs in the northern portion of the study area.
None of the stations on the Hudson Plains showed any change pattern. This indicates that there is a spatial basis for the Streamflow Regime change that is influenced by location and ecozone rather than by Streamflow Regime Type.  Table 4 near here.

Landscape and Hydrological Storage Trends
The trends in the mean values of the three normalized difference indices are presented in Figure  triangles, trends whose significance is less than 0.1 are shown as red or blue dots, and those 405 with no trend in black. There is a stronger association of the trends in the three indices with spatial location and with ecozones than with Streamflow Regime type or trend pattern.
Frequently, the landscape and hydrological storage trends occur in a spatial domain that spans two or more ecozones ( Figure 10, Figure S21 to S23). The fraction of stations having statistically significant trends in NDVI (landscape trends) are 415 greater than the 5% expected by chance alone across most Streamflow Regime Types (Table   5). Table 5 shows the combination of increasing and decreasing trends and each separately.
For example, the fraction of all significant trends for maximum NDVI exceeds 5% in Streamflow Regime Types 1, 2, 3, 5, 9, and 10. The fraction of significant decreasing trends for maximum NDVI is greater than 5% for Streamflow Regime Types 1, 3, 5, 9 and 10. The 420 fraction of increasing trends in maximum NDVI exceeds 5% in Streamflow Regime Types 1, 2, and 5. All significant trends in mean NDVI are increasing and occur in Streamflow Regime Types 1, 9, 11, and 12. Significant trends in minimum NDVI are predominately decreasing in Streamflow Regime Types 2, and 5 and increasing in Streamflow Regime Types 1, 11 and 12 and both increasing and decreasing trends in Streamflow Regime Type 4. 425 There are also more significant trends in NDVI than would be expected by chance alone across all Trend Patterns (Table 6). The fraction of basins with significant trends in maximum NDVI exceeds 5% in all Trend Patterns, except Trend Pattern 3 which had no basins with significant trends. All mean NDVI trends are increasing and occur only in Trend Patterns 1, 4, and 6.
Minimum NDVI trends are predominately decreasing in all six Patterns and increasing in Trend 430 Patterns 4 and 6.
The association of trends in mean NDVI between 1985 and 2011 with ecozone are each shown in Figure 10a ( Figure S21) and Table 7. There is a stronger association of NDVI trends with ecozones than with either Streamflow Regime Types or Trend Patterns. Increasing trends in mean NDVI occur in the Taiga Plains, Taiga Shield, Boreal Shield, Boreal Cordillera and Taiga   435 Cordillera ecozones, decreasing trends in mean NDVI are found more often than expected by chance alone in the Montane Cordillera ( Figure 10a, Table 7). The spatial patterns of trends in mean NDVI are similar to those of the maximum and minimum NDWI; however, there are more basins with significant trends in maximum and minimum NDVI than for mean NDVI.
Basins with significant increasing trends in maximum NDVI are found in the western portion increasing trends were found in the Boreal Shield, Montane Cordillera, Boreal Cordillera, and Taiga Cordillera ( Figure S21c). Basins with significant trends are not randomly distributed through any ecozone, as spatial clustering is evident in each of the three NDVI trends in Figure   S21.

