Predicting future US water yield and ecosystem productivity by linking an ecohydrological model to WRF dynamically downscaled climate projections

Introduction Conclusions References

evaluated the future annual and monthly changes of hydrology and ecosystem productivity for the 18 Water Resource Regions (WRRs) or 2-digit HUCs. Across the CONUS, the future multi-year means show increases in annual precipitation (P ) of 45 mm yr −1 (6 %), 1.8 • C increase in temperature (T ), 37 mm yr −1 (7 %) increase in ET, 9 mm yr −1 (3 %) increase in Q, and 106 g C m −2 yr −1 (9 %) increase in GPP. Response to climate 15 change was highly variable across the 82, 773 watersheds, but in general, the majority would see consistent increases in all variables evaluated. Over half of the 82 773 watersheds, mostly found in the northeast and the southern part of the southwest would have an increase in annual Q (>100 mm yr −1 or 20 %). This study provides an integrated method and example for comprehensive assessment of the potential impacts of 20 climate change on watershed water balances and ecosystem productivity at high spatial and temporal resolutions. Results will be useful for policy-makers and land managers in formulating appropriate watershed-specific strategies for sustaining water and carbon sources in the face of climate change. 12705

Introduction
The Earth's climate system has been significantly altered over the past 100 years due to human activities, such as emissions of greenhouse gas, aerosol and land use/cover change (LUCC). The Intergovernmental Panel on Climate Change (IPCC, 2014) concluded that global mean surface temperature (T ) has increased by 0.78 • C between 5 1850-1900 and 2003-2012. Additionally, extreme precipitation and droughts have increased (Tebaldi et al., 2006;Trenberth, 2011;Bony et al., 2013;Hegerl et al., 2014). The global climate is projected to continue to change over this century and beyond (IPCC, 2014). Comparing to the period of 1986-2005, the period 2018-2100 is projected to see 0.3 to 4.8 • C increase in global surface temperature (IPCC, 2014). Projec- 10 tions of future changes in precipitation suggest a small increase in the global average, but a substantial shift in where and how intensely precipitation falls (Noake et al., 2012;Scheff and Frierson, 2012;J. Liu et al., 2013). In response, the hydrological cycle and ecosystems have been and will be markedly changed through various physical, chemical and biological processes (Labat et al.,15 2004; Milly et al., 2005;Dai et al., 2009;Harding et al., 2011;Sedláček and Knutti, 2014). Mounting evidence has suggested that climate change played an important role in controlling the water cycle by affecting evaporation, transpiration and runoff (McCabe et al., 2002;Hamlet et al., 2007;Syed et al., 2010;Wang and Hejazi, 2011;Chien et al., 2013;Hegerl et al., 2014;Huntington and Billmire, 2014;McCabe and Wolock, 2014;Piontek et al., 2014;Schewe et al., 2014;Zhang et al., 2014;Aparício et al., 2015).
In the US, average temperature has dramatically increased since the record keeping began in 1895. The most recent decade was believed to be the nation's warmest on record (see the website: http://www.nasa.gov/home/hqnews/2010/jan/HQ_10-017_ Warmest_temps.html). The mean precipitation over the US has increased overall since 1900; some areas have increased with a higher rate than the national average, and some areas have decreased (Groisman et al., 2004;Meehl et al., 2005;Anderson et al., 2015). Over the past century, climate change in the US has caused severe water stress, floods and droughts as well as forest morality (Xu et al., 2013), consequently 15 leading to serious economic losses in some regions. Quantifying the impacts of future climate change on water and ecosystem productivity has become a major research area in hydrology and ecosystem sciences (Lettenmaier et al., 1994;Lins and Slack, 1999;Groisman et al., 2001;McCabe and Wolock, 2011;Sagarika et al., 2014).
