Modeled PET for the southwest United States has been calibrated to measured PET from California Irrigation Management Information System and Arizona Meteorological Network stations. Using PET and gridded precipitation, maximum and minimum air temperature, and the approach of the National Weather Service Snow‐17 model , snow is accumulated, sublimated, and melted to produce available water . These driving forces for the water balance have been calibrated regionally to solar radiation and PET data, and snow cover estimates have been compared to Moderate Resolution Imaging Spectroradiometer snow cover maps . However, the final calibrations of snowmelt and runoff have illustrated goodness‐of‐fit, as will be shown in the results.Available water occupies the soil profile, where water will become actual evapotranspiration, and may result in runoff or recharge, depending on the permeability of the underlying bedrock. Total soil‐water storage is calculated as porosity multiplied by soil depth. Field capacity is the soil water volume below which drainage is negligible, and wilting point is the soil water volume below which actual evapotranspiration does not occur . Once available water is calculated, it may exceed total soil storage and become runoff, or it may be less than total soil storage but greater than field capacity and become recharge. Anything less than field capacity will be calculated as actual evapotranspiration at the rate of PET for that month until it reaches wilting point. When soil water is less than total soil storage and greater than field capacity,black plastic planting pots soil water greater than field capacity equals recharge. If recharge is greater than bedrock permeability , then recharge = K and excess becomes runoff, else it will recharge at K until field capacity.

Runoff and recharge combine to calculate basin discharge, and actual evapotranspiration is subtracted from PET to calculate climate water deficit.The BCM can be used to identify locations and climatic conditions that generate excess water by quantifying the amount of water available either as runoff generated throughout a basin, or as in‐place recharge . Because of the grid‐based, simplified nature of the model, with no routing of runoff to downstream cells, long time series for very large areas can be simulated easily. However, if local unimpaired stream flow is available, estimated recharge and runoff for each grid cell can be used to calculate basin discharge that can be extrapolated through time for varying climates. In addition, the application of the model across landscapes allows for grid‐based comparisons between different areas. Because of the modular and mechanistic approach used by the BCM, it is flexible with respect to incorporating new input data or updating of algorithms should better calculations be derived. A flow chart indicating all input files necessary to operate the BCM, and the output files resulting from the simulations, is shown in Appendix A. After running the BCM, the 14 climate and hydrologic variables were produced in raster format for every month of every year modeled . To evaluate hydrologic response to climate for all basins in hydrologic California, we used the BCM to calculate hydrologic conditions across the landscape for 1971–2000 and to project them for the two GCMs and two emission scenarios for 2001–2100. Trends in climate, hydrologic derivatives of runoff and recharge, and climatic water deficit are separately analyzed for both historical‐to‐baseline, and baseline‐to‐future time periods . Although recharge and runoff were calculated for every grid‐cell and summarized as totals for basins, the estimate of basin discharge as a time series requires a further calculation of stream flow. Calculation of stream flow uses a series of equations that can be calibrated with coefficients from existing streamgage data, that then permit estimation of basin discharge for time periods when there are no stream flow measurements. We calculated basin discharge for each of 138 basins for which we also obtained streamgage data, and used the 138 streamgage datasets for calibration and validation.

The regional BCM developed for the southwest United States was applied to California following regional calibrations for solar radiation, PET, snow cover, , and groundwater . The California calibration is based on study areas with ongoing studies that were designed to provide runoff and recharge for historic, baseline, and future climatic conditions. Generally the watersheds used for calibration basins were identified on the basis of lack of impairments, such as urbanization, agriculture, reservoirs, or diversions, although this was not always possible.We used 68 basins for which bedrock permeability was iteratively changed to optimize the match between calculated basin discharge and measured stream flow. Calibration basins represent 9 of the 14 dominant geologic types in California, and have been calibrated to bedrock permeability on the basis of mapped geology for California . The BCM performs no routing of stream flow, which is done as post‐processing to produce total basin discharge for any basin outlet or pour point of interest, such as streamgages or reservoirs. The 68 calibration basins were calibrated to optimize the match between BCM‐derived discharge and stream flow by iteratively adjusting the bedrock permeability corresponding to the geologic types located within the basins to alter the proportion of excess water that becomes recharge or runoff. This part of the calibration process is followed by accounting for stream channel gains and losses to calculate basin discharge, optimize the fit between total measured volume and simulated volume for the period of record for each gage, and maintain a mass balance among stream flow and BCM recharge and runoff.For comparison to the calibration basins, and to evaluate model performance representing the state, additional validation basins were identified for the calculation of discharge on the basis of general lack of impairments, as well as statewide coverage of landscapes and geology. Hydrologic results for these basins were developed on the basis of the calibration to bedrock permeability performed using the calibration basins. The calibrations and validation basins are distributed across the range of elevation, aridity , and bedrock permeability, in comparison to all basins in California , and we also show the relationship between them for the same three environmental conditions . Study basins generally cover the range of elevations for the state .

