Spatial Knowledge and Information Canada, 2019, 7(2), 1 A novel method to analyze trends of extreme precipitation distribution- A case study over the Arctic regions of Canada CHIRANJIB CHAUDHURI COLIN ROBERTSON Dept. of Geography and Environmental Studies Dept. of Geography and Environmental Studies Wilfrid Laurier University Wilfrid Laurier University cchaudhuri@wlu.ca crobertson@wlu.ca inconclusive. Our analysis emphasizes the ABSTRACT need for robust trend estimation methods in the arctic regions where data uncertainty is Climate change is thought to be changing very high due to low station density. precipitation regimes in northern Canada, Standard trend analysis of the gridded particularly in terms of magnitude and the observation data may lead to false positive frequency of extreme events. Given the large trends. geographic size of the Arctic and the relative sparsity of meteorological stations with 1. Introduction: historical weather observations, characterizing large scale climate trends Extreme precipitation events are one of the remains challenging. This paper analyzes the major influencing factors in the design, trend of extreme precipitation over the analysis, and operation of various water Arctic regions of Canada through the resource infrastructure. Climate change has analysis of a gridded (10km x 10km) potential to alter precipitation patterns, precipitation dataset to calculate the annual intensities, and extremes. Various global maximum daily precipitation between 1950- climate modelling studies indicate that 2010. However, the different grid points in global warming can have more prominent any derived gridded datasets have spatial effects on extreme precipitation than on auto-correlations which can potentially mean precipitation (Mailhot et al., 2007; induce bias in the trend estimation, when Bates et al., 2008). considering all the grid points. To mitigate these problems, we proposed a novel spatial- The Arctic regions of Canada are warming pooling method which creates a spatially much faster than the lower latitudes through decorrelated variable distribution a process referred as Arctic Amplification considering the effective correlation radius (AA) (Francis and Vavrus, 2012; Serreze of the variable. Furthermore, we analyze the and Barry 2014; Overland et al. 2015). spatially decorrelated precipitation time- Glisan and Gutowski (2014) indicate that the series using the extreme value theory in the regions of the Arctic which have low-level context of long-term low-frequency convergence of moisture are prone to variability. The trend pattern suggests extreme precipitation events. Furthermore, increasing frequency and variability of the recent increase of occurrences of extreme precipitation over the Southern cyclonic activities can possibly add to the Arctic regions. Analysis of extreme moisture content of the atmospheric column precipitation over the northern Arctic was and which in turn may lead to stronger 2 Trends of extreme precipitation distribution extremes such as extreme precipitation method (Hutchinson et al., 2009) with scales to total column moisture content latitude, longitude and elevation as (IPCC, 2013). predictors. Furthermore, a Canada ecozone shapefile from the CGDI National Thus, in the context of present and future Frameworks Data was used to define the possible climate changes, it is important to boundaries of the Northern and Sothern study the change characteristics of extreme Arctic regions. precipitation over the Arctic regions of Canada. However, the direct application of The analysis methodologies consisted of a analytical methods developed and used sequence of steps. First, the annual where station density is high are likely to maximum precipitation is calculated for perform poorly in areas where the station each of the grid points for the duration 1950- density is much lower. Careful consideration 2010 using the gridded precipitation dataset. of both process spatial dependence and Secondly, the precipitation grids belonging induced dependence through interpolation to each of the eco-regions are identified models is needed. Accordingly, the using ecozone boundaries. The semi- objectives of our study are as follows: variogram of the median extreme i. To provide a novel method to precipitation for each of the eco-regions is analyze the trend over gridded then calculated and used to determine the datasets where nearby grid points correlation range of extreme precipitation can be highly