Climate change impacts on the water cycle and ecosystem productivity vary from 20 region to region because climate change patterns are not uniform across space or time (Sankarasubramanian et al., 2001;Sankarasubramanian, 2003;Wang and Hejazi, 2011;Xu et al., 2013;Brikowski, 2014), and the spatial variability across regions increases as the spatial resolution increases from large river basins to small catchments. To support future watershed management and to develop sound climate change Introduction Two major research gaps exist in past climate change studies that aim to quantify the interactions among climate, hydrology and ecosystem productivity. First, few studies provided projections of future climate change impacts on water and carbon balances at watershed scale using a consistent approach. Various land surface models (LSMs) simulate and predict water fluxes for a large region, but the scale is often too coarse 5 with a spatial resolution ranging from 0.25 to 2.5 • , and the spatial representation of the projections (e.g., grids) are not readily consumable by resource managers who operate at the watershed scale. In addition, key hydrological processes (e.g., lateral surface and sub-surface flows among grid boxes) embedded in LSMs have not been considered, potentially resulting in uncertainties in water balance projections (Overgaard et al., 2006;Li et al., 2011). Second, future climate projections have high uncertainty. To save computational resources and enhance the computational efficiency, statistical (or empirical) downscaling methods have been used to generate climate forcing to land surface models or watershed ecosystem models. This type of methods does not consider the effects of atmospheric dynamical processes (Xue et al., 2014) and could 15 introduce uncertainties into the crucial land surface variables.
The objective of this study is to address these gaps in our understanding of watershed scale responses to climate change by exploring the potential benefits of linking dynamically downscaled climate projections with a common ecosystem model for climate change assessment at a fine spatial scale (82 773 12-digit HUC watersheds 20 of the CONUS). The specific objectives of this study are to (1) examine future climate changes in precipitation, and temperature during 1979-2007 and 2031-2060 for one plausible emission scenario over the CONUS using WRF (Weather Research and Forecasting) dynamically downscaled climate projections, (2) predict future changes of the CONUS water yield (Q), ET, and GPP by linking the WRF dynamically downscaled 25 climate change scenarios and the WaSSI model. The goal is to generate information that can be useful for resource managers and policy makers to plan for potential shifts in water resources and ecosystem productivity at scales that are relevant from the watershed to national level.  (Fig. 1a). The 12-digit HUC watershed is defined in a national standard, six-level 5 hierarchical system of hydrologic units in the US, ranging from 18 WRRs in the conterminous US at the first level to approximately 83 000 sub-watersheds, or 12-digit HUC watersheds, at the sixth level (Seaber et al., 1987). Hydrologic Units in each level are nested within the next higher level, and are assigned a unique code consisting of two to 12 digits for WRR and 12-digit HUC watersheds, respectively. The WRRs vary in 10 size with the maximum of 1.3 × 10 6 km 2 (WRR10) and the minimum of 1.1 × 10 5 km 2 (WRR6), while the 12-digit HUCs are 95 km 2 on average. Land surface characteristics ( Fig. 1b) vary dramatically among these WRRs (Fry et al., 2011). The WRRs in the east had the larger percentages (around 10 %) of urban use with WRR2 (13 %) and WRR4 (11 %) ranked as the top two. The wetlands are mainly located in the WRRs 15 in the eastern US, while the western regions had the higher percentages of shrubland (> 30 %). The WRRs in the east had higher forest (including mix, evergreen and deciduous forests) percentages (> 33 %) than the west (< 33 %). The deciduous and the evergreen forests were mainly found in the east and the west, respectively. Most of the crop lands were located in the east and central CONUS (Fig. 1b).

