Evaluation and Uncertainty Analysis of the Land Surface Hydrology in LS3MIP Models Over China

Abstract

The Land Surface, Snow and Soil moisture Model Intercomparison Project (LS3MIP) offers valuable land surface hydrology products from the land modules of current Earth system models (ESMs). Historical hydrological variables from six ESMs driven by four meteorological forcing data sets (GSWP, WFDEI, CRU‐NCEP, and Princeton) in Land Model Intercomparison Project (LMIP) have been extensively evaluated with various high‐quality reference data sets over Chinese mainland. Compared with the reference data sets, the multi‐model ensemble means (MMEs) of most hydrological variables are underestimated, while their annual trends show high spatial consistency, with sign consistency over 56%–85% of land area. After computing and ranking four statistical metrics (bias, correlation coefficient, normalized standard deviation, and unbiased root‐mean‐square biases) between simulations and references, it is found that the CLM5 has the best performance, while the GSWP3 exhibits the highest quality. Furthermore, the analysis of variance method (ANOVA) is then used to trace sources (model, atmospheric forcing data sets and their interactions) of the uncertainty of those modeling hydrological variables for 1900–2012 (1948–2012 for runoff) over China. The results indicate that the total uncertainty and its composition vary with time and decrease significantly in recent decades, reflecting the enhanced forcing data quality. Larger forcing uncertainty existed during the early twentieth century because less available observation data sets have been adopted to constrain climate variables. For all modeling hydrological variables, the model uncertainty plays the dominant role, suggesting that the quality of LMIP products largely relies on Land surface models.

Full text

Evaluation and Uncertainty Analysis of the Land Surface
Hydrology in LS3MIP Models Over China
Xin Ma
1
and Aihui Wang
1,2
1
Nansen‐Zhu International Research Centre, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing,
China,
2
Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters (CIC‐FEMD), Nanjing
University of Information Science & Technology, Nanjing, China
Abstract The Land Surface, Snow and Soil moisture Model Intercomparison Project (LS3MIP) offers
valuable land surface hydrology products from the land modules of current Earth system models (ESMs).
Historical hydrological variables from six ESMs driven by four meteorological forcing data sets (GSWP,
WFDEI, CRU‐NCEP, and Princeton) in Land Model Intercomparison Project (LMIP) have been extensively
evaluated with various high‐quality reference data sets over Chinese mainland. Compared with the reference
data sets, the multi‐model ensemble means (MMEs) of most hydrological variables are underestimated, while
their annual trends show high spatial consistency, with sign consistency over 56%–85% of land area. After
computing and ranking four statistical metrics (bias, correlation coefficient, normalized standard deviation, and
unbiased root‐mean‐square biases) between simulations and references, it is found that the CLM5 has the best
performance, while the GSWP3 exhibits the highest quality. Furthermore, the analysis of variance method
(ANOVA) is then used to trace sources (model, atmospheric forcing data sets and their interactions) of the
uncertainty of those modeling hydrological variables for 1900–2012 (1948–2012 for runoff) over China. The
results indicate that the total uncertainty and its composition vary with time and decrease significantly in recent
decades, reflecting the enhanced forcing data quality. Larger forcing uncertainty existed during the early
twentieth century because less available observation data sets have been adopted to constrain climate variables.
For all modeling hydrological variables, the model uncertainty plays the dominant role, suggesting that the
quality of LMIP products largely relies on Land surface models.
Plain Language Summary Land surface models (LSMs) have served as essential tools for
simulating the response of land surface processes under changing climate. This study focuses on the
performance and uncertainty of hydrological variables from historical (1900–2012) simulations in the Land
Model Intercomparison Project (LMIP), which is a part of the Land Surface, Snow, and Soil Moisture Model
Intercomparison Project (LS3MIP). Using various reference data sets over Chinese mainland, we evaluated
precipitation, evapotranspiration, soil moisture, total runoff and snow cover fraction products from six LSMs
driven by four meteorological forcing data sets. Our findings reveal that, on average, all hydrological variables
are underestimated, but they exhibit a high spatial consistency of trend signs with reference data sets. Among all
simulations, CLM5 stands out for its superior performance and GSWP3 forcing demonstrates the highest
quality. Additionally, the Analysis of Variance (ANOVA) method is adopted to separate the simulation
uncertainties into three sources from the model, the meteorological forcing data set and their interactions. It is
indicated that the total uncertainty has substantially decreased in recent decades, and the model uncertainty is the
dominant factor for these hydrological variables. This study may serve as some valuable references in selecting
LSMs and forcing data sets in the future.
1. Introduction
Terrestrial hydrology is pivotal in the climate system, directly affecting the ecosystem, industry, agriculture, and
human activities. In the context of global climate change, the trends in most hydrological variables have a
warming‐induced intensification at regional to continental scales (Huntington, 2006). Numerical models are
essential tools for simulating historical climate and projecting future climate change (Jiang et al., 2016; Sillmann
et al., 2013). Land surface models (LSMs) are a vital tool for simulating the dynamics of the land processes and
their role in the Earth system. Land surface models involve a series of interrelated processes among bio‐
geophysics, biogeochemistry, hydrology, and carbon exchange, as well as human and societal impacts on the
land surface (Fisher & Koven, 2020; Koster et al., 2009). As the numerical models become increasingly complex,
RESEARCH ARTICLE
10.1029/2023EA003391
Key Points:
•The precipitation, evapotranspiration,
soil moisture, total runoff, and snow
cover fraction in LS3MIP are
extensively evaluated in China
•For LS3MIP historical hydrological
variables over China, model
uncertainty is the dominant factor
overall
Supporting Information:
Supporting Information may be found in
the online version of this article.
Correspondence to:
A. Wang,
wangaihui@mail.iap.ac.cn
Citation:
Ma, X., & Wang, A. (2024). Evaluation
and uncertainty analysis of the land surface
hydrology in LS3MIP models over China.
Earth and Space Science,11,
e2023EA003391. https://doi.org/10.1029/
2023EA003391
Received 30 OCT 2023
Accepted 13 JUN 2024
Author Contributions:
Conceptualization: Aihui Wang
Data curation: Xin Ma
Formal analysis: Xin Ma
Investigation: Xin Ma
Methodology: Xin Ma, Aihui Wang
Project administration: Aihui Wang
Resources: Xin Ma
Software: Xin Ma
Supervision: Aihui Wang
Validation: Xin Ma
Visualization: Xin Ma
Writing – original draft: Xin Ma
Writing – review & editing: Xin Ma,
Aihui Wang
© 2024. The Author(s).
This is an open access article under the
terms of the Creative Commons
Attribution‐NonCommercial‐NoDerivs
License, which permits use and
distribution in any medium, provided the
original work is properly cited, the use is
non‐commercial and no modifications or
adaptations are made.
MA AND WANG 1 of 22

