1 Introduction

A warming of the climate has been observed globally and in the western United States (e.g., California and Nevada), mainly due to anthropogenic forcing (Intergovernmental Panel on Climate Change, IPCC 2007; U. S. Global Change Research Program 2009). Climate-related changes include increased surface air temperature and atmospheric water vapor, more severe weather extremes, changes in precipitation patterns and intensity, more common and intense floods and droughts, reduced snowpack, earlier melting of the snowpack, shifting of the runoff timing, and a rise in sea level (e.g., IPCC 2007). Climate change will stress water resources, impact energy supply and demand, challenge crop and livestock production, increase risk to human health, and affect the environment and ecosystems.

Trenberth et al. (2007) showed that the global mean surface air temperature has risen around 0.74°C over the last 100 years (1906–2005), and the rate of increase has accelerated from 0.07°C/decade over the last 100 years to 0.13°C/decade over the last 50 years. Surface air temperatures have increased more over land (0.27°C/decade) than over ocean (0.13°C/decade). Therefore, increases in the frequency of extreme high average temperatures are expected (Easterling et al. 2000; DeGaetano and Allen 2002; Meehl and Tebaldi 2004). As the result of rising temperature, more precipitation falls as rain instead of snow for locations and seasons with climatological mean temperatures around 0°C. Observational studies have shown an earlier onset of spring and longer liquid precipitation seasons in some high latitude regions in the last 50 years (Cayan et al. 2001; Groisman et al. 2001, 2005, 2007; Easterling 2002).

Increasing air temperature also increases the water holding capacity of the atmosphere. Increased temperatures have raised water vapor over oceans by about 5% over the last century for the same relative humidity (e.g., IPCC 2007). More moisture in the atmosphere has generally increased precipitation intensity and the risk of heavy rain and snow events. The events of heavy (upper 5%) and very heavy (upper 1%) precipitation have increased by 14 and 20%, respectively, in the U.S. during the last three decades of the twentieth century, especially over the eastern parts of the country (Kunkel et al. 2003; Groisman et al. 2004). However, the total precipitation and/or the frequency of days with an appreciable amount of precipitation (wet days) may be unchanged or decreasing (Easterling et al. 2000; Sun and Groisman 2004; Groisman et al. 2005; Im et al. 2009). It was also reported that there has been an increase in both severe droughts and heavy rains in many regions of the world (Huntington 2006; Allan and Soden 2008; Kendon et al. 2008). The variation of precipitation in the second half of the twentieth century was dominated by variations on interannual to inter-decadal time scales (Manton et al. 2001; Peterson et al. 2002; Griffiths et al. 2003; Herath and Ratnayake 2004).

Some studies have focused on the influence of climate change on the western United States (Dettinger and Cayan 1995; Cayan 1996; Leung and Ghan 1999; Barnett et al. 2004, 2008; Dettinger et al. 2004; Leung et al. 2004; Pierce et al. 2008; Mote and Salathé 2010). Results show the snow water equivalent (SWE) for the western U.S., as measured on April 1, has decreased from 1950 (60 cm; Mote 2003) to 2000 (50 cm), except in the southern Sierra Nevada (Mote 2003, 2006; Vanrheenen et al. 2004; Mote et al. 2005). More winter precipitation fell as rain instead of snow in the period 1949–2004 (Knowles et al. 2006), and earlier snowmelt runoff was also observed (Dettinger and Cayan 1995; Cayan et al. 2001; Stewart et al. 2004; Rauscher et al. 2008). Moreover, warming trends of the minimum daily temperatures over western and central North America were observed (Robeson 2004). Precipitation has generally increased during the winter and spring, the wettest seasons in states such as California and Nevada, over the past 50 years, while summer and fall have decreased in precipitation (NOAA’s National Climatic Data Center 2008; U. S. Global Change Research Program 2009). A more arid climate in the southwestern part of the North America is predicted (Seager et al. 2007). However, there are still some debates on how the precipitation pattern will change for these regions in the future (e.g., Weare 2008).