Dynamically downscaled climate by WRF
The IPCC Special Report on Emissions Scenarios (SRES) scenarios were designed to project future global environment with a special reference to the production of greenhouse gases and aerosol precursor emissions (Nakicenvoic et al., 2000). The SRES scenarios include four narrative storylines (i.e., A1, A2, B1 and B2) describing the rela- 25 tionships between the forces affecting greenhouse gas and aerosol emissions and their Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | evolution in the 21st century. Each storyline represents a specific and typical demographic, economic, technological, social and environment progresses with divergence in increasingly irreversible ways. The A2 SRES scenario was selected because it represents a potentially worst-case scenario, and because post-2000 global carbon emissions estimates indicate that current emissions are tracking the higher of the SRES 5 emissions projections (Raupach et al., 2007) making the A2 scenario potentially more likely given current trends. The Global Circulation Models (GCMs) have significant issues in representing local climates due to their coarse spatial resolution (Leung and Qian, 2003). Two types of downscaling method are available: dynamical and statistical (or empirical) downscaling 10 (Huang et al., 2011) to downscale the GCMs climate data to a higher spatial resolution for regional and local applications. Dynamical downscaling was used here for generating the current and the future climate due to better representation of finer scale physical processes in climate variables (Gao et al., 2011).
Among the many GCMs available in the World Climate Research Programme's 15 (WCRP) Coupled Model Intercomparison Project phase 3 (CMIP3) Climate Projections (Meehl et al., 2007), the HadCM3 model has been regarded as having the most realistic description of the ENSO mechanisms in the current climate, and reasonably capture ENSO-associated precipitation anomalies over the North America (van Oldenborgh et al., 2005;Joseph and Nigam, 2006;Dominguez et al., 2009 (Kain and Fritsch, 1993), Goddard Shortwave radiation (Chou and Suarez, 1994), Rapid Radiative Transfer Model (RRTM), Longwave (Mlawer et al., 1997), Eta surface layer (Janjic, 2002), Mellor-Yamada-Janjic (MYJ) planetary boundary layer (Janjic, 2002), and the Noah land surface model Version 1.0 (Chen and Dudhia, 2001). To ensure the maintenance of synoptic-scale circulation features (e.g., ridges and troughs) 5 in the RCM, we performed spectral nudging on the zonal and meridional winds, the temperatures and the geo-potential height fields for all pressure levels below 0.36 of the surface pressure (for a surface pressure of 1000 mb it would be all pressures below 360 mb) effectively nudging only at very high elevations above the surface.

Climate data bias corrections
Although the WRF projections were found to be adequate for a hydrological study  in the Colorado River Basin (Wi et al., 2012), we found some large regional biases that required bias correction prior to coupling with the ecosystem model. We used the monthly Bias Correction Spatial Disaggregation (BCSD; Wood et al. (2002Wood et al. ( , 2004 approach that has been applied for hydrologic forecasting in the 15 eastern US (Wood et al., 2002). Steps in the BCSD procedure includes (1) scale up historical Precipitation Elevation Regression on Independent Slopes Model (PRISM) monthly precipitation (P ) and temperature with 4 km×4 km resolution (Daly et al., 1994;PRISM Climate Group, 2013)  HUC watershed scale because the WaSSI model operated on the 12-digit HUC watershed level.

