What is the importance of climate model bias when projecting the impacts of climate change on land surface processes?

Abstract

Regional climate change impact (CCI) studies have widely involved
downscaling and bias-correcting (BC) Global Climate Model
(GCM)-projected climate for driving land surface models. However, BC may
cause uncertainties in projecting hydrologic and biogeochemical
responses to future climate due to the impaired spatiotemporal
covariance of climate variables and a breakdown of physical conservation
principles. Here we quantify the impact of BC on simulated
climate-driven changes in water variables (evapotranspiration, ET;
runoff; snow water equivalent, SWE; and water demand for irrigation),
crop yield, biogenic volatile organic compounds (BVOC), nitric oxide
(NO) emissions, and dissolved inorganic nitrogen (DIN) export over the
Pacific Northwest (PNW) Region. We also quantify the impacts on net
primary production (NPP) over a small watershed in the region (HJ
Andrews). Simulation results from the coupled ECHAM5/MPI-OM model with
A1B emission scenario were firstly dynamically downscaled to 12 km
resolutions with WRF model. Then a quantile mapping based statistical
downscaling model was used to downscale them into 1/16th degree
resolution daily climate data over historical and future periods. Two
series climate data were generated according to the option of
bias-correction (i.e. with bias-correction (BC) and without
bias-correction, NBC). Impact models were then applied to estimate
hydrologic and biogeochemical responses to both BC and NBC
meteorological datasets. These impact models include a macro-scale
hydrologic model (VIC), a coupled cropping system model (VIC-CropSyst),
an ecohydrologic model (RHESSys), a biogenic emissions model (MEGAN),
and a nutrient export model (Global-NEWS). Results
demonstrate that the BC and NBC climate data provide consistent
estimates of the climate-driven changes in water fluxes (ET, runoff, and
water demand), VOCs (isoprene and monoterpenes) and NO emissions, mean
crop yield, and river DIN export over the PNW domain. However,
significant differences rise from projected SWE, crop yield from dry
lands, and HJ Andrews's ET between BC and NBC data. Even though BC
post-processing has no significant impacts on most of the studied
variables when taking PNW as a whole, their effects have large spatial
variations and some local areas are substantially influenced. In
addition, there are months during which BC and NBC post-processing
produces significant differences in projected changes, such as summer
runoff. Factor-controlled simulations indicate that BC post-processing
of precipitation and temperature both substantially contribute to these
differences at region scales. We conclude that there are
trade-offs between using BC climate data for offline CCI studies vs.
direct modeled climate data. These trade-offs should be considered when
designing integrated modeling frameworks for specific applications;
e.g., BC may be more important when considering impacts on reservoir
operations in mountainous watersheds than when investigating impacts on
biogenic emissions and air quality (where VOCs are a primary indicator).

Full text

Biogeosciences, 11, 2601–2622, 2014
www.biogeosciences.net/11/2601/2014/
doi:10.5194/bg-11-2601-2014
© Author(s) 2014. CC Attribution 3.0 License.
Biogeosciences
Open Access
What is the importance of climate model bias when projecting the
impacts of climate change on land surface processes?
M. Liu1, K. Rajagopalan1, S. H. Chung1, X. Jiang2, J. Harrison3, T. Nergui1, A. Guenther1,4, C. Miller3, J. Reyes1,
C. Tague5, J. Choate5, E. P. Salathé6, C. O. Stöckle7, and J. C. Adam1
1Civil and Environ Engineering, Washington State University, Pullman, WA, USA
2Atmospheric Chemistry Division, NCAR Earth System Laboratory, Boulder, CO, USA
3School of the Environment, Washington State University, Vancouver, WA, USA
4Atmospheric Sciences and Global Change, Pacific Northwest National Laboratory, Richland WA, USA
5Bren School of Environmental Science & Management, University of California, Santa Barbara, CA, USA
6School of Science Technology Engineering and Mathematics, University of Washington, Bothell, WA, USA
7Department of Biological Systems Engineering, Washington State University, Pullman, WA, USA
Correspondence to: J. C. Adam (jcadam@wsu.edu)
Received: 22 August 2013 – Published in Biogeosciences Discuss.: 4 November 2013
Revised: 14 March 2014 – Accepted: 28 March 2014 – Published: 16 May 2014
Abstract. Regional climate change impact (CCI) studies
have widely involved downscaling and bias correcting (BC)
global climate model (GCM)-projected climate for driving
land surface models. However, BC may cause uncertainties
in projecting hydrologic and biogeochemical responses to fu-
ture climate due to the impaired spatiotemporal covariance
of climate variables and a breakdown of physical conserva-
tion principles. Here we quantify the impact of BC on simu-
lated climate-driven changes in water variables (evapotran-
spiration (ET), runoff, snow water equivalent (SWE), and
water demand for irrigation), crop yield, biogenic volatile
organic compounds (BVOC), nitric oxide (NO) emissions,
and dissolved inorganic nitrogen (DIN) export over the Pa-
cific Northwest (PNW) region. We also quantify the impacts
on net primary production (NPP) over a small watershed in
the region (HJ-Andrews). Simulation results from the cou-
pled ECHAM5–MPI-OM model with A1B emission sce-
nario were first dynamically downscaled to 12km resolution
with the WRF model. Then a quantile-mapping-based statis-
tical downscaling model was used to downscale them into
1/16◦resolution daily climate data over historical and future
periods. Two climate data series were generated, with bias
correction (BC) and without bias correction (NBC). Impact
models were then applied to estimate hydrologic and bio-
geochemical responses to both BC and NBC meteorological
data sets. These impact models include a macroscale hydro-
logic model (VIC), a coupled cropping system model (VIC-
CropSyst), an ecohydrological model (RHESSys), a biogenic
emissions model (MEGAN), and a nutrient export model
(Global-NEWS).
Results demonstrate that the BC and NBC climate data
provide consistent estimates of the climate-driven changes in
water fluxes (ET, runoff, and water demand), VOCs (isoprene
and monoterpenes) and NO emissions, mean crop yield, and
river DIN export over the PNW domain. However, signifi-
cant differences rise from projected SWE, crop yield from
dry lands, and HJ-Andrews’s ET between BC and NBC data.
Even though BC post-processing has no significant impacts
on most of the studied variables when taking PNW as a
whole, their effects have large spatial variations and some
local areas are substantially influenced. In addition, there are
months during which BC and NBC post-processing produces
significant differences in projected changes, such as sum-
mer runoff. Factor-controlled simulations indicate that BC
post-processing of precipitation and temperature both sub-
stantially contribute to these differences at regional scales.
We conclude that there are trade-offs between using BC
climate data for offline CCI studies versus directly modeled
climate data. These trade-offs should be considered when
designing integrated modeling frameworks for specific ap-
plications; for example, BC may be more important when
considering impacts on reservoir operations in mountain-
Published by Copernicus Publications on behalf of the European Geosciences Union.

2602 M. Liu et al.: What is the importance of climate model bias?
ous watersheds than when investigating impacts on biogenic
emissions and air quality, for which VOCs are a primary in-
dicator.
1 Introduction
“To bias-correct or not?” is debated in the scientific commu-
nity (Ehret et al., 2012; Hagemann et al., 2011; Muerth et
al., 2013). Bias correction (BC) discussed here is the pro-
cess of adjusting global climate model (GCM) or regional
climate model (RCM) output – mainly temperature (T) and
precipitation (P) – depending on discrepancies between ob-
served and modeled results over the period of observation.
While BC is a post-processing step that is a widely applied
for climate change impact (CCI) studies, there are several
known issues. One concern is that most studies that use
BC GCM/RCM data without adequate quantification of the
effects of BC, thereby introducing additional uncertainties
(Ehret et al., 2012; Muerth et al., 2013; Teutschbein and Seib-
ert, 2012).
Bias in climate models can be attributed to uncertainties
in representations of atmospheric physics (Maraun, 2012),
boundary conditions and initialization (Bromwich et al.,
2013), inadequate reference data sets such as reanalysis data
(Dee et al., 2011a, 2011b; Thorne and Vose, 2010), climate
variability (Ehret et al., 2012), limitations in input data res-
olution (Wood et al., 2011), and simplifications required due
to limited computing capacity. BC of GCM/RCM output as a
post-processing step has been used to address this deficiency
(Wood et al., 2004). While GCMs/RCMs should continue
to be developed to improve predictability, current deficien-
cies in GCMs/RCMs often necessitate correction of result-
ing climate biases to make the data useful as input for CCI
studies.
Ehret (2012) reviewed the problems in using BC
GCM/RCM data, including problematic assumptions of sta-
tionarity of the error statistics, independently adjusting cli-
mate variables, and lack of a physical basis. Several recent
studies have questioned the validity of common assumptions
for the BC process (Berg et al., 2009; Christensen et al.,
2008; Haerter et al., 2011; Hagemann et al., 2011; John-
son and Sharma, 2012; Maraun, 2012; Piani et al., 2010;
Vannitsem, 2011). Recent advancement in BC methodology
addresses some of these shortcomings. For example, there
are methods that allow for correction of biases in Tand P
while preserving the relationships between them (Hoffmann
and Rath, 2012; Piani and Haerter, 2012) and attempts to
allow non-stationary BC (Buser et al., 2009). Vannitsem et
al. (2011) question the utility of BC particularly in the con-
text of decadal forecasts of a transient climate, which is a
timescale of importance in many impact studies; for example,
many state planning agencies operate on a 20-year horizon,
irrigation infrastructure and farm machinery often have 10-
year investment payback periods, and perennial crops have
investment horizons of 10 to 30 years.
Although the deficiencies of BC are known, the effects
of BC on the climate change signal and hence the conse-
quences of BC on hydrometeorology, biogeochemistry, eco-
logical, and agricultural estimates are still unclear. Recently,
studies have attempted to quantify the effects of bias correct-
ing input climate data on model outcomes (Chen et al., 2011;
Hagemann et al., 2011; Muerth et al., 2013). Results indi-
cate that although BC better reproduces historical observa-
tions, it can also alter the climate change signal for certain
locations (Hagemann et al., 2011), and/or for certain indica-
tors (Muerth et al., 2013). However, these studies are cur-
rently limited to stand-alone hydrologic models. There is a
need to characterize how BC of modeled climate data affects
the projection of land surface response and impacts to wa-
ter quality/quantity, ecosystem productivity, and emissions of
reactive species that influence air quality.
As Earth system models (EaSMs) currently stand, there
are trade-offs associated with the decision to use offline (with
BC) or online (without BC) simulations. While BC enhances
model skill with respect to observations, it often violates
laws for conservation of mass and energy that are funda-
mental to non-BC models. Although NBC climate predic-
tions are often inconsistent with observations in terms of ab-
solute magnitudes, CCI studies based on these data sets or
online models often carry the assumption that these models
may still be useful to characterize changes between current
and future conditions. The central questions this paper ad-
dresses are as follows: to what extent do CCI impacts (repre-
sented by the difference between future and historical hydro-
logic/biogeochemical simulations) differ when driving land
surface models by simulated meteorology with and without
BC, and what are the implications of these differences? In
short, to what extent does BC matter in terms of projecting
the response of the land surface?
The objective of this work is to understand and quantify
the sensitivity of multiple decision-relevant variables (related
to hydrology, agriculture, ecosystems, air quality, and nu-
trient export) to the BC post-process on climate data. To
achieve this objective, we use both BC and NBC meteorolog-
ical variables as input to drive impact models that cover re-
gional and watershed-scale hydrology, crop growth and phe-
nology, river nutrient export, and biogenic emissions. This
allows EaSM teams to make better-informed decisions on the
tradeoffs that exist when developing an integrated modeling
application for a specific natural or agricultural management
question.
Biogeosciences, 11, 2601–2622, 2014 www.biogeosciences.net/11/2601/2014/

