Trend analysis of watershed-scale annual and seasonal precipitation in Northern California based on dynamically downscaled future climate projections

Precipitation in Northern California is the main source of water for the State of California. Approximately 70% of California's average annual runoff occurs in the northern half of the state, while about 75% of the state's urban and agricultural water needs are for the southern half (CADWR 2003). Precipitation is closely related to the severity of droughts and floods. Lower amounts of annual precipitation may cause droughts, while intense precipitation may produce floods. Therefore, it is necessary to investigate the impacts of future climate change on precipitation in Northern California due to its critical role in the planning and management of California water resources.

Many future climate change studies have been conducted by simulations of global climate models (GCMs) based on future climate scenarios. GCMs simulate the atmospheric variables, including precipitation, for the whole globe using a typical horizontal grid resolution between 1 to 2 degrees (Flato et al. 2013). Although such a grid resolution may be sufficient for a global or continental-scale analysis, it may be too coarse for a regional-scale study. In fact, usually only a few GCM grid points fall within a study area at the regional scale. In such a case, GCMs cannot reliably simulate circulation patterns that lead to hydrological extreme events (Christensen & Christensen 2003). Therefore, their results need to be downscaled to a finer resolution to reliably assess regional impacts of climate change. Regional climate models (RCMs), which provide a better representation of the orographic effects, land-sea contrast, and land surface characteristics (Jones et al. 1995; Christensen & Christensen 2007), are widely used to downscale GCM projections to regional or local scales. Since an RCM simulation is usually conducted at time intervals of minutes or seconds, the dynamical downscaling technique is capable of producing outputs at finer temporal resolutions, even at less than hourly time-scale, when compared to outputs from GCMs.

Since the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) in 2007, the typical horizontal grid resolution of RCMs has increased from 50 km to 25 km (Christensen et al. 2010; Maraun et al. 2010; Flato et al. 2013). Furthermore, RCM simulations with grid resolutions below 20 km have recently become available (e.g., Dankers et al. 2007; Hohenegger et al. 2008; Hollweg et al. 2008; Kanada et al. 2008; Kavvas et al. 2009; Tomassini & Jacob 2009; Früh et al. 2010; van Roosmalen et al. 2010; Chen et al. 2011; Ohara et al. 2011a; Kendon et al. 2012). However, due to computational restrictions, multi-decadal and centennial RCM applications at high grid resolutions are still scarce.

For instance, Gao et al. (2008) analyzed multi-decadal monsoon precipitation over China at a 20-km grid resolution during historical (1961–1990) and future climate (2071–2100) conditions forced by the IPCC A2 scenario. Wakazuki et al. (2008) evaluated the climatological reproducibility of extreme precipitation events around Japan with a 5-km resolution RCM, using boundary conditions from a 20-km resolution GCM, for 10 years in historical and future climates. Salathé et al. (2010) performed two 100-year simulations with an RCM forced by the CCSM3 A2 emission scenario and the ECHAM5 A1B emission scenario at 36- and 20-km resolutions over the State of Washington. Diffenbaugh et al. (2011) nested a 25-km resolution RCM within the CCSM3 model (Collins et al. 2006) to create a five-member ensemble simulation of the period from 1950 to 2099 in the SRES A1B scenario for the full continental United States. Chen et al. (2011) reconstructed the historical climate conditions during the historical critical period 1957–1969 over the 960,000 km² Tigris–Euphrates river basin at 15-km grid resolution by nesting the MM5 RCM into the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis data. Similarly, Ohara et al. (2011a) reconstructed the historical climate conditions over the Mekong river basin at 15-km grid resolution by nesting the MM5 RCM into the NCEP/NCAR reanalysis data. Kjellstrom et al. (2011) analyzed seasonal mean temperature, precipitation, and wind speed over Europe by an ensemble of 16 RCM simulations of 50-km horizontal resolution for the time period from 1961 to 2100 by using boundary conditions from seven GCMs under four emission scenarios.