Dynamically downscaling from coarse-resolution global-model simulations to higher-resolution regional-scale outputs has been conducted in previous studies (e.g., Jones et al. 1997; Christensen et al. 1998; Giorgi and Shields 1999; Leung and Ghan 1999; Small et al. 1999; Snyder et al. 2004; Kim et al. 2002; Bell et al. 2004; Leung et al. 2004; Kanamitsu and Kanamaru 2007; Caldwell et al. 2009). However, most of these studies (e.g., Bell et al. 2004; Leung et al. 2004; Caldwell et al. 2009) used a relatively coarse resolution (e.g., about 10–40 km), which is not sufficient to study climatic changes at regional scales in mountainous regions, such as the Sierra Nevada. Therefore, higher resolution (4 km) dynamical-scaling simulations over California and Nevada regions were carried out here.

In this study we refined our understanding of regional changes in California and Nevada, especially those related to precipitation and extreme events (e.g., heat waves, heavy precipitation), through WRF (Weather Research Forecast) model dynamical downscaling. The model and data are described in Sect. 2, followed by a comparison between the dynamical downscaling results and statistical downscaling results in Sect. 3. The influence of climate change on California and Nevada is discussed in Sect. 4. A summary and conclusions are given in Sect. 5.

2 Description of model, data, and experiment design

2.1 Model

The Weather Research and Forecasting (WRF) Model, a community-developed mesoscale model, was used to dynamically downscale the global model outputs, including Global Forecast System (GFS) reanalysis and Parallel Climate Model (PCM) simulations, to finer resolution in the California and Nevada region. PCM can capture ENSO (El-Niño Southern Oscillation) and PDO (Pacific Decadal Oscillation) dominant features very well (e.g., Joseph and Nigam 2006; Meehl and Hu 2006). As a mesoscale model, WRF can simulate the local response for a given ENSO or PDO environmental forcing. WRF is a compressible, non-hydrostatic model, using terrain-following coordinates and governing equations written in flux-form, which conserves mass and dry entropy. The Runge–Kutta third-order time scheme was used and the fifth and third-order advection schemes were chosen for the horizontal and vertical directions, respectively. The advanced research WRF (ARW) model version 2.2 (Skamarock et al. 2005) was adopted for all simulations.

2.2 Data

Future initial and boundary conditions were provided by PCM simulations under the business as usual (BAU) scenario. PCM is a coupled climate model. The atmospheric component is the National Center for Atmospheric Research (NCAR) Community Climate Model; the ocean component is the Parallel Ocean Program with 32 vertical levels and a horizontal resolution of 2/3°. The sea ice component is the Naval Postgraduate School model using a resolution of 27 km. The horizontal resolution of PCM data used here was about 2.8° × 2.8°, and the time interval for most of the PCM variables was 6 h. To validate the PCM data, current PCM data was downscaled and compared to downscaled GFS reanalysis data. GFS is a medium-range, global model. The horizontal resolution of reanalysis is 1° × 1° and the time interval of the data set is 6 h.

For dynamical downscaling, 7 years of GFS reanalysis (2000–2006) and 10 years of both current PCM climate simulations (1997–2006) and future PCM climate simulations (2047–2056) were used. Ten-year time periods of PCM data were chosen for climate study, as suggested by Jones et al. (1997). However, for GFS reanalysis, only 7 years data were used because the model resolution was upgraded from T126L28 to T170L42 (T refers to horizontal spectral truncations, L refers to vertical level) in 2000 (http://www.emc.ncep.noaa.gov/gmb/STATS/html/model_changes.html).

The dynamical downscaling results were also validated through comparison with the statistical downscaling results from PRISM (Parameter-elevation Regressions on Independent Slopes Model) (Daly et al. 1994; DiLuzio et al. 2008). PRISM data are gridded interpolations from station data based on weighting functions using distance, elevation, topographic facet, atmospheric layer, orographic effectiveness, and coastal proximity as parameters. Thus the spatial scale and pattern of each variable (e.g., precipitation) are well preserved in PRISM, and the data can even be used in areas with complex terrain. PRISM data are available monthly with a horizontal resolution of 4 km, similar to the scale of dynamical downscaling simulations. The data period used here is from 1997 to 2006 (the same as the model time period).

