Dynamical Downscaling Data for Studying Climatic Impacts on Hydrology , Permafrost , and Ecosystems in Arctic Alaska

Abstract. Climatic changes are most pronounced in northern high latitude regions. Yet, there is a paucity of observational data, both spatially and temporally, such that regional-scale dynamics are not fully captured, limiting our ability to make reliable projections. In this study, a group of dynamical downscaling products were created for the period 1950 to 2100 to better understand climate change and its impacts on hydrology, permafrost, and ecosystems at a resolution suitable for northern Alaska. An ERA-interim reanalysis dataset and the Community Earth System Model (CESM) served as the forcing mechanisms in this dynamical downscaling framework, and the Weather Research & Forecast (WRF) model, embedded with an optimization for the Arctic (Polar WRF), served as the Regional Climate Model (RCM). This downscaled output consists of multiple climatic variables (precipitation, temperature, wind speed, dew point temperature, and surface air pressure) for a 10 km grid spacing at three-hour intervals. The modeling products were evaluated and calibrated using a bias-correction approach. The ERA-interim forced WRF (ERA-WRF) produced reasonable climatic variables as a result, yielding a more closely correlated temperature field than precipitation field when long-term monthly climatology was compared with its forcing and observational data. A linear scaling method then further corrected the bias, based on ERA-interim monthly climatology, and bias-corrected ERA-WRF fields were applied as a reference for calibration of both the historical and the projected CESM forced WRF (CESM-WRF) products. Biases, such as, a cold temperature bias during summer and a warm temperature bias during winter as well as a wet bias for annual precipitation that CESM holds over northern Alaska persisted in CESM-WRF runs. The linear scaling of CESM-WRF eventually produced high-resolution downscaling products for the Alaskan North Slope for hydrological and ecological research, together with the calibrated ERA-WRF run, and its capability extends far beyond that. Other climatic research has been proposed, including exploration of historical and projected climatic extreme events and their possible connections to low-frequency sea-atmospheric oscillations, as well as near-surface permafrost degradation and ice regime shifts of lakes. These dynamically downscaled, bias corrected climatic datasets provide improved spatial and temporal resolution data necessary for ongoing modeling efforts in northern Alaska focused on reconstructing and projecting hydrologic changes, ecosystem processes and responses, and permafrost thermal regimes. The dynamical downscaling methods presented in this study can also be used to create more suitable model input datasets for other sub-regions of the Arctic. Supplementary data are available at https://doi.org/10.1594/PANGAEA.863625 .


Introduction
Climate change is most pronounced in high latitude regions (Johannessen et al., 2004;Serreze and 5 Francis, 2006;Hinzman et al., 2005).Although the exact mechanism is still under vivid discussion, Arctic amplification has been strengthening since the late 1970s, resulting in a stronger surface temperature increase than at lower latitudes, and thus a more interactive land-surface background of the Arctic (Alexeev et al., 2005;Serreze and Francis, 2006).
The physical and ecological components of the Arctic are strongly affected by the regional and global Manuscript under review for journal Earth Syst.Sci.Data Published: 25 August 2016 c Author(s) 2016.CC-BY 3.0 License.
wind speed and snow redistribution (Black, 1954;Liston and Sturm, 2002;Rasmussen et al., 2012).These difficulties coincide with observational winter precipitation climatology, yielding close to zero amounts for some of the stations.
On the other hand, the temperature measuring instruments has been proven trustworthy (Vose et al., 2007).Daily maximum temperature (TMAX) and minimum temperature (TMIN) are retrieved from the 5 three-hourly ERA-WRF output and the six-hourly ERA-interim output to fit NCDC GHCN-D temperature variables.Since maximum and minimum values for temperature from ERA-interim and WRF are filtered out from daily temperature with stationary time intervals, while NCDC GHCN-D records truly daily maximum and minimum temperatures, this manner of extraction may lead to some biases during the comparison.
TMAX in ERA-interim and WRF are extracted from the temperature at 0000 UTC (3 pm local time), 10 while TMAX in NCDC is measured as the true daily temperature maximum.WRF slightly underestimates TMAX climatologically, compared to observation (Fig. 4).This cold bias is obvious mostly during the warmest months (June to August) and the coldest months (November to February).The only exception is the Deadhorse site, at which WRF produces small warm biases (less than 3 K) from February to May.For most stations, ERA-interim also presents cold biases compared to observations, especially in the summer.In winter, 15 however, cold biases between ERA-interim and observation are generally not as much as those between WRF and observation.
Similarly, TMIN in ERA-interim and WRF are extracted from the temperature at 1200 UTC (3:00 am local time).Unlike TMAX, TMIN monthly climatology generally shows a warm bias between ERA-interim and observation, and a cold bias between WRF and observation (Fig. 5).These biases are illustrated year 20 round, except for March to May, when the cold bias of WRF becomes negligible for all five stations.TMAX and TMIN jointly reflect the diurnal temperature cycle.ERA-interim is found to have less diurnal temperature variation over the North Slope.WRF, on the other hand, produces cold biases for both TMAX and TMIN during the warmest months.However, the TMAX bias of WRF in the winter is so small that it helps even the cold bias of TMIN during the same period, representing a bigger diurnal temperature