Bedrock permeability as a representation of geology is dominated by lower permeability basins because very high permeability basins, such as those with alluvial valley fill, do not generate stream flow .The range of climates in the state, represented by the UNESCO Arid Zone Research program aridity categories , is covered less well by the study basins and neglects the hyper‐arid and arid locations due to lack of stream flow data . The representation of study basins within the ecoregions in the state also reflects the lack of streamgage data in the desert areas, as well as in the eastern side of the Sierra Nevada, and in the deep soils of the Central Valley , where any gaged streams are very impaired.Calibration statistics are shown in Appendix C and spatially in Figure 6, with the linear regression r2 for monthly and yearly comparison of measured and simulated basin discharge, and the Nash‐Sutcliffe efficiency statistic calculated as 1 minus the ratio of the mean square error to the variance. The NSS is widely used to evaluate the performance of hydrologic models, generally being sensitive to differences in the observed and modeled simulated means and variances, but is overly sensitive to extreme values, similarly to r2 . The NSS ranges from negative infinity to 1, with higher values indicating better agreement. Average calibration statistics for all basins are NSS = 0.65, monthly r2 = 0.70, and yearly r2 = 0.86.In our study, calibration basins have a mean NSS of 0.71 , with the higher values for the Russian River basin, just north of the San Francisco Bay Area,drainage planter pot and lower values for the Santa Cruz basins, just south of the Bay Area, where there are many urban impacts . There are several cases where urbanization and agriculture were identified as factors resulting in the inability to calculate a mass balance. The measured stream flow at Aptos Creek at Aptos had very high peaks that were not reproduced by the BCM. This basin is dominated by urbanization, suggesting that the high peak flows were a result of urban landscapes enhancing runoff, both during precipitation events where there is reduced infiltration and during the summer when urban runoff is enhanced—neither of which is taken into account in the BCM. In order to match measured volumes and stream flow patterns, the runoff is reduced by 80 percent, and the recharge is reduced by 50 percent. An example of diversions and groundwater pumping for public use can be seen in the difference between the Merced River at Happy Isles, upstream of Yosemite Village, and the Merced River at Pohono, downstream of Yosemite Village, where the percentage of runoff is reduced to 45 percent to match measured flows .

The basin discharge for the validation basins, not used for calibration, was developed using the adjusted bedrock permeability values developed during calibration. The mean NSS for these basins is 0.61 , with the upper Klamath and small basins in the Modoc Plateau volcanics performing the poorest . This is likely due to the large groundwater reservoir in the volcanics that has very long travel times from precipitation input to outflow in streams. An example of a calibration in the volcanics for the Sprague River basin illustrates the large base flow component with high base flow exponent . The Sprague River basin also has a large agricultural component and return flows, so the attempt to maintain a match in volumes results in an overestimate of the peak flows. The presence of a groundwater reservoir also shows in the differences between the r2 values for the monthly and yearly values , which identifies lags in the monthly calibration between measured and simulated discharge that are negated when calculated yearly. There is a large difference for the Kings River above the North Fork near Trimmer, for example, indicating the potential for a lag in groundwater flows becoming base flows that appear at the base of the basin and not being accounted for in a monthly model; whereas, the yearly r2 is very high. The basins in the volcanics consistently show a larger range in the two r2 values, which is also illustrated in the Sprague River near the Beatty, Oregon, calibration by the mismatch in the timing of the peaks. For California, we produced 270 m grids to represent historic and future climates from 1900 to 2100, resulting in 6,594,862 grid cells statewide, and a map for each of the 14 variables for each month. For the historic data and four future scenarios, this produced over 11 terabytes of data. We then created water year summaries of the 14 variables. The water year starts in October and ends in September. For the two temperature variables we averaged the temperature over the water year, and for the other 12 variables we summed all data for 12 months Since retaining yearly values for this region results in unwieldy large files, we reduced the data size for distribution and analysis to 30‐year summaries, providing monthly average values for variables historically for 1911–1940, 1941–1970, and 1971–2000. Future climate values are based on 100‐year simulations, with 2010–2039, 2040–2069, and 2070–2099 time slices produced. We also developed summaries for 10‐year periods based on time slices starting with 1911–1920 and running through 2090‐2099. Appendix D has a list of all available variables, file size, format, and acronym. We wrote a program to summarize the 30‐year datasets by various statistical measures, to create a manageable dataset for analysis of long‐term trends. We calculated these statistics for both annual and monthly average values. Statistics were developed for each 30‐year time period by applying a linear regression model to the input rasters, which produced the seven statistics for each variable for each 30‐year time period. The linear regression model used equations from Zar . Change over the historical baseline period 1971–2000 was described as the slope of the regression model multiplied by 30 years. We characterized the variables calculated by the BCM for watersheds and for ecoregions, and compared historical summaries and patterns to future projections.