2.3 Experimental design

For the downscaling, WRF simulations were run twice a month using GFS reanalysis or PCM outputs as initial and boundary conditions. Three domains with two-way interaction were simulated and the resolutions were 36 km, 12 km, and 4 km for domains 1–3, respectively. The numbers of grid points for domains 1, 2, and 3 were 91 × 95 × 31, 145 × 145 × 31 and 271 × 310 × 31 points, respectively, in the x–y–z directions. The highest-resolution domain covers (33°N–42°N, 125°W–115°W). The Lambert projection was used in the model, which gives even grid spacing, unlike lat-lon coordinates. Four dimensional data assimilation (i.e., nudging; see Liu et al. 2008 for details) was applied to domain 1 using PCM/GFS data to retain the large-scale properties from the global model simulations or reanalysis. The physics schemes that were used for simulations were Grell-Devenyi cumulus parameterization (Grell and Devenyi 2002), WRF Single-Moment 6-class (WSM6) cloud microphysics (Hong et al. 2004), Yonsei University (YSU) planetary boundary parameterization (Hong et al. 2006), Rapid Radiation Transfer Model long-wave radiation (Mlawer et al. 1997) and Dudhia short-wave radiation (Dudhia 1989). No cumulus parameterization was applied to domain 3 because of its fine resolution. The model was integrated with a time step of 180 s for domain 1.

3 Comparison of the dynamical downscaling results with the statistical downscaling PRISM dataset

The dynamical downscaling results from GFS reanalysis (called GFS-WRF runs) and PCM simulations (called PCM-WRF runs) were compared with the statistical downscaling (PRISM) results. The variables investigated here include precipitation, the surface maximum and minimum temperature, and the surface wind field. Due to the shorter time period of data sets from GFS (i.e., 2000–2006), only results from 2000 to 2006 were used for comparison among these three data sets (i.e., GFS-WRF, PCM-WRF, and PRISM).

3.1 Precipitation

The western U.S. has its maximum precipitation in winter and minimum in summer due to a north–south seasonal shift of the Pacific high and mid-latitude storm track. Topography also plays an important role in precipitation patterns; mountain blocking and orographic lifting effects produce more precipitation on the windward sides of the mountain ranges. Snowmelt from winter snow stored in the Sierra Nevada mountain range provides the major water source for irrigation in the next growing season in CA (http://www.wrcc.dri.edu/narratives/CALIFORNIA.htm).

Dominant precipitation features for California were captured in both the WRF simulations and PRISM results (Fig. 1). Maximum precipitation occurred at the mountain tops (Fig. 2) with values around 500–600 mm/month for PRISM, 600–700 mm/month for PCM-WRF runs, and 700–800 mm/month for GFS-WRF runs. The maximum values simulated from the dynamical downscaling results were slightly larger than those from PRISM. Statistical downscaling may have difficulty extrapolating to higher elevations due to fewer available observations. The rainfall patterns of these three results were consistent except that PCM-WRF runs had a slight wet bias (around 50 mm) in the southern part of California.

Fig. 1
figure 1

Monthly averaged precipitation (mm) during the winter season (i.e., December, January, and February) from 2000 to 2006 for a PRISM, b PCM-WRF outputs, and c GFS-WRF outputs

Full size image
Fig. 2
figure 2

Terrain of the model domain 3, where locations mentioned in the text are highlighted with: (1) Carquinez Strait, (2) Sacramento Delta, (3) Central Valley, (4) Sierra Nevada, (5) mountain ridge (higher elevation), (6) Mojave Desert, and (7) Palm Desert

Full size image

PCM-WRF precipitation predictions were not only larger than PRISM results, but also extended farther southward (Fig. 3a). The maximum difference between PCM-WRF and PRISM can vary from 150 mm/month to 250 mm/month (yellow area in the figure) at the mountaintop. The southward shift is not obvious in the GFS-WRF simulations (Fig. 3b), but they show more precipitation in higher elevations. The standard deviations of precipitation were calculated for PRISM, PCM-WRF, and GFS-WRF (not shown), which also have their maximum value of around 400 mm/month at the mountaintops. A large standard deviation implies large interannual variation in precipitation. The average annual number of precipitation days is shown in Fig. 4. The threshold for a rainy day is 1 mm. Results from PCM-WRF and GFS-WRF were highly correlated (the pattern correlation is around 0.83, and is significant at a 95% confidence level), with higher precipitation frequency over the oceans and at higher elevations. Compared to GFS-WRF runs, PCM-WRF runs produced more rainy dates over the southern Sierra Nevada and fewer rainy days over the ocean near the coastal region, particularly near the coast of southern CA.