5
Other than monthly climatology comparisons of precipitation and temperature between observations, reanalysis data, and RCM simulation, statistics further reveal an in-depth picture of RCM performance.
Taylor diagrams are presented for these five stations, showing the correlation coefficients of monthly precipitation (green), TMAX (red), and TMIN (blue) climatology of ERA-interim ( ╳) and WRF (+) compared to observational data (Fig. 6).

10
Both ERA-interim and WRF demonstrate monthly/seasonal precipitation and temperature variabilities.
Correlation coefficients are higher than 0.7 in all cases.Among these three variables, TMAX and TMIN are more closely correlated to observation than is precipitation.Temperature coefficients are all higher than 0.95, while precipitation coefficients are in the range of 0.7 to 0.9.Regarding comparison between data sets, however, WRF-modeled precipitation at these five stations show higher coefficients than ERA-modeled 15 precipitation at Barrow, Wainwright, and Nuiqsut, and similar to Deadhorse and Umiat.The TMIN coefficients are also slightly higher than the TMAX coefficient, especially in comparison between WRF and observations.
Another important statistical parameter these Taylor diagrams illustrate is normalized standard deviation (STD), representing the monthly/seasonal variability in its reference (observation).Both ERA-interim and 20 ERA-WRF precipitation amounts have a higher standard deviation.The only exception is the STD of ERAinterim precipitation in Barrow, which is similar to observations.Regarding ERA-interim and WRF, WRF produces about 1.5 times the STD of both the ERA-interim and observation.WRF precipitation STDs are higher than those of ERA-interim in Deadhorse and Umiat, while the two differ little in Wainwright and Nuiqsut.

25
Earth Syst.Sci. Data Discuss., doi:10.5194/essd-2016-31, 2016 Open the North Slope.Further, the data quality for Nuiqsut, located just outside the northeast portion of the Fish Creek Watershed, is critical to the reasonability and accuracy of the hydrological model forced by CESM-WRF runs.The purpose of this evaluation is not only to validate CESM-WRF simulation, but also to produce bias correction parameters that will be used for bias correcting the projected CESM-WRF simulation.Fig. 7 shows monthly biases for precipitation and temperature at Barrow from 1980 to 2005.The left 5 panel shows biases before bias correction, and the right panel presents biases after applying linear scaling bias correction.For precipitation, original CESM-WRF precipitation has a wetness bias, generally, which is higher in summer (JJA) than in winter.After bias correction, the plot is distinctly "lighter" in color, indicating lower biases throughout the period.Statistically, linear scaling drags mean bias down from 0.4681 to 0.0018, and RMSE down from 0.8135 to 0.3865.Warm biases can be found in monthly mean temperature, occurring 10 mostly during winter.After bias correction, months with high warm biases (>8 K) decrease from the original.
Statistically, mean bias decreases from 1.5729 to 0.4357, and RMSE from 5.1694 to 4.6587.Fig. 8 includes the same type of plots as Fig. 7, but for the Nuiqsut station.Precipitation and temperature biases are similar to Barrow.Linear scaling effectively corrects the wet biases for precipitation and warm biases for temperature.
What is different is that wet biases for precipitation and warm biases for temperature are higher in spring 15 (MMA) than are those in Barrow.Linear scaling also fixes those successfully.Fig. 9 and Fig. 10 contour seasonal and annual differences in precipitation and temperature between CESM-WRF and ERA-WRF, both bias corrected.Differences between these two data sets are small: precipitation differs less than 0.1 mm/day and temperature less than 1 K, showing the reasonability of dynamical downscaling and the effectiveness of bias correction.CESM-WRF climatology shows a slightly 20 lower precipitation rate (< 0.02 mm/day) and slightly higher temperature (< 0.4 K) across the study region in northern Alaska.Some seasonal variation features are also found for precipitation and temperature differences.
Although the bias is small enough for a reasonable modeled data set, seasonal variability is evident in both the CESM-WRF and the ERA-WRF.CESM-WRF precipitation in spring (Fig. 9) and temperature in  Linear scaling has proven effective for correcting biases but still retaining the short-term variability of the original CESM-WRF.Bias correction parameters for historical simulation are thus applied for bias correction on the projected CESM-WRF run.After this, CESM forced dynamical downscaling products for both the historical and the projected periods are completed.These data sets, as well as the reference ERA-WRF simulation, can be applied for various research topics in climatology, hydrology, and ecology over the 10 Alaskan North Slope, thanks to their fine grid spacing and reasonable capture of a set of climatic variables.