there is a growing demand for comprehensive and multifaceted evaluations of modeling products. Since the
World Climate Research Programme initiated the first Coupled Model Intercomparison Project in 1995, it has
progressed to its sixth phase, referred to as CMIP6 (Eyring et al., 2016). The modeling products of the CMIP6
have provided a scientific foundation for studying the earth system and formulating policies to address climate
change.
The land surface water balance variables from CMIP6 products have been widely used in many studies.
Previous studies have extensively evaluated the key CMIP6 land surface hydrological variables, including
precipitation (Masud et al., 2021; Xin et al., 2020; Zhu & Yang, 2020), evapotranspiration (Bjarke et al., 2023;
Lu et al., 2021), soil moisture (Qiao et al., 2022; Wang, Kong, et al., 2022; Wang, Miao, et al., 2022; Yuan
et al., 2021), runoff (Guo et al., 2022; Wang, Kong, et al., 2022; Wang, Miao, et al., 2022), and snow (Mudryk
et al., 2020; Zhang et al., 2022). Besides, these CMIP6 simulations have also been employed to study extreme
climate events, for instance, extreme precipitation events (Srivastava et al., 2020; Xu et al., 2021), droughts
(Chen & Yuan, 2022; Xu et al., 2021), floods (Di Sante et al., 2021; Meresa et al., 2022), and heatwaves
(Hirsch et al., 2021; Sun et al., 2023; Xie et al., 2022). However, even the state‐of‐the‐art Earth system models
(ESMs) cannot accurately describe the complex land surface processes (van den Hurk et al., 2016). Usually, the
land surface processes in the ESMs contain systematic biases and uncertainties, for example, biases in hy-
drological characteristics (Hagemann et al., 2011; Papadimitriou et al., 2017), distribution of water and energy
(Materia et al., 2022; Mueller & Seneviratne, 2014) and feedback strengths (Qu & Hall, 2014). These biases
may have a nonlinear effect on the forecast skill of key climate variables (Koster et al., 2010), patterns of
climate change (Campoy et al., 2013; Seneviratne et al., 2013), as well as the trends in water resources
(Lehning & KATUL, 2013). The uncertainties in the model simulations have also been explored by using
various statistical methods, and most of these studies are focused on the uncertainty in climate projection.
Ashraf Vaghefi et al. (2019) summarized the previous studies on uncertainty analysis and then broadly cate-
gorized them into two groups: characterizing model performance with uncertainty band (Faramarzi
et al., 2009), and identifying and tracking the compositions of uncertainty using statistical methods (Ashraf
Vaghefi et al., 2019; Bonan et al., 2019; Chen & Yuan, 2022; Hawkins & Sutton, 2009; Yip et al., 2011). For
example, Hawkins and Sutton (2009,2011) proposed an analysis of variance (ANOVA) method to separate and
quantify uncertainty in regional climate predictions. They found that the model uncertainty was more important
than internal variability uncertainty in the precipitation and surface air temperature predictions. Yip
et al. (2011) extended this approach used in Hawkins and Sutton (2009) into a simple and coherent framework
and then applied it to explore the partitions of uncertainties in the CMIP3 surface air temperature predictions.
The ANOVA method is a model‐based method that decomposes total variance into components from different
sources and their interactions (Hawkins & Sutton, 2009; Yip et al., 2011). Because it requires fewer assumptions
than other approaches, such as the classical Bayesian approaches (Haydon & Deletic, 2009) and the pseudo‐
Bayesian methods (Freni et al., 2009), ANOVA can effectively quantify the uncertainties from different sour-
ces and their interactions (Ashraf Vaghefi et al., 2019; Hawkins & Sutton, 2009; Yip et al., 2011). Bosshard
et al. (2013) presented a subsampling procedure based on the ANOVA method to explore the sources of un-
certainty in an ensemble of hydrological climate‐impact projections. Chen and Yuan (2022) quantified the un-
certainty of internal variability in future projections of seasonal soil moisture droughts over China. Abbaspour
et al. (1997,2004) developed and extended the second version of the Sequential Uncertainty Fitting (SUFI‐2)
Program for inverse modeling. The SUFI‐2 uses a sequence of steps to progressively reduce the initial un-
certainties until a certain calibration requirement is reached. Ashraf Vaghefi et al. (2019) further developed a
method that combined the ANOVA and the SUFI‐2 approach. This method was used to investigate how the
effects of hydrological model parameterization and regionalization, along with other uncertainty drivers,
contribute to the overall uncertainty in climate impact projections. Most of the above previous studies agree that
the primary source of uncertainty in hydro‐climatic impacts modeling is the choice of model (i.e., ESM, General
Circulation Model, or hydrological model).
Building on the above context, CMIP6 has designed the LandMIP series experiments to complement the land‐
based experiments similar to the ocean‐ and atmospheric‐based ones (van den Hurk et al., 2016). The LandMIP
contains two series of experiments: the Land Surface, Snow and Soil Moisture Model Intercomparison Project
(LS3MIP) and the Land Use Model Intercomparison Project. LS3MIP consists of two series of simulations: (a)
the land‐only experiment (Land Model Intercomparison Project (LMIP) and (b) the land‐atmosphere coupled
suites (Land Feedback Model Intercomparison Project, LFMIP). More details of LS3MIP and its LMIP
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 2 of 22

experiments can be seen in van den Hurk et al. (2016). However, the experiments of the LSMIP have not been
effectively utilized thus far. Although some studies have evaluated the performance of the main international
ESMs over China domain, their land surface modules have not been fully assessed. Based on the LMIP
simulations in the LS3MIP, this study aims to identify and diagnose the systematic biases contained in the land
modules of current ESMs (van den Hurk et al., 2016). The land‐only LS3MIP Land‐Hist experiments are
conducted from the offline running LSMs within ESMs framework driven by the observation‐based meteo-
rological data sets. In order to enhance the understanding of the water budget simulations of LSMs, this study
also explores the uncertainties in the land surface hydrological variables of the Land‐Hist LS3MIP simulations
within continental China. We first employ high‐quality station‐ and satellite‐based observations as the reference
data sets to systematically evaluate the land surface hydrological variables of the Land‐Hist experiment in
LMIP. Then, the uncertainties of each hydrological variable from both the model structure and the meteoro-
logical forcing data, as well as their interaction, are quantitatively analyzed using the ANOVA method. This
paper is organized as follows. Section 2provides details about the LS3MIP hydrologic simulations, the
reference data, and the ANOVA method. Section 3displays the comparison results of LS3MIP with reference
data sets. Section 4presents the composition of uncertainty in LS3MIP. The conclusions are summarized in
Section 5.
2. Data and Methods
This study analyzes the monthly precipitation (PR), evapotranspiration (ET), soil moisture within 0–10 cm soil
depth (SM), snow cover fraction (SCF), and total runoff (Ro) of the multi‐model simulations in the LS3MIP
Land‐Hist experiment.
2.1. LS3MIP Land‐Hist Experiment
Table 1lists the information on six ESMs, including their land modules, resolutions, and institutions. Meteo-
rological variables (i.e., air temperature and precipitation) are used to force LSMs to derive land surface water and
energy quantities. The meteorological forcing data sets used in LS3MIP include the GSWP3 (the 3rd phase of the
Global Soil Wetness Project) in Tier1 and three alternate forcings: WFDEI (Water and Global Change Forcing
Data Methodology Applied to ERA‐Interim data, Weedon et al., 2014), CRU‐NCEP (Viovy & Ciais, 2009), and
Princeton (Princeton Global Forcing data set, Sheffield et al., 2006) in Tier 2. Apart from utilizing different
forcing data sets, Tier 1 and Tier 2 have no distinctions in the configurations of models in LS3MIP. Comparison
between Tier 1 and Tier 2 can illustrate the sensitivity of hydrological vari-
ables to different forcing data sets. Table 2shows the time range and spatial
and temporal resolution of the four meteorological forcing data sets. The
GSWP3 was originally developed from the downscaled twentieth Century
Reanalysis and then bias‐corrected with the multiple observational and sat-
ellite data (Compo et al., 2011). The WFDEI combines WATCH (Water and
Global Change, Weedon et al., 2011) and the ERA‐Interim reanalysis prod-
ucts in different time ranges. The CRU‐NCEP merges the CRU TS v3.2 with
the NCEP reanalysis data set, then bias correction is applied to enhance the
data quality (Viovy & Ciais, 2009). The Princeton forcing was originally
Table 1
Summary of Earth System Models in LS3MIP Used in This Study
Earth system model Land surface model Spatial resolution Institution & country Reference
BCC‐CSM2‐MR BCC_AVIM2 1.125° ×1.125° CMA, China Wu et al. (2019)
CESM2 CLM5.0 1.25° ×0.94° NCAR, USA Danabasoglu et al. (2020)
CMCC‐ESM2 CLM4.5 1.25° ×0.94° CMCC, Italy Lovato et al. (2022)
EC‐Earth3‐Veg HTESSEL 0.70° ×0.70° SMHI and 26 other European institutes Döscher et al. (2021)
MPI‐ESM1‐2‐LR JSBACH 1.875° ×1.875° MPI‐M, Germany Müller et al. (2018)
MIROC6 MATSIRO6.0 1.41° ×1.41° AORI, University of Tokyo/JAMSTEC/
National Japan Institute for Environmental Studies
Tatebe et al. (2019)
Table 2
Introduction of the Meteorological Forcing Data Sets Used in the LS3MIP
Simulations for This Study
Name Temporal resolution Spatial resolution Time range
GSWP3 3h 0.5° ×0.5° 1901–2014
WFDEI 3h 0.5° ×0.5° 1901–2001
CRU‐NCEP 6h 0.5° ×0.5° 1901–2010
Princeton 3h 0.5° ×0.5° 1901–2012
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 3 of 22