Fig. 3
figure 3

Differences of monthly averaged precipitation (mm) during the winter season from 2000 to 2006 a between PCM-WRF runs and PRISM, and b between GFS-WRF runs and PRISM

Full size image
Fig. 4
figure 4

Averaged yearly number of rainy days from 2000 to 2006 for a PCM-WRF runs, b GFS-WRF runs, and c their difference. The threshold for a rainy day is 1 mm

Full size image

3.2 Temperature

The dominant features of the averaged 2-m maximum and minimum air temperature fields from PRISM were reasonably captured by dynamical downscaling during the winter season. Compared with GFS-WRF runs, the values of the maximum 2-m temperature from PCM-WRF were closer to those from PRISM over the Central Valley, while those from GFS-WRF runs were around 1° closer than PCM-WRF modeled values to those from PRISM for the rest of domain 3 (Fig. 5). For averaged daily minimum temperature in winter (not shown), results from PCM-WRF were closer to PRISM than GFS-WRF results were. This is unexpected and requires further study (e.g., studying the data used in GFS/PCM simulation). GFS-WRF runs should have been closer to PRISM since actual observations were assimilated into GFS reanalysis. The analysis on diurnal and seasonal data also showed lower diurnal and seasonal variation of the temperature on the coast compared to inland regions due to the adjustment from the ocean (figure not shown). Higher elevations also experienced larger temperature variation.

Fig. 5
figure 5

Averaged daily maximum 2-m temperature (K) during the winter season from 2000 to 2006 for a PRISM, b PCM-WRF runs, and c GFS-WRF runs

Full size image

3.3 Wind field

California is located in the path of the mid-latitude westerlies; the wind direction can be west, northwest or southwest over ocean. However, the inland wind field is very complex due to topographic effects (Fig. 6). PRISM does not contain wind field. Therefore, only the dynamical downscaling 10-m winds from PCM (Fig. 6a) and GFS (Fig. 6b) were compared. Overall, their dominant features were similar, except over parts of the ocean. However, winds over mountain ridges were higher from GFS-WRF runs than from PCM-WRF runs. For the zonal wind, most regions had positive values (westerly component) over ocean and mountain, and the maximum value was around 5–7 m/s at the mountain top; negative values are found in the Sacramento Valley. For the meridional wind, negative values (northerly component) were found along the south part of the coast region. The northerly wind along the coast can produce oceanic upwelling due to Ekman pumping. The maximum value of meridional wind was around 4–6 m/s at higher elevations.

Fig. 6
figure 6

Averaged a zonal wind and b meridional wind during the winter season from 2000 to 2006 for PCM-WRF runs. c and d are the same as (a) and (b), except for GFS-WRF runs

Full size image

Overall, these experiments show that the dominant features of precipitation, temperature, and wind fields are well simulated by WRF runs using PCM based initial and boundary conditions. The magnitude, spatial pattern, and seasonal variations of these variables are well reproduced by PCM-WRF runs when comparing with the PRISM dataset and reanalysis results (GFS-WRF runs).

4 The potential influence of climate change on California

4.1 The influence of climate change to the mean fields

The influence of climate change on California can be predicted by comparing current and future meteorological fields. Comparison of WRF models using current (1997–2006) and future (2047–2056) PCM inputs showed increases in both precipitation (Fig. 7a, b) and temperature (Fig. 8). An upward shift of the freezing level in the future climate was also predicted (Fig. 7a, b). Here, the freezing level is defined by the 0°C contour line of the mean 2-m temperature. More rainfall also occurred below the freezing level. Although more precipitation will occur in the future, more falls as rain and cannot be stored for summer use (US Climate Change research program 2009). The new snow pack predicted above the freezing level (i.e., gain of snow in the future) is approximately 15% less than the lost snow pack between the two freezing levels, (i.e., loss of snow in the future due to climate warming).

Fig. 7
figure 7

a Difference of averaged monthly precipitation (mm) during the winter season between future PCM-WRF runs (2047–2056) and current PCM-WRF runs (1997–2006), future–current. The red line is the freezing line of future PCM-WRF runs during winter and the purple line is the freezing line of current PCM-WRF runs. b Averaged monthly precipitation (mm) during the winter season for future PCM-WRF runs (2047–2056). The red line is the freezing line of future PCM-WRF runs during winter and the purple line is the freezing line of current PCM-WRF runs (1997–2006)

Full size image
Fig. 8
figure 8

Difference of averaged daily 2-m temperature (K) during the a winter and b summer seasons between future PCM-WRF runs (2047–2056) and current PCM-WRF runs (1997–2006), future–current

Full size image

Surface wind speed change is moderate, under 1–2 m/s in most of the region, though larger (>2 m/s) at ridge tops, in southern CA, and over the southern part of the ocean (Fig. 9).