M. Liu et al.: What is the importance of climate model bias? 2603
2 Methods
2.1 Study domain description
The domain of this study is the US Pacific Northwest region
(PNW), which includes the Columbia River basin (CRB)
and coastal watersheds in the states of Washington and
Oregon (Fig. 1). The region supports a vast array of agri-
cultural and natural resources. However, due to its winter-
dominated precipitation and associated large seasonal stor-
age of water as snowpack, this region faces substantial risk
from global warming (Adam et al., 2014; Liu et al., 2013;
Mote, 2003). Farmland occupies 11% of the CRB, with a
highly diverse mix of crops, including alfalfa, hay, winter
wheat, apples, sweet corn, potatoes, and sugar beets (Na-
tional Research Council, 2004). Across the PNW region,
31 % of all farmland is irrigated, and 70 % of this land area
is irrigated from surface water (USDA National Agricultural
Statistics Service, 2008).
For investigating impacts on forest ecosystems, we per-
form a watershed-scale simulation on the National Science
Foundation (NSF) Long-Term Ecological Research (LTER)
HJ-Andrews site in central Oregon (see inset in Fig. 1). Lo-
cated along the western slopes of the Cascade Mountain
Range, the site encompasses 64 km2and extends to the Look-
out Creek watershed boundaries, which drains to the McKen-
zie River.
2.2 Land surface model descriptions
Impact models being used for this study are major compo-
nents from a regional Earth system model (BioEarth) cur-
rently in development that aims to improve understanding
of interactions among carbon, nitrogen, and water at the re-
gional scale, in the context of global change, to inform de-
cision makers’ strategies regarding natural and agricultural
resource management (Adam et al., 2014). These models in-
clude a macroscale hydrologic model (VIC), a coupled crop-
ping system model (VIC-CropSyst), an ecohydrologic model
(RHESSys), a biogenic emissions model (MEGAN), and a
nutrient export model (Global-NEWS). Each of these mod-
els is described briefly below.
The Variable Infiltration Capacity (VIC; Liang et al., 1994)
model is a fully distributed, physically based, macroscale
model that solves the water and energy budgets for every
grid cell in the study domain. It was developed for large-
scale applications (1/16◦–2◦), and subgrid heterogeneities in
land cover and topography are considered. VIC account for
key moisture and energy fluxes between the land surface and
the atmosphere, and includes algorithms for shallow subsur-
face (frozen and unfrozen) moisture, snow, lake, and wetland
dynamics (Andreadis et al., 2009; Bowling and Lettenmaier,
2010; Cherkauer and Lettenmaier, 2003). VIC has been ap-
plied over all continental land areas, and has been used exten-
sively over the PNW (Adam et al., 2009; Elsner et al., 2010;
Fig. 1. Study area and simulation domain for this study. Pacific
Northwest: VIC, VIC-CropSyst, and MEGAN domains; Columbia
River basin: NEWS-DIN domain; HJ-Andrews: RHESSys domain.
Hamlet et al., 2007, 2012; Hamlet and Lettenmaier, 2005;
Liu et al., 2013; Maurer et al., 2002). In this application, we
use the implementation (version 4.0.7) described by Hamlet
et al. (2012).
CropSyst (Stöckle et al., 1994, 2003) is a field-scale,
multi-year, multi-crop model developed to serve as an an-
alytical tool to study the effect of climate, soils, and man-
agement on cropping systems productivity, nutrient cycling
and fate, and the environment. Management options include
crop rotation, cultivar selection, irrigation, nitrogen fertiliza-
tion, tillage operations, and residue management. CropSyst
has been evaluated and used in the PNW (e.g., Peralta and
Stöckle, 2002) and around the world (e.g., Stöckle et al.,
2003). A simplified version of CropSyst that focuses on wa-
ter use and productivity was extracted for coupling with the
VIC hydrology model (VIC-CropSyst v1.1; Rajagopalan et
al., in preparation). VIC passes meteorological and hydro-
logical parameters to CropSyst and CropSyst handles crop
growth and passes irrigation demand to VIC. The crop distri-
bution and irrigation extension was generated from cropland
data of the Washington State Department of Agriculture and
the Cropland Data Layer from the United States of Depart-
ment of Agriculture (USDA) (USDA National Agricultural
Statistics Service Cropland Data Layer, 2011). Irrigation ex-
tent outside Washington state is identified from survey data.
All irrigated croplands are assumed to be managed with ideal
irrigation practices, i.e., without drought-induced interrup-
tions to water rights (Yorgey et al., 2011).
The Regional Hydro-Ecologic Simulation System
(RHESSys v5.15; Tague and Band, 2004) is a physically
based watershed-scale ecohydrological model designed to
simulate climate and land use change impacts on ecosystem
carbon and nutrient cycling and hydrology. It uses an
www.biogeosciences.net/11/2601/2014/ Biogeosciences, 11, 2601–2622, 2014

2604 M. Liu et al.: What is the importance of climate model bias?
adaptation of BIOME-BGC (White and Running, 1994) and
a modified version of the Century-NGAS model (Parton et
al., 1993) to simulate above- and belowground carbon and
nitrogen processes. RHESSys fully couples these biogeo-
chemical processes with a spatially distributed hydrologic
model. RHESSys has been applied in a number of different
environments, including watersheds in the PNW (e.g.,
Christensen et al., 2008; Meentemeyer and Moody, 2002;
Tague and Grant, 2009; Tague et al., 2007, 2008a, b, 2009,
2013; Zierl and Bugmann, 2005).
The Model of Emissions of Gases and Aerosols from Na-
ture (MEGAN v2.1; Guenther et al., 2012) incorporates re-
cent advances in the understanding of the processes con-
trolling biogenic emissions (e.g., solar radiation, tempera-
ture, soil moisture, carbon dioxide concentration, vegeta-
tion type, leaf age, and LAI) at a resolution suitable for
regional modeling. While MEGANv2.1 can be run as an
offline model or as an integrated component of land sur-
face and atmospheric chemistry models, we used the of-
fline version for this study. As we are most interested in
examining the climate change impacts on biogenic emis-
sions, the land cover used in all simulations was kept con-
stant (i.e., year 2008 conditions) based on MODIS LAI
and plant functional types (PFTs) (Guenther et al., 2012;
http://acd.ucar.edu/~guenther/MEGAN/MEGAN.htm).
The Global Nutrient Export from Water(S)heds (Global
NEWS v2.0) model predicts annual average export of mul-
tiple forms of carbon and multiple nutrients as a function
of climate, basin characteristics, and human activities within
watersheds. NEWS sub-models have been applied broadly to
understand land-to-ocean transport of carbon and nutrients at
regional and global scales (Dumont et al., 2005; Harrison et
al., 2005a, b, 2010; Mayorga et al., 2010; Seitzinger et al.,
2005, 2010). More recently, NEWS sub-models have been
successfully applied at regional and sub-basin spatial scales
and at monthly timescales (Harrison et al., 2010; McCrackin
et al., 2013; Thieu et al., 2010) as well as in hindcast and sce-
nario modes to examine historic and potential future changes
in coastal nutrient loading (Seitzinger et al., 2010). For this
study, we apply the NEWS-DIN model to simulate the dis-
solved inorganic nitrogen (DIN) export from the CRB as a re-
sult of human activities, natural processes, and in-stream re-
moval processes (Dumont et al., 2005; Mayorga et al., 2010;
Seitzinger et al., 2002). Nutrient loading and reservoir in-
formation used by NEWS is derived from prior global-scale
analyses (Bouwman et al., 2010; Van Drecht et al., 2009).
2.3 Climate data
The climate data for this study are output from a regional cli-
mate model simulation using the Weather Research and Fore-
casting (WRF; Skamarock et al., 2008) model to dynamically
downscale a coupled global general circulation model con-
sisting of ECHAM5 (the atmospheric component) (Roeck-
ner et al., 1999, 2003) and the Max Planck Institute Ocean
Model (MPI-OM; Marsland et al., 2003) (the ocean compo-
nent). These simulations are described in detail by Salathé
et al. (2013) and briefly summarized here. For the historical
period (1970–1999), the ECHAM5–MPI-OM simulations of
the 20th century forced by historical greenhouse gas con-
centration, aerosol, and solar forcing were used; for the 21st
century, simulated results with the A1B emissions scenario
of the Special Report on Emissions Scenarios (SRES) were
used. The A1B is a medium-high greenhouse gas emission
scenario that is for “business as usual” in the first half of the
21st century with greater mitigation in the second half, and a
balanced energy system (Nakicenovic and Swart, 2000).
The implementation of the WRF model used in this study
for the PNW is described and evaluated in Zhang et al. (2009)
and Duliere et al. (2011). The simulations used nested grids
at 36 and 12km spacing. The 12 km nest covers the region
shown in Fig. 1. The outer 36 km nest receives boundary con-
ditions and interior nudging from the global ECHAM5–MPI-
OM fields; the inner 12km nest is forced only at its bound-
ary by the outer nest (i.e., without nudging). Gridded analysis
nudging was applied only to the outer nest, only on upper air
wind, temperature, and moisture fields, and only on wind in
the planetary boundary layer. The simulations use the follow-
ing parameterization choices in WRF: WRF Single-Moment
5-class (WSM5) microphysics (Hong et al., 2004), Kain–
Fritsch convection (Kain and Fritsch, 1993), the NOAH land-
surface model (Chen and Dudhia, 2001), the YSU (Hong and
Pan, 1996) boundary layer, and NCAR CAM shortwave and
longwave radiation (Collins et al., 2004).
Although regional-scale climate models represent the im-
portant mesoscale weather and climate features far better
than global models, the simulations are still subject to biases
resulting from deficiencies in both the global forcing fields
and the regional model (Christensen et al., 2008; Wood et al.,
2004). To obtain acceptable hydrologic simulations, these bi-
ases must be removed. In addition, to link the WRF results
to the VIC hydrologic model, the simulations require addi-
tional downscaling from the 12km WRF grid to the 1/16◦
(0.0625◦) VIC grid. These downscaling and BC steps are per-
formed as follows: first the WRF simulations of temperature
(T ), precipitation (P ), and wind speed at 12 km×12 km grid
cells and 6-hour time step were aggregated to daily average
Tand wind speed and daily total P. The daily maximum
and minimum Twere identified from these four sub-daily
records. These data were then downscaled to 1/16◦with
the Symap algorithm, which uses a four-nearest-neighbor
inverse-distance weighting approach (Maurer et al., 2002;
Shepard, 1984). Re-gridded Tand Pwere then subjected
to BC by using a quantile-mapping approach applied at
daily time step, and detailed descriptions can be found in
Salathé et al. (2013) and Wood et al. (2002); this method is
termed bias correction–statistical downscaling (BCSD). The
quantile-mapping BC is achieved by replacing simulated val-
ues for Tand Pwith values having the same percentiles
(non-exceedance probabilities) (Wood et al., 2002). For each
Biogeosciences, 11, 2601–2622, 2014 www.biogeosciences.net/11/2601/2014/

M. Liu et al.: What is the importance of climate model bias? 2605
grid cell, cumulative distribution functions (CDF) of Tand
Pwere generated for each calendar month and climate sce-
nario for both observation and projections. For a given pe-
riod of GCM-WRF projections, the quantile-mapping algo-
rithm looks up the simulated CDF during the training period
(1970–1999). The corresponding quantile value from the ob-
served CDF then replaces the simulated value for that period.
For future periods, systematic shifts in the magnitude and
variance result in projected Tand Pfalling outside of the
original CDF of the training period. Thus, for each 30-year
future period, the average shift in Tis removed before BC but
is added back after the corresponding value is identified in
the simulated CDF of the draining period. However, extrap-
olation of the CDF during the training period is still needed
because future simulated Tor Pfrequently fall outside its
range. For this study, values that fall outside the simulated
training-period CDF were mapped to an equivalent observed
value via the multiplicative (P) or additive (T) anomaly from
the observed mean.
Re-gridded wind speed from WRF was applied directly
without BC. The observed training data for the quantile map-
ping were from gridded historical Tand P(Elsner et al.,
2010, Hamlet et al., 2012; also used in this study as baseline
runs for the 1980s) and applied to WRF-simulated variables
for both historical and future periods (Table 1). The wind
speed data in the observed historical climate were re-gridded
from reanalysis data (Elsner et al., 2010). The assumption
for this quantile-based BC approach is that historical bi-
ases in WRF simulations are comparable to those that would
occur over future climates (Salathe et al., 2013). BCSD
downscaling largely preserved the spatial details of precip-
itation and temperature from the regional climate model,
while removing systematic biases (determined by compari-
son with observations) without losing the simulated spatial
correlation between Tand P(Salathe et al., 2013; Wood
et al., 2002). Themessl et al. (2011) provided a detailed
review on seven major statistical-downscaling and BC ap-
proaches and concluded that quantile mapping has advan-
tages in removing regional climate model deficiencies in the
entire Pdistribution (including mean, day-to-day variabil-
ity, and extremes). Recently, by using BCSD approach, a
new archive of downscaled Coupled Model Intercomparison
Project Phase 5 (CMIP5) (Taylor et al., 2012) climate projec-
tions for the conterminous United States, called NASA Earth
Exchange (NEX) Downscaled Climate Projections at 30 arc-
seconds (NEX-DCP30), has been generated from NEX plat-
form and distributed through the NASA Center for Climate
Simulation (NCCS) (Thrasher et al., 2013). NEX-DCP30
contains more than 100 downscaled climate projections from
33 CMIP5 GCMs and 4 RCP scenarios (Thrasher et al.,
2013). This case study on the hydrologic and biogeochemical
consequences from the BCSD quantile-mapping approach
can provide valuable information to communities using this
data set and the typical statistical-downscaling method.
Table 1. List of climate scenarios for this study. All WRF sim-
ulations used as boundary conditions results from the coupled
ECHAM5–MPI-OM model run with the IPCC SRES A1B scenario
(Salathé et al., 2010, 2013).
Time Source Bias-corrected Variable
period variable name
1970–1999 Gridded – OBS
observed
WRF T&PBCT&P ,1980s
WRF TBCT ,1980s
WRF PBCP ,1980s
WRF None NBCT&P ,1980s
2010–2039 WRF T&PBCT&P ,2020s
WRF TBCT ,2020s
WRF PBCP ,2020s
WRF None NBCT&P ,2020s
2040–2069 WRF T&PBCT&P ,2050s
WRF TBCT ,2050s
WRF PBCP ,2050s
WRF None NBCT&P ,2050s
The VIC model uses daily T(including daily maximum
and minimum T ) and Pto simulate other meteorological
variables including short- and longwave radiation, surface
temperature, and relative humidity by solving surface en-
ergy budget equations and using algorithms from the Moun-
tain Microclimate Simulation Model (MTCLIM; Bohn et al.,
2013; Hungerford et al., 1989; Kimball et al., 1997; Thorn-
ton and Running, 1999; Thornton et al., 2000) (Liang et
al., 1994; Maurer et al., 2002). RHESSys uses a similar ap-
proach to estimate other meteorological variables relying on
Tand P(Tague and Band, 2004). The meteorological driv-
ing forces for MEGAN were from VIC-modeled variables in-
cluding surface temperature, short- and longwave radiation,
and humidity (Fig. 2).
2.4 Simulation experiments
To quantify the effects of BC (individually on Tand P, as
well as the combined effects) on projected land surface re-
sponse, different combinations of BC and NBC climate data
are used to drive these models (Table 1). Figure 2 demon-
strates the information and work flow between each model
and data source. For the RHESSys simulations, we estimate
forest net primary productivity (NPP) in response to climate
variation, without accounting for disturbances and changes
in nitrogen limitation. Global NEWS uses annual outputs of
surface runoff and baseflow from VIC offline simulations as
well as irrigation water demand from VIC-CropSyst.
The following terrestrial responses to climate change and
BC are investigated: hydrological processes (evapotranspi-
ration, ET; runoff; and snowpack water equivalent, SWE),
www.biogeosciences.net/11/2601/2014/ Biogeosciences, 11, 2601–2622, 2014