The WaSSI model
The Water Supply Stress Index (WaSSI) model is an integrated, water-centric processbased ecohydrological model designed for modeling water and carbon balance and wa- 5 ter supply stress at a high resolution across a broad scale (Sun et al., 2011a(Sun et al., , 2015aCaldwell et al., 2012). It operates on a monthly time step at the 8-digit HUC or 12-digit HUC watershed scale for the CONUS. The WaSSI model simulates the full monthly water (ET, Q and soil moisture storage) and carbon balances (GPP, ecosystem respiration and net ecosystem productivity) for each land cover class at the given watershed scale. This model has been tested in a variety of geographical regions, and has been widely used for quantitatively assessing combined or individual effects of climate change, land use/cover change (LUCC), and population dynamics on water supply stress and ecosystem productivity (i.e., carbon dynamics) over the CONUS (Sun et al., 2008(Sun et al., , 2011a(Sun et al., , 2015a The key algorithms of the WaSSI model were derived from accumulated knowledge of ecosystem carbon and water cycles gained through the global eddy covariance flux monitoring networks and watershed-based ecohydrologhical studies across the US 20 The ecosystem ET sub-module, the core of the WaSSI model, is described as a function of potential ET (PET), LAI, precipitation, and soil water availability by land cover type (Sun et al., 2011a). The snow model embedded with WaSSI (McCabe and Wolock, 1999;McCabe and Markstrom, 2007) estimates snow melt rates and mean monthly snow water equivalent (SWE) mean watershed elevation and monthly air temperature. 25 Infiltration, surface runoff, soil moisture and baseflow processes for each watershed are simulated by the Sacramento Soil Moisture Accounting Model (SAC-SMA; Burnash, 1995). The ecosystem productivity module computes carbon dynamics (GPP and res-12712 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | piration) using linear relationships between ET and GPP derived from global eddy covariance flux measurements (Sun et al., 2011a, b). The User Guide of WaSSI Ecosystem Services Model-Version 2.1 (http://www.forestthreats.org/research/tools/WaSSI) provides detailed description of model algorithms and data requirements (Caldwell et al., 2012). 5 The necessary inputs to run the WaSSI model include monthly precipitation, monthly mean air temperature, monthly mean leaf area index (LAI) by land cover, land cover composition within each watershed, and 11 SAC-SMA soil parameters. The historical (1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997) and the future (2031-2060) climate data dynamically downscaled by WRF model, were bias corrected, and thus were scaled to the 12-digit HUC level. The 10 2006 National Land Cover Dataset (NLCD; http://www.mrlc.gov/nlcd06_data.php) with 17 land cover classes were aggregated into 10 classes (Fry et al., 2011): crop, deciduous forest, evergreen forest, mixed forest, grassland, shrubland, wetland, water, urban and barren. WaSSI The monthly LAI time series data required by WaSSI for each land cover type were derived from the Moderate Resolution Imaging Spectro-15 radiometer (MODIS) -MOD15A2 FPAR/LAI 8 day product (Myneni et al., 2002). The 1 km × 1 km SAC-SMA soil dataset provided by the State Soil Geographic Data Base (STATSGO) -based on the Sacramento Soil Moisture Accounting Model Soil Parameters was aggregated to the 12-digit HUC watershed. No WaSSI model parameters were calibrated during the model evaluation process. 20 The WaSSI has been evaluated at multiple scales using gaging station data for streamflow and remote sensing products for evapotranspiration across the US (Sun et al., 2011a(Sun et al., , 2015aCaldwell et al., 2012). At the 12-digit HUC scale, the model was validated using monthly and annual water yield data collected at 72 selected USGS watersheds, and ET and GPP data for 170 National Forests over the CONUS (Sun et al.,25 2015a). Overall, the validation results suggested that this model could capture characteristics of water and carbon balances at the selected spatial levels under various climatic conditions (Sun et al., 2015a, b

Impact analysis
We first examined modeled changes in monthly ET and GPP at the 12-digit HUC watershed scale using the WRF dynamically downscaled, bias corrected historical and future climate data, respectively. We next computed future annual changes at three spatial levels: the entire CONUS as whole, the 12-digit HUC watershed, and the in-5 dividual WRR. The multi-year means of annual precipitation, temperature, ET, Q, and GPP averaged across the whole CONUS, WRR, or each 12-digit HUC watershed for the 1979-2007 time period were compared to those for the 2031-2060 period. The absolute or percent (except for temperature) changes for each variable were calculated. The absolute differences were expressed as the future means minus those in the historical period, while the percent differences were calculated using the absolute difference divided by baseline mean in the 1979-2007. The future monthly changes of these ecosystem flux variables were also assessed for the whole CONUS and each WRR.  The 12-digit HUC watersheds with higher values (> 1000 g C m −2 yr −1 ) of GPP (Fig. 4a) were mainly located in the areas of the southeast and the Pacific Northwest. By contrast, the average annual temperature climatology of the CONUS presented a clear latitudinal increase ranging from −0.8 • C in the north to 22 • C in the south. Temperature in the Rocky Mountains was lower than 4 • C relative to the surrounding regions 5 due to topographical effects. The area weighted average precipitation, temperature, ET, Q and GPP across CONUS in the period of 1979-2007 was 801 mm yr −1 , 11.2 • C, 515, 290 mm yr −1 and 1232 g C m −2 yr −1 , respectively ( were also found in the WRR3, 6 and 8, but the western WRRs (e.g., WRR13-16 and 18) exhibited lowest values (< 800 g C m −2 yr −1 ).
The baseline CONUS-wide intra-annual precipitation presented a complicated pattern (Fig. 5). Except in February, precipitation in all the months was more than in August, and then increased. The intra-annual distribution was different (e.g., phases and magnitudes) among the 18 WRRs, due to the complex differences in topography and climate among them (not shown). For WRR16-18, most precipitation fell in HESSD 12,2015 Predicting future US water yield and ecosystem productivity S. Sun et al. January-April and October-December, while precipitation in other WRRs mainly concentrated in May-September. In all the WRRs, the intra-annual temperature followed a unimodal curve, with peaks in July or August and the lowest values in January or December. For ET and GPP, the higher values were mainly found from May to November, except for the WRR18. Comparing the monthly ET or GPP distributions among the 18 5 WRRs, they could be divided into three categories: unimodal, sine and trough curves.