Fig. 9
figure 9

Difference of averaged 10-m wind speed (m/s) during the winter season between future PCM-WRF runs (2047–2056) and current PCM-WRF runs (1997–2006), future–current

Full size image

The trend of climate change was also seen in PRISM data. Here trend is defined as an increasing or decreasing tendency in a climate variable’s behavior over a specific period of time at a given place. An increasing (decreasing) tendency is called a positive (negative) trend (http://www.nws.noaa.gov/climate/help/glossary.php). Comparing current (1999–2008) and past meteorological fields (1979–1988) showed an increase in precipitation in some regions (Figs. 10, 11) (e.g., northern side of coast region) as well as changing freeze lines (Fig. 11).

Fig. 10
figure 10

Difference of monthly mean precipitation during the winter season between current (1999–2008) and past PRISM results (1979–1988), current–past, and the freezing line of current mean temperature from PRISM (1999–2008) (red line) during winter

Full size image
Fig. 11
figure 11

Current (1999–2008) monthly mean precipitation from PRISM during the winter season and freezing line of the current (1999–2008, red line) and past PRISM (1979–1988, purple line) mean temperature during winter

Full size image

4.2 Impact of climate change on extreme events in California and Nevada regions

The possible influence of climate change on extreme events (e.g., extreme hot days, heavy precipitation) in California and Nevada regions can also be inferred from the current and future WRF simulations. Both extreme hot days (>100°F or 37.8°C) (Fig. 12) and precipitation days (>1 mm/day) (Fig. 13) are expected to increase. More hot days are expected in the Central Valley and Mojave Desert (Fig. 12). Increases in precipitation events were mainly located in the coastal regions of central California (Fig. 13c). One possible reason is that increasing temperature will increase the capacity of atmospheric moisture acquisition over the ocean (IPCC 2007; U. S. Global Change Research Program 2009). The coastal region therefore receives more rainfall, with smaller moisture increases inland due to topographic blocking effects. However, around San Francisco, the moisture can intrude further inland by passing through the Carquinez Strait and cause increased rainfall on the Sacramento Delta. Medium rainfall events (daily precipitation > 10mm) change mostly in the coastal region (Fig. 14a), with heavy rainfall events (daily precipitation >100 mm) occurring more often in the Southern coast and the mountainous regions of northwestern California (Fig. 14b). This may relate to large-scale forcing (http://www.wrcc.dri.edu/narratives/CALIFORNIA.htm), e.g. cold fronts from the north or tropical storms from the south. However, the signal is weak, and further studies are needed.

Fig. 12
figure 12

Average number of extreme hot days (>100° F or 37.8°C) from PCM-WRF runs during the warm season (i.e., May to October) for a 1997–2006, b 2047–2056, and c their differences (future–current)

Full size image
Fig. 13
figure 13

Number of rain days (daily rainfall > 1 mm) for a current (1997–2006), b future (2047–2056), and c their difference (future–current)

Full size image
Fig. 14
figure 14

Same as in Fig. 13c except for a daily rainfall > 10 mm, and b daily rainfall > 100 mm

Full size image

4.3 Changing probability distribution functions (PDF) due to a warming climate

The changes in probability distribution functions of temperature, precipitation and wind speed due to climate change were further analyzed for California and Nevada. Based on hourly-averaged data, we obtained the grid-averaged 2-m temperature, precipitation, and 10-m wind speed for each of the states, and then performed a probability distribution analysis on the mean temperature, total precipitation, and mean wind speed for winter (Fig. 15) and summer (Fig. 16). The temperature probability distribution function (PDF) peak shifted 2.5°C and 1.5°C higher for winter, from current to future, for California and Nevada, respectively (Fig. 15a, b). The summer temperature shift in Nevada was around 3°C, larger than the 2.2°C shift of California (Fig. 16a, b). The warming in California is relatively greater in winter than in the summer, however, in Nevada this is reversed. Both California and Nevada are anticipated to have drier summers (Fig. 16c, d). In winter, California becomes wetter with heavier rainfall events, and there will be little change in Nevada, except that the PDF extends to a larger precipitation value (Fig. 15c, d). Wind speed PDFs do not show any large changes (Figs. 15e, f, 16e, f).