Chan et al. (2014) and Kendon et al. (2012) carried out 1.5-km and 12-km RCM simulations to model precipitation in southern United Kingdom between the years 1990 and 2008, and investigated their performance. de Elía et al. (2013) analyzed the interannual climate variability through simulations of climate of the recent past (driven by reanalysis for 1980–2004) and climate change simulations (driven by GCMs, for 1971–2000 and 2041–2070, using the A2 emission scenario) over North America at a horizontal resolution of approximately 50 km. Gao et al. (2014) dynamically downscaled a single member of historical climate (1975–2004) and two scenarios of future climate (2005–2100) at 20-km resolution over North America and compared these simulations with a multi-model ensemble of global climate simulations to investigate the changes in water availability. However, Jang et al. (2017) have shown recently that in order to achieve realistic simulations of the regional climate over the mountainous terrain of Northern California, it is necessary to refine the dynamical downscaling grid resolutions to 9 km or less. Accordingly, recently Ishida et al. (2017) dynamically downscaled six future climate projection realizations over Northern California at 9-km spatial grid resolution and hourly intervals. However, the number of projections in the ensemble was still relatively small, and they focused on trends in annual volume and annual peak precipitation.

In climate change evaluation studies, it is important to assess the underlying uncertainty. Uncertainties in climate projections may arise from three sources: internal variability of the climate system, model uncertainty, and greenhouse emission scenario uncertainty (Hawkins & Sutton 2009). An accepted way of dealing with such uncertainties is to perform an ensemble of simulations (e.g., Hewitt 2004; Christensen et al. 2007; Déqué et al. 2007; Kjellstrom et al. 2011).

In this context, in order to investigate the impacts of climate change on watershed-scale precipitation over Northern California, 13 future climate projections from two GCMs, based on four greenhouse emission scenarios (SRES A1B, A1FI, A2, and B1; Nakicenovic & Swart 2000) were dynamically downscaled over eight watersheds in Northern California for a future period of 90 water years from 2010 to 2100 A relatively fine horizontal resolution (9 km) was utilized in these simulations. For watershed-scale analysis, eight Northern California watersheds were selected. These are: the Sacramento River watershed (SRW), the American River watershed (ARW), the Yuba River watershed (YRW), the Upper Feather River watershed (FRW), the Cache Creek watershed (CCW), the Shasta Dam watershed (SHA), the Trinity River watershed (TRW), and the Northern Central Valley (NCV).

To investigate climate change impacts on watershed-scale precipitation over Northern California, a trend analysis, which requires long duration data, was employed. Then, trends in annual basin-average precipitation, which are important for the water resources management of the eight study watersheds, were analyzed based on the ensemble of the 13 realizations. Furthermore, trends in seasonal basin-average precipitation were similarly examined due to the importance of such seasonal variability in water management.

The following section will describe the study area. In the third section, future climate projection realizations that were used in this study will be presented. Then, the implementation of an RCM over the study area for dynamical downscaling will be explained. The fourth section will show the results and the discussions for the performed trend analyses. Finally, concluding remarks will be provided in the last section.