2606 M. Liu et al.: What is the importance of climate model bias?
ECHAM5‐A1B
1.9° (1980s,2020s,2050s)
WRF
12km
Observed‐Gridded
1/16° (1980s)
BC
1/16°
NBC
1/16°
SD
1/16°
Abbreviations&Legend:
SD:StatisticalDownscaling;BC:Bias Correction;NBC:No Bias Correction;CY:CropYield;WD:Water
Demand;DIN:DissolvedInorganicNitrogen;ET:Evapotranspiration;NPP:NetPrimaryProductivity;
SWE:SnowpackWaterEquivalent;NO:NitrogenMonoxide
OutputvariablesfrommodelsModel
DataflowData
VIC
VIC‐CropSyst
RHESSys
GlobalNEWS
MEGAN
Dailymetdata
(Tmax,Tmin,Precipitation)
Subset
HJ‐Andrews
Croplands
Allland
SurfaceRunoff
Baseflow
Radiation
Humidity
DIN
ISOPRENE
MONOPRENES
NO
ET
NPP
CY
WD
ET
Runoff
SWE
WD
Fig. 2. Offline simulations and data flow for this study.
agricultural processes (crop yields, CY; irrigation water de-
mand, WD), emissions of gases that contributes to ozone
and aerosol formation in the atmosphere (isoprene; monoter-
penes; and nitrogen monoxide, NO), river export of DIN,
and forest NPP (Fig. 2). Except for MEGAN, all model
simulations are performed for three 30-year periods: 1970–
1999 (hereafter, 1980s), 2010–2039 (2020s), and 2040–2069
(2050s). For each 30-year period, only the most recent 25
years of simulated results are used in analyses, providing a
5-year model spin-up period. MEGAN simulations are con-
ducted with 25-year-averaged climate data for each period.
2.5 Attributing individual and combining effects of T
and Pbias corrections
The projected impact of climate change on a given variable
is quantified as a percentage change (Eqs. 1 and 2).
1BCT&P ,t (%)=BCT&P , t −BCT&P , 1980s
BCT&P , 1980s
×100% (1)
1NBCT&P ,t (%)=(2)
NBCT&P , t −NBCT&P , 1980s
NBCT&P , 1980s
×100%,
where tis time period in the future (i.e., 2020s or 2050s),
1BC (%) is the estimated change in percentage based on BC
climate data, while 1NBC (%) is the percentage change from
NBC data; BCT&P ,t (or NBCT&P ,t )represents the simulated
results during period twith BC (or NBC) data; and T&P
indicates that both Tand Pare treated the same (either both
BC or both NBC). The total BC-derived discrepancy in the
projected impact was calculated with Eq. (3).
effT&P,t =1BCT&P ,t (%)−1NBCT&P ,t (%)(3)
The rationale for calculating this discrepancy is that, while
there might be significant differences in simulated variables
driven by BC and NBC climate data, the estimated percent-
age change between the future and historical periods may
possibly not be as dissimilar because the BC process is de-
signed to conserve the changes in Tand Pbetween future
and historical periods. In this case, effT&P,t from Eq. (3)
would be small enough that the BC process would not be
necessary for this certain variable if its relative change is the
major consideration in a particular decision-making process.
We also consider the individual roles of BC of Pand BC of
Thave on percentage change. Equation (6) is used to quan-
tify the impact due to BC of Tbut not P, and Eq. (7) was
used to quantify the impact due to BC of Pbut not T.
1BCT ,t (%)=BCT ,∧P , t −BCT,∧P , 1980s
BCT ,∧P , 1980s
×100%,(4)
1BCP ,t (%)=BC∧T ,P , t −BC∧T ,P , 1980s
BC∧T ,P , 1980s
×100%,(5)
efft
T=1BCT&P,t (%)−1BCP,t (%)+1BCT ,t (%)−1NBCT&P ,t (%)
2
,(6)
efft
P=1BCT&P,t (%)−1BCT ,t (%)+1BCP ,t (%)−1NBCT&P ,t (%)
2
,(7)
Biogeosciences, 11, 2601–2622, 2014 www.biogeosciences.net/11/2601/2014/

M. Liu et al.: What is the importance of climate model bias? 2607
where 1BCT ,t (%)and 1BCP ,t (%)are the percentage
change in a response variable caused by BC of Tonly and
BC of Ponly, respectively; efft
Tand efft
Pare average effects
of BC of Tand BC of P, respectively, under the context of
interactions between T&P; and the symbol ∧indicates that
the variable was not bias-corrected.
2.6 Statistical method for quantifying significance
In this study, we emphasize the regional responses of land
surface processes to BC and NBC climate data by examining
the projected changes on summe total changes over the study
domains as well as spatial variations within the domains by
examining each grid cell. For the region-wide estimation, we
consider the significance of BC effects on total water avail-
ability, yield, SWE, biogenic volatile organic compounds
(BVOC), NO, and DIN export by summing over the study
domains (i.e., the PNW, the CRB, or the HJ-Andrews) before
applying the Student ttest on each response variable. In con-
sidering spatial correlations in meteorological variables, the
Walker test has been widely used to estimate the significance
of rejecting a global null hypothesis across a study domain
by first testing the significance of each grid cell (Folland and
Anderson, 2002; Livezey and Chen, 1983; von Storch, 1982;
Wilks, 2006). However, strong local effects (i.e., the lowest p
value in the ttest) can dominate and result in an easier rejec-
tion of the global null hypothesis (Wilks, 2006). To examine
spatial variations, we first quantified the differences and pro-
jected changes (future minus historical) in land surface pro-
cesses as driven by BC and NBC climate over each grid cell.
Next, we counted the percentage of grid cells over which the
differences in projected changes are significant (based on rel-
ative change in percentage). In this study, we take α=0.05
in the ttest as the index to claim the difference is significant.
In the t-test analysis, the two groups of simulated results
(i.e., BC and NBC climate) were preprocessed by subtracting
the mean of the historical period (i.e., mean of 1975–1999).
As mentioned earlier, we report the percentage change for all
targeted variables except T, for which we report the absolute
change.
3 Results
We first discuss the changes to the climate signal (Sect. 3.1)
and then discuss how this climate change signal translates
into impacts on hydrologic and biogeochemical processes at
both annual (Sect. 3.2) and seasonal (Sect. 3.3) timescales.
Finally, we discuss the relative effects of BC on Tand Pto
the overall change signal (Sect. 3.4).
3.1 BC and NBC climate data over the historical and
future periods
When spatially averaged over the study domain, there is
a significant difference between NBC downscaled climate
‐0.03°C
-10%
0%
10%
20%
30%
T
P
ET
Runoff
SWE
Total YD
Irrig.YD
Dry.YD
WD
Isoprene
Monoterpenes
NO
HJ-ET
HJ-NPP
DIN
Differences in the % rate of change
(T:°C; Others: %)
2020s‐1980s 2050s‐1980s
Fig. 3. Differences in the percentage change between BC and NBC
climate data and the simulated outputs driven by bias-corrected
(BC) and non-bias-corrected (NBC) climate. They are calculated as
[1BC20s-80s (%) −1NBC20s-80s (%)] for the change between the
1980s and 2020s and [1BC50s-80s (%) −1NBC50s-80s (%)] for the
change between the 1980s and 2050s. For T, it is total differences in
degrees Celsius, i.e., (1BC20s-80s −1NBC20s-80s)for the change
between the 1980s and 2020s, and (1BC50s-80s −1NBC50s-80s)
for the change between the 1980s and 2050s. T: annual mean tem-
perature; P: average annual precipitation; ET: average annual evap-
otranspiration; Runoff: total runoff; SWE: snowpack water equiva-
lent on 1 April; Total YD: total yield from all croplands; Irrig. YD:
yield from irrigated cropland; Dry YD: yield from dryland (non-
irrigated cropland); WD: total irrigation water demand over irri-
gated cropland, HJ-ET: RHESSys modeled ET over HJ-Andrews
watershed; HJ-NPP: RHESSys modeled net primary production
(NPP) over HJ-Andrews watershed; DIN: NEWS modeled dis-
solved inorganic nitrogen yield over the Columbia River basin. The
small stars under or above each column represent pvalue <0.05 for
the Student ttest of differences between BC anomalies and NBC
anomalies during the period of the 2020s and 2050s, respectively.
data of the 1980s (i.e., NBC climate) and observations.
NBC has a mean annual Tand Pthat is 2.7◦C lower
and 156 mm year−1(or 17%) higher than observations, re-
spectively (Table 2). After BC, the modeled climate closely
matches the observations (Table 2) as expected (i.e., Salathe
et al., 2013; Wood et al., 2002). The absolute projected cli-
mate change signal in T (1T ) in the 2020s and 2050s com-
pared to the 1980s is preserved in the BC process (Table 2).
1T is approximately 1 ◦C for the period 1980s–2020s and
2.5◦C for the period 1980s–2050s for both BC and NBC
cases. The projected climate change signal of Pis a little
higher under BC. Pincreases by 45 mm year−1(or 5.2 %)
under BC and 37 mm year−1(or 3.5 %) under NBC for the
period 1980s–2020s, and 97 mm year−1(or 11.1 %) under
BC and 89 mm year−1(or 8.5 %) under NBC for the pe-
riod 1980s–2050s. Overall, the average differences between
BC and NBC in 1T and percentage rate of change in P
[1P (%)] over these two time periods (i.e., 1980s–2020s and
1980s–2050s) are insignificant, i.e., 0.03 ◦C (Student’s ttest
on anomaly in annual mean, p=0.89) and 2.1 % (Student’s
ttest on anomaly in annual precipitation, p=0.72) (Fig. 3).
www.biogeosciences.net/11/2601/2014/ Biogeosciences, 11, 2601–2622, 2014