NDWI
There are more significant trends in NDWI (hydrological storage trends) than would be expected by chance alone in most Streamflow Regime Types (Table 5) and this is more prominent in the mean and minimum NDWI than in maximum NDWI. The fraction of significant trends in maximum NDWI exceed 0.05 in Types 2, 5, 9, 11, and 12 with Streamflow 455 Regime Types 9, 11, and 12 showing increasing trends much greater than the threshold (>0.20), while Types 2, 3, and 5 show decreasing trends but only near the threshold value.
Significant trends in mean NDWI include both decreasing in Streamflow Regime Types 1, 3, 4, 9, 11 and 12 and increasing in Streamflow Regime Types 2, 3, and 5. Significant trends in minimum NDWI are decreasing in Streamflow Regime Types 9 and 11 and increasing in 460 Streamflow Regime Types 2 and 12 with both increasing and decreasing trends in Streamflow Regime Types 1, 3, 4, and 5. The largest fraction of significant trends (> 0.33) are decreasing trends in mean NDWI in Streamflow Regime Types 9, 11, and 12.
There are more significant trends in NDWI than would be expected by chance alone for all Trend Patterns (Table 6). Significant decreasing trends in maximum NDWI exceed the 465 threshold fraction in Trend Patterns 1 and 6, and increasing trends in Trend Patterns 3 and 4 (Table 6). Decreasing trends in mean NDWI exceed the threshold in Trend Patterns 1 and 6, as do increasing trends in Trend Patterns 1 to 5. The fraction of basins with decreasing trends in minimum NDWI exceeded the threshold in Trend Pattern 5 and for increasing trends in Trend Patterns 1, 4, 5, and 6. Only Trend Pattern 5, i.e. without a trend pattern, was found to 470 have both increasing and decreasing trends in minimum NDWI.
The association of trends in mean NDWI with ecozones are shown in Figure 10b and Table 7 ( Figure S22 shows results for maximum, mean, and minimum NDWI). Similar to NDVI, there is a stronger spatial association of NDWI trends with ecozones than with either Streamflow Regime Types or Trend Patterns. Decreasing trends in mean NDWI occur in the northern There are also more significant trends in NDSI (hydrological storage trends) than would be expected by chance alone across Streamflow Regime Types (Table 5) and this is most prominent in the minimum NDSI. Significant decreasing trends in maximum NDSI exceed the 0.05 threshold in types 9, 11, and 12; increasing trends exceed the threshold in Types 2, and 5.
Only decreasing trends in mean NDSI were detected in Streamflow Regime types 1,4,9,11,495 and 12. Both decreasing and increasing trends in minimum NDSI occurred in Streamflow Regime types 1, 3, 4, and 11; in Streamflow Regime Type 9 only decreasing trends in minimum NDSI were found, and only increasing trends in Types 2, 10, and 12. The fraction of stations showing increasing trends in minimum NDSI is much greater than decreasing trends.
There are more significant trends in NDSI than would be expected by chance alone across all 500 Streamflow Regime trend patterns (Table 6) with increasing and decreasing having similar frequencies. The fraction of significant decreasing trends in maximum NDSI exceed the threshold in Streamflow Regime trend patterns 1 and 6 (as was true for NDWI), and increasing trends in Trend Patterns 3, 4, and 5 ( The association of trends in NDSI with ecozones are each shown in Figure 10c, Figure S23, and  Table 7); there were insignificant increasing trends in mean NDSI, but significant increases are evident in Figure 10c in the Prairies and Boreal Plains. There are many more significant trends in minimum NDSI than in either maximum or mean NDSI (Table 7, Figure S23). The spatial 515 patterns for mean NDSI are similar to those for maximum and minimum NDSI. Basins with significant decreasing maximum NDSI trends are found in the Taiga Plains, Taiga Shield, and Boreal Cordillera ( Figure S23a, Table 7); decreasing maximum NDSI are found in the eastern portion of the Boreal Plains, and in the Prairies. Trends in minimum NDSI are more numerous than for maximum or mean NDSI (Table 7) and include both increasing and decreasing trends 520 in the Taiga Plains, Boreal Shield, and Boreal Plains ( Figure S23c). Only decreasing minimum NDSI trends were found in the Montane Cordillera, and increasing trends in minimum NDSI only occurred in the Prairies, Boreal Cordillera, and Taiga Cordillera. Basins with significant trends are not randomly distributed through any ecozone, as spatial clustering is evident in each of the three NDSI trends in Figure S23. 525

Trend Patterns and Landscape Change
Trend pattern 6 ( Figure S20), with prominent winter increases in streamflow across the north (Taiga and subarctic) was originally described by Whitfield and Cannon (2000). This region also has significant increases in mean NDVI (Figure 10a), and decreases in mean NDWI and 530 NDSI (Figure 10b). At this scale, one possible interpretation of the observed trends would be that warming has altered the seasonal pattern of permafrost and increased winter flows and also the greenness of these basins which suggests higher evapotranspiration, (Figures 10a and S21) and reduced the amount of standing water (Figures 10b and S22), and the snowcovered period   Table 8. Typically the first recession phase is steeper than the second phase where there is one.
In Streamflow Regime Type 11 the third phase is steeper than the second, but not as steep as the first. After a rapid recession, Streamflow Regime Type 2 becomes nearly horizontal, 605 probably due to Prairie streams typically having no base flows. In Streamflow Regime Type 5 the recession has two linear phases, and also terminates in a horizontal section, which is again

Landscape and Hydrological Storage changes
The primary interest in landscape and hydrological storage change in this study is to determine if changes, as evidenced by trends in normalized difference indices (NDVI, NDWI, and NDSI) The subarctic climate of the Boreal region has large inter-annual variability and will be prone 745 to future climate change (Woo et al 2008); models suggest that future winter flows will increase, spring melt will advance, but peak and summer flows will decline because ET will increasing wetness near that meridian is shifting westward in Canada.

Mountains
Most of the basins (73%) in the Cordillera (Montane, Boreal, and Taiga) fall into Change Pattern 5 (Table 4, Figure S19) which has a general lack of structure in changes. Many of these basins show a few periods with increases or decreases in flow for several periods 800 consistent with freshet timing changes (e.g. Figure 1 and Figure 2), but there is a lack of consistency as indicated by the inability for a cluster of similar patterns to be formed   Figure 11. Types with * have only one member and are excluded here.