In this study, the impact of future climate on precipitation was investigated over eight watersheds in Northern California, as depicted in Figure 1. Northern California is the primary water source for the state of California, with the Sacramento River as its major river having a drainage area of approximately 70,000 km². This drainage area, otherwise known as the SRW, receives two thirds of the state's precipitation on its own. The SRW contains a portion of the Central Valley, a huge valley region in California, which will be referred to as the NCV in this study. The Central Valley is surrounded by several mountain ranges: the Coastal Range from the west, the Sierra Nevada mountain range from the east, and the Klamath Mountains and the Cascade Range from the north. During winter, these mountain ranges receive plenty of precipitation due to a combination of the orographic effect and high-moisture atmospheric flows from the Pacific Ocean, which are referred to as an ‘Atmospheric River’ (Zhu & Newell 1994). The CCW is a representative hydrologic unit over the Coastal Range within the SRW. Meanwhile, the ARW and the YRW are located on the western slopes of the Sierra Nevada mountain range. The ARW shares its northern border with the YRW, and flows into the Sacramento River at Sacramento, the capital city of California. There are two major reservoirs in California: Lake Oroville and Shasta Lake. The drainage areas of Lake Oroville and Shasta Lake are referred to as the Upper FRW and the SHA, respectively. While these five watersheds are located within the SRW, another important watershed, known as the TRW, lies outside of the SRW. The TRW is the drainage area of the Trinity Dam whose reservoir water is diverted into the Central Valley. The drainage areas of the above eight study watersheds (including the NCV) are given in Table 1.

This study uses 13 different future climate projection realizations from two GCMs based on four greenhouse gas emission scenarios of the Special Report on Emissions Scenarios (SRES, Nakicenovic & Swart 2000). The two GCMs are the fifth-generation atmospheric global climate model (ECHAM5, Roeckner et al. 2003) by the German Max-Planck Institute (MPI) and the third-generation community climate system model (CCSM3, Collins et al. 2006) by the University Corporation for Atmospheric Research in the United States. Nine and four realizations are available from ECHAM5 and CCSM3, respectively, for dynamical downscaling, which requires at least 6-hourly temporal resolutions to reasonably reflect diurnal changes in atmospheric conditions to the simulations. The nine ECHAM5 realizations were obtained based on three SRES scenarios (A1B, A2, and B1), each with three different initial conditions resulting in the following realizations: A1B-1, A1B-2, A1B-3; A2-1, A2-2, A2-3; B1-1, B1-2, and B1-3. On the other hand, the four CCSM3 realizations were based on four SRES scenarios (A1B, A1FI, A2, and B1) with only one single initial condition. The SRES scenarios were created based on different assumptions of greenhouse gas emissions, population growth, land use, fossil fuel consumption, as well as economic and technological developments. Among the four scenarios, A1FI is considered to be the worst case scenario, A2 is slightly milder but similar to A1FI, A1B is the most likely scenario, while B1 is the best-case scenario. A complete description of these scenarios can be found in Nakicenovic & Swart (2000).

As previously mentioned, a reliable assessment of the regional impact of climate change on precipitation necessitates the downscaling of GCM results to finer spatial resolutions. One possible technique to achieve this is dynamical downscaling, which requires the use of an RCM. This study employs the fifth generation Penn State/NCAR Mesoscale Model (MM5, Grell et al. 1994) as the RCM used to dynamically downscale future climate projection realizations. The MM5 has been utilized by many studies to reconstruct the precipitation in the West Coast of the United States (Colle & Mass 2000; Mass et al. 2002; Garvert et al. 2005; Moss et al. 2010; Lin et al. 2013), where the study watersheds are located. Furthermore, other studies successfully validated and used the MM5 to analyze precipitation on mountainous regions in California (Grubišić et al. 2005; Reeves et al. 2008; Ohara et al. 2011b; Ishida et al. 2015b, 2015c). It should be noted that even though the Weather Research and Forecasting Model (WRF, Skamarock et al. 2005) is a newer RCM, the MM5 provides simulated precipitation results in Northern California that are comparable to those of the WRF model, while at the same time having the capability of running much faster than the WRF model, according to Ishida et al. (2015a). Hence, since this study dynamically downscales more than 1,000 years of future climate projection realization data to a 9-km grid resolution, computational speed is an important factor for such a large number of simulations.

Three two-way nested domains were set up for the MM5 dynamical downscaling, as shown in Figure 1(b). The outer domain (D1) has 24 × 22 horizontal grid points at an 81-km resolution, the second domain (D2) has 37 × 31 horizontal grid points at a 27-km resolution, and the innermost domain (D3) has 73 × 49 horizontal grid points at a 9-km resolution with 24 vertical sigma levels. The model configuration of the MM5 used in this study is shown in Table 2.