based on the surface climate of NCEP/NCAR reanalysis products and then bias corrected at 6‐hourly, daily, and
monthly scales with various observation‐based data sets (Sheffield et al., 2006). To facilitate the evaluation of
different model products, all hydrological variables from LS3MIP simulations are firstly interpolated to a
0.25° ×0.25° resolution using the bilinear interpolation method.
2.2. Reference Data Sets
Reference data sets from diverse sources are adopted to evaluate the performance of LS3MIP Land‐Hist simu-
lations. Those data sets are primarily from observational‐based records and have been proven to be of high
quality. Table 3provides detailed information on those reference data sets. The CN05.1 was constructed using in‐
situ observations from more than 2400 stations in China (Wu & Gao, 2013), and it has been widely employed as
basis climate data in previous studies (Gao et al., 2013; Miao & Wang, 2020; Wang, Kong, et al., 2022; Wang,
Miao, et al., 2022; Wu et al., 2017). The daily PR in CN05.1 is aggregated to monthly to compare with the
LS3MIP simulations.
The ET reference data set is obtained from the monthly Global Land Evaporation Amsterdam Model (GLEAM)
v3.3a, which is a combination of data sets from algorithms and multi‐satellite remote sensing observations
(Martenss et al., 2017; Miralles et al., 2011) covering the period from 1980 to 2012. Yang et al. (2017) reported
that the GLEAM ET shows reasonable accuracy in representation of actual ET over different time scales after
comparing it with the in‐situ observed ET at eight Chinese stations.
The SM reference data set is from monthly station observations collected via the gravimetric method during the
growing seasons (April to September) from 1992 to 2013 (Wang & Shi, 2019, denoted as WS2019). Considering
the availability of SM in WS2019, we selected 0–10 cm observations available for at least 4 months within each
growing season, along with a minimum of 5 years for 1992–2013. After applying the above data quality control
measures, we retained the SM observations at 651 stations, which were then compared with the standard near‐
surface SM (i.e., 0–10 cm) output in CMIP6.
For the Ro reference data set, we adopted the Global River Flow and Continental Discharge Data set (Dai &
Trenberth, 2002; Dai et al., 2009, Dai, 2017. Hereafter referred to as D17), which contains a time series of
monthly river flow rates observed at the largest 925 rivers in the world. In this study, three Chinese river basins
(the Yangtze, Yellow River, and Songliao) in D17 were used. Referring to the International Land Model
Benchmarking project (ILAMB, https://www.ilamb.org/), we convert the D17 river flow data from river gauge
stations to river mouths, and the unit is converted to mm/day. The coefficient of variation and linear trend of Ro
are not shown due to the limited number of river basins in D17 over China used in this study. The monthly Ro of
D17 for 1948–2012 is then used to evaluate LS3MIP simulations.
For SCF, the Moderate‐Resolution Imaging Spectroradiometer (MODIS) monthly product version 6 utilized an
improved detection algorithm known as the normalized difference snow index, which enhances the resulting
quality compared to previous versions (Hall & Riggs, 2016). The MODIS SCF product during boreal winter
months (December‐January‐February) for 1980–2012 was used as the reference data set in this study.
Table 3
Information About the Reference Data Sets Used to Evaluate LS3MIP Hydrological Variable Simulation in China
Variable
Time
range
Station information and spatial &
temporal resolution Names References
Precipitation 1980–2012 0.25° ×0.25°, monthly CN05.1 Wu and Gao (2013)
Evapotranspiration 1980–2012 0.25° ×0.25°, monthly GLEAM v3.3a Miralles et al. (2011), Martenss et al. (2017), https://www.gleam.eu/
Soil moisture 1992–2012 651 stations at 10 cm depth, monthly WS2019 Wang and Shi (2019)
Runoff 1948–2004 The Yangtze, Yellow River, and
Songliao river basins
D17 Dai and Trenberth. (2002), Dai et al. (2009), Dai (2017)
Snow cover
fraction
2002–2012 0.05° ×0.05°，winter MODIS
（MYD10CM）
Hall and Riggs (2016)
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 4 of 22