Discussion
This paper introduces the birth of two dynamical downscaling products forced by ERA-interim reanalysis data and CESM model output.After computational work was completed, we evaluated these modeled variables and corrected their bias based on ERA-interim climatology and observational datasets.

15
ERA-WRF models produce reasonable precipitation and temperature fields compared to ERA-interim.The mean precipitation amount and the seasonal variability of ERA-WRF are close to those of ERA-interim, though both of them have nearly double the annual precipitation amount relative to observational data.
Temperature is, unsurprisingly, simulated better than precipitation.ERA-WRF TMAX and TMIN are especially well-matched to observations throughout the year, although slight cold biases are found, mostly 20 during winter, over the Alaskan North Slope, compared to ERA-interim.On the North Slope, the short and weak solar radiation in winter drag the diurnal solar radiation fluctuation down to a low level, due to the high latitude.This disappearance of variability makes solar radiation less important to driving the daily temperature cycle over the North Slope.On the other hand, cloud cover and wind advection jump out as the important factors for surface temperature, in both summer and winter (Dai et al., 1999;Przybylak, 2000).product (Fig. 7 & 8).Previous research on CESM1 temperature modeling has found that it underestimates the seasonal cycle over the Arctic, which produces warmer winters and colder summers compared to reanalysis data does (Walston et al., 2014).CISL RDA ds316.1 applies Reynold averaging of CESM variables, based on ERA-interim that rescales 35-year climatology of CESM but maintains the perturbation term completely (Bruyè re et al., 2014;Bruyè re et al., 2015).We can assume this underestimation remains in 10 CESM-WRF, brought by its forcing.A linear scaling method for rescaling the monthly climatology/seasonal cycle is applied instead, for better bias corrections than the Reynold averaging method of both ERA-WRF and CESM-WRF.Also, it is notable that not all variables are able to be bias-corrected in this way, and the limitation results from the coarse grid of ERA-interim.For some variables that are not spatially continuous, like the snow depth which is only over the land, the interpolation of variable field from ERA's grid to WRF's 15 grid limits data accuracy over the coastal area, and the fact that ERA's grid does not follow the coast well makes more problematic, since ERA mistakenly recognizes some part of coastal land as part of the ocean, like Barrow.Thus, these kind of variables are not recommended to be bias-corrected before new approach of calibration is developed.
Linear scaling of CESM-WRF diminished monthly average precipitation and temperature biases, reflect 20 the decreases of mean bias and RMSE.Taking Barrow and Nuiqsut, for example, the original CESM-WRF generally exhibits a wet bias during summer and a warm bias during winter, compared to bias-corrected ERA-WRF (Fig. 7 and Fig. 8).These are also clearly diminished by the bias correction of CESM-WRF.
Precipitation correction has a relatively better effect than temperature correction, with both exhibiting virtual biases and statistics.The majority of the climatic variables from both the ERA-WRF and CESM-WRF have Spatial variability of temperature climatology over the Alaskan North Slope has been found to be very small due to its relatively flat topography, though precipitation climatology increases from the coast to the interior because of the orographic impediment caused by the Brooks Range (Zhang et al., 1996;Serreze and Hurst, 2000;Wendler et al., 2009).Comparison between CESM-WRF and ERA-WRF seasonal climatology 5 coincides with this feature, showing some north-to-south gradient for temperature comparison.Precipitation comparison also yields some signals that may be relative to topography, though their existence is still uncertain, as the topographical background is offered by WRF rather than the forcing, and the land-surface backgrounds of these two runs are identical.
The linear scaling method maintains the majority of spatial distribution and temporal climate variability 10 from the original data set, thus retaining their advantage from fine grid spacing and favorability of regional climate impact research.These two dynamical downscaling products, using the polar WRF model and forced respectively by reanalysis data and the earth system model offer major climatic variables, with high spatial resolution over our study domain in northern Alaska (Fig. 2).