Before analyzing trends in the watershed-scale precipitation under future climate change conditions, model bias was corrected by comparing dynamically-downscaled precipitation data during a historical period with the corresponding observation data. The historical simulated data were obtained from the control runs of ECHAM5 and CCSM3 which were dynamically downscaled to a 9-km grid resolution for 50 water years (October 1949 through September 1999) by means of the MM5. As for the observation data, monthly precipitation data of the Parameter-elevation Relationships on Independent Slopes Model (PRISM; http://prism.oregonstate.edu, Daly et al. 1994; Daly et al. 2008) were utilized for the same historical period above. PRISM data are derived from interpolating gauge-observed precipitation data by means of a moving-window regression function based on a linear relationship between precipitation and elevation. To correct the biases in the dynamically-downscaled precipitation, first, the PRISM monthly precipitation at 4-km resolution was interpolated onto the same 9-km grids of the MM5 model. Then, the mean monthly values over the determined 50-year historical period were calculated based on the interpolated PRISM precipitation and the dynamically-downscaled precipitation data for ECHAM5 and CCSM3. After comparing the dynamically-downscaled and the PRISM mean monthly precipitation data, bias correction factors for ECHAM5 and CCSM3 were obtained at the 9-km resolution grids of the MM5's inner domain. In this study, the bias correction factors ranged between 0.2 and 2.0. Finally, the bias correction factors were applied to the dynamically-downscaled precipitation for each future climate projection realization based on an assumption that the bias correction factor will be consistent throughout the future period.

Since trends in peak precipitation are analyzed in this study, watershed-scale dynamically-downscaled precipitation data were evaluated by comparing them to PRISM daily precipitation data. While the control runs of ECHAM5 and CCSM3 are available from 1900 to 1999, PRISM daily precipitation data are only available starting from 1981 to the present. Therefore, it was only possible to compare the observed and downscaled watershed-scale daily precipitation for the 18 overlapping water years (October 1, 1981 through September 30, 1999). Table 3 shows the mean and standard deviation of the observed and downscaled daily basin-average precipitation over the eight study watersheds, while Figure 2 plots their corresponding cumulative distribution functions, also over each of the study watersheds. The two-sample Kolmogorov–Smirnov nonparametric test was used to evaluate the differences between the cumulative distribution functions of the observed and the downscaled daily basin-average precipitation. The resulting p-values in Figure 2 show that the downscaled daily basin-average precipitation from both ECHAM5 and CCSM3 passed the Kolmogorov–Smirnov test at the 99% confidence level over all eight study watersheds.

This study focused on future climate change impacts on the annual and seasonal accumulated basin-average precipitation over the eight study watersheds. Since the water year in California starts in October, the water year is divided into the following four seasons: OND months (October, November, and December), JFM months (January, February, and March), AMJ months (April, May, and June), and JAS months (July, August, and September). Analyses in this study were done during the future period of 90 water years (2010 to 2100) based on the dynamically downscaled bias-corrected future climate projection realizations.

Trends in basin-average precipitation were investigated based on several ensemble averages over future climate projection realizations. The first ensemble average was based on all 13 realizations. It includes both of the GCMs (ECHAM5 and CCSM3) and all of the four scenarios (A1B, A1FI, A2, and B1). The ensemble average over all the realizations shows the general trend of basin-average precipitation during the 21st century. Secondly, the ensemble averages over the nine ECHAM5 realizations and the four CCSM3 realizations were separately investigated to consider GCM model uncertainties. Finally, the ensemble averages over the four A1B realizations, the four A2 realizations, the four B1 realizations, and the A1FI realization were utilized to reveal differences in precipitation trends among the emission scenarios for the future period.