2.3. Methods
2.3.1. Statistical Methods for Intercomparison
To ensure compatibility with the coarser resolutions of PR and ET reference data sets, all monthly PR, ET, and
SCF of LS3MIP Land‐Hist simulations are remapped to a 0.25° ×0.25° resolution using the bilinear interpolation
method before the following analyses. For SM, the gridded LS3MIP simulations are interpolated into the station
location of the reference data using the near‐neighbor interpolation method. The basin‐level mean Ro is computed
by the mean of Ro simulations across the Yangtze, Yellow River, and Songliao River basins. The multi‐model
ensemble mean (MME) of each variable is calculated from all the available model simulations in LS3MIP. To
assess the spatial variability of LS3MIP hydrological variables across continental China, we compute four sta-
tistical metrics between LS3MIP simulations and reference data sets. Those metrics include the bias (BIAS),
correlation coefficient (CC), normalized standard deviation (NorSTD), unbiased root‐mean‐square biases
(ubRMSE), and the coefficient of variation (CV, the ratio of STD and mean). Normalized standard deviation
represents the degree of agreement between model simulation results and reference data regarding spatial
variability:
NorSTD =σs
σr(1)
where σ denotes the STD, with subscripts sand rrepresenting simulation and reference data set. NorSTD is
generally greater than 0. The closer NorSTD is to 1, the better the agreement is. The ubRMSE is defined as:
ubRMSE =
RMSE2Bias2
√(2)
The CV represents the interannual variabilities of hydrological variables from LS3MIP simulations or reference
data sets.
A ranking method is employed to synthesize the performance of each variable based on individual statistical
characteristics (Brunke et al., 2003; Ma & Wang, 2022; Wang & Zeng, 2012). For each hydrological variable,
four statistical metrics (i.e., bias, spatial CC, NorSTD, or ubRMSE) are ranked among all LS3MIP simulations,
and the score is assigned starting with 1 to the best simulation and then increase the number in turn to the worse
one. For example, if there are three LSM simulations available for SM, the simulation with the smallest bias is
assigned a score of 1, while the one with the largest bias is assigned a score of 3. The final ranking score is
obtained by averaging all scores across four statistical metrics, with the lower score indicating superior perfor-
mance. Take JSBACH precipitation driven by GSWP3 as an example, the ranking score is 1 for bias, NorSTD,
and ubRMSE, and it is 3 for the CC (Figure 6). The final mean ranking score for this case is 1.5, which is the
lowest compared to other simulations and then can be regarded as the best overall performance.
2.3.2. ANOVA Method
We implement the ANOVA method (Ashraf Vaghefi et al., 2019) to discuss the different sources of uncertainty in
the LS3MIP hydrological variables. Through ANOVA, the source of variances of each hydrological variable can
be decomposed into the original contributing sources (model structure and atmospheric forcing variables) and
their nonlinear interactions. This method also applies the Sequential Uncertainty Fitting algorithm to capture the
uncertainty associated with the meteorological forcing data set and the model (including surface parameter, model
structure, and factors other than the forcing data set, Ashraf Vaghefi et al., 2019). According to the ANOVA
method, the total sum of square errors (SST) of a hydrological variable xin LS3MIIP simulations is defined as:
SST =∑NModel
i=1∑NForcing
j=1(xi,jx)(3)
where x
i,j
is the hydrological variable xfrom model idriven by a meteorological forcing data set j,N
Model
is the
number of ESMs, and N
Forcing
is the number of forcing data sets. To avoid the influence of missing values, we
chose six simulations from three models: CESM2, MPI‐ESM1‐2‐LR, and MIROC6, driven by two forcing data
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 5 of 22

sets: GSWP3 and Princeton. xis the overall mean. SST can be split into the sum of squares due to the uncertainties
of model, forcing data set and their interactions:
SST =SSModel +SSForcing +SSI (4)
SSI =SSModel&Forcing (5)
SSModel =NForcing ∑
NModel
i=1(xi,∗x)2(6)
SSForcing =NModel ∑
NForcing
j=1(x∗,jx)2(7)
SSI =∑
NModel
i=1
∑
NForcing
j=1(xi,jxi,∗x∗,j+x)2(8)
where SS
model
, SS
Forcing
, and SSI are the uncertainty of model, forcing, and their interactions, respectively. The
symbol “*” represents averaging over a particular index. x
i,*
is the mean of all forcing data sets for model i,x
*, j
is
the mean of all models for forcing j, and x
i,j
is the mean of all ensemble members. To prevent interference from
extraneous factors and ensure accuracy, the total uncertainty, and its decomposition are determined by applying a
sliding 5‐year window to these six simulation groups.
3. Evaluation of LS3MIP Land Surface Hydrological Variables
Figure 1compares annual historical hydrological variables among reference data sets, individual LS3MIP
simulations, and MME for different periods over China. PR is one of the most important hydrological variables
because it is a critical link among different elements of the surface water cycle (Held & Soden, 2006; Trenberth &
Guillemot, 1998). In offline LSM simulations, hydrological variables are directly affected by the quality of PR
forcing (Wang et al., 2016), which usually required to be preprocessed in the model initialization. Consequently,
even with the same atmospheric forcing data set, the PR outputs from different LSMs may be diverse. Therefore,
it is deserved to evaluate the PR outputs because they are real land moisture input in the LSM integration. Further
discussions on PR can be found in Section 5. Simulations of PR show small spreads, with a temporal CC between
MME and CN05.1 reaching 0.99, indicating a high consistency of PR in different forcing data sets. The biases in
PR between individual models and CN05.1 are also small, varying from 0.07 to 0.1 mm/day (Figure S1 in
Supporting Information S1). The CC of ET, SCF, SM, and Ro between MME and reference data sets are all
relatively high (0.74, 0.62, 0.91, and 0.84, respectively). The differences between individual simulation and
reference data sets of each hydrological variable are shown in Figure S1–S4 in Supporting Information S1. The
MME of LS3MIP hydrological simulations consistently exhibits lower values than their respective reference data.
However, the SCF stands as an exception due to the substantial overestimation of the land model (HTESSEL) of
EC‐Eearth3‐Veg (see also Figure S3 in Supporting Information S1). In HTESSEL, the SCF depends on a simple
relationship with snow water equivalent and snow density, and the above overestimation may be induced by the
biases of these two snow condition variables in the snow scheme (Balsamo et al., 2009,2011).
Figure 2shows the interannual variabilities of CV for PR, ET, SCF, and SM from both LS3MIP and reference
data sets. The LS3MIP well captures the spatial pattern of CV in four hydrological variables. The spatial mean
CVs of these hydrological variables are 6.11, 20.14, 9.09, and 22.94 for LS3MIP MME, and 6.35, 12.34, 4.79, and
10.36 for reference data sets, respectively. The STDs of PR, ET, SM, and SCF are 0.23 mm/day, 0.06 mm/day,
0.01 mm
3
/mm
3,
and 5.53%, which are slightly lower than those from the reference data sets (0.26 mm/day,
0.09 mm/day, 0.03 mm
3
/mm
3
and 8.57%, respectively). The lower interannual variabilities of LS3MIP MME
reflect the over‐smoothing error to some extent, which is a common systemic bias in CMIP6 ESMs, such as in the
CESM (Baker et al., 2016).
Figure 3shows the linear trend of historical PR, ET, SCF, and SM from MME and reference data sets over
different time ranges in China. Both PR and ET exhibit a consistent decreasing trend in northeast China and an
increasing trend in the Tibetan Plateau. In both MME and GLEAM, a significant increasing trend of ET exists in
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 6 of 22

