Future climate data:
A set of benchmark emission scenarios referred to Representative Concentration Pathways or RCPs (Moss, Babiker et al. 2008) are possible development trajectories for the main climate change drivers (Van Vuuren, Edmonds et al. 2011). Research on the multi-gas emission scenarios were the base of the RCPs development (Fujino, Nair et al. 2006; Smith and Wigley 2006; Clarke, Edmonds et al. 2007; Riahi, Grübler et al. 2007; Van Vuuren, Den Elzen et al. 2007; Wise, Calvin et al. 2009). They collectively encompass (extending to year 2100) radiative forcing values from 2.6 to 8.5 W/m2 relative to year 1750 (59, 27). These scenarios are as follows: RCP2.6, RCP4.5, RCP6, and RCP8.5. RCP 2.6 is a mitigation scenario and its goal is to keep the global mean temperature rise under 2℃ (Moss, Babiker et al. 2008; Moss, Edmonds et al. 2010; Van Vuuren, Stehfest et al. 2011). The radiative forcing for RCP2.6 increases up to around 3 Watts per square meter (W/m2) before 2100 and then declines (Meinshausen, Smith et al. 2011; Van Vuuren, Stehfest et al. 2011). Under RCP4.5 and RCP6, concentration of GHGs are stabilized (without overshoot) after 2100 (Moss, Edmonds et al. 2010). RCP4.5 stops increasing radiative forcing at 4.5 W/m2 by year 2100 and the forcing becomes constant afterward (Thomson, Calvin et al. 2011). RCP6 pathway controls the increasing radiative forcing at 6 W/m2without exceeding the value afterward (Masui, Matsumoto et al. 2011). GHGs emission increase by around 2060 and then decline till 2100 (64). RCP8.5 assumes high population and slow economic growth which leads to increasing GHGs emissions resulting in radiative forcing as high as 8.5 W/m2 by end the 21st century and it is assumed to rise afterward (Riahi, Rao et al. 2011). Additional actions are required to halt continuously rising level of GHG concentrations which are due to the growth of global population and economic activities (Pachauri, Allen et al. 2014). These actions are dependent upon the political and socio-economic conditions on the global scale (Van Vuuren, Den Elzen et al. 2007; Van Vuuren, Stehfest et al. 2011). With taking the current global political condition and its possible future pathway into account, we selected the RCP 4.5 and RCP6.0 as moderate and severe pathways, respectively.
The General Climate Models (GCMs) use these RCPs to produce future climate data. The main source of climate projections is the modeling results of the Coupled Model Intercomparison Projects (CMIP3 & CMIP5) (Sunde, He et al. 2017). Since GCMs’ horizontal resolution is low, it is difficult to derive regional scale climate information from them (Flato, Marotzke et al. 2014). In general, GCM results are not reliable for models with resolution less than 200 km (Meehl, Stocker et al. 2007). Hydrological processes occur on a scale (in order of 10km) at which GCMs (resolution of 1o to 2.5olatitude-longitude) cannot provide reliable results (Kundzewicz, Mata et al. 2007; Pierce and Cayan 2016). Moreover, GCMs are not able to capture frequency and magnitude of extreme events (Christensen and Christensen 2007; Fowler, Blenkinsop et al. 2007). Therefore, for important climate variables like precipitation and temperature it is necessary to use higher resolution. Downscaling techniques have been used rigorously to produce climate variables from GCMs on the desired scale for hydrological modeling of climate change impact studies (Maraun, Wetterhall et al. 2010; Fu, Charles et al. 2013; Sunde, He et al. 2017). Between two types of the existing downscaling techniques which are dynamical and statistical, we used the statistical downscaling method. Statistical method downscales GCMs’ output based on the historical relationship between large- and small-scale conditions (Pierce, Cayan et al. 2014). In this study we used a statistical downscaling called Localized Constructed Analogs (LOCA). LOCA chooses analog days from observed data and applies a multiscale spatial matching scheme to estimate suitable downscaled climate variables (Pierce, Cayan et al. 2014). More realistic regional patterns of precipitation, better estimates of extreme events, and reduced number of light-precipitation days are the advantages of LOCA (Pierce, Cayan et al. 2014). More information on LOCA can be found here: http://loca.ucsd.edu/ , (Ficklin and Barnhart 2014).
