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.