correlated. within each ecozone. Hexagonal grids with ii. Analyze the trends of spatial size (center to center distance) equal to the mean and standard deviation of correlation range are then generated. The annual mean precipitation over hexagonal grid ensures the equal distance of Northern and Southern Arctic neighbors in each direction. regions of Canada. iii. Analyze the spatial variation of The precipitation data are then mapped to trends of extreme precipitation. the center of the grid using a nearest iv. Analyze trends of return period neighbor approach. This procedure finally level for extreme precipitation produces the hexagonally gridded de- within an extreme value analysis correlated extreme precipitation time-series framework. for each of the eco-regions. The temporal trend of spatial mean and spatial standard 2. Methods and Data: deviation are estimated using the mean and standard deviations of hexagonal grid The Pacific Climate Impacts Consortium centres over each of the eco-regions. In (PCIC) NRCANmet gridded (10km x 10km) addition, the trends of individual hexagonal precipitation dataset (Hopkinson et. al., grid points are also calculated. Finally, 2011; Hutchinson et. al., 2009) were return period precipitation levels (RL) are obtained for all of Canada; years 1950-2010. estimated with 30-year moving windows The NRCANmet observational dataset was using Generalized Extreme Value (GEV) produced by Natural Resources Canada distributions fitted on each of these (NRCan). Gridding was accomplished with windows. Then, we estimate the return the Australian National University Spline period precipitation levels from the GEV (ANUSPLIN) implementation of the tri- distribution. The trends of 30-years RL and variate thin plate splines interpolation 100-years RL are analyzed in this paper. Trends of extreme precipitation distribution 3 The empirical variogram models (Figure 2) give the estimated range parameters of 2.1 Study Area around 32,000 km for Northern Arctic and around 286 km for Southern Arctic region. The comparison focused on in this paper is The sparse station density on the Northern the two arctic ecozones of Canada. The Arctic region likely contributes to the larger Northern Arctic Ecozone is the coldest and range of this region. Figure 3 and 4 presents driest landscape in the Arctic which the trends of spatial-mean of annual extreme comprises the non-mountainous portions of precipitation over Northern Arctic and the Arctic Islands as well as the Southern Arctic regions using both the grid northernmost areas of Quebec. The mean points and sub-samples. For the Northern annual precipitation is very low, ranging Arctic region, the trend computed from all from 10–20 cm. The Southern Arctic the grid points is significant at 99% Ecozone covers much of the northern confidence level and has median estimate of mainland of Canada, from the the Yukon 0.03 mm/year. Considering around 10 mm Territory to northern Quebec. An annual of intercept this trend indicates around 24% mean precipitation of 20–50 cm is observed increase of precipitation in 60 years. here. However, the sub-sampling method reveals no significant trend. The low station density in this region can lead to this type of 3. Results contrasting trend estimation when we consider the spatial correlation into the trend Figure 1 presents the eco-zone boundaries of estimation method. The Southern Arctic the Northern and Southern Arctic overlaid region also has 99% confidence level trend with the hexagonal grid which are with median estimate of 0.05 mm/year when constructed as part of this study and the considering all the grid points (Figure 3). Environmental Canada meteorological Like the Northern Arctic, this trend indicates stations within the study zones. In the inset an increase of 23% during the studied 60- the boundaries of all the eco-zones are year period. The higher station density in shown and the Northern Arctic and Southern this region enables us to put greater Arctic are highlighted in yellow. The confidence in this estimate. This is also hexagonal mesh over each of the ecoregions reflected in sub-sampling estimate of with sizes equal to the range parameters of precipitation which had also detected a the corresponding variogram. The hexagonal significant trend. grids are clipped at the boundary of the region and the centers are recalculated as the The trends of spatial-standard deviation of centroid of the clipped region. This ensures annual extreme precipitation