Using hydroclimatic extremes to guide future hydrologic predictions

14 There are growing numbers of studies on climate change impacts on forest hydrology but limited 15 attempts have been made to use current hydroclimatic extremes to constrain future climatic 16 conditions. Here we used historical wet and dry years as a proxy for expected future extremes in a 17 boreal headwater catchment. Hydrologic modelling assessments showed that runoff could be 18 underestimated by at least 35% when dry year parameterization was used for wet year conditions. 19 Uncertainty analysis showed that behavioural parameter sets from wet and dry year separated 20 mainly on precipitation related parameters and to a lesser extent on parameter sets related to 21 landscape processes. While inherent uncertainty in climate models still drives the overall uncertainty 22 in runoff projections, hydrologic model calibration for climate impact studies should be based on 23 years that best approximate future conditions to constrain uncertainty in projecting future 24 conditions. 25


29
There are growing numbers of studies on climate change impacts on forest hydrology but these are 30 usually based on long-time series that depict average system behaviour (Bonan, 2008 hydrologic models has been based on average long term natural rainfall-runoff processes. However, 48 average conditions may not best reflect processes operating under changing conditions. As a result, 49 all models have their inherent uncertainties that can be amplified when projecting future conditions. 50 The predictive uncertainties resulted from hydrologic models is due in part to issues of 51 conceptualization, scaling and connectivity of processes between the landscape mosaic of a 52 increasing frequency of storm events observed in different parts of the world (Dai, 2011;Trenberth, 59 years as a proxy for the future conditions expected as climate changes. Here we used hydrological 66 and meteorological observations in dry and wet years in a long term monitored headwater 67 catchment in northern Sweden. The objectives of this study were to: 1) to utilize long term field 68 observations to gain insights into present extreme hydroclimatic behaviour; 2) to model the extreme 69 behaviour using multi-criteria goodness-of-fit metrics; 3) to quantify the uncertainty in our current 70 predictive practices that is based on long term series; 4) to conduct a robust parameter uncertainty 71 assessment that will help to gain further insights into plausible differences in hydrologic behaviour in 72 dry and wet years; and 5) to use an ensemble of climate change scenarios to test whether our 73 current predictive uncertainty regarding future extremes could be attributed to inherent 74 uncertainties in climate models or be driven by differences in hydrologic model calibration strategies. 75  95 We used 15 different regional climate models (RCMs) from the ENSEMBLES project (Van der Linden  96 and Mitchell, 2009) in the downscaling and analysis presented here (Table 1)

108
PERSiST is a semi-distributed bucket type rainfall-runoff model with a flexibility that allows modelers 109 to specify the routing of water following the perceptual understanding of their landscapes (Futter et  110 al., 2014). This feature makes PERSiST a useful tool to simulate streamflow from landscape mosaic 111 patches at a watershed scale. The model operates on a daily time scale with inputs of precipitation 112 and air temperature. The spatial interface requires an estimate of area, land cover proportion and 113 reach length/width of the hydrologic response units. In the PERSiST application presented here, we 114 used three buckets to represent the hydrology of Svartberget. These include snow, upper soil and 115 lower soil buckets. In the snow routine bucket, the model utilized a simple degree day 116 evapotranspiration and degree day melt factor (Futter et al., 2014). Although the maximum rate of 117 evapotranspiration could be independent of wet and dry years as used in this study, the actual rate 118 of evapotranspiration could be influenced by the amount of water in the soil and by an 119 evapotranspiration adjustment parameter. The latter is an exponent for limiting evapotranspiration 120 that adjusts the rate of ET (depending on water depth in the bucket or how much is 121 evapotranspired). The snow threshold partitions precipitation as either rain or snow. The model also 122 simulates canopy interception for snowfall and rainfall to the uppermost bucket. 123 The quick flow bucket simulates surface or direct runoff in response to the inputs of rainfall or 124 snowfall as a function of soil moisture saturation. Partitioning of the runoff generation process 125 between the quick flow and lower soil buckets (upper and lower) is defined in the square matrix 126 (Table 2). The evapotranspiration adjustment parameter sets the rate at which ET can occur when 127 the soil is no longer able to generate runoff and this was set to 1 in the upper soil box. Maximum 128 capacity is the field capacity of the soil that determines the maximum soil water content held. The 129 time constant specifies the rate of water drainage from a bucket and requires a value of at least 1 in 130 PERSiST. The relative area index determines the fraction of area covered by the bucket and is also set 131 to 1 for our simulations. Infiltration parameters in each bucket determine the rate of water 132 movement through the soil matrix. The model is based on series of first order differential equations 133 that are solved sequentially following the bucket order in the square matrix. Parameter values and ranges used in the Monte Carlo analysis are listed in Table 3. 136 The model was calibrated against streamflow to generate present day runoff conditions. Initial 137 manual calibration was performed on the entire time series to minimize the difference between the 138 simulated and observed runoff. The manual calibration also helps to identify a suite of parameters 139 and their ranges to be used in the Monte Carlo analysis by varying each parameter value such that 140 parameter(s) that separate the hydrology of wet from dry years. Wet years were defined as the 151 hydrologic years with runoff exceeding 430 mm/yr or 40% higher than average annual runoff (1995, 152 2002, 2005 and 2010). Dry years were defined as the hydrologic years with runoff less than 150 153 mm/yr or less than 50% of average annual runoff (1987, 1992, 2000 and 2001  Preliminary analysis showed that the Svartberget hydroclimate was highly variable and thus helped 160 to partition the long term series into dry and wet years (SI 1). As a result, both dry and wet year 161 conditions were different in terms of climate and cumulative runoff patterns. The cumulative 162 distribution of the dry/wet year series (Fig 2a) showed that dry year precipitation (462 ± 102 mm) 163 was only 64% of precipitation observed in wet year (716 ± 56 mm). Similar patterns were observed in 164 runoff dynamics (Fig. 2b) where total runoff in dry years (129 ± 35 mm) was 29% of total runoff 165 observed in wet years (449 ± 19 mm). Runoff response was 63% of total precipitation that fell in wet 166 years and 28% of precipitation in the dry year regime. These were summarized in Table 4 Result also showed that temperature in wet and dry years were similar on average, while winter 175 months were generally slightly warmer during wet years and summers slightly warmer in dry year 176 (Fig 3c). 177