For the trend analyses, the least-squares regression method was utilized to calculate the slopes and standard errors of the trend lines. Moreover, the significance in the trends was investigated by means of the Mann–Kendall trend test (Mann 1945; Kendall 1975), which is a non-parametric statistical test on the significance of time series trends.

Annual accumulated basin-average precipitation over the eight study watersheds based on each of the 13 future climate projection realizations over the 90-year future period (2010–2100) is illustrated in Figure 3. The annual basin-average precipitation fluctuates over all the eight watersheds in all of the 13 realizations due to the internal variability. Higher peaks of the annual basin-average precipitation can be found over the SRW, the FRW, and the SHA during the late 21st century compared to the early 21st century. A single large peak of the annual basin-average precipitation is detected over the ARW during the late 21st century and over the NCV during the middle of the 21st century, respectively. Nevertheless, the ensemble average of the annual basin-average precipitation over the 13 realizations does not change through the 21st century as illustrated in Figure 3. Table 4 shows that the slope of the trend line of the annual basin-average precipitation is negative over all the watersheds except the NCV, although the downward trend is not significant.

Figure 4 shows the ensemble averages, along with the 10-year moving averages, of the annual basin-average precipitation over the nine ECHAM5 realizations and the four CCSM3 realizations for the eight study watersheds during the 21st century. Generally, ECHAM5 projected higher annual basin-average precipitation over the study watersheds. Only over the TRW, the annual precipitation result from CCSM3 is higher than that from ECHAM5 during the early 21st century, but both results become similar during the late 21st century. The 10-year moving average of the annual basin-average precipitation from ECHAM5 remains unchanged until the middle of the 2060s, and then gradually increases through a decade over the study watersheds, except the FRW. On the other hand, the 10-year moving average of the annual basin-average precipitation from CCSM3 shows a dry period around the year 2050. Based on the trend analysis by the least-squares regression method, the annual basin-average precipitation from ECHAM5 increases during the 21st century over the study watersheds (except the FRW), while that from CCSM3 decreases through the 21st century over all the study watersheds (Table 4). Thus, the slopes of the trend lines of the annual basin-average precipitation are opposite between ECHAM5 and CCSM3, except over the FRW. However, it should be noted that the Mann–Kendall trend test detected no trend in annual precipitation projections from both of the GCMs.

The ensemble averages of the annual basin-average precipitation over the eight study watersheds over four A1B realizations, four A2 realizations, four B1 realizations, and one A1FI realization are illustrated with their 10-year moving averages in Figure 5. The ensemble averages of the annual basin-average precipitation fluctuate over all the study watersheds. However, there are no clear differences in the 10-year moving averages among the scenarios except A1FI. As shown in Table 4, the annual basin-average precipitation over the SRW, the CCW, and the NCV increases in A1B and A2, but decreases in B1. Conversely, the annual precipitation over the SHA decreases in A2, but increases in B1. The annual precipitation over the ARW and the YRW decreases in A1B, but increases in B1. It may be noted that A2 is the second most severe scenario, and B1 is the mildest scenario among the four emission scenarios. However, changes in the ensemble averages of the annual precipitation over each of the scenarios during the 21st century do not depend on the severity of the scenarios. Furthermore, the results of the Mann–Kendall trend test show no trend in the annual basin-average precipitation over each of the watersheds in A1B, A2, and B1. On the other hand, the annual basin-average precipitation shows a significant downward trend over the study watersheds (except the TRW) in A1FI. A1FI is the most severe emission scenario with respect to global warming, which might be the reason for its significant downward trend in the annual precipitation. However, it should be noted that this study uses a single realization of A1FI, only from CCSM3. Thus, it might also be inappropriate to conclude that the annual precipitation will significantly decrease under A1FI scenario.

Consequently, no significant trends in the ensemble averages of the annual basin-average precipitation were detected over the study watersheds except in a single realization of the A1FI scenario from CCSM3. These results indicate that there may be no significant change in annual water resources in Northern California through the 21st century.