for Northern the mapped nearby grid points are taken Arctic region the trend is not significant at from the ecozone itself. Notice the sparsity 99% confidence level and has median of the stations in these areas. This low estimate of 0.02 mm/year (Figure 5). The density can be attributed to lower population sub-sample estimation of standard deviation and economic activities in this region, but is trend is also not significant and is almost in contrast to the anticipated need for close to zero. The Southern Arctic region detailed weather data to track climatic was significant at 99% confidence level changes in the North. trend with median estimate of 0.05 mm/year considering all the grid points (Figure 6). 4 Trends of extreme precipitation distribution This is also consistent with the sub-sample for Northern and Southern Arctic and can be estimation of trend of spatial standard compared well in terms of trend. However, deviation. The trends of annual maximum we have reservation in commenting on the precipitation over the individual grids is trend of grid points where few or no stations given in Figure 7. Notice, the Southern are present in the vicinity, Arctic only has few grid points with significant trends. The spatial distribution of Northern and Southern Arctic Zones with trend shows an overall increase of extreme the hexagonal grids and Meteorological precipitation. stations Thirty-year return period extreme precipitation over the individual hexagonal grids show some significant spatial trends. This indicates more severe extreme events at this level. Also, notice there are a few grid points with significant decreasing trend indicating less severe extreme events at this RP level. This further gives a notion of a stable unchanged moisture balance over the region where increase precipitation in one region will result into decrease in other regions. For the 100-year return period extreme precipitation over the individual hexagonal grids, there was a change of significance level of few grid points in contrast to the 30-years return period plots. This signifies increased extreme precipitation on the grid points where confidence level changed from low to high. Figure 1: The boundaries of the Northern Furthermore, the grid points where and Southern Arctic zones overlaid with the confidence levels changed from high to low hexagonal grids and meteorological stations. possibly indicates a convergence of the distribution for higher return period levels. Ea) Northern Arctic b) Southern Arctic Overall, Figure 8 and 9 indicate an increase in severity of extreme events. Comparisons of station level annual maximum precipitation against the same from grid cell are reasonably good to comment on statistical properties of precipitation on the grid cell where a number of stations are present. For example, Figure 10 presents the two stations in studied zones where the magnitude of precipitation are in Figure 2: Semi-variogram and empirical good agreement with the nearby grid points variogram model of annual extreme with correlation 0.58 and 0.93 respectively precipitation over (a) Northern Arctic and (b) Southern Arctic regions Trends of extreme precipitation distribution 5 a) Full Gridded b) Sub-sampled a) Full Gridded b) Sub-sampled Full gridded (b)Sub-samples Figure 3: Trend in spatial-mean of annual extreme precipitation over Northern Arctic Figure 6: Trend in spatial-standard deviation for (a) Full Gridded data and (b) Sub- of annual extreme precipitation over samples Southern Arctic for (a) Full Gridded data and (b) Sub-samples a) Full Gridded b) Sub-sampled Trend in annual maximum precipitation Figure 4: Trend in spatial-mean of annual extreme precipitation over Southern Arctic for (a) Full Gridded data and (b) Sub- samples a) Full Gridded b) Sub-sampled (b)Sub-samples Figure 7: The trend of annual maximum precipitation on individual grid points. The grid points with 99% confidence level trends are highlighted with bold boundaries. Figure 5: Trend in spatial-standard deviation of annual extreme precipitation over Northern Arctic for (a) Full Gridded data and (b) Sub-samples 6 Trends of extreme precipitation distribution Trend in 30-years RP precipitation confidence level trends are highlighted with bold boundaries. (a) Northern (b) Southern Arctic Arctic Figure 10: Comparison of station level annual maximum of grid level annual maximum for two stations in Northern and Southern Arctic regions. 