Future climate change
Projected future precipitation and temperature followed a similar pattern as the baseline (Fig. 2). Precipitation showed a longitudinal decrease from the east to the west, but temperature presented a clear latitudinal decrease. However, for each 12-digit HUC 10 watershed, these two climate variables would increase or decrease by different magnitudes in the future ( Fig. 2c and d for precipitation, and Fig. 2g). Annual precipitation is predicted to increase in 82 % of the 82 773 12-digit HUC watersheds, those watersheds that were predicted to have a decrease in precipitation were mainly located in the southeast and the west coastal regions. The northeast and the northwest coastal re- 15 gions were predicted to have a greater absolute increase (> 150 mm yr −1 ) or decrease (> 200 mm yr −1 ) in precipitation, respectively (Fig. 2c). The greater percent increases in precipitation (> 18 %) were found in some watersheds in the southwest and the northeast regions (Fig. 2d). Future temperature was predicted to increase consistently across watersheds, ranging from 1.0 o C to 3.0 • C. The northwest and the north-central 20 regions were predicted to have temperature increase more than 2.1 • C (Fig. 2g).
For the CONUS as a whole, the area weighted mean annual precipitation and temperature for 2013-2060 was 844 mm yr −1 and 13.1 • C, respectively (Table 1). The mean annual P for the entire CONUS was predicted to increase by 45 mm yr −1 (6 %) and T increase by 1.8 • C (Table 2). All WRRs exhibited increases in P with the ex- 25 ception of WRR17 with a slight decrease in P (13 mm yr −1 or 1 %). The large absolute change in precipitation (> 60 mm yr −1 ) was in WRR2, 4, 5 and 7, while the WRR8 and 14 had lower predicted increases (< 15 mm yr −1 ). The higher relative increases in pre- 12,2015 Predicting future US water yield and ecosystem productivity S. Sun et al. cipitation (≥ 10 %) existed in the WRR2, 5, 15 and 16, however, WRR1 and 8 had lower predicted increases (≤ 1 %). For the future temperatures, all the 18 WRRs were predicted to increase in temperature relative to the baseline period, especially in the WRR9, 10, 14 and 16 (≥ 2 • C).

HESSD
Both future P and T had similar intra-annual fluctuations to those of the baseline 5 period (top panels in Fig. 5a and b). However, the magnitudes of differences in both P and T differed in different seasons were different (the bottom of Fig. 5a and b). In most months, precipitation was predicted to increase from 3 to 11 mm yr −1 , especially in January, May and September (> 7 mm yr −1 ). For February, March, October and November, P was predicted to decrease from −5 to −1 mm yr −1 . The temperatures for each month were predicted to increase by at least 1.5 • C, particularly for January and June-October (> 2.0 • C) (Fig. 5b).
The comparisons of seasonal climatic change patterns among the 18 WRRs suggested the timings of change were similar among WRRs (not shown), but the magnitudes of changes varied greatly. The future monthly precipitation was predicted to in- 15 crease in January and May-October in more than 10 WRRs. The increases were most pronounced in January, July and September (Fig. 6a). In other months, however, the future monthly precipitation would decrease in most of the WRRs. The future monthly temperature for all the WRRs was predicted to increase from 0.5 to 3.0 • C. January and June-October temperatures in most WRRs were predicted to increase more than 20 1.5 • C for most WRRs.