Trend analysis of accumulated basin-average precipitation during the four divided seasons (OND, JFM, AMJ, and JAS) over the eight study watersheds was conducted based on the 13 future climate projection realizations over the future period of 90 water years (2010–2100). Figure 6 plots the accumulated basin-average precipitation over the eight study watersheds for the OND months based on each of the 13 future climate projection realizations with their ensemble average and the 95% confidence band. The highest peaks of the accumulated basin-average precipitation for the OND months can be found during the middle 21st century from 2050 to 2070 over all of the eight study watersheds. At least two large peaks of the basin-average precipitation are detected over each study watershed. Compared to the early and the middle 21st century, there are less peaks during the late 21st century over the study watersheds except for TRW and NCV. Consequently, a downward trend at the 95% confidence level is detected over those six watersheds while there is no trend over TRW and NCV, as tabulated in Table 5. The results of the trend analysis based on the other ensemble averages can also be found in Table 5. While no trend was detected for CCSM3, the basin-average precipitation for the OND months from ECHAM5 shows a significant downward trend over YRW, FRW, and SHA, which are located in the northeast part of California. A1B scenario projects that the downward trend in the basin-average precipitation for the OND months will be significant over five study watersheds (SRW, ARW, YRW, FRW, and SHA). In A1FI scenario, the downward trend is significant over all study watersheds. On the other hand, there is no trend in the basin-average precipitation for the OND months for A2 and B1 scenarios.

For the JFM months, the accumulated basin-average precipitation over the eight study watersheds based on each of the 13 future climate projection realizations is illustrated in Figure 7. ARW and YRW have a single large peak of the accumulated basin-average precipitation during the late 21st century while the other watersheds have several peaks. The ensemble average over the 13 future climate projection realizations fluctuates throughout the future period, but the fluctuation is within a certain range over each study watershed, resulting in no trend in basin-average precipitation over all study watersheds for the JFM months (Table 6). Furthermore, no trend is found in the basin-average precipitation of ECHAM5 and CCSM3, and for all future climate scenarios.

As for the AMJ months, the accumulated basin-average precipitation over the eight study watersheds based on each of the 13 future climate projection realizations is illustrated in Figure 8. Over most study watersheds, there are lower numbers of higher peaks of the basin-average precipitation for the AMJ months during the middle 21st century. Although a single large peak can be found over FRW during the middle 21st century, it can be found over the CCW and NCV during the late 21st century. As shown in Table 7, a significant downward trend is detected in the basin-average precipitation for the AMJ months over FRW, CCW, SHA, and NCV. Similar to the OND months, there is a significant downward trend in the basin-average precipitation for ECHAM5 for the OND months over ARW, YRW, FRW, and SHA while there is no significant trend over all study watersheds for CCSM3. Among the future climate scenarios, only A2 projects a significant downward trend over the study watersheds except TRW.

Finally, Figure 9 illustrates the accumulated basin-average precipitation over the eight study watersheds for the JAS months based on each of the 13 future climate projection realizations. This figure shows that there is a lower number of higher peaks of the accumulated basin-average precipitation for the JAS months compared to the other seasons. As tabulated in Table 8, the results of the Mann–Kendall trend test indicate that increase in the accumulated basin-average precipitation for the JAS months is significant over all the eight study watersheds unlike the other seasons. Opposing trends are detected in the basin-average precipitation for the JAS months for ECHAM5 and CCSM3. The basin-average precipitation of ECHAM5 shows a significant upward trend over all the study watersheds during the 21st century. In contrast, that of CCSM3 has a significant downward trend over the study watersheds except CCW. A1B and A2 project a significant increase in the basin-average precipitation for the JAS months over SRW, CCW, SHA, and TRW, and over SRW, FRW, CCW, SHA, TRW, and NCV, respectively. On the other hand, A1FI shows a significant downward trend in the basin-average precipitation over all of the eight study watersheds for the JAS months.