Applications 15
The dynamically down-scaled datasets presented in this study provide a framework for enhancing previous research efforts in Northern Alaska.For example, Scenarios Network for Alaska and Arctic components of atmosphere, land, and ocean, despite of their high resolutions.Some studies that involve surface-air interaction, such as projecting the runoff of a watershed, have to rely on multiple data sets that are independently built from each other.Some inconsistency between variables of atmosphere, land, or ocean therefore may occur.These inconsistency may lead to biases when other numerical simulations are driven by this dataset. 5 The downscaled products developed by this study combine the advantages of reanalysis data set/ESM and RCM.It not only downscales the ESM's coarse grid spacing to enable regional climate studies, but also revises its lack of temperature/precipitation variabilities and extremes.Not to mention that it has gridded coverage that offsets the difficulty presented by sparse availability of observations over the Alaskan North Slope.What makes this product outweigh others is that it offers climatic variables from multiple major 10 components of the earth system, including the atmosphere, the land, and the ocean.All the provided variables are reasonably correlated and dependent with each other within the Polar WRF modeling framework.Thus, it is especially suitable for regional climate impacts studies that involves land-air interactions.
Our downscaled and bias-corrected product is being used to drive a grid-based Water Flow and Balance Simulation Model (WaSiM) at watershed scales ranging from 30 to 5,000 km 2 in northern Alaska.Recent Simulation of lake ice growth using temperature and snow depth data from Polar WRF is being compared to multi-temporal synthetic aperture radar (SAR) analysis of lake ice regimes, which is helping to understand changes in sub-lake permafrost, overwintering fish habitat, and availability of winter water supply for industry (Arp et al. 2012).
The advantage of using our data set for this analysis is the ability to provide continuous data for specific regions corresponding to SAR image acquisitions, whereas previous studies 5 using station data often prove very limiting in terms of data gaps and particularly representing snow at a regional distribution.Evidence suggests that shallow lakes along the outer Arctic Coastal Plain are most sensitive to reduced ice growth (Arp et al., 2012;Surdu et al., 2014), yet this proximity is often poorly captured with coarse resolution climate datasets or station data.
Finer grid and optimized parameterization schemes of Polar WRF enable the recapturing of climatic 10 extremes, such as that occurred in 2007 (Jones et al., 2009;Alexeev et al. 2014).Other future studies may include historical and projected extreme climatic events across the North Slope, the projected frequency and intensity of extreme climatic events under the changing climate can impact more than that from the shift of the mean, and the better capability of these downscaling products of capturing the extremes suitably facilitates the extreme study, as well as the teleconnection of low-frequency sea-atmosphere oscillations.The climatic 15 impact to ecosystem is difficult to estimate over the arctic since the lack of detailed the routinely observation (Post et al., 2009).These high-resolution products are able to serve as a high-quality alternative climatic background.The applications may include exploring the degrading permafrost and its deepening active layer to the releasing carbon from underground and vegetation production over the arctic tundra, and then the impact to the habitat change of insects and large animals living based on this arctic environment.Earth Syst.Sci. Data Discuss., doi:10.5194/essd-2016-31, 2016 Open 25 variation during the coldest months.Temperature evaluation experiments by the Polar WRF group also found Earth Syst.Sci.Data Discuss., doi:10.5194/essd-2016-31forjournal Earth Syst.Sci.Data Published: 25 August 2016 c Author(s) 2016.CC-BY 3.0 License.a cold bias in winter and warm bias in summer on the North Slope since Polar WRF version 3.1.1(Hines et al., 2009; Hines et al.; 2011).As found here, these biases remain in version 3.5.1.The variabilities of both TMAX and TMIN are very restricted, especially in summer when the longer period of sunlight decreases the diurnal and daily temperature variations.
for journal Earth Syst.Sci.Data Published: 25 August 2016 c Author(s) 2016.CC-BY 3.0 License.Manuscript under review for journal Earth Syst.Sci.Data Published: 25 August 2016 c Author(s) 2016.CC-BY 3.0 License.
25 summer (Fig. 10) exhibit opposite features from the annual difference in ERA-WRF.Also, among CESM-Earth Syst.Sci.Data Discuss., doi:10.5194/essd-2016-31forjournal Earth Syst.Sci.Data Published: 25 August 2016 c Author(s) 2016.CC-BY 3.0 License.WRF, precipitation over mountainous areas and inland lakes remain elevated relative to ERA-WRF.Since the configuration of these two WRF runs are identical, this small scale fluctuation in precipitation can follow only from the input field-CESM data.The cause of this heterogeneity in spatial precipitation distribution is beyond the scope of this paper, although it's interesting to witness how large-scale input causes this heterogeneous feature for precipitation in WRF.
Planning (SNAP) offers downscaled high-resolution daily temperature/precipitation based on CMIP3/5 GCMs (https://www.snap.uaf.edu/).The Geophysical Institute Permafrost Laboratory model (GIPL) models soil dynamics and offers important variables on permafrost research, such as soil temperature at multiple 20 layers, active layer thickness, freeze-up time, etc. (Marchenko et al. 2008).The GIPL, together with other two ecosystem models, comprise the Integrated Ecosystem Model (IEM), offering various ecosystem projections for all of Alaska and Northwest Canada (Rupp et al., 2015).The previous efforts have been limited by inadequately downscaled and bias-corrected climatic datasets.One shortcoming of the above mentioned model-based data sets is that they are lack of a complete set of climatic variables that coupled with 25 Earth Syst.Sci.Data Discuss., doi:10.5194/essd-2016-31forjournal Earth Syst.Sci.Data Published: 25 August 2016 c Author(s) 2016.CC-BY 3.0 License.
15 droughts, such as occurred in the summer of 2007, along with uncertainty regarding hydrologic intensification(Rawlins et al., 2010)  make the use of such hydrologic models valuable for understanding complex climatepermafrost-hydrology interactions(Liljedahl et al. 2016) and for simulating runoff for specific locations which lack gauging records.In the latter case, WaSiM is being applied to a small catchment in the National Petroleum Reserve-Alaska where petroleum development is planned and baseline streamflow records are 20 insufficiently short to evaluate any impact for changes in land-use (i.e., permanent roads, drilling pad, and lake-water extraction for operations)(Heim et al., 2014).Changes in hydrologic connectivity among rivers, streams, and lakes and how this affects fish migration and habitat use is of great interest regarding changes in land-use and even more so regionally with changing or variable climate.Its high-resolution and reasonable precipitation, as well as other key variables, empower the WASiM recapturing the historical and projected 25 variability spatially and temporally over a complex watershed.Earth Syst.Sci.Data Discuss., doi:10.5194/essd-2016-31forjournal Earth Syst.Sci.Data Published: 25 August 2016 c Author(s) 2016.CC-BY 3.0 License.