Annual change
The spatial patterns in ET and Q for the baseline were similar to those in the future (Fig. 3). However, the changes of annual ET ( Fig. 3c and d) and Q ( Fig. 3g and  reductions in annual ET were mainly concentrated in the northwest coastal region. Annual ET was predicted increase more (> 32 mm yr −1 ) in the northeast region ( Fig. 3c and d), especially in the southeast coastal region and the southern extent of the northeast region (> 48 mm yr −1 ). Relative changes (%) in ET were highest (> 6 %) in most of the western regions (excluding the west coast) and the northeast, and the highest 5 increases (> 12 %) were predicted in the southernmost portion of the southwest region. Annual Q was predicted to increase in 52 % of the 82 773 12-digit HUC watersheds by 2031-2060 ( Fig. 3g and h). In general, annual Q in the northeast and the southernmost portion of the southeastern region was predicted increase, while other regions were predicted to decrease ( Fig. 3g and h). The largest increases (> 100 mm yr −1 ) and 10 the decreases (> 100 mm yr −1 ) in Q were mainly found in the northeast and the west coastal and the southeast regions, respectively. Q in the southernmost portion of the southwest region was predicted to increase more than 20 %, while the central part of the west CONUS would generally decrease more than 20 %.
Projected future multi-year mean annual ET and Q across CONUS were 551 and 15 297 mm yr −1 , respectively (Table 1), representing an increase in ET by 37 mm yr −1 or 7 %, and in Q by 9 mm yr −1 or 3 % ( Table 2). All WRRs were predicted to have an increase in ET. The WRR2, 5 and 7 were predicted to have the largest absolute increases in ET (> 45 mm yr −1 ), while the WRR17 (18 mm yr −1 ) was predicted to have the lowest increases. The highest predicted relative increases in ET (≥ 10 %) were in WRR5,9,20 16 and 17, however, WRR17 was predicted to have the lowest increases (4 %). Annual Q was predicted to increase in nine WRRs, decrease in eight WRRs, and remain unchanged in one WRR relative to the baseline period (

Seasonal change
The variations of future CONUS-wide multi-year mean monthly ET and Q were presented in Fig. 5c and d. The predicted future monthly ET increased the most (> 2 mm yr −1 ) in January. The April-October ET was predicted to increase more than other four months. Monthly Q was predicted to increase in most months (9 months), espe-5 cially in January and September (increase > 3 mm yr −1 ). We also found that the seasonal patterns of ET and Q for the baseline and the future periods were similar (not shown here), but a magnitudes have changed. Figure 6c and d presents the number of the WRR within a given change interval for ET and Q by month, respectively. Generally, the future monthly ET was predicted to increase by dif-10 ferent rates for each month at each WRR (Fig. 6c). Moreover, ET from May to September (roughly the growing season) was predicted to increase more (> 2.4 mm yr −1 ) than other months in most of the 18 WRRs. Q was predicted to increase in January, February, July, September and December in most of WRRs, but was predicted to decrease in April and November.

Annual change
The overall spatial distribution of GPP did not change in the future (Fig. 4b) when compared to the baseline (Fig. 4a). However, the magnitudes of GPP changes in the future varied spatially (Fig. 4c and d). Annual GPP was predicted to increase in the 20 majority (98 %) of the 82 773 12-digit HUC watersheds. The few watersheds that were predicted to experience reductions in annual GPP were mainly located in the northwest coastal region. A relatively high increase (> 120 g C m −2 yr −1 ) in GPP was predicted in the northeast, especially in the southern part of the region (> 180 g C m −2 yr −1 ; Fig. 4c).
In contrast to the absolute difference, most of the west CONUS (excluding the coastal 25 regions) were predicted to have the largest relative increase (> 12 %) in annual GPP. The highest changes (> 20 %) were mainly located in southern portion of the southwest region.

HESSD
Over the CONUS, multi-year mean annual GPP was predicted to be 1339 g C m −2 yr −1 in the future (Table 1), representing an increase of 106 g C m −2 yr −1 or 9 % (Table 2). Future annual GPP in every WRR was predicted to increase from 49 g C 5 m −2 yr −1 to 202 g C m −2 yr −1 or from 5 to 12 % ( Table 2). The WRR2-WRR10 were predicted to have the larger absolute increases in GPP (> 100 g C m −2 yr −1 ), especially for WRR5 with the maximum of 202 g C m −2 yr −1 , while the WRR13 (49 g C m −2 yr −1 ) had the lowest increases. Relative increases in GPP ranged from 5 to 17 % among all the WRRs. The higher GPP increases (> 10 %) occurred in WRR4, 5,7,9, 16, with the largest of 17 % in WRR16, while other WRRs had the lower increments than 10 %, particularly in WRR3 and 8 with the minimum of 5 %. mean monthly GPP during the baseline and the future periods, the future magnitude in each month was predicted to change to some degree (the bottom of Fig. 5e). Overall, the future monthly ET was projected to have the larger increments (> 9 g C m −2 yr −1 )