Fig. 15
figure 15

Probability distribution function (PDF) for California (a, c, e) and Nevada regions (b, d, f) for temperature (a, b), rainfall (c, d), and wind speed (e, f) in winter. The units are K, mm per hour for the whole domain, and m/s for temperature, rainfall, and wind speed, respectively

Full size image
Fig. 16
figure 16

Same as in Fig. 15 except for summer

Full size image

The downscaled simulation results are very sensitive to topography. Precipitation is mainly caused by orographic lifting and mountain blocking. Therefore, model resolution must be high enough to depict topography correctly in order to produce good simulation results.

Previous studies on dynamical downscaling were mainly carried out with low-resolution simulations (e.g., 40 km). Here, comparison of precipitation in January 2006 from GFS-PCM runs with different resolutions (i.e., domains) was made (Fig. 17) to evaluate the effect of model resolution on conclusions. Note that in this study, precipitation from a coarser resolution (e.g., domain 1) was obtained from upscaling results due to two-way interaction during the model integration. This may be slightly different from the use of a coarse resolution only (i.e., one-way nesting). Results showed that simulated precipitation strength and fine-scale features were highly dependent upon model resolution, especially in mountainous regions. The information from fine-scale features is very important to some regional-scale studies, such as of hydrological cycles in watersheds. Although there was a wet bias in southern CA from WRF-PCM runs, which can be contributed from both PCM and WRF simulations, the bias will be canceled in the difference field if the same bias is assumed for current and future PCM-WRF runs. Thus, the climate trend obtained in this study still provides some useful information.

Fig. 17
figure 17

Simulated January precipitation (mm/month) in 2006 from WRF-PCM runs for horizontal resolutions at a 36 km, b 12 km, and c 4 km

Full size image

5 Summary and conclusions

The possible influences of climate change on California were investigated with WRF dynamical downscaling on the outputs of PCM under the Business As Usual (BAU) scenario with a 4 km horizontal resolution. Current PCM-based dynamical downscaling results were compared with the dynamical downscaling from GFS inputs and PRISM statistically downscaled station data. Dominant features were consistent among WRF-PCM runs, WRF-GFS runs, and PRISM. The statistical downscaling results have a tendency to estimate less variability in the parameter values, perhaps due to data sparseness resulting in a smoothing effect when compared with the dynamical downscaling, especially in the complex topographic region of California.

The possible influences of climate change on California and Nevada regions were investigated with both observational data analysis and model simulations. According to PRISM results, increasing precipitation was found in northern California by comparing current (1999–2008) and past (1979–1988) observational results. Comparisons between model current runs (1979–2006) and future runs (2047–2056) also suggested possible precipitation increases in California, especially in the north. Using the BAU scenario, the surface air temperature increases by 1–1.5°C (2–3°F) for California and 1.5–2°C (3–4°F) for Nevada. Winter precipitation is expected to increase in northern California under a warming climate, but California becomes drier in the summer. The surface winds only change noticeably at higher elevations. Increasing surface air temperature leads to more winter precipitation falling as rain instead of snow, earlier snow melt, and more extreme events (heat waves, heavy precipitation, etc.). PRISM data suggest that the freezing level has already moved to higher elevations in the past 30 years. Model simulations also suggest a similar change in the future snowpack. More hot days in the Sacramento Valley and the southern part of California and Nevada are also expected. Rainfall frequency along the coastal region and the region from San Francisco inland are predicted to increase. Increasing temperature leads to an atmosphere that can hold more moisture, and provides more fuel for rain in the coastal region. However, increasing temperature also increases the height of the lifting condensation level, having a negative effect on precipitation. The Coast Mountains block moisture at the coast, resulting in little change in rain frequency in the inland region. Heavy precipitation events occur in northern California and the southern Californian coast, which seems related to large scale forcing (e.g., cold front or tropical storm invasion). Predictions depend on the model resolution, and downscaling with a higher resolution may be needed for areas with complex topography.

It must be mentioned that our conclusions are solely based on downscaling results of one regional climate model (RCM) driven by the outputs of one global climate model (GCM) under the BAU emission scenario. It is expected that the use of different RCMs, GCMs and/or different emission scenarios might give different conclusions. Therefore, an ensemble approach, which uses combinations of different RCMs, GCMs, and emission scenarios with a longer downscaling time period, to estimate the uncertainties and the impacts of climate change on California is encouraged for future work when more computing resources are available.