Figure 1 :
Figure 1: The flowchart of study approach.
review for journal Earth Syst.Sci.Data Published: 25 August 2016 c Author(s) 2016.CC-BY 3.0 License.

Figure 2 : 5 Earth
Figure 2: The red box in figure a is the simulation domain, and figure (b) is the detailed topography of the

Figure 3 :
Figure 3: Monthly mean precipitation rate (mm/day) with error bars from NCDC (green dashed line), WRF (red

Figure 4 :
Figure 4: Monthly mean maximum temperature (degree Celsius) with error bars from NCDC (green dashed

Figure 5 :
Figure 5: Monthly mean minimum temperature (degree Celsius) with error bars from NCDC (green dashed

Figure 6 :
Figure 6: Taylor diagram displaying the correlation coefficients and normalized standard deviations of

Figure 7 :
Figure 7: The comparison between differences in raw CESM-WRF and bias-corrected ERA-WRF (left panel), as

Figure 8 :
Figure 8: The comparison between differences in raw CESM-WRF and bias-corrected ERA-WRF (left panel), as

Figure 9 :
Figure 9: Seasonal and annual differences in precipitation rate (mm/day) between CESM-WRF and ERA-WRF,

Figure 10 :
Figure 10: Seasonal and annual differences in daily mean temperature (degree Celsius) between CESM-WRF