Considering the complexity of the GCMs, CMIP5 outputs are inevitably biased (Teutschbein and Seibert 2010; Taylor, Stouffer et al. 2012). Bias correction (BC) is the process of transforming GCM outputs using algorithms in order to adjust the outputs (Teutschbein and Seibert 2010; Chen, Marek et al. 2019). Basically, biases are detected by comparing the observation and simulation results and then they are used to correct baseline and projections (Teutschbein and Seibert 2010; Chen, Marek et al. 2019). Bias-corrected inputs for hydrological modeling improve the result, hence bias correction is needed for GCMs output (Wilby, Hay et al. 2000; Pierce, Cayan et al. 2015). LOCA as a downscaling technique improved based on constructed analogs (CA) process contains a bias correction step (Hidalgo León, Dettinger et al. 2008; Pierce, Cayan et al. 2014). The BC in LOCA includes 3 steps. First, a preconditioning technique is used to correct the annual cycle and then two different distribution techniques are used, one for temperature and one for precipitation , and finally a frequency-dependent bias correction (FDBC) is used to adjust the sequencing of variation for different time scale, since the sequencing for GCM outputs potentially differ from observed ones (Li, Sheffield et al. 2010; Pierce, Cayan et al. 2015). We obtained and analyzed CMIP5 output the LOCA dataset for three models, CCSM4,GISS-E2-R, and GFDL-CM3, under RCP4.5 and RCP6.0 from Downscaled CMIP5 Climate and Hydrology Projections (https://gdo-dcp.ucllnl.org/) (Schmidt, Ruedy et al. 2006; Donner, Wyman et al. 2011; Gent, Danabasoglu et al. 2011; Taylor, Stouffer et al. 2012; Bureau of Reclamation 2013). The downloaded data are bias-corrected 1/16th degree latitude-longitude (~6km ×6km) daily precipitation (mm/day), and maximum and minimum temperature (°C) projections. Hereafter the downloaded dataset, which is downscaled, and bias corrected by LOCA, is referred as “the CMIP5 multi-model ensemble LOCA”. The LOCA dataset contains future projections under RCP4.5 and RCP6.0 for 32 GCMs for the conterminous US from 1950 to 2099.
In hydrological projection process using GCMs, their initial condition, future scenarios, and hydrological model incorporate uncertainties to the result (Chen, Brissette et al. 2011). Ouyang et al. (2015) have concluded that different result of the future projections are partially due to the different climate models (Ouyang, Zhu et al. 2015). Considering the numerous numbers of the GCMs and the variability they could cover based on the model skill and independency, we selected the three models to be able to analyze broad extents of changing climate variables within the UCS; in this way we were able to address the uncertainty (Sanderson, Knutti et al. 2015; Sanderson, Knutti et al. 2015; Sunde, He et al. 2017). Locating and Selecting Scenarios Online (LASSO) tool from Environmental Protection Agency (EPA) ( https://lasso.epa.gov/) was used to filter out the selected model from 32 GCMs. Through the different steps of the tool, we have examined climate parameters variabilities with two time periods (annual and seasonal) and selection strategies to reach the goal of three representative models. Models’ name, their associated institution, type of experiment, and ensemble member are shown in Table 4. For the ensemble member names, the number after r is the ‘realization’ number and is used to identify the initial condition. The number after i is ‘initialization method indicator’ and the number after p refers to model versions with the same perturbed physics (Taylor, Balaji et al. 2011). The listed ensemble members are different only in their initial condition.