2608 M. Liu et al.: What is the importance of climate model bias?
Table 2. Changes in climate and simulated variables during the periods of 1980s–2020s and 1980s–2050s, and differences in the changes
between results with bias correction (BC) and without bias correction (NBC) of climate data.
Period Variable T
(◦C) P
(mmyr−1)VIC offline VIC-CropSyst (cropland)
ET
(mmyr−1)Runoff
(mmyr−1)SWE
(mmH2O) Irrigation
demand
(billionm3yr−1)
Irrigated
cropland
yield
(Mtyr−1)
Dryland
yield
(Mtyr−1)
Total
yield
(Mtyr−1)
1980s
(1970–1999) OBS 6.45 893 410 482 127 21.31 17.7 11.9 29.6
BC 6.40 877 421 456 105 25.36 20.0 11.1 31.1
NBC 3.68 1049 393 652 358 17.75 19.7 26.3 46.1
2020s
(2010–2039) BC 7.35 922 437 486 100 25.36 20.9 13.6 34.6
1BC1
20s–80s 0.95 45 15 30 −5 0.00 0.9 2.6 3.5
Eq. (1): 1BC20s–80s(%) 5.2 % 3.6 % 6.7 % −4.8 % 0.0 % 4.5 % 23.1 % 11.1 %
NBC 4.60 1086 406 677 316 17.94 21.0 30.0 50.9
1NBC2
20s–80s 0.92 37 13 24 −42 0.19 1.2 3.7 4.9
Eq. (2): 1NBC20s–80s(%) 3.5 % 3.3 % 3.7 % −11.7 % 1.3 % 6.3 % 13.9 % 10.6 %
Eq. (3): 1BC(%)−1NBC(%)1.6% 0.3% 2.9% 6.9 %66 −1.3%−1.8%9.2%60.5 %
2050s
(2040–2069) BC 8.92 974 469 506 82 24.91 22.3 17.7 40.1
1BC1
50s–80s 2.52 97 47 51 −23 −0.45 2.3 6.7 8.9
Eq. (1): 1BC50s–80s(%) 11.1 % 11.2% 11.1 % −21.6% −2.2 % 11.4 % 60.0% 28.7 %
NBC 6.17 1138 444 695 199 17.55 22.3 36.5 58.8
1NBC2
50s–80s 2.48 89 52 43 −159 −0.20 2.5 10.2 12.7
Eq. (2): 1NBC50s–80s(%) 8.5 % 13.1 % 6.6 % −44.3 % −1.4 % 12.9 % 38.6 % 27.6 %
Eq. (3): 1BC(%)−1NBC(%)2.6% −1.9%4.6% 22.7%6−0.8%−1.5%21.5 %61.1 %
Avg. (1BC−1NBC)3
2020s,2050s 0.03 8.29 −1.04 6.99 86.30 −0.18 −0.22 −2.30 −2.60
Avg. (1BC(%)−1NBC(%))4
2020s,2050s 2.1 % −0.8 % 3.8 % 14.8 % −0.8 % −1.0 % 15.4 % 8.4 %
Avg. (1BC−1NBC)2020s,2050s/Obs.(%)50.9% −0.3%1.5 % 67.7% −1.0%−1.3%−19.3%−8.8%
Periods Climate data/items MEGAN RHESSys
(HJ-Andrews) NEWS
(Columbia River basin export)
Isoprene
(µgm−2h−1)Mono-
terpenes
(µgm−2h−1)
NO
(µgm−2h−1)T
(◦C) P
(mmyr−1)NPP
(gC m−2yr−1)ET
(mmyr−1)T
(◦C) P
(mmyr−1)DIN
(kgN km−2yr−1)
1980s
(1970–1999) OBS 66.06 59.28 0.31 9.06 2218 1181 831 6.01 785 100.9
BC 65.27 58.60 0.30 9.01 2201 1068 876 5.96 768 91.6
NBC 33.47 40.99 0.21 6.62 2253 1356 743 3.14 961 153.7
2020s
(2010–2039) BC 74.80 64.38 0.34 9.86 2229 998 847 6.93 807 98.3
1BC1
20s–80s 9.53 5.78 0.04 0.85 28 −70 −29 0.97 39.17 6.7
Eq. (1): 1BC20s–80s(%) 14.6 % 9.9 % 14.6 % 1.3 % −6.5 % −3.3 % 5.1 % 7.3 %
NBC 38.16 44.70 0.23 7.41 2248 1373 755 4.08 995 159.3
1NBC2
20s–80s 4.69 3.71 0.03 0.79 −5 17 12 0.94 33 5.6
Eq. (2): 1NBC20s–80s(%) 14.0 % 9.1 % 14.2 % −0.2 % 1.2 % 1.6 % 3.5 % 3.6 %
Eq. (3): 1BC(%)−1NBC(%)0.6% 0.8% 0.4 % 1.5% −7.8%−4.9%61.6% 3.6 %
2050s
(2040–2069) BC 93.47 75.67 0.43 11.15 2316 1021 852 8.55 861 104.8
1BC1
50s–80s 28.20 17.07 0.13 2.14 115 −47 −24 2.59 94 13.2
Eq. (1): 1BC50s–80s(%) 43.2 % 29.1 % 43.8 % 5.2% −4.4 % −2.7 % 12.2% 14.4 %
NBC 47.91 52.31 0.30 8.69 2316 1365 741 5.70 1050 165.0
1NBC2
50s–80s 14.44 11.32 0.09 2.07 63 9 −2 2.56 89 11.3
Eq. (2): 1NBC50s–80s(%) 43.1 % 27.6 % 43.6 % 2.8 % 0.7% −0.3 % 9.2 % 7.3 %
Eq. (3): 1BC(%)−1NBC(%)0.1% 1.5% 0.2 % 2.4% −5.1%−2.5%3.0% 7.1 %
Avg. (1BC−1NBC)3
2020s,2050s 9.30 3.91 0.03 0.06 42 −71.24 −31.64 5.40 1.50
Avg. (1BC(%)−1NBC(%))4
2020s,2050s 0.3 % 1.2 % 0.3 % 2.0 % −6.4% −3.7 % 0.03 2.3% 5.4 %
Avg. (1BC−1NBC)2020s,2050s/Obs.(%)514.1 % 6.6% 9.3% 1.9% −6.0%−3.8%0.7 % 1.5%
11BC is the absolute change derived from BC climate between target periods; in other words, 1BC20s–80s represents the change between the 1980s and 2020s and 1BC50s–80s represents the change between the 1980s and 2050s.
21NBC is the absolute change derived from NBC climate between target periods; in other words, 1NBC20s–80s represents the change between the 1980s and 2020s and 1NBC50s–80s represents the change between the 1980s and 2050s.
3The difference in absolute change between BC and NBC climate averaged over the changes between the 1980s and 2020s and between the 1980s and 2050s, i.e., [(1BC20s–80s−1NBC20s–80s )+(1BC50s–80s −1NBC50s–80s)]/2.
4The difference in percentage change between BC and NBC climate averaged over the changes between the 1980s and 2020s and between the 1980ss and 2050s, i.e., {[1BC20s–80s(%)−1NBC20s–80s(%)] + [1BC50s–80s(%)−1NBC50s–80s (%)]}/2.
5The percentage difference between BC and NBC climate averaged over the changes between the 1980s and 2020s and between the 1980s and 2050s and relativeto observed climate (or the simulated outputs driven by observed data), i.e.,
[(1BC20s–80s −1NBC20s–80s)+(1BC50s–80s −1NBC50s–80s)]/(2×Obs1980s )×100%.
6This signifies a pvalue <0.05 for the Student ttest (not applicable for MEGAN results).
Biogeosciences, 11, 2601–2622, 2014 www.biogeosciences.net/11/2601/2014/

M. Liu et al.: What is the importance of climate model bias? 2609
9.9% 0.0%
10.9% 13.9%
94.8%
0.0%
45.8%
28.2%
0.6%
23.5%
94.2%
2.1%
14.1%
0.2%
0%
20%
40%
60%
80%
100%
ET
P
Runoff
SWE*
HJ-ET-
RHESSys
HJ-ET-VIC
HJ-NPP
2020S 2050S
Fig. 4. Percentage of grid cells with significant differences (p <
0.05) between the projected changes using BC and NBC climate
data. Note: for 1 April SWE, the sample size is the number of grid
cell that has a minimum of 25 mm SWE on 1 April according to
observed climate for the historical period.
The downscaled ECHAM A1B simulation suggests that
projected increases in Tand Pare greater in the southern
PNW than in the western and central PNW in the next half
century (Fig. 5a and b). Overall, BC post-processing of the
downscaled climate has generally conserved the spatial pat-
terns of 1T and 1P (%) over the study domain (as compared
to the original NBC climate) (Figs. 3, 4, 5a and b). However,
in certain regions, such as the central floodplain between the
Cascade and Rocky Mountains, the BC climate change sig-
nal 1P (%) is 5–10 % higher than the NBC climate signal
over the period of 1980s–2050s (Fig. 5b).
3.2 Impacts of BC on annual-scale hydrological and
biogeochemical processes
3.2.1 Impacts on large-scale hydrology (VIC simula-
tions)
Evapotranspiration (ET)
ET, including soil evaporation, canopy evaporation, and plant
transpiration, is the total water vapor leaving the land sur-
face to the atmosphere and is controlled by the availability
of energy and water. Climate change affects ET by altering
both energy and moisture availability. VIC simulations indi-
cate that both BC and NBC climate data result in 3 and 12 %
increases in ET during the periods 1980s–2020s and 1980s–
2050s, respectively (Table 2). There are no significant dif-
ferences in overall ET projections (p=0.52 and 0.83 over
ET anomalies in the 2020s and 2050s, respectively) (Fig. 3).
However, there is a large spatial variation in the ET climate
change signal between BC and NBC climate inputs, espe-
cially in the 2050s. Across the PNW, 28.2 % of the grid
cells shows significant differences in estimated ET percent-
age change between BC and NBC data over the period of
1980s–2050s (Fig. 4). Figure 5c shows that the use of BC
climate results in reduced 1ET (%) in high mountainous re-
gions such as the northern Cascade Ranges and the Rocky
Mountains, and higher 1ET (%) in flat-terrain areas such
as the floodplain of the Snake River and the Harney Basin,
as compared to NBC climate inputs. This difference indi-
cates that over higher mountainous terrain, similar relative
warming from cooler climates (NBC data) may produce a
higher ET loss than from hotter climates (BC data). While
over lower altitude regions, the magnitude of ET is mostly
controlled by moisture availability; that is, a higher relative
Pincrease can produce a higher relative ET increase (Fig. 5b
and c).
Runoff
Runoff, including baseflow and surface runoff, is the to-
tal water that flows to rivers and reservoirs. Most of the
runoff in the PNW is generated from the mountainous re-
gions (Fig. 5d). Both BC- and NBC-driven simulations pre-
dict increased runoff in the future. Use of BC climate data
leads to projected increases of 6.7 and 11.1 % in the periods
of 1980s–2020s and 1980s–2050s, respectively, while use
of NBC climate leads to lower increases (of 3.7 and 6.6%,
respectively) (Table 2, Figs. 3 and 4). Spatial patterns and
overall changes in total runoff over the PNW demonstrates
that changes in runoff are mostly controlled by the relative
change of P; that is, a higher 1P (BC data) can generate
higher 1Runoff (BC data), particularly in the context of a
drier climate (BC data) (Table 2). The largest increase (up
to 20%) in runoff occurs near the confluence of the Yakima,
Snake, and Columbia rivers in Washington, the Harney Basin
in Oregon, the Salmon River Mountains in Idaho, and the
mountainous areas in the northern PNW (Fig. 5d). Overall,
11 % of the grid cells of the PNW show significant effects of
BC treatment for the 2020s and this increases to 23.5% for
the 2050s (Fig. 4).
Snowpack water equivalent (SWE)
April 1st SWE is a commonly used indicator of water re-
sources availability in the western US because melting of the
snowpack generates spring–summer peak flows (Adam et al.,
2009; Barnett et al., 2005; Hamlet et al., 2005). Model esti-
mates show that the mountainous areas of the northern Cas-
cade and Rocky Mountains have larger SWE storage than
any other region in the PNW (Fig. 5e). With a projected
warming trend in the future, SWE will continuously decrease
between the 1980s and 2050s. The BC climate change sig-
nal of 1 April SWE is significantly weaker than the NBC
signal. Use of BC climate results in SWE decreases of 4.8
and 21.6 % for the periods 1980s–2020s and 1980s–2050s,
respectively, which are much lower than the change signal
projected by NBC climate (11.7 and 44.3%, respectively)
(Table 2, Figs. 3 and 4). The differences between BC- and
NBC-derived 1SWE (%) are significant in both time periods
www.biogeosciences.net/11/2601/2014/ Biogeosciences, 11, 2601–2622, 2014

2610 M. Liu et al.: What is the importance of climate model bias?
Fig. 5. Spatial patterns of differences in climate and modeled variables driven by observed data (Obs.), bias-corrected (BC), and non-bias-
corrected (NBC) climate. 1BC20s−80sand 1BC50s−80sare total change between the 1980s and 2020s and between the 1980s and 2050s,
respectively, under BC climate. 1BC20s−80s(%) and 1BC50s−80s(%) are the percentage change between the 1980s and 2020s and between
the 1980s and 2050s, respectively, and they are calculated as 1BC2020s/BC1980s×100 % and 1BC2050s/BC1980s×100 %, respectively.
“1BC20s−80s(%)−1NBC20s−80s(%)” is the difference in the percentage change between the 1980s and 2020s in simulated results driven
by BC and NBC climate; similarly “1BC50s−80s(%)−1NBC50s−80s(%)” is the difference in the percentage change between the 1980s
and 2050s. “1BC20s−80s−1NBC20s−80s” is the difference in the total change between the 1980s and 2020s in simulated results driven by
BC and NBC.
Biogeosciences, 11, 2601–2622, 2014 www.biogeosciences.net/11/2601/2014/

M. Liu et al.: What is the importance of climate model bias? 2611
Fig. 5. Continued.
(p=0.03 for anomalies in the 2020s and p < 0.001 for
anomalies in the 2050s) and BC treatment affects 94.2% of
the grid cells in estimating relative change in SWE during
the period of 1980s–2050s (Fig. 3). Our simulations indicate
that under hotter and dryer conditions (BC data), continu-
ous warming may have fewer relative impacts on snowpack
www.biogeosciences.net/11/2601/2014/ Biogeosciences, 11, 2601–2622, 2014