178
Results showed that there was less agreement between the observed series and uncorrected 179 individual RCMs (SI 2a, b). However, bias correction helped to reduce the uncertainty by providing a 180 better match for the ensemble median of the air temperature and precipitation with their 181 corresponding observed series (SI 2c, d). Results showed that ensemble median performed better in 182 fitting the observed air temperature than precipitation. Results also showed a possible increase in air 183 temperature by 2.8-5 o C (median of 3.7 o C) and possible increase in precipitation by 2-27% (median of 184 17%). Although precipitation and temperature were projected to increase throughout the year, the 185 temperature changes would be more pronounced during winter months irrespective of whether it 186 was a dry or wet year (Fig. 3c). However, projected changes in precipitation followed similar patterns 187 to historical wet year with more precipitation expected between late winter months through spring 188 ( Fig. 3a). Result also showed that the winter period with temperature below 0 o C could be shortened 189 as climate warms in the future (SI 2). 190

191
Model behavioral performance followed similar patterns when metrics such as R 2 , NS and log NS 192 were used (SI 3a-c) and could be used interchangeably to measure model performances.

202
Using the best performing parameter sets based on the NS statistic as an example, the model 203 performed well in simulating the interannual runoff patterns but underestimated the peaks (SI 4). 204 When resolved to their respective dry and wet year components, the model performed better in 205 simulating runoff conditions in wet year despite its larger data spread and higher spring peaks than 206 the dry year regime (SI 5). When parameterization for dry year was used for runoff prediction in wet 207 years, runoff was underestimated by 35% due to significant uncertainty that stemmed from growing 208 season months (Fig. 4). Modelling analysis presented here also showed that no single metric can be 209 an effective measure of model performance under extreme conditions depicted in dry and wet years 210 (Fig 5a-c). However, utilizing a behavioural mean of these different performance metrics (Fig. 5d-f) 211 appeared to be a more effective way of calibrating to extreme hydroclimatic conditions. While the 212 behavioural mean performed better in simulating runoff dynamics in winter through spring in the 213 long term record and significantly reduced the uncertainty in dry and wet years, larger uncertainty 214 existed in summer through autumn months in dry and wet year compared to the long term record. 215

216
While we observed a wide prediction range from behavioural parameter sets (Fig. 5), we have limited 217 information on the underlining processes. Therefore, we subjected the behavioural parameter sets 218 to further analysis to identify sensitive parameters and plausible patterns of hydrologic processes 219 that differentiate dry and wet years (Fig. 6). The cumulative distribution function (CDF) of 220 behavioural parameter sets showed both rain and flow multipliers were sensitive parameters in dry 221 year and tended toward lower ranges. The rain multiplier was less sensitive in wet years unlike the 222 flow multiplier. Long term simulations showed no sensitivity to the rain multiplier but were sensitive 223 to the flow multiplier. We observed similar patterns of behaviour to flow multiplier in all the three 224 hydrologic regimes (Fig. 6b) conditions (Fig. 7). Result showed that both dry and wet years separated well in canonical space. 230 However, the separation was driven mainly on quantitative parameters related to precipitation, 231 interception and evapotranspiration on canonical axis 1 (Rmult, Int and DDE). The parameters 232 separated to a lesser extent on processes related to snow parameters on canonical axis 2 (Smult, SM 233 and DDM).

324
Despite the fundamental issues of parameter equifinality (Beven, 2006)  suggested that both extreme conditions followed similar runoff generation processes. These 340 suggested that the main physical mechanism to explain parameter sensitivity and hydroclimatic 341 behaviour to extreme conditions were related to differences in their precipitation patterns rather 342 than landscape-driven hydrologic processes. 343

344
Even though equifinality limits the use of CDFs alone in identifying all sensitive parameters, DFA of 345 behavioural parameters gave further insights on plausible differences in wet/dry hydrologic 346 behaviour when projected on canonical space. This suggested that hydrological model 347 parameterizations calibrated to high flow associated with wet year differ from parameterizations for 348 long term or dry conditions. Therefore, parameter separation primarily on quantitative parameters 349 (Rmult, Int and DDE) related to rainfall and evapotranspiration on canonical axis 1 suggested that 350 climate is a first order control of hydroclimatic extremes in the boreal forest. This is consistent with 351