4. Conclusion In summary, the conclusions of this study Figure 8: The trend of 30-years return period are as follows; precipitation level estimated using moving window approach. The grid points with 99% i. This study indicates the necessity confidence level trends are highlighted with bold of considering spatial correlation boundaries. of extreme precipitation when Trend in 100-years RP precipitation analyzing trend. This consideration is extremely important for the regions where station density is low and the gridded dataset presents a false sense of data coverage. We can qualitatively draw an inverse relationship between the station density and the range parameter of the semi-variogram. ii. The trend in Southern Arctic regions is consistent with the modeling studies indicating increase in Arctic extreme precipitation (Bintaja R. et. al., 2017). However, the trend in Northern Arctic is in contrast with the trends computed considering all the grid points. Figure 9: The trend of 100-years return period Uncertainties of the trend precipitation level estimated using moving estimation can be attributed to window approach. The grid points with 99% the low station density in that region. Trends of extreme precipitation distribution 7 iii. The trend in spatial standard- Francis, J. A. & Hunter, E. (2007). Drivers of deviation indicates the possible declining sea ice in the Arctic winter. Geophysical Research Letters, 34, changes of local drivers such as; L17503. doi:10.1029/2007GL030995 landcover and/or disturbances Glisan, J M. & Gutowski, W. J. (2014). WRF during the study period over the summer extreme daily precipitation over the CORDEX Arctic. Journal of Southern Arctic region. Geophysical Research Atmospheres, 119(4), 1720-1732. doi: We have produced an unbiased spatial-time 10.1002/2013JD020697 series of annual maximum precipitation for Hopkinson, R.F., McKenney, D.W., Milewska, E.J., Hutchinson, M.F., Papadopol, P., Vincent, Northern and Southern Arctic regions and L.A., 2011. Impact of Aligning detected trends in the Southern Arctic at Climatological Day on Gridding Daily eco-zone level. However, the sparse station Maximum–Minimum Temperature and density in the region prevented us from Precipitation over Canada. J. Appl. Meteorol. Climatol.50, 1654– drawing major conclusion regarding the 1665. https://doi.org/10.1175/2011JAMC2 trends at the local level. Our analysis is 684.1 effective when we are trying to analyze the Hutchinson, M.F., McKenney, D.W., Lawrence, K., eco-zone scale, however, this analysis Pedlar, J.H., Hopkinson, R.F., Milewska, E., Papadopol, P., 2009. Development and cannot capture fine-scale structure of Testing of Canada-Wide Interpolated precipitation. Linkage to both instrumented Spatial Models of Daily Minimum– field plot data and additional climate Maximum Temperature and Precipitation for 1961–2003. J. Appl. Meteorol. modelling studies might provide further Climatol. 48, 725– confidence in the findings presented here. 741. https://doi.org/10.1175/2008JAMC19 79.1 Acknowledgements IPCC (2013). Climate change 2013: The physical science basis. Contribution of working We thank Global Water Futures for group I to the fifth assessment report of the intergovernmental panel on climate providing us the funding to carry out this change. T. F. Stocker, D. Qin, G.-K. study. We thank PCIC for distribution and Plattner, M. Tignor, S.K. Allen, J. storage of the NRCANmet gridded Boschung, A. Nauels, Y. Xia, V. Bex & precipitation dataset, used in this study. P.M. Midgley (Eds.). Cambridge, UK & New York, NY: Cambridge University Press. doi:10.1017/CBO9781107415324 References Mailhot, A., S. Duchesne, D. Caya, and G. Talbot. 2007. Assessment of future change in Bates, B. C., Z. W. Kundzewicz, S. Wu, and J. P. intensity– duration–frequency (IDF) curves Palutikof. Eds. 2008. Climate Change and for Southern Quebec using the Canadian Water. Technical Paper of the Regional Climate Model (CRCM). Journal Intergovernmental Panel on Climate of Hydrology 347: 197- 210. Change. Geneva: IPCC Secretariat, 210 Overland, J., Francis, J. A., Hall, R., Hanna, E., pp. Kim, S. -J., & Vihma, T. (2015). The Bintanja, R., & Andry, O. (2017). Towards a rain‐ melting Arctic and midlatitude weather dominated Arctic. Nature Climate patterns: Are they connected? Journal of Change, 7(4), 263– Climate, 28(20), 7917-7932. doi: 267. https://doi.org/10.1038/NCLIM 10.1175/JCLI-D-14-00822.1 Serreze, M. C. & Barry, R. G. (2014). The Arctic ATE3240 Climate System (2nd ed.). Cambridge, UK: Cambridge University Press