2612 M. Liu et al.: What is the importance of climate model bias?
accumulation than under cooler and wetter background con-
ditions (NBC data), particularly over high mountainous areas
(Fig. 5e). The SWE of a cooler baseline (NBC data) has more
potential to be affected by additional warming as compared
to a warmer baseline (BC data) subject to the same addi-
tional warming (Fig. 5e), especially if the additional warming
moves grid cells from below freezing temperatures to above
freezing temperatures. Also, a larger expanse of negligible
snowpack areas in the BC scenario as compared to the NBC
scenario can result in larger areas being categorized as hav-
ing no change in SWE for the BC scenario.
3.2.2 Impacts on croplands (VIC-CropSyst simulations)
Crop yield and irrigation water demand are important fac-
tors in a farmer’s decision-making process. In the study do-
main, over 40 types of crops, including tree fruits, grains,
cereals, vegetables, and crops for forage are simulated by
VIC-CropSyst. The crop yield in each grid cell and the re-
gional mean crop yield are calculated as area-weighted aver-
age yields in this analysis (Fig. 5f).
Crop yield
BC has a large effect on estimated crop yield by VIC-
CropSyst. Over the historical period, the total crop yield from
BC and observed climate data are similar (31.1 and 29.6 mil-
lion tons (Mt), respectively), while the NBC climate data re-
sulted in a much higher crop yield (46.1 Mt) (Table 2). The
major difference in estimated yields among NBC, BC, and
observed climate data occurs over dryland (non-irrigated)
crops (Fig. 5f). Both BC- and NBC-projected climate cause
large increases in total yield between present-day and the
2020s or 2050s (Table 2, Fig. 5f). In contrast to the projec-
tions for dryland crops, BC does little to change yield pro-
jections for irrigated crops (Table 2; Figs. 3, 4, 5f and g).
The behavior of aggregate yields is the net response of mul-
tiple crop types that can have different responses to climate
change. In addition to the relative increases in temperature,
crop yields can be affected by the actual magnitude of tem-
perature itself, which can lead to differences in responses in
a cooler NBC scenario versus a warmer BC scenario. The
warmer BC scenario can have a longer available growing sea-
son than the cooler NBC scenario, leading to higher relative
yield increases in crop types that can take advantage of mul-
tiple cuttings. The warmer BC scenario can also have lower
relative yield increases in crops that are harvested just once
in the season by an accelerated growing degree-day accu-
mulation. In addition, for dryland crops, a higher relative in-
crease in precipitation can lease to higher relative increases
in yields. Irrigated crop growth is simulated assuming that
there is no water limitation and differences in yields tend to
be lower than that of dryland crops.
Water demand (WD) over the irrigated area
Water demand is defined as the irrigation water required by
crops to reach their potential yield. Observed, BC, and NBC
climate produce significantly different estimates on water
demand during the 1980s, i.e. 21.31, 25.36, and 17.75 bil-
lion m3per year, respectively (Table 2). Over the entire PNW,
both BC and NBC climate data predict insignificant changes
in WD during periods of 1980s–2020s and 1980s–2050s (Ta-
ble 2, Figs. 3 and 4). However, there are large spatial varia-
tions in projected WD changes over these periods (Fig. 5g).
Generally, WD decreases in central Washington and south-
ern Idaho during the 2020s. In the 2050s, the irrigated areas
near the mountains show an increase in WD, whereas there
are strong decreases elsewhere, particularly when NBC cli-
mate data are used (Fig. 5g). The magnitude and variations
of WD depend on changes in ET and P. With a higher pro-
jected P, the PNW has a slight decrease in WD over of the
PNW. Because BC data predict a lower ET increase and a
higher P increase, they project a much lower WD during the
2050s period than when using NBC data.
3.2.3 Impacts on biogenic VOC and NO emissions
(MEGAN simulations)
One of several mechanisms by which climate change affects
air quality is by changing emissions of ozone and aerosol
precursors, such as isoprene, monoterpene, and nitrogen ox-
ides nitrogen oxides (NOx=NO+NO2), from the terres-
trial ecosystems. Under projected warmer PNW climate, iso-
prene, monoterpenes, and NO emissions are expected to in-
crease from current emission levels as they are highly de-
pendent on T(Guenther et al., 2012). The observed climate
resulted in annual and area-averaged isoprene emissions of
67 µg m−2h−1over the simulation domain (Table 2). The
highest emissions (>120 µg m−2h−1) occur in the conifer-
dominated forests on the western side of the Cascade Moun-
tains, where temperatures are also the highest (Fig. 5h). The
lowest emissions occur in high-elevation areas, where tem-
peratures are lowest (Fig. 5h). The magnitude of annual
monoterpene emissions is on the same order as isoprene
emissions (59 µg m−2h−1for the whole domain) (Table 2).
The highest monoterpene emissions appear on the west side
of the Cascades, while the lowest occur in central and south-
ern part of the domain, which are dominated by crop, shrub,
and grasslands (Fig. 5i). NO emissions are the highest over
the agricultural areas, with an average rate of 0.3µg m−2h−1
over the whole domain (Fig. 6j).
Driven by BC climate data, MEGAN estimates slightly
lower isoprene, monoterpenes, and NO emission rates by
1.2 % than by observed meteorological data (Table 2). In
contrast, when compared to observed meteorological data,
NBC climate underestimates isoprene, monoterpenes, and
NO emissions by 49, 31, and 36%, respectively.
Biogeosciences, 11, 2601–2622, 2014 www.biogeosciences.net/11/2601/2014/

M. Liu et al.: What is the importance of climate model bias? 2613
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
5.5
JFMAMJJASOND
P(mm/day)
Observed BC1980s
BC2020s BC2050s
NBC1980s NBC2020s
NBC2050s
‐2%
0%
2%
4%
6%
8%
10%
12%
14%
JFMAMJJASOND
P(%)
Diff‐20s Diff‐50s
‐10
‐5
0
5
10
15
20
25
JFMAMJJASOND
T(°C)
‐0.04
‐0.02
0.00
0.02
0.04
0.06
0.08
0.10
0.12
JFMAMJJASOND
T(°C)
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
JFMAMJJASOND
ET(mm/day)
‐6%
‐4%
‐2%
0%
2%
4%
6%
JFMAMJJASOND
ET(%)
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
JFMAMJJASOND
Runoff(mm/day)
‐20%
‐15%
‐10%
‐5%
0%
5%
10%
15%
20%
25%
30%
JFMAMJJASOND
Runoff(%)
0
50
100
150
200
250
300
350
400
JFMAMJJASOND
SWE(mm)
‐10%
0%
10%
20%
30%
40%
50%
JFMAMJJASOND
SWE(mm)
0
1
2
3
4
5
6
7
JFMAMJJASOND
WD(Billionm3)
‐1.0
‐0.5
0.0
0.5
1.0
1.5
2.0
JFMAMJJASOND
WD(Billionm3)
a
b
c
d
e
fl
k
j
i
h
g
Fig. 6. Seasonal patterns of differences in climate and simulated hydrological variables driven by bias-corrected (BC) and non-bias-
corrected (NBC) data. The left column is monthly mean (from January to December) over different scenarios and time periods; the
right column is the differences between BC- and NBC climate and modeled variables in two periods, i.e., 2020s–1980s and 2050s–
1980s. The unit of water demand (WD) is billionm3. Note: plotted for precipitation, ET, runoff, and SWE are differences in per-
centage changes, i.e., (1BC20s−80s/BC1980s×100%)−(1NBC20s−80s/NBC1980s×100%) for changes between 1980s and 2020s and
(1BC50s−80s/BC1980s×100%)−(1NBC50s−80s/NBC1980s×100%) for changes between the 1980s and 2050s; plotted for temperature
and water demand are differences in absolute changes, i.e., 1BC20s−80s−1NBC20s−80sfor changes between the 1980s and 2020s and
1BC50s−80s−1NBC50s−80sfor changes between the 1980s and 2050s. The small stars under or above each column mean pvalue < 0.05
for the Student ttest of differences between BC anomalies and NBC anomalies for each month.
www.biogeosciences.net/11/2601/2014/ Biogeosciences, 11, 2601–2622, 2014

2614 M. Liu et al.: What is the importance of climate model bias?
Both BC and NBC cases project increasing emission
trends in future climate, but the projected emission rates vary
significantly. For example, isoprene emission is projected to
increase from 65 µg m−2h−1in the 1980s to 75 µg m−2h−1
in the 2020s with BC climate vs. 33 µgm−2h−1in the 1980s
to 38 µg m−2h−1in the 2020s with NBC climate. Even
though the magnitude of the estimated emission rates dif-
fer by a factor of 2, the projected percent increases are simi-
lar. Isoprene, monoterpene, and NO emissions are projected
to increase by 14, 9, and 10 % during 1980s–2020s and in-
crease by 43, 28, and 13 % during 1980s–2050s, respectively,
under both BC and NBC climate. Hence, in comparison to
other CCI variables, BC has a small effect on the climate
change signal for the biogenic emissions considered in this
study (Table 2).
3.2.4 Impacts on export of DIN from the land to the
ocean (NEWS simulations)
The concentration of DIN in streams and reservoirs is an
important indicator for water quality and health of aquatic
ecosystems. In this study, Global NEWS simulates the DIN
export from the CRB, which covers 85% of the study domain
(Fig. 1). For the 1980s, NEWS estimates an average DIN
yield (that is, the average DIN leaching from the land that
is eventually exported to the ocean) of 92, 101, and 154kg
N km−2year−1, using observed, BC, and NBC climate in-
puts, respectively (Table 2). Both BC and NBC climate in-
puts result in predicted increases in DIN export for the 2020s
and the 2050s, which closely match changes in Pand runoff
(Table 2). Using NBC climate data results in somewhat lower
percentage increases (+3.6 and +7.3 % for the 2020s and
2050s, respectively), although the differences are not statis-
tically significant (p > 0.7 for all periods examined) (Figs. 3
and 4). Overall, variations of DIN yield and its transport are
highly correlated with changes in Pbecause Pinfluences
runoff, which in turn affects the efficiency with which DIN
is transferred from soils to surface waters, and subsequently
downstream to the coast (Table 2). Of course, temperature
and irrigation demand also have the potential to influence
runoff, but the effects of each of these drivers appear to be
smaller than the effect of precipitation over the region and
time period examined.
3.2.5 Impacts on water and carbon fluxes in a small
forested watershed
Evapotranspiration
For this small watershed study site, HJ-Andrews, observed
meteorological data, and BC and NBC estimates give sim-
ilar Pfor the 1980s baseline. NBC data, however, indicate
substantially lower T(6.6 and 9◦C from NBC and observed
data, respectively). By using observed, BC, and NBC climate
data, RHESSys estimates the mean annual ET to be 831, 876,
and 743 mm year−1, respectively, for the 1980s. The lower T
in NBC climate data (6.6 ◦C) results in lower modeled ET
(Table 2). For the projection of future 1ET (%), BC and
NBC climate produce significant differences over the period
of the 2020s (Figs. 4 and 5, p=0.02); that is, using BC cli-
mate data leads to a decrease of 3.3 %, while using NBC data
leads to an increase of 1.6 % (Table 2, Fig. 6k). In comparing
ET in the 2050s relative to the 1980s baseline, the BC case
predicts a lower ET by 2.7%, while the NBC case shows no
significant change in ET (Table 2, Fig. 6k). RHESSys results
indicate that differences and changes in Pover the wet re-
gions have no significant effects on the estimated ET, while
the difference in Tbaseline could lead to significant differ-
ences.
Net primary production
NPP is commonly used to provide an estimate of the carbon
gained by an ecosystem and develop a carbon balance be-
tween the terrestrial biosphere and the atmosphere (Chapin
et al., 2002; Clark et al., 2001). In HJ-Andrews, NBC cli-
mate data (colder) produce higher NPP than BC (warmer)
and observed climate data (warmer) during the 1980s (Ta-
ble 2). Following the 1ET (%), using BC and NBC inputs
result in large differences in modeled 1NPP (%). With BC
climate data, RHESSys predicts decreases in NPP by 6.5 and
4.4 % during the periods of 1980s–2020s and 1980s–2050s,
respectively (Table 2). In contrast, using NBC climate data,
RHESSys predicts slight increases in NPP for both future pe-
riods (Table 2, Fig. 6l). Although the differences in modeled
NPP as a function of differences between BC and NBC cli-
mate are not statistically significant (p=0.12 and 0.46 for
the 2020s and 2050s, respectively), they show a clear spatial
pattern. Figure 6l shows that large differences in estimated
1NPP (%) between BC and NBC climate data are concen-
trated at higher elevations, whereas the NBC case predicts
increases while BC results in decreases. This result may in-
dicate that contemporary warmer temperatures in the region
(i.e., observed and BC climate data) have reached a critical
threshold where further increases in Twill cause decreases in
NPP. Conversely, under cooler conditions (NBC data), warm-
ing can still enhance vegetation productivity, especially at
higher elevations.
3.3 Impact of BC in estimates of seasonal patterns
To investigate the seasonal shift of CCI on water resources,
we analyze the monthly water fluxes and water demand by
using VIC offline and VIC-CropSyst simulations (Fig. 2).
Biogeosciences, 11, 2601–2622, 2014 www.biogeosciences.net/11/2601/2014/

M. Liu et al.: What is the importance of climate model bias? 2615
-4%
-2%
0%
2%
4%
6%
8%
10%
ET
Runoff
SWE
Irrig. YD
Dry. YD
Total YD
WD
BC P's effects
BC T's effects
Total Differences
-10%
-5%
0%
5%
10%
15%
20%
25%
ET
Runoff
SWE
Irrig. YD
Dry. YD
Total YD
WD
BC P's effects
BC T's effects
Total Differences
2020s-1980s 2050s-1980s
Fig. 7. Contributions of bias corrections (BC) on temperature (T) (Eq. 5) and precipitation (P) (Eq. 7) to the total differences (Eq. 3) of
modeled changes in major hydrologic variables and crop yield between BC and NBC climate driving forces. Left panel: changes between
the 1980s and 2020s; right panel: changes between the 1980s and 2050s. ET: average annual evapotranspiration; Runoff: total runoff; SWE:
snowpack water equivalent on 1 April; Total YD: total yield from all croplands; Irrig. YD: yield from irrigated cropland; Dry. YD: yield from
dryland (non-irrigated cropland); WD: total irrigation water demand over irrigated cropland.
3.3.1 Differences in BC and NBC climate data
Figure 6 depicts the average monthly climate (P&T ), sim-
ulated water fluxes (ET and runoff), SWE, and water de-
mand (WD) with different climate data and periods. BC and
NBC climate data exhibit large discrepancies in summer P,
while they have similar patterns in monthly T(Fig. 6a and
b). BC data for 1P (%) result in greater increases than for
NBC data in almost every month, particularly between May
and October (Fig. 6g). However, the difference between BC
and NBC 1P (%) is not statistically significant (p > 0.05)
due to large interannual variations in monthly P. BC and
NBC cases show only small discrepancies in predictions of
monthly 1T over the periods of 1980s–2020s and 1980s–
2050s (Fig. 6h).
3.3.2 Seasonal patterns of discrepancies between BC
and NBC climate-data-driven changes in water
fluxes
Figure 6d demonstrates that using NBC climate data (in com-
parison to BC and observed data) result in a large overestima-
tion of runoff from May to August due to the high SWE for
this scenario. Lower Tand higher Pfor the NBC case (rel-
ative to the BC case) result in a larger area of snow cover as
well as larger snowpack volumes (Fig. 6e). Irrigation WD is
greater for the BC case than for NBC, particularly from May
to August, mostly due to the much higher projected NBC P
(Fig. 6a and f).
The differences in predicted percentage change in ET,
runoff, SWE, and WD driven by BC and NBC climate data
vary seasonally, as depicted in the right panel of Fig. 6. Gen-
erally, use of BC (vs. NBC) climate data leads to lower 1ET
(%) for the months of June–August in the 2050s (Fig. 6i).
However, the differences are not statistically significant ex-
cept for 2050s July (p=0.04) (Fig. 6i). Use of BC (vs. NBC)
produces significant discrepancies in many months for long-
term predictions of 1Runoff (%) (Fig. 6j); for example, us-
ing 2050s BC climate results in higher 1Runoff (%) in Au-
gust (p < 0.001) and September (p < 0.001) by more than
25 %, whereas this difference is negative for most of the other
months. We report very large differences in monthly 1SWE
(%) projections, which are much larger for the 2050s period
than for the 2020s (Fig. 6k). BC results in larger SWE in-
creases (in comparison to NBC) throughout the cold season,
particularly for the 2050s. For changes in irrigation water
demand, 1WD (%), BC climate results in much larger in-
creases than NBC for the 2020s, although this increase is
significant only for September (Fig. 7l). While the magni-
tude of these differences between BC and NBC is smaller for
the 2050s and vary by month in the sign of this difference,
they are generally statistically significant.
3.4 Relative contributions of Tand Pto the overall
differences between BC and NBC climate change
impacts
We separate out the individual roles of Tand Pto differ-
ences between climate change projections using BC versus
NBC input data sets (Table 1). These tests showed that BC
post-processing of WRF-simulated Pplays a more impor-
tant role than BC post-processing of Tin impacting changes
in runoff, SWE, and dryland crop yield; that is, using NBC
Pfor CCI analyses can lead to underestimation of the in-
creases in runoff and dry land crop yield and overestimation
of the decrease in SWE (Fig. 7). Figure 7 also demonstrates
that BC of Tis the dominant factor in causing the BC and
NBC differences in the projected changes of irrigated crop
yield (recall that over irrigated cropland, we assume that all
crop water requirements are met, reducing the potential role
of P) in the 2020s and 2050s and long-term ET changes in
www.biogeosciences.net/11/2601/2014/ Biogeosciences, 11, 2601–2622, 2014

2616 M. Liu et al.: What is the importance of climate model bias?
ET
SWE
IRR_WD
DIN
-30%
-20%
-10%
0%
10%
20%
30%
40%
50%
60%
-1012345
-30%
-20%
-10%
0%
10%
20%
30%
40%
50%
60%
-20% 0% 20% 40%
-30%
-25%
-20%
-15%
-10%
-5%
0%
5%
10%
15%
20%
25%
-1012345
-30%
-25%
-20%
-15%
-10%
-5%
0%
5%
10%
15%
20%
25%
-20% 0% 20% 40%
-10%
-5%
0%
5%
10%
15%
20%
25%
30%
-1012345
-10%
-5%
0%
5%
10%
15%
20%
25%
30%
-20% 0% 20% 40%
-80%
-60%
-40%
-20%
0%
20%
40%
60%
80%
-1012345
-80%
-60%
-40%
-20%
0%
20%
40%
60%
80%
-20% 0% 20% 40%
a) PNW:∆ET(%)
c)PNW:∆SWE(%)
e)PNW:∆WD(%)
g)CRB:∆DIN(%)
b)PNW:∆ET(%)
d)PNW:∆SWE(%)
f)PNW:∆WD(%)
h)CRB:∆DIN(%)
∆T(°C) ∆P(%)
Fig. 8. Land surface response to climate change with bias-corrected
(BC) and non-bias-corrected (NBC) data between the 1980s and
2020s and between the 1980s and 2050s. Horizontal axes indicate
the anomaly of T(left column, ◦C) or the anomaly of P(right
column, %). Vertical axes indicate anomalies of evapotranspiration
(ET), snowpack water equivalent (SWE), irrigation water demand
(WD), export of dissolved inorganic nitrogen (DIN), and net pri-
mary productivity (NPP) in responding to BC and NBC climate
data. Diamonds represent anomalies according to BC climate and
rectangles represent estimated anomalies driven by NBC climate in
each future year comparing with the base period of the 1980s. The
black lines represent second-order polynomial regression curves for
responses and BC climate, and dashed lines represent second-order
polynomial repression curves for responses and NBC climate. PNW
(Pacific Northwest) represents the domain for ET, SWE, and WD
estimations from VIC or VIC-CropSyst models; CRB represents
the simulation domain of NEWS model; and HJ-Andrews repre-
sents the simulation domain of RHESSys on ET and NPP (Fig. 1).
the 2050s, indicating a primarily energy-limited ET regime
(Liu et al., 2013).
The impact of BC and NBC depends on the distance of
baseline and climate change to environmental threshold (e.g.,
0◦C for snowpack accumulations) that determines the lin-
-60%
-40%
-20%
0%
20%
40%
60%
80%
-1012345
-60%
-40%
-20%
0%
20%
40%
60%
80%
-20% 0% 20% 40% 60%
-30%
-25%
-20%
-15%
-10%
-5%
0%
5%
10%
15%
20%
25%
-1012345
-30%
-25%
-20%
-15%
-10%
-5%
0%
5%
10%
15%
20%
25%
-20% 0% 20% 40% 60%
∆T(°C) ∆P(%)
i)HJ‐Andrews:∆ET(%) j)HJ‐Andrews:∆ET(%)
k)HJ‐Andrews:∆NPP(%) l)HJ‐Andrews:∆NPP(%)
Fig. 8. Continued.
ear or non-linear responses of land surface and ecosystems
to climate change. Because of non-linear responses of terres-
trial ecosystems to climate change, the same amount of abso-
lute change in Tand Pmay produce significant differences
in the response of hydrologic and biogeochemical processes
when they start from different baselines. Figure 8 demon-
strates that in responding to anomalies of Tand Pin the fu-
ture, the land surface shows different sensitivities to BC and
NBC climate. According to modeled results, ET, SWE, and
DIN show higher sensitivities to the baseline climate condi-
tion than other variables. As expected, the Tbaseline plays
a strong role in determining SWE changes (Fig. 8c), while
DIN changes are most sensitive to Pgradients (Fig. 8h),
and ET changes are equally sensitive to gradients in both
Tand P(Figs. 8a and b). Furthermore, BC and NBC cli-
mates produce different response curves for each of the im-
pact variables. For example, with increasing T, using NBC
climate data produces a linearly increasing trend in total ET,
while using BC climate data results in a relative leveling-
off of changes at high-Tanomalies over the PNW (Fig. 8a);
with increasing P, NBC shows a leveling-off in increasing
ET, while NB climate predicts a continuing linear increase
along with increased P(Fig. 8b). The responding curve,
however, is spatial-explicit and scale-depended. For instance,
The slope of the response curve of ÄET to ÄT and ÄP over
the small watershed (i.e. HJ-Andrews) is much smaller than
that over the entire PNW, while the difference in responding
to BC and NBC climate data is larger over the small wa-
tershed than the entire region (Fig. 8a-b and i–j). These re-
sults indicate that the responses of the land surface to climate
change depend on the magnitude of the change, the baseline
climate, and location.
Biogeosciences, 11, 2601–2622, 2014 www.biogeosciences.net/11/2601/2014/

M. Liu et al.: What is the importance of climate model bias? 2617
4 Discussion
Our analysis indicates that the choice of bias correcting or
not bias correcting downscaled global GCM projected cli-
mate could affect estimates of climate change impacts in the
future. Non-linear responses of hydrologic and biogeochem-
ical processes to climate change can partly explain this phe-
nomenon. The spatial variations in different quasi-equilibria
among local biotic and abiotic environments such as climate,
vegetation, geology, topology, etc. introduce more complex-
ities into these uncertainties in regional assessments. There-
fore, when using projected results from GCM and regional
climate models, uncertainty analysis should be conducted,
particularly in situations where some post-processing proce-
dures such as BC, which is based on contemporary observa-
tional data, are involved.
Unrealistic land surface responses from not performing
BC of modeled climate prior to driving impact models may
be obvious in some cases. Failing to account for BC im-
pacts on snow processes could produce misleading projec-
tions on future water availability, an issue of importance to
both ecosystems and society. Using BC (vs. NBC) climate
data in our study would result in fewer perceived threats
to society and ecosystems, less shrinkage of the snowpack,
higher crop yields, and lower water demand. Furthermore,
while our study suggested that BC may not be necessary for
projecting the impacts of climate change on the magnitude
of some biogenic emissions, there are important air quality
implications to not performing BC. For example, underesti-
mation of biogenic VOCs emissions may affect the predic-
tion of ozone as well as secondary organic aerosol (SOA)
and thus fine particulate matter (PM2.5)concentrations, with
ozone and PM2.5concentrations being the key decision-
making variables for air quality management. Assessing how
bias in climate model results affects projected changes in
ozone, SOA, and PM2.5requires detailed analysis with the
use of chemical transport models that not only accounts for
changing emission rates but also explicitly simulate the non-
linear, meteorology-dependent atmospheric chemistry, trans-
port, and deposition processes. To our knowledge, no tech-
niques have been developed to bias correct three-dimensional
RCM output, and therefore no studies have applied bias cor-
rection to the input of chemical transport models.
As often the case, CCI simulations performed here use
models that are run offline from each other and thus feed-
backs between subcomponents of the Earth system are not
addressed. The following are some examples of feedbacks
to the atmosphere that are not are considered in CCI studies
of this kind: ET and WD affect soil moisture and thus la-
tent heat transfer, SWE affects surface albedo and thus radia-
tion, VOC and NOxaffect SOA formation and thus aerosol–
radiation and aerosol–cloud–precipitation interactions, and
NPP affects carbon and thus atmospheric CO2levels. In-
deed, not accounting for feedbacks may contribute to some of
the biases that have necessitated the need for bias correction
in the first place. To consider these feedbacks requires fully
coupled models for which bias correction would break fun-
damental laws of physics, defeating the purpose of process-
based models. Coupled models are useful for formulating hy-
potheses on which of the feedback processes have important
consequences for decision-making variables and for design-
ing experimental studies that can quantify these feedbacks
and improve models.
There are other limitations in this study. For example, we
only quantified the effects of BC of Tand Pon CCI of
regional hydrology, agricultural activities, DIN export, and
biogenic VOC and NO emissions. Other related climate fac-
tors including wind speed, relative humidity, and radiation
are either based on reanalysis data or modeled by the VIC
model (Hamlet et al., 2007; Maurer et al., 2002; Salathe
et al., 2013). Even though the derived climate factors from
Tand Pby using MT-CLIM model have been evaluated
against field observations (Bohn et al., 2013; Thornton and
Running, 1999), the spatiotemporal relationships within cli-
mate variables in contemporary periods may change over the
long-term future. Therefore, by only using information of T
and Pchanges, CCI studies may produce large uncertainties.
Each bias correction and downscaling approach may have in-
dividual strengths and weaknesses in reconstructing the spa-
tial and temporal patterns of Tand P, as well as extreme
events (Maurer and Hidalgo, 2008; Quintana Segui et al.,
2010; Themessl et al., 2011). Estimating the sensitivities of
different downscaling methods on hydrologic and ecological
impacts is meaningful and necessary for quantifying the un-
certainties from certain downscaled climate data in regional
applications. This effort could be an extension of current in-
tercomparisons on global GCMs and activities on model–
data intercomparisons over carbon, water, and crops at con-
tinental and global scales (Asseng et al., 2013; Huntzinger et
al., 2012; Rosenzweig et al., 2013; Taylor et al., 2012).
5 Conclusions
Herein, we quantify the effects of bias correction of climate
model output on climate change impact projections of re-
gional and watershed-scale hydrologic and biogeochemical
processes. As expected (due to the BC methodology), us-
ing BC climate data produced almost the same simulated
results as using gridded meteorological observations for the
historical time period. Without BC, however, the direct use
of modeled climate data by land surface models produced
very different hydrologic and biogeochemical results over
the historical period. While we anticipated that these differ-
ences would be large, an interesting question is the degree to
which the response to a climate change signal is preserved,
even if the baseline climate conditions are not. This is an im-
portant question because, in fully coupled land–atmosphere
schemes, BC is generally not performed, so that dynamical
consistency in simulated variables is retained. In doing this,
www.biogeosciences.net/11/2601/2014/ Biogeosciences, 11, 2601–2622, 2014