the east and southwest regions of China. The multi‐year downtrends of SCF shows from both MODIS and MME,
with values of 2.07%/10 years and 1.56%/10 years, respectively. In northern and northwestern China, the SCF
exhibits a noteworthy downtrend (1.6%/10 years, p＞95%). Conversely, there is an ascending trend in certain
areas of northeastern China. The 0–10 cm SM at 651 sites during 1992–2012 overall displays a wetting trend, and
the mean linear trend for WS2019 and MME are 0.13 ×10
2
mm
3
/mm
3
/10 years and 0.19 ×10
2
mm
3
/mm
3
/
10 years, respectively. SM mainly exhibits a wetting trend in the Yellow River Basin, while it shows a drying
trend in the Yangtze River Basin and northeast China (Figure 3), consistent with the results of CMIP6 multi‐
model simulation in Wang, Kong, et al. (2022), Wang, Miao, et al. (2022). To quantitatively describe the
trend sign consistency between simulations and the respective reference data sets, we also calculate the per-
centage of the area/stations with the same positive or negative trend. The high consistency rates (83% for PR, 74%
for ET, 85% for SCF, and 56% for SM) demonstrate generally high performance of LS3MIP simulations in terms
of trends in Figure 3. It is important to note that the limited number of SM stations and their uneven distribution
across the Chinese mainland in WS2019 may have influenced the evaluation results.
Figure 1. Comparisons of annual hydrological variables relative to reference data set (black curve), individual LS3MIP
simulations (gray curve), and MME (red curve) for different time ranges in China. (a) PR from LS3MIP MME and CN05.1
for 1980–2012, (b) ET from LS3MIP MME and GLEAM for 1980–2012, (c) SCF from LS3MIP MME and MODIS for
boreal winter during 2002–2012, (d) 0–10 cm SM at 651 stations from LS3MIP MME and WS2019 for 1992–2012, (e) the
average of Ro from LS3MIP MME and D17 for 1948–2004 for the Yangtze, Yellow River, and Songliao River basins. The
values in the bottom left corner represent the correlation coefficients between the MME and reference data, and all the results
passed the statistical significance test at p=0.05.
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 7 of 22

The Taylor diagrams (Taylor, 2001) in Figure 4summarize the NorSTD and spatial CC of LS3MIP hydrological
variable simulations over different periods across the Chinese mainland. The monthly climatology mean of each
variable was removed before those statistical metrics were computed. This would ensure that the comparison of
surface hydrological variables (except for PR) is not affected by its seasonal cycle, which is mainly determined by
the seasonal magnitude of forced PR. The spatial CC between PR simulations and CN05.1 ranged from 0.89 to
0.92, indicating a high consistency among four forcing data sets and reference data sets. Because the quality of PR
is crucial to other hydrological variables (e.g., ET and near‐surface SM) in offline LSMs (Ma & Wang, 2022; Vano
et al., 2012; Wang et al., 2016), the above result implies that there are no significant differences in PR inputs for
models. For ET anomalies, the spatial CCs from different models vary from 0.26 to 0.36. The 0–10 cm SM
Figure 2. The coefficients of variation of historical LS3MIP PR, ET, SCF, and SM in China. (a, b) PR from CN05.1 and MME during 1980–2012, (c, d) ET from
GLEAM and MME during 1980–2012, (e, f) SCF from MODIS and MME in boreal winter during 2002–2012, (g, h), 0–10 cm SM at 651 stations from WS2019 and
MME during 1992–2012. The values at the bottom left corner represent the spatial mean.
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 8 of 22

anomalies exhibit more clustered distributions, with spatial CC and NorSTDs varying for a range of 0.23–0.42, and
0.46–0.91, respectively. The spatial CC of SCF is distributed between 0.29 and 0.70, indicating a pronounced
model‐dependent pattern. For instance, SCF and Ro simulations of JSBACH with different forcing data sets
(represented by dots and six in different colors) are closer in Figures 4d and 4e. It also should be noted that NorSTD
values for both JSBACH SCF and Ro are less favorable compared to the results of other sets of simulations.
Figure 3. Spatial distribution of linear trend in historical LS3MIP PR, ET, SCF, and 0–10 cm SM in China for different periods. (a, b) PR from CN05.1 and MME during
1980–2012, (c, d) ET from GLEAM and MME during 1980–2012, (e, f) SCF from MODIS and MME in boreal winter during 2002–2012, (g, h), 0–10 cm SM at 651
stations from WS2019 and MME during 1992–2012. The values in the bottom left corner represent the regional mean values. The percentages in the upper middle of the
right column indicate the ratio of grids/areas with the same positive and negative trends across all the records. We conducted student‐t tests to determine the statistical
significance of these trends. Dots in panels (a–f) denotes statistically significant trends at p=0.05, while * in panels (g, h) indicate SM sites that passed the statistical
significance test at p=0.05.
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 9 of 22

Figure 4. The Taylor diagrams summarized the NorSTD and spatial CC of LS3MIP hydrological variable simulations over
different periods in mainland China. (a) PR from MME and CN05.1 for 1980–2012, (b) ET from MME and GLEAM for
1980–2012, (c) 0–10 cm SM from MME and WS2019 at 651 stations for 1992–2012, (d) SCF from MME and MODIS in
boreal winter during 2002–2012, (e) Ro from MME and D17 for the Yangtze, Yellow River, and Songliao River basins for
1948–2004. All the CC values passed the statistical significance test at p=0.05.
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 10 of 22

To draw a comprehensive conclusion regarding the representation of LS3MIP hydrological variable simulations
by different models driven with diverse meteorological forcing data sets in China, we also calculated four sta-
tistical metrics: bias, spatial CC, NorSTD, and ubRMSE. To facilitate comparisons, we employ a ranking method
to assess the statistical results of each model and meteorological forcing data set. Lower ranking scores indicate
better simulation performance compared to other results. Figure 5presents the results in terms of those statistical
metrics along with different time ranges. Overall, ET and Ro from CESM2, and SCF and SM from MATSIRO6
(the land module of MIROC6) achieved the best bias scores (ranking score of 1), indicating a better performance
compared to other models in Figure 5. The simulations of MATSIRO6 driven by all meteorological forcings have
lower ranking scores for NorSTD, indicating that MATSIRO6 performs well in simulating the spatial variability
of hydrological variables. The JSBACH and HTESSEL models show poorer performance in SCF and SM among
all simulation sets. The BCC_AVIM2, in combination with Princeton, achieves the best score for Ro simulations
in NorSTD. According to the ranking scores based on the spatial CC in Figure 5, the CLM5 driven by GSWP3
exhibits the best performance for PR and Ro simulations. The HTESSEL, in combination with GSWP3, achieves
the best score for ET simulations. It also indicates that simulations forced by GSWP3 overall have the best
performance in terms of the spatial CC compared to other forcings. Simultaneously, the minimum values for
ubRMSE in SCF, ET, PR, and Ro simulations are all obtained from various models driven by GSWP3. These
results highlight the advantage of GSWP3 as the meteorological forcing for driving LSMs, particularly over
China.
Figure 6shows the overall mean ranking scores derived from these four statical metrics for the combination of
model and forcing data set. The details of the calculation of the mean ranking scores can be seen in Section 2.3.1.
For PR, the results from GSWP3 show clear advantages over other forcing data sets, with ranking scores ranging
from 1.50 to 3.75 (Figure 6). Of all the model outputs driven by GSWP3 forcing, JSBACH achieves the best
ranking score for PR, indicating that this model performs better in handling PR (i.e., interpolation of 3‐hourly PR
into the model integration time step, partitioning of the total PR into rain and snow) than other models. The
WFDEI also performs relatively well (Figures 5and 6) in handling PR because its bias correction preserves the
spatial continuity of “large‐scale” or frontal precipitation events spanning multiple half‐degree grid boxes
(Weedon et al., 2011,2014). ET simulations from the MATSIRO6 model driven by all four forcing data sets have
better performance than other model‐data combinations (Figures 5and 6). Considering the dependence of the SM
and SCF simulations on the LSMs, the CLM4.5, CLM5.0, and MATSIRO have a reasonable ability to represent
the SM compared to other LSMs. The Ro of CLM5 driven with Princeton and the GSWP3 both have better
ranking scores (3.75, and 4.50), indicating that CLM5 has an advantage in simulating Ro than other LSMs. In
conclusion, based on the results of the four statistical metrics employed in this evaluation of the LS3MIP hy-
drological variables in China, the CLM5 model exhibits the highest simulation performance, while the GSWP3
data set demonstrates superior quality in these simulations. Better performance is also achieved with the com-
bination of MATSIRO and WFDEI.
4. Uncertainty Analysis of LS3MIP Hydrological Variables Simulations
The goals of LS3MIP extend beyond providing a comprehensive assessment of land surface variables; they also
encompass diagnosing simulation bias and uncertainty in the land modules of current ESMs (van den Hurk
et al., 2016). The uncertainty in climate prediction is typically decomposed into scenario uncertainty, model
uncertainty, and internal variability (Hawkins & Sutton, 2009; Yip et al., 2011). For LSM simulations, the model
uncertainty and external uncertainty from the meteorological forcing data set are significant factors influencing
model simulation performance (Bonan et al., 2019).
In this section, we employed a combination of the ANOVA method and the SUFI‐2 program to identify the sources
of uncertainty in LS3MIP hydrological variable simulations over China. Given that not all LS3MIP hydrological
variables are available from different ESMs driven with distinct forcing data sets, we utilized model outputs from
three models: CLM5, MATSIRO6, and JSBACH, driven by two distinct forcing data sets: GSWP3 and Princeton,
to ensure sufficient data participation. The 5‐year moving average of these six groups of simulations is used in
ANOVA to mitigate the impact of randomness. Figure 7illustrates the annual total uncertainty and their com-
ponents from the model, forcing, and their interaction (M&F) for LS3MIP hydrological variables during 1900–
2012 (for Ro, the time range is 1948–2012). As a key forcing variable, PR is the dominant uncertainty source
for the LSM hydrological variables, and its uncertainty reaches 1.5 (mm/day)
2
with obvious fluctuation before the
1950s. In the case of the same forcing data sets, the model uncertainty may arise from the fact that different ESMs
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 11 of 22