As described above, there are differences in the trends of the accumulated basin-average precipitation among the different seasons. In fact, there is no significant trend during the winter months (JFM), some significant downward trends during the spring (AMJ) and fall (OND) months, and several significant upward trends for the summer months (JAS), all being considered with respect to the ensemble average of the basin-average precipitation over all 13 future climate projection realizations. Furthermore, when looking at the specific scenarios, A1B scenario projects some significant downward trends in the seasonal basin-average precipitation for the OND months, and some significant upward trends for the JAS months. The trend analysis of the seasonal basin-average precipitation from the A1FI scenario indicates that basin-average precipitation over most of the study watersheds will likely decrease in the future for all seasons, except for the AMJ months. For the A2 scenario, no significant trend in basin-average precipitation is detected during colder seasons (OND and JFM), but a significant downward trend is detected over some study watersheds during warmer seasons (AMJ and JAS). As for the B1 scenario, it projects no trend over all the study watersheds and for all seasons.

Thirteen different future climate projection realizations from two GCMs based on four SRES scenarios were dynamically downscaled to 9-km resolution grids over Northern California for 90 water years, from 2010 to 2100, by means of the MM5 regional climate model. Basin-average precipitation was calculated for historical and future conditions from the downscaled results at the 9-km resolution grids for the eight study watersheds: the SRW, the ARW, the YRW, the FRW, the CCW, the SHA, the TRW, and the NCV. Analysis of the mean, standard deviation and cumulative distribution function behavior of the daily precipitation over the eight watersheds showed that precipitation values obtained from the dynamical downscaling of the control runs of ECHAM5 and CCSM3 models, from 1981 to 1999, compared well with the PRISM data. Then, long-term future trends of annual and seasonal basin-average precipitation were investigated over the eight watersheds in Northern California by thirteen climate projections from 2010 to 2100.

The trend analyses of annual basin-average precipitation projections showed that there was no significant trend in the annual basin-average precipitation over the eight study watersheds except for the A1FI scenario. A significant downward trend was detected over the study watersheds for the A1FI scenario, except over TRW. Moreover, ECHAM5 generally projected higher annual basin-average precipitation over the study watersheds compared to CCSM3, although there were no significant trends in the ensemble average of the annual basin-average precipitation for both the ECHAM5 and CCSM3 realizations over all of the study watersheds.

Unlike the trend analysis of the annual basin-average precipitation, some significant upward and downward trends were detected in the seasonal basin-average precipitation results. For example, a downward trend in the basin-average precipitation for the OND months and for the AMJ months was significant at the 95% confidence level over six study watersheds (SRW, ARW, YRW, FRW, CCW, and SHA) and over five study watersheds (SRW, FRW, CCW, SHA, NCV), respectively, for the ensemble average of 13 realizations. Moreover, all future climate scenarios, except the B1 scenario, projected some significant trends in the seasonal basin-average precipitation for one or more seasons. In addition, differences in trends in the accumulated basin-average precipitation were detected among seasons as discussed in the previous section.

Therefore, the results provided in this study provide several important insights about the future climate change impacts on precipitation over the eight study watersheds in Northern California. In fact, the projected trends in the seasonal basin-average precipitation were found to be different among each of the study watersheds. This implies that it is important to assess the impacts of future climate change for each watershed separately, even though the study watersheds may be located close to each other. Furthermore, it was clear that the analyses of the seasonal and annual precipitation can both be equally important. Even if no significant trend can be detected in annual precipitation, a significant trend may be found in the corresponding seasonal precipitation. Finally, the results revealed the large variability in the precipitation obtained from the different GCMs and scenarios. In fact, this study showed how different GCMs and different scenarios project differing precipitation trends, thus confirming the importance of taking into account the uncertainties in such projections.