Seasonal change
in January and May-October relative to other months. The future intra-annual fluctuation patterns of GPP for each WRR were similar to the baseline periods (not shown 20 here). As indicated by the number of the WRR within a given GPP difference interval (Fig. 6e)

Uncertainties
In the present study, we assumed that the water balance and ecosystem productivity at each 12-digit HUC watersheds were unaffected by human activities as represented by a fixed land cover (year 2006), and changes in ecosystem fluxes were fully attributed 5 to climate change alone. However, most catchments in the US will experience some levels of human influences (National Research Council, 2002). Hydrology and ecosystems can be influenced significantly by human activities on various temporal and spatial scales (Foley et al., 2005;Harding et al., 2011). Hydraulic projects such as dam constructions, reservoir management (Hu et al., 2008), groundwater withdrawals for 10 irrigation and domestic use, and land use/cover change all affect watershed balances (Foley et al., 2005;Piao et al., 2007;Wang and Hejazi, 2011;Schilling et al., 2008) and ecosystem productivity (Zhang et al., 2014). Similarly, natural disturbances (e.g., wildfire, climate extremes, and pest and pathogen outbreak) could also impact water balance and ecosystem productivity in 15 the past and the future. For example, the direct effects of wildfire include plant mortality and thus exert adverse impacts on vegetation productivity, consequently leading to a decrease in carbon uptake and stocks (Lenihan et al., 2008;Dore et al., 2010;Lee et al., 2015). Wildfires alter the watershed hydrologic processes by reducing vegetation canopy interception, transpiration, and infiltration rates (Yao, 2003;Neary et al., 2005;20 Bond-Lamberty et al., 2009;Brookhouse et al., 2013;Nolan et al., 2014Nolan et al., , 2015. As an important natural disturbance, droughts generally increase vapor pressure gradient between leaves and atmosphere and thus cause stress on plant hydraulic systems (Anderegg et al., 2012;Reichstein et al., 2013). As a result, high tension in the xylem can trigger embolism and partial failure of hydraulic transport in the stem, and may result 25 in vegetation mortality, which can aversely impact water yield and carbon sequestration (Cook et al., 2007;Allen et al., 2010;Guardiola-Claramonte et al., 2011;Adams et al., 2012). In addition, droughts often lead to pest and pathogen outbreaks (Over-12721 Introduction  , 1990;Hason and Weltzin, 2000;Marengo et al., 2008;DeRose and Long, 2012;Jactel et al., 2012), and thus predispose an individual plant species to disease or mortality (Schoeneweiss, 1981;Ayers and Lombarder, 2000). Although our modeling approach considered water stress on productivity, tree mortality was not considered and the impacts of droughts on Q and GPP might be underestimated. 5 Elevated CO 2 can also affect water yield and ecosystem productivity by changing water use efficiency and vegetation processes (e.g., stomatal conductance and LAI; Sun et al., 2014). However, we did not consider the CO 2 fertilization effects, potentially resulting in errors in estimating ET, GPP or water yield (Cox et al., 2000;Gedney et al., 2006;Oki et al., 2006;Betts et al., 2007;Piao et al., 2007). Neglecting human activities 10 and natural disturbances and their combined effects may introduce uncertainties into our results. However, the potential errors are largely dependent on specific trajectories of climate change and land cover change (Qi et al., 2009;Thompson et al., 2011;Alkama et al., 2013). The complex interactions of climate, disturbance, ecohydrological processes require a more mechanistic integrated modeling approach that is beyond 15 the scope of this study.