treat forcing data sets differently, such as employing various interpolation methods and distinct allocations of input
water into rain and snow (Kiehl, 2007; Yip et al., 2011). There is substantial forcing uncertainty in LS3MIP
hydrological cycle simulations during the early twentieth century, in which less available observation data sets
were adopted to constrain climate variables. This also implies that we should be cautious when utilizing those
forcing data sets during the early twentieth century. The forcing variables exhibit considerable variability, as
depicted in Figure 7a. For LS3MIP hydrological simulations, except for PR, model uncertainty is the dominant
factor and varies considerably from year to year. The dependence of uncertainty on the model differs among
hydrological variables. The interaction of different sources is an important concept in the ANOVA method, which
demonstrates the proportion of nonlinear sources of uncertainty (Ashraf Vaghefi et al., 2019; Yip et al., 2011). The
uncertainty associated with M&F is the smallest and the most stable component of the total uncertainty. It pri-
marily reflects how models nonlinearly respond to forcing data sets (Ashraf Vaghefi et al., 2019).
Figures 8and 9show the annual variance decompositions of total uncertainty in LS3MIP hydrological variable
simulations and their spatial distribution. For PR simulations, the forcing term is the dominant source of
Figure 5. Ranking scores of four statistical metrics: bias, NorSTD, spatial CC, and ubRMSE for LS3MIP hydrological
variable simulations. See Section 2.3.1. The first column of the vertical axis represents the LSMs (i.e., BCC_AVIM), and the
second column represents the forcing data set with abbreviations (PN: Princeton, CR: CRU‐NCEP, GS: GSWP3, WF:
WFDEI). The gray dashed lines with a score of 1 indicate the optimal results for each group of hydrological variables. All the
CC values passed the statistical significance test at p=0.05.
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 12 of 22

uncertainty, accounting for 78.27% of the total uncertainty variance. It should be noted that PR in offline LSM
simulations is directly derived from the input forcing data sets. However, the contribution of model uncertainty
for PR is still not negligible, reaching up to 15.63% of the total uncertainty during 1902–2012. This suggests that
the uncertainty induced by processing forcing data sets in ESMs is also noteworthy. Model uncertainty constitutes
the primary component of the total uncertainty for SM, SCF, and Ro simulations, and their impact varies among
different variables. In the case of ET, the model uncertainty and forcing uncertainty account for approximately
44.47% and 48.74% of the total simulation uncertainty, respectively, indicating a nearly equal influence on the
overall uncertainty. For LS3MIP SM simulation, the primary source of uncertainty lies in the model term
(84.64%), with the smallest fraction attributed to forcing uncertainty compared to other variables. This suggests
that variations in SM parameterization schemes in LSMs can lead to large discrepancies in model outputs (Zeng
et al., 2021). The M&F term contributes more to the uncertainty in the SCF simulation than other variables. The
dominant uncertainty contribution for Ro simulations in LS3MIP arises from the model term, up to 68.73% of the
total. However, the uncertainty contribution from forcing term in Ro simulation is also important, accounting for
22.43% in Figure 8.
Regarding the spatial distribution of uncertainty in annual LS3MIP hydrological variable simulations shown in
Figure 9, ET exhibits a dipolar pattern in China, extending from southeast to northwest. In the eastern, southern,
and southwestern regions of China, the uncertainty in ET simulation is predominantly due to the model term.
Conversely, in the northeastern, northwestern, and the Tibetan Plateau regions, the uncertainty in ET simulation is
primarily influenced by meteorological forcing data sets. The contribution of these two sources to the total un-
certainty in these regions exceeds 90%, emphasizing their dominant role. In the relatively flat and economically
developed eastern and southern areas of China, there are sufficient stations to provide long‐term meteorological
observations to develop forcing data sets. However, in western China and the Tibetan Plateau, the relatively sparse
and uneven distribution of meteorological stations impacts the quality of the forcing data set. Also, not all
meteorological stations have measurements for hydrological variables such as SM (Wang et al., 2016; Wang &
Shi, 2019). It can be seen in Figure S1 in Supporting Information S1 that the PR in LS3MIP from individual
Figure 6. Mean ranking scores of four statistical metrics (bias, spatial CC, ubRMSE, and NorSTD) for LS3MIP hydrological
variable simulations. See Section 2.3.1. The horizontal axis represents the meteorological forcing data set used for
simulation, while the vertical axis represents the model. All the CC values passed the statistical significance test at p=0.05.
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 13 of 22