2618 M. Liu et al.: What is the importance of climate model bias?
a fundamental assumption is that models can predict the rel-
ative change of a variable of interest in response to climate
change reasonably well, even if model prediction of the ac-
tual value of the variable is biased. Herein, we test whether
or not this assumption holds true and for which land surface
variables.
Due to the conservation of absolute change of Tand per-
centage rate of change in Pduring BC post-processing, pro-
jected BC and NBC climate data produce somewhat similar
results in the percentage rate of change in many of our re-
sponse variables, including ET, runoff, total crop yield, irri-
gated water demand, VOC emissions (isoprene and monoter-
penes), NO emission, and DIN river export over the PNW
over the time periods of 1980s–2020s and 1980s–2050s.
However, there a few important variables for which BC does
have a large impact in the response to climate change, no-
tably SWE and dryland (non-irrigated) crop yield. Both of
these variables are key decision variables for managing our
natural and agricultural resources. Overall, not performing
BC would result in an overestimation of the decrease of SWE
and an underestimation of the increase in dryland crop yield
due to climate change, thus painting a more dire portrait of
future conditions than would be suggested by using BC data.
Considering all variables under BC in projecting responses to
climate change, there is the potential for a large range of ef-
fects when performing land surface simulations with biased
climate inputs.
We conclude that there are trade-offs between using BC
climate data for offline CCI studies versus applying coupled
regional Earth system models that retain dynamical consis-
tency between variables and capture feedback effects. These
trade-offs should be considered when designing integrated
modeling frameworks for specific applications; for exam-
ple, BC may be more important when considering impacts
on reservoir operations in mountainous watersheds (where
1 April SWE is an important decision factor) than when
investigating impacts on biogenic emissions and air qual-
ity (where VOCs are a primary indicator). However, even
in these instances where BC may not be deemed important,
there may be some important negative consequences to not
correcting for bias, such as a host of air quality effects and
aerosol–radiation–cloud interactions caused by projecting bi-
ased values of biogenic emissions.
Acknowledgements. This study has been supported by the United
States Department of Agriculture (grant no. 20116700330346 for
the Biosphere-relevant Earth system modeling project, BioEarth)
and the National Science Foundation (grant no. 0903714 for Nitro-
gen Systems: Policy-oriented Integrated Research and Education
program, NSPIRE).
Edited by: S. M. Noe
References
Adam, J. C., Hamlet, A. F., and Lettenmaier, D. P.: Implications of
global climate change for snowmelt hydrology in the twenty-first
century, Hydrol. Proc., 23, 962–972, 2009.
Adam, J. C., Stephens, J. C., Chung, S. H., Brady, M. P., Evans,
R. D., Kruger, C. E., Lamb, B. K., Liu, M. L., Stöckle, C. O.,
Vaughan, J. K., Rajagopalan, K., Harrison, J. A., Tague, C. L.,
Kalyanaraman, A., Chen, Y., Guenther, A., Leung, F. Y., Le-
ung, L. R., Perleberg, A. B., Yoder, J., Allen, E., Anderson, S.,
Chandrasekharan, B., Malek, K., Mullis, T., Miller, C., Nergui,
T., Poinsatte, J., Reyes, J., Zhu, J., Choate, J. S., Jiang, X., Nel-
son, R., Yoon, J. H., Yorgey, G. G., Johnson, K., Chinnayakana-
halli, K. J., Hamlet, A. F., Nijssen, B.,and Walden, V.: BioEarth:
Envisioning and developing a new regional earth system model
to inform natural and agricultural resource management, Clim.
Change, doi:10.1007/s10584-014-1115-2, 2014.
Andreadis, K. M., Storck, P., and Lettenmaier, D. P.: Modeling snow
accumulation and ablation processes in forested environments,
Water Resour. Res., 45, W05429, doi:10.1029/2008WR007042,
2009.
Asseng, S., Ewert, F., Rosenzweig, C., Jones, J. W., Hatfield, J. L.,
Ruane, A. C., Boote, K. J., Thorburn, P. J., Rotter, R. P., Cam-
marano, D., Brisson, N., Basso, B., Martre, P., Aggarwal, P. K.,
Angulo, C., Bertuzzi, P., Biernath, C., Challinor, A. J., Doltra, J.,
Gayler, S., Goldberg, R., Grant, R., Heng, L., Hooker, J., Hunt,
L. A., Ingwersen, J., Izaurralde, R. C., Kersebaum, K. C., Muller,
C., Naresh Kumar, S., Nendel, C., O/’Leary, G., Olesen, J. E.,
Osborne, T. M., Palosuo, T., Priesack, E., Ripoche, D., Semenov,
M. A., Shcherbak, I., Steduto, P., Stockle, C., Stratonovitch, P.,
Streck, T., Supit, I., Tao, F., Travasso, M., Waha, K., Wallach, D.,
White, J. W., Williams, J. R. and Wolf, J.: Uncertainty in simu-
lating wheat yields under climate change, Nature Clim. Change,
3, 827–832, 2013.
Avise, J., Chen, J., Lamb, B., Wiedinmyer, C., Guenther, A.,
Salathé, E., and Mass, C.: Attribution of projected changes
in summertime US ozone and PM2.5 concentrations to global
changes, Atmos. Chem. Phys., 9, 1111–1124, doi:10.5194/acp-
9-1111-2009, 2009.
Avise, J., Abraham, R. G., Chung, S. H., Chen, J., Lamb, B.,
Salathe, E. P., Zhang, Y., Nolte, C. G., Loughlin, D. H., Guen-
ther, A., Wiedinmyer, C., and Duhl, T.: Evaluating the effects of
climate change on summertime ozone using a relative response
factor approach for policymakers, J. Air Waste Manage. Assoc.,
62, 1061–1074, 2012.
Barnett, T., Adam, J., and Lettenmaier, D.: Potential impacts of
a warming climate on water availability in snow-dominated re-
gions, NATURE, 438, 303–309, 2005.
Berg, P., Haerter, J. O., Thejll, P., Piani, C., Hagemann,
S., and Christensen, J. H.: Seasonal characteristics of the
relationship between daily precipitation intensity and sur-
face temperature, J. Geophys. Res.-Atmos., 114, D18102,
doi:10.1029/2009JD012008, 2009.
Bohn, T. J., Livneh, B., Oyler, J. W., Running, S. W., Nijssen, B.,
and Lettenmaier, D. P.: Global evaluation of MTCLIM and re-
lated algorithms for forcing of ecological and hydrological mod-
els, Agr. Forest. Meteorol., 176, 38–49, 2013.
Bouwman, A. F., van Grinsven, J. J. M., and Eickhout, B.: Conse-
quences of the cultivation of energy crops for the global nitrogen
cycle, Ecol. Appl., 20, 101–109, doi:10.1890/08-0608.1, 2010.
Biogeosciences, 11, 2601–2622, 2014 www.biogeosciences.net/11/2601/2014/

M. Liu et al.: What is the importance of climate model bias? 2619
Bowling, L. C. and Lettenmaier, D. P.: Modeling the Effects of
Lakes and Wetlands on the Water Balance of Arctic Environ-
ments, J. Hydrometeorol., 11, 276–295, 2010.
Bromwich, D. H., Otieno, F. O., Hines, K. M., Manning, K., and
Shilo, E.: Comprehensive evaluation of polar weather research
and forecasting model performance in the Arctic, J. Geophys.
Res.-Atmos., 118, 274–292, 2013.
Buser, C. M., Kuensch, H. R., Luethi, D., Wild, M., and Schaer, C.:
Bayesian multi-model projection of climate: bias assumptions
and interannual variability, Clim. Dynam., 33, 849–868, 2009.
Chapin, F. S., Matson, P. A., and Mooney, H. A.: Principles of ter-
restrial ecosystem ecology, Springer-Verlag, New York, USA,
2002.
Chen, C., Haerter, J. O., Hagemann, S., and Piani, C.: On the con-
tribution of statistical bias correction to the uncertainty in the
projected hydrological cycle, Geophys. Res. Lett., 38, L20403,
doi:10.1029/2011GL049318, 2011.
Chen, F. and Dudhia, J.: Coupling an advanced land surface-
hydrology model with the Penn State-NCAR MM5 modeling
system, Part I: Model implementation and sensitivity, Monthly
Weather Rev., 129, 569–585, 2001.
Chen, J., Avise, J., Guenther, A., Wiedinmyer, C., Salathe, E., Jack-
son, R. B., and Lamb, B.: Future land use and land cover in-
fluences on regional biogenic emissions and air quality in the
United States, Atmos. Environ., 43, 5771–5780, 2009a.
Chen, J., Avise, J., Lamb, B., Salathé, E., Mass, C., Guenther, A.,
Wiedinmyer, C., Lamarque, J.-F., O’Neill, S., McKenzie, D., and
Larkin, N.: The effects of global changes upon regional ozone
pollution in the United States, Atmos. Chem. Phys., 9, 1125–
1141, doi:10.5194/acp-9-1125-2009, 2009b.
Cherkauer, K. and Lettenmaier, D.: Simulation of spatial variability
in snow and frozen soil, J. Geophys. Res.-Atmos., 108, D228858,
doi:10.1029/2003JD003575, 2003.
Christensen, L., Tague, C. L., and Baron, J. S.: Spatial patterns of
simulated transpiration response to climate variability in a snow
dominated mountain ecosystem, Hydrol. Proc., 22, 3576–3588,
2008.
Clark, D., Brown, S., Kicklighter, D., Chambers, J., Thomlinson,
J., and Ni, J.: Measuring net primary production in forests: Con-
cepts and field methods, Ecol. Appl., 11, 356–370, 2001.
Collins, W. D., Rasch, P. J., Boville, B. A., Hack, J. J., McCaa, J.
R., Williamson, D. L., Kiehl, J. T., Briegleb, B., Bitz, C., Lin,
S.-J., Zhang, M., and Dai, Y.: Description of the NCAR Commu-
nity Atmospheric Model (CAM 3.0), National Center for Atmo-
spheric Research, Boulder, CO, 2004.
Dawson, J. P., Racherla, P. N., Lynn, B. H., Adams, P. J., and Pandis,
S. N.: Impacts of climate change on regional and urban air quality
in the eastern United States: Role of meteorology, J. Geophys.
Res.-Atmos., 114, D05308, doi:10.1029/2008JD009849, 2009.
Dee, D. P., Kaellen, E., Simmons, A. J., and Haimberger, L.: Com-
ments on “Reanalyses Suitable for Characterizing Long-Term
Trends,” Bull. Am. Meteorol. Soc., 92, 65–70, 2011a.
Dee, D. P., Uppala, S. M., Simmons, A. J., Berrisford, P., Poli,
P., Kobayashi, S., Andrae, U., Balmaseda, M. A., Balsamo, G.,
Bauer, P., Bechtold, P., Beljaars, A. C. M., van de Berg, L., Bid-
lot, J., Bormann, N., Delsol, C., Dragani, R., Fuentes, M., Geer,
A. J., Haimberger, L., Healy, S. B., Hersbach, H., Holm, E. V.,
Isaksen, L., Kallberg, P., Koehler, M., Matricardi, M., McNally,
A. P., Monge-Sanz, B. M., Morcrette, J.-J., Park, B.-K., Peubey,
C., de Rosnay, P., Tavolato, C., Thepaut, J.-N., and Vitart, F.:
The ERA-Interim reanalysis: configuration and performance of
the data assimilation system, Quart. J. Roy. Meteorol. Soc., 137,
553–597, 2011b.
Duliere, V., Zhang, Y., and Salathe, Jr., E. P.: Extreme Precip-
itation and Temperature over the U.S. Pacific Northwest: A
Comparison between Observations, Reanalysis Data, and Re-
gional Models, JOURNAL OF CLIMATE, 24(7), 1950–1964,
doi:10.1175/2010JCLI3224.1, 2011.
Dumont, E., Harrison, J. A., Kroeze, C., Bakker, E. J., and
Seitzinger, S. P.: Global distribution and sources of dissolved
inorganic nitrogen export to the coastal zone: Results from a
spatially explicit, global model, Global Biogeochem. Cy., 19,
GB4S02, doi:10.1029/2005GB002488, 2005.
Ehret, U., Zehe, E., Wulfmeyer, V., Warrach-Sagi, K., and Liebert,
J.: Should we apply bias correction to global and regional climate
model data?, Hydrol. Earth Syst. Sci., 16, 3391–3404, 2012,
http://www.hydrol-earth-syst-sci.net/16/3391/2012/.
Elsner, M. M., Cuo, L., Voisin, N., Deems, J. S., Hamlet, A. F.,
Vano, J. A., Mickelson, K. E. B., Lee, S.-Y., and Lettenmaier, D.
P.: Implications of 21st century climate change for the hydrology
of Washington State, Clim. Change, 102, 225–260, 2010.
Folland, C. and Anderson, C.: Estimating changing extremes using
empirical ranking methods, J. Climate, 15, 2954–2960, 2002.
Guenther, A. B., Jiang, X., Heald, C. L., Sakulyanontvittaya, T.,
Duhl, T., Emmons, L. K., and Wang, X.: The Model of Emissions
of Gases and Aerosols from Nature version 2.1 (MEGAN2.1): an
extended and updated framework for modeling biogenic emis-
sions, Geosci. Model Develop., 5, 1471–1492, 2012.
Haerter, J. O., Hagemann, S., Moseley, C., and Piani, C.: Climate
model bias correction and the role of timescales, Hydrol. Earth
Syst. Sci., 15, 1065–1079, 2011,
http://www.hydrol-earth-syst-sci.net/15/1065/2011/.
Hagemann, S., Chen, C., Haerter, J. O., Heinke, J., Gerten, D., and
Piani, C.: Impact of a Statistical Bias Correction on the Projected
Hydrological Changes Obtained from Three GCMs and Two Hy-
drology Models, J. Hydrometeorol., 12, 556–578, 2011.
Hamlet, A. and Lettenmaier, D.: Production of temporally consis-
tent gridded precipitation and temperature fields for the continen-
tal United States, J. Hydrometeorol., 6, 330–336, 2005.
Hamlet, A., Mote, P., Clark, M., and Lettenmaier, D.: Effects of
temperature and precipitation variability on snowpack trends in
the western United States, J. Climate, 18, 4545–4561, 2005.
Hamlet, A. F.: Assessing water resources adaptive capacity to cli-
mate change impacts in the Pacific Northwest Region of North
America, Hydrol. Earth Syst. Sci., 15, 1427–1443, 2011,
http://www.hydrol-earth-syst-sci.net/15/1427/2011/.
Hamlet, A. F., Mote, P. W., Clark, M. P., and Lettenmaier, D. P.:
Twentieth-century trends in runoff, evapotranspiration, and soil
moisture in the western United States, J. Climate, 20, 1468–1486,
2007.
Hamlet, A. F., Carrasco, P., Deems, J., Elsner, M. M., Kamstra,
T., Lee, C., Lee, S.-Y., Mauger, G., Salathe, P., Tohver, I., and
Binder, L. W.: Final Project Report for the Columbia Basin Cli-
mate CHange Scenarios Project. available at: http://www.hydro.
washington.edu/2860/report, 2010.
Hamlet, A. F., Elsner, M. M., Mauger, G., Lee, S.-Y., and Tohver,
I.: An overview of the Columbia basin climate change scenarios
www.biogeosciences.net/11/2601/2014/ Biogeosciences, 11, 2601–2622, 2014

2620 M. Liu et al.: What is the importance of climate model bias?
project: approach, methods, and summary of key results, Atmos.-
Oc., 2012.
Harrison, J. A., Caraco, N., and Seitzinger, S. P.: Global patterns
and sources of dissolved organic matter export to the coastal
zone: Results from a spatially explicit, global model, Global
Biogeochem. Cy., 19, GB4S04, doi:10.1029/2005GB002480,
2005a.
Harrison, J. A., Seitzinger, S. P., Bouwman, A. F., Caraco, N. F.,
Beusen, A. H. W., and Vorosmarty, C. J.: Dissolved inorganic
phosphorus export to the coastal zone: Results from a spatially
explicit, global model, Global Biogeochem. Cy., 19, GB4S03,
doi:10.1029/2004GB002357, 2005b.
Harrison, J. A., Bouwman, A. F., Mayorga, E., and Seitzinger,
S.: Magnitudes and sources of dissolved inorganic phospho-
rus inputs to surface fresh waters and the coastal zone: A
new global model, Global Biogeochem. Cy., 24, GB1003,
doi:10.1029/2009GB003590, 2010.
Hoffmann, H. and Rath, T.: Meteorologically consistent bias cor-
rection of climate time series for agricultural models, Theoret.
Appl. Climatol., 110, 129–141, 2012.
Hong, S. and Pan, H.: Nonlocal boundary layer vertical diffusion in
a Medium-Range Forecast Model, Monthly Weather Rev., 124,
2322–2339, 1996.
Hong, S., Dudhia, J., and Chen, S.: A revised approach to ice mi-
crophysical processes for the bulk parameterization of clouds and
precipitation, Monthly Weather Rev., 132, 103–120, 2004.
Hungerford, R., Nemani, R., Running, S. W., and Coughlan, J. C.:
MTCLIM – A mountain microclimate simulation-model, USDA
Forest Service Intermountain Research Station Research Paper,
1–52, 1989.
Huntzinger, D. N., Post, W. M., Wei, Y., Michalak, A. M., West,
T. O., Jacobson, A. R., Baker, I. T., Chen, J. M., Davis, K. J.,
Hayes, D. J., Hoffman, F. M., Jain, A. K., Liu, S., McGuire, A.
D., Neilson, R. P., Potter, C., Poulter, B., Price, D., Raczka, B.
M., Tian, H. Q., Thornton, P., Tomelleri, E., Viovy, N., Xiao, J.,
Yuan, W., Zeng, N., Zhao, M., and Cook, R.: North American
Carbon Program (NACP) regional interim synthesis: Terrestrial
biospheric model intercomparison, Ecol. Modell., 232, 144–157,
2012.
Johnson, F. and Sharma, A.: A nesting model for bias correc-
tion of variability at multiple time scales in general circula-
tion model precipitation simulations, Water Resour. Res., 48,
W01504, doi:10.1029/2011WR010464, 2012.
Kain, J. S. and Fritsch, J. M.: Convective parameterization for
mesoscale models: the Kain–Fritsch scheme, in The representa-
tion of cumulus convection in numerical models, Am. Meteorol.
Soc., 24, 165–170, 1993.
Kimball, J., Running, S., and Nemani, R.: An improved method for
estimating surface humidity from daily minimum temperature,
Agr. Forest Meteorol., 85, 87–98, 1997.
Lam, Y. F., Fu, J. S., Wu, S., and Mickley, L. J.: Impacts of fu-
ture climate change and effects of biogenic emissions on surface
ozone and particulate matter concentrations in the United States,
Atmos. Chem. Phys., 11, 4789–4806, 2011,
http://www.atmos-chem-phys.net/11/4789/2011/.
Liang, X., Lettenmaier, D. P., Wood, E. F., and Burges, S. J.: A sim-
ple hydrologically based model of land-surface water and energy
fluxes for general-circulation models, J. Geophys. Res.-Atmos.,
99, 14415–14428, 1994.
Liu, M., Adam, J. C., and Hamlet, A. F.:Spatial-temporal variations
of evapotranspiration and runoff/precipitation ratios responding
to the changing climate in the Pacific Northwest during 1921–
2006, J. Geophys. Res., 118, 380–394, 2013.
Livezey, R. E. and Chen, W.: Statistical field significance and its de-
termination by monte-carlo techniques, Monthly Weather Rev.,
111, 46–59, 1983.
Maraun, D.: Nonstationarities of regional climate model biases in
European seasonal mean temperature and precipitation sums,
Geophys. Res. Lett., 39, L06706, doi:10.1029/2012GL051210,
2012.
Marsland, S., Haak, H., Jungclaus, J., Latif, M., and Roske, F.: The
Max-Planck-Institute global ocean/sea ice model with orthogo-
nal curvilinear coordinates, Ocean Modell., 5, 91–127, 2003.
Maurer, E. P. and Hidalgo, H. G.: Utility of daily vs. monthly large-
scale climate data: an intercomparison of two statistical down-
scaling methods, Hydrol. Earth Syst. Sci., 12, 551–563, 2008,
http://www.hydrol-earth-syst-sci.net/12/551/2008/.
Maurer, E. P., Wood, A. W., Adam, J. C., Lettenmaier, D. P., and
Nijssen, B.: A long-term hydrologically based dataset of land
surface fluxes and states for the conterminous United States, J.
Climate, 15, 3237–3251, 2002.
Mayorga, E., Seitzinger, S. P., Harrison, J. A., Dumont, E., Beusen,
A. H. W., Bouwman, A. F., Fekete, B. M., Kroeze, C., and Van
Drecht, G.: Global Nutrient Export from WaterSheds 2 (NEWS
2): Model development and implementation, Environ. Modell.
Software, 25, 837–853, 2010.
McCrackin, M. L., Harrison, J. A., and Compton, J. E.: A com-
parison of NEWS and SPARROW models to understand sources
of nitrogen delivered to US coastal areas, Biogeochemistry, 114,
281–297, 2013.
Meentemeyer, R. K. and Moody, A.: Distribution of plant life his-
tory types in California chaparral: the role of topographically-
determined drought severity, J. Vegetat. Sci., 13, 67–78, 2002.
Mote, P.: Trends in snow water equivalent in the Pacific North-
west and their climatic causes, Geophys. Res. Lett., 30, 1601,
doi:10.1029/2003GL017258, 2003.
Muerth, M. J., Gauvin St-Denis, B., Ricard, S., Velázquez, J. A.,
Schmid, J., Minville, M., Caya, D., Chaumont, D., Ludwig, R.,
and Turcotte, R.: On the need for bias correction in regional cli-
mate scenarios to assess climate change impacts on river runoff,
Hydrol. Earth Syst. Sci., 17, 1189–1204, doi:10.5194/hess-17-
1189-2013, 2013.
Nakicenovic, N. and Swart, R., (Eds.): IPCC Special Report Emis-
sion Scenarios, Cambridge University Press, UK, 2000.
National Research Council: Managing the Columbia River: In-
stream flows, water withdrawals, and salmon survival, The Na-
tional Academies Press, Washington, DC, 2004.
Nolte, C. G., Gilliland, A. B., Hogrefe, C., and Mickley, L. J.: Link-
ing global to regional models to assess future climate impacts
on surface ozone levels in the United States, J. Geophys. Res.-
Atmos., 113, D14307, doi:10.1029/2007JD008497, 2008.
Parton, W., Scurlock, J., Ojima, D., Gilmanov, T., Scholes, R.,
Schimel, D., Kirchner, T., Menaut, J., Seastedt, T., Moya, E.,
Kamnalrut, A., and Kinyamario, J.: Observations and modeling
of biomass and soil organic-matter dynamics for the grassland
biome worldwide, Global Biogeochem. Cy., 7, 785–809, 1993.
Peralta, J. M. and Stockle, C. O.: Nitrate from an irrigated crop
rotation at the Pasco-Quincy area (Washington, USA) available
Biogeosciences, 11, 2601–2622, 2014 www.biogeosciences.net/11/2601/2014/

M. Liu et al.: What is the importance of climate model bias? 2621
for groundwater contamination: A long-term simulation study,
Agriculture, Ecosys. Environ., 88, 23–24, 2002.
Piani, C. and Haerter, J. O.: Two dimensional bias correction of tem-
perature and precipitation copulas in climate models, Geophys.
Res. Lett., 39, L20401, doi:10.1029/2012GL053839, 2012.
Piani, C., Weedon, G. P., Best, M., Gomes, S. M., Viterbo, P., Hage-
mann, S., and Haerter, J. O.: Statistical bias correction of global
simulated daily precipitation and temperature for the application
of hydrological models, J. Hydrol., 395, 199–215, 2010.
Quintana Segui, P., Ribes, A., Martin, E., Habets, F. and Boe, J.:
Comparison of three downscaling methods in simulating the im-
pact of climate change on the hydrology of Mediterranean basins,
J. Hydrol., 383, 111–124, 2010.
Roeckner, E., Bengtsson, L., Feichter, J., Lelieveld, J., and
Rodhe, H.: Transient climate change simulations with a coupled
atmosphere-ocean GCM including the tropospheric sulfur cycle,
J. Climate, 12, 3004–3032, 1999.
Roeckner, E., Bauml, G., Bonaventura, L., Brokopf, R., Esch,
M., Giorgetta, M., Hagemann, S., Kirchner, I., Kornblueh,
L., Manzini, E., Rhodin, A., Schlese, U., Schulzweida, U.,
and Tompkins, A.: The atmospheric general circulation model
ECHAM5, Part I: Model descroption, Max Planck Institute for
Meteorology, Hamburg, 2003.
Rosenzweig, C., Jones, J. W., Hatfield, J. L., Ruane, A. C., Boote,
K. J., Thorburne, P., Antle, J. M., Nelson, G. C., Porter, C.,
Janssen, S., Asseng, S., Basso, B., Ewert, F., Wallach, D., Baig-
orria, G.. and Winter, J. M.: The Agricultural Model Intercom-
parison and Improvement Project (AgMIP): Protocols and pilot
studies, Agricult. Forest Meteorol., 170, 166–182, 2013.
Salathe, P., Hamlet, A. F., Stumbaugh, M., Lee, S.-Y., and Steed,
R.: Estimates of 21st Century flood risk in the Pacific Northwest
based on regional scale climate model simulations, J. Hydrome-
teorol., in press, doi:10.1175/JHM-D-13-0137.1, 2013.
Seitzinger, S., Kroeze, C., Bouwman, A., Caraco, N., Dentener,
F. and Styles, R.: Global patterns of dissolved inorganic and
particulate nitrogen inputs to coastal systems: Recent condi-
tions and future projections, ESTUARIES, 25(4B), 640–655,
doi:10.1007/BF02804897, 2002.
Seitzinger, S. P., Harrison, J. A., Dumont, E., Beusen, A. H.
W., and Bouwman, A. F.: Sources and delivery of carbon,
nitrogen, and phosphorus to the coastal zone: An overview
of Global Nutrient Export from Watersheds (NEWS) models
and their application, Global Biogeochem. Cy., 19, GB4S01,
doi:10.1029/2005GB002606, 2005.
Seitzinger, S. P., Mayorga, </