Management implications
Numerous modeling studies around the world have showed that climate change could increase or decrease water availability to some ecosystems and human populations under different climate scenarios (Arnell, 1999;Blanc et al., 2014;Ingjerd et al., 2014;20 Kundzewicz and Gerten, 2015). Our analyses showed that, over the whole CONUS, P would increase by 45 mm (6 %) leading to a small increases in Q by 9 mm yr −1 (3 %).
However, there are large regional differences in Q responses to future climate change among the 18 WRRs. The range in the magnitude of response is large, ranging from a decrease of −32 mm yr −1 to an increase of 113 mm yr −1 or from −12 to 21 %. Despite 25 the increase in annual P, annual Q in WRR1, 3, 8, 11, 14, 16 and 18 was predicted to decrease by various degrees, due to the increased ET. Consequentially, under the climate scenario studied water supply stress will likely increase in these WRRs. In ad-12722 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | dition, monthly Q responses to future climate also vary among watersheds. Water yield in about half of the 18 WRRs (mainly located in the west CONUS) was predicted to decrease and increase in WRR2-8. The increased Q in the wet months could increase flooding risk, while decreased Q in dry months could aggravate water shortage conditions. For example, the monthly Q in California (mostly in the WRR18) was predicted to decrease by around 5 mm during spring through early summer (the major runoff generation season) due to coupling of changes in P and ET and/or precipitation forms in the mountainous regions in Sierra Nevada mountains. The decrease in flow may cause severe water shortage similar to what is happening in 2014-2015 in California (Aghakouchak et al., 2014;Mao et al., 2015). Hydrological changes could bring many impacts on water-related economic sectors. For example, droughts would reduce low flows and degrade water quality (high water temperature and nutrient concentrations), and thus have negative effects on fisheries (Magoulick et al., 2003;Dolbeth et al., 2008;Gillson et al., 2009), navigation (Theiling et al., 1996;Roberts, 2001), and recreation (Thomas et al., 2013). 15 The modeling results suggested that GPP over the whole CONUS would increase 106 g C m −2 yr −1 (9 %) in the future. The increase by WRR ranged from 49 g C m −2 yr −1 to 202 g C m −2 yr −1 or from 5 to 17 % among the 18 WRRs. These findings suggested that carbon stock and vegetation capacity to sequester atmospheric CO 2 for the entire CONUS and each WRR could be enhanced under the SRES A2 climate scenario. 20 Most WRRs were predicted to have GPP increases during late spring to summer, which implied that the capability of ecosystems to sequester carbon in these months will be significantly enhanced in future. By contrast, GPP was predicted to decrease in several WRRs for several months. For example, during August and September, predicted GPP in WRR17 decreased. The ecosystem sequestration carbon capability would be 25 reduced in these months under the SRES A2 climate scenario. For forests, variations of GPP caused by climate change will be ultimately reflected in timber production, soil carbon storage, and other ecosystem processes such as dissolved carbon loading in aquatic ecosystems. According to this study, the forest biomass and timber production  (Sohngen and Mendelsohn, 1998;Irland et al., 2001;Alig et al., 2004). At the same time, forest densification of forest lands under a warming climate may provide conditions of increased wildfire potential (Y. Liu et al., 2013).

5
We assessed the impacts of future climate change on hydrological cycle and ecosystem productivity over the CONUS by linking an ecohydrology model (i.e., WaSSI) with WRF dynamically downscaled the HadCM3 model climate data under the IPCC SRES A2 emission scenario. The current study represents a coupling of bias-corrected, dynamically downscaled climate data with an ecohyrological model to address regional ecosystem issues. The study provides a potential scenario of likely impacts of future climate change on watershed hydrology and productivity across the CONUS, including 82 773 12-digit HUC watersheds. Although only one future climate scenario (the SRES A2 emission scenario) and one GCM (HadCM3 model) was employed here, the methodology applies to other scenarios when more climate change scenarios gener-15 ated from WRF are available. Future climate change will not likely change the overall spatial patterns of precipitation, temperature, ET, Q and GPP. However, a large spatial variability in the hydrological and ecosystem productivity responses is expected at both the 12-digit and 2-digit HUC scales. The assessment results provide a benchmark of water yield and ecosystem 20 productivity response across the CONUS. This type of information will be useful for prioritizing watershed restoration and developing specific measures to mitigate the negative impacts of future climate to sustain the terrestrial ecosystem on different spatial scales (i.e., 12-digit HUC and WRR). Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version