simulations exhibits large spreads in these regions with higher forcing uncertainty decomposition. Additionally,
the contribution of forcing uncertainty over the Tibetan Plateau and western China exceeds 90% in Figure 9b. The
uncertainty in SM simulation is primarily attributed to the model term, especially in southeastern China where the
quality of forcing data sets is relatively consistent. In the case of SCF simulation, the uncertainty related to the
M&F term makes a significant contribution, particularly in the southern region of China, amounting to over 60%.
This suggests that in warm areas, such as southern China, where SCF is minimal, there are large variations in the
nonlinear response of the model to the forcing variables. It would be beneficial to consider employing filter
methods to process the SCF simulations, as this may help mitigate uncertainty in the results in these regions (Ma &
Wang, 2022). A significant portion of the 22.43% forcing uncertainty contribution originates from the Tibetan
Plateau in China. The contrast between the forcing and model uncertainties in the Tibetan Plateau region, as shown
in Figure 9, indicates that uncertainty variations resulting from different models are not significantly distinct. The
primary source of hydrological variable simulation uncertainty in this area arises from forcing factors.
5. Conclusion and Discussions
The LMIP in LS3MIP experiment within CMIP6 offers valuable insights through offline simulations conducted
by multiple LSMs embedded in ESMs. These simulations provide a foundation for conducting a thorough
Figure 7. The annual total uncertainty and its components (model, forcing, and their interaction M&F, see Section 2.3.2) are derived from a 5‐year moving average of
LS3MIP hydrological variables in China from 1900 to 2012 (1948–2012 for Ro).
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 14 of 22

assessment of LSM performance. This study has diagnosed the bias and then explored the sources of uncertainty
in the land surface hydrological simulations over Chinese mainland from the LS3MIP Land‐Hist experiment. The
hydrological variables, including PR, ET, 0–10 cm SM, Ro, and SCF, from six ESMs driven by four meteoro-
logical forcing data sets are compared against various high‐quality reference data sets. The results indicate that the
MMEs of those variables are overall underestimated. For PR, ET, and SM, simulations have a better seasonal
cycle. Compared with the reference data sets, the sign consistency of annual trend respectively occupies 83%,
74%, 85% land area for PR, ET, SCF, and 56% stations for SM. These results suggest a good reproduction of the
linear trends in LS3MIP simulations. Based on the assessment of bias, spatial CC, NorSTD, and ubRMSE for the
Figure 8. The annual variance decompositions of total uncertainty (see Section 2.3.2), expressed in percentage (%), are derived from a 5‐year moving average of LS3MIP
hydrological variable simulations in China during 1900–2012 (1948–2012 for Ro).
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 15 of 22

LS3MIP hydrological variable, CLM5 demonstrates the best simulation performance, while GSWP3 exhibits the
highest quality among four meteorological forcing data sets. We then use the ANOVA combined with the SUFI‐2
approach to trace the sources of uncertainty and their composition in the LS3MIP hydrological variables. The
impact of the model, forcing, and their interaction varies across variables, with model uncertainty being the
predominant factor. When using the same forcing data sets, model uncertainty may arise from ESMs handling
forcing data sets, for example, temporal interpolation methods, the allocation of total precipitation into rain/snow.
For instance, large forcing uncertainty existing over the Tibetan Plateau is from the disparity among different
Figure 9. The spatial distribution of variance decompositions of the total uncertainty (see Section 2.3.2) in LS3MIP hydrological variable simulations during 1900–2012
in China (1948–2012 for Ro). Unit: %. The values at bottom left corner represent the spatial mean.
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 16 of 22

meteorological forcing data sets, which may weaken the ratio of uncertainty from LSMs. The above performance
assessment and uncertainty analysis have systematically revealed the applicability of LSMs from multiple ESMs
in China, which provides a reference for understanding the simulation biases in land surface water variables.
Identifying the various sources of uncertainty in offline simulations of land surface hydrological variables can
assist decision‐makers and scientists in addressing uncertainty‐related issues more effectively. This under-
standing can also guide the necessary research and data acquisition efforts aimed at reducing uncertainty.
In general, the hydrological variables in LSM are updated at each time step based on the land surface water
balance equation. Different hydrological variables are closely linked, but the degree of linkage depends on the
parameterization scheme and the structure of LSMs. To illustrate this further, Figure 10 compares the biases of
SM versus ET and SM versus PR in CLM5, HETSSEL, and MATSIRO6 forced by GSWP3 at 651 stations during
1992–2012. For SM bias versus ET bias (Figures 10a, 10c, and 10e), positive SM bias is often associated with
overestimated ET in CLM5. However, in the HETSSEL, a combination of dry SM bias and high ET was pre-
dominant. In MATSIRO6, positive ET seems decoupled with the dry/wet SM bias. For SM bias versus PR bias in
(Figures 10b, 10d, and 10f), positive bias in PR in CLM5 generally causes wet bias in SM, while HETSSEL
mainly shows negative bias in PR with dry soil. In MATSIRO6, PR is dominated by positive bias, while the
relationship between SM and PR is much weaker compared to the two other models.
In LSMs, SM plays a vital role in the numerical solution for the momentum, sensible heat, and water vapor fluxes.
Take CLM5 as an example, the surface albedo, which is related to the surface radiation as well as the surface
hydrological states, is calculated by the two‐stream radiation scheme based on SM and vegetation (Lawrence
et al., 2018). Moisture in the shallow snow/soil layer provides the water sources to support the ground evaporation
and updates during model integration. Additionally, to reduce the biases in ET and total water storage in semiarid
regions, CLM5 includes a soil resistance parameterization scheme in which the thickness of the dry surface layer
is a function of SM and soil type in the top layer (Swenson & Lawrence, 2014). These relationships among
hydrological variables for different seasons change nonlinearly. Lawrence et al. (2007) reported that the parti-
tioning of ET (transpiration, soil evaporation, and canopy evaporation) can influence the hydrological cycles.
When the ratio of transpiration to ET increases, the soil is wetter than normal, and interseasonal soil water storage
and annual variations will also be enhanced. Through land‐atmosphere interaction, surface SM memory time
Figure 10. ET biases versus SM biases (a, c, e) and SM biases versus PR biases (b, d, f) in CLM5, HETSSEL, and
MATSIRO6 forced by GSWP3 at 651 stations over China during 1992–2012.
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 17 of 22

scales decrease, and subsurface SM will then exert a slightly greater influence on PR. In addition to ET, the CLM5
surface Ro parameterization scheme relies on SM saturation (Lawrence et al., 2018). These results indicate that
hydrologic variables are closely correlated with each other, and their relationships are strongly dependent on the
form of parameterization schemes in each LSM.
In this study, PR is examined along with other output variables from the LMIP in LS3MIP offline simulation for
two main reasons. First, PR is an important meteorological forcing variable, and it directly and strongly influences
the land surface hydrological variables such as SM, ET, Ro, and SCF (Wang et al., 2016). When studying the
multi‐model offline simulation of hydrological variables in LS3MIP over China, it is also necessary to discuss PR
to facilitate the understanding of performance and variations in hydrological simulations. Second, according to
Figures 1a, 4a and 5, and 6a, even if the same meteorological forcing data set is used to run different ESMs for
offline LSM simulation, there are still spreads in the PR outputs. In addition, according to the uncertainty analysis
in Figure 9a, the proportion of the model structure uncertainty in the total uncertainty of PR is 15.75% during
1900–2012 over China. The total uncertainty and forcing‐induced uncertainty decrease as the quality of PR
observations improves recently, while the proportion of uncertainty generated by the model increases and can
exceed 20% in many years. LSMs, as part of ESMs, are affected by the structure of their ESMs due to the
processing of the meteorological forcing data set, such as the way the grid space is decomposed. There are also
differences in the treatment of the input PR in different LSMs. Using rain/snow partitioning as an example, CLM5
repartitions the input total PR using a linear ramp. In CLM5, all PR is treated as snowfall below 0°C, as rainfall
above 2°C, and as a mix of rainfall and snowfall for intermediate temperatures. In JSBACH, PR in the form of rain
or snow depends on the fluxes of precipitate phases, which are obtained by integrating the relevant processes from
the top of the model to the pressure level in ECHAM (Lawrence et al., 2018; Reick et al., 2021; Roeckner
et al., 2003). Therefore, the evaluation of PR outputs will help to explore sources of the hydrological simulation
uncertainties in current LSMs. Thus, this historical hydrological assessment will enhance the understanding of
LMIP performance under climate change and provide scientific guidance for future water resource management
and climate adaptation strategies.
Appendix A
Table A1 lists all the abbreviations used in this paper and their definitions.
Table A1
A Table of Acronyms
Abbreviation Definition
ANOVA ANalysis of Variance Method
CC Correlation Coefficient
CMIP6 Sixth Phase of the Coupled Model Intercomparison Project
CN05.1 The gridded precipitation reference data set over China
CV Coefficient of Variation
D17 Global River Flow and Continental Discharge Data set
ESM Earth system model
ET Evapotranspiration
GLEAM Global Land Evaporation Amsterdam Model
GSWP3 The 3rd phase of the Global Soil Wetness Project
ILAMB International Land Model Benchmarking project
LFMIP Land Feedback Model Intercomparison Project
LMIP Land Model Intercomparison Project
LS3MIP Land Surface, Snow and Soil moisture Model Intercomparison
Project
LSM Land Surface Model
LUMIP Land Use Model Intercomparison Project
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 18 of 22

Data Availability Statement
The hydrological variables from LS3MIP in CMIP6 are available for downloading on the ESGF node: https://
esgf‐node.llnl.gov/search/cmip6/. The gridded CN05.1 precipitation data set is available at https://ccrc.iap.ac.cn/
resource. The GLEAM ET can be obtained at https://www.gleam.eu/. The WS2019 soil moisture data set is
available at https://nzc.iap.ac.cn/content?cid=24&aid=928. The runoff reference data set produced originally by
Dai and Trenberth (2002) and Dai et al. (2009) is available at https://rda.ucar.edu/datasets/ds551.0/. The MODIS
snow cover product (MYD10CM) is downloadable at the official website (https://nsidc.org/data/MYD10CM/
versions/6). All processed data have been deposited in Zenodo (Ma, 2023).
References
Abbaspour, K. C., Johnson, C. A., & Van Genuchten, M. T. (2004). Estimating uncertain flow and transport parameters using a sequential
uncertainty fitting procedure. Vadose Zone Journal,3(4), 1340–1352. https://doi.org/10.2113/3.4.1340
Abbaspour, K. C., Van Genuchten, M. T., Schulin, R., & Schläppi, E. (1997). A sequential uncertainty domain inverse procedure for estimating
subsurface flow and transport parameters. Water Resources Research,33(8), 1879–1892. https://doi.org/10.1029/97WR01230
Ashraf Vaghefi, S., Iravani, M., Sauchyn, D., Andreichuk, Y., Goss, G., & Faramarzi, M. (2019). Regionalization and parameterization of a
hydrologic model significantly affect the cascade of uncertainty in climate‐impact projections. Climate Dynamics,53(5–6), 2861–2886. https://
doi.org/10.1007/s00382‐019‐04664‐w
Baker, A. H., Hammerling, D. M., Mickelson, S. A., Xu, H., Stolpe, M. B., Naveau, P., et al. (2016). Evaluating lossy data compression on climate
simulation data within a large ensemble. Geoscientific Model Development,9(12), 4381–4403. https://doi.org/10.5194/gmd‐9‐4381‐2016
Balsamo, G., Beljaars, A., Scipal, K., Viterbo, P., van den Hurk, B., Hirschi, M., & Betts, A. K. (2009). A revised hydrology for the ECMWF
model: Verification from field site to terrestrial water storage and impact in the Integrated Forecast System. Journal of Hydrometeorology,
10(3), 623–643. https://doi.org/10.1175/2008JHM1068.1
Balsamo, G., Pappenberger, F., Dutra, E., Viterbo, P., & Van den Hurk, B. J. J. M. (2011). A revised land hydrology in the ECMWF model: A step
towards daily water flux prediction in a fully‐closed water cycle. Hydrological Processes,25(7), 1046–1054. https://doi.org/10.1002/hyp.7808
Bjarke, N., Barsugli, J., & Livneh, B. (2023). Ensemble of CMIP6 derived reference and potential evapotranspiration with radiative and advective
components. Scientific Data,10(1), 417. https://doi.org/10.1038/s41597‐023‐02290‐0
Bonan, G. B., Lombardozzi, D. L., Wieder, W. R., Oleson, K. W., Lawrence, D. M., Hoffman, F. M., & Collier, N. (2019). Model structure and
climate data uncertainty in historical simulations of the terrestrial carbon cycle (1850–2014). Global Biogeochemical Cycles,33(10), 1310–
1326. https://doi.org/10.1029/2019GB006175
Bosshard, T., Carambia, M., Goergen, K., Kotlarski, S., Krahe, P., Zappa, M., & Schär, C. (2013). Quantifying uncertainty sources in an ensemble
of hydrological climate‐impact projections. Water Resources Research,49(3), 1523–1536. https://doi.org/10.1029/2011WR011533
Table A1
Continued
Abbreviation Definition
M&F Uncertainty from the Interaction of Model and Forcing
MME Multi‐model mean
MODIS Moderate‐Resolution Imaging Spectroradiometer
NorSTD Normalized Standard Deviation
PR Precipitation
Princeton Princeton Global Forcing data set
Ro Total runoff
SCF Snow cover fraction
SM Soil moisture
SST Sum of square errors
STD Standard deviation
SUFI‐2 Sequential Uncertainty Fitting version 2
ubRMSE Unbiased root‐mean‐square biases
WATCH Water and Global Change
WCRP World Climate Research Program
WFDEI Water and Global Change Forcing Data Methodology Applied
to ERA‐Interim data
WS2019 The soil moisture reference data set from (Wang & Shi, 2019)
Acknowledgments
This work is jointly supported by the
National Science Fund for Distinguished
Young Scholars (Grant 41925021) and the
Science Fund for Creative Research
Groups of China (Grant 42221004). We
thank the support of the National Key
Scientific and Technological Infrastructure
project “Earth System Science Numerical
Simulator Facility (EarthLab)”. Two
anonymous reviewers are also thanked for
their constructive comments.
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 19 of 22

Brunke, M. A., Fairall, C. W., Zeng, X., Eymard, L., & Curry, J. A. (2003). Which bulk aerodynamic algorithms are least problematic in
computing ocean surface turbulent fluxes? Journal of Climate,16(4), 619–635. https://doi.org/10.1175/1520‐0442(2003)016<0619:
WBAAAL>2.0.CO;2
Campoy, A., Ducharne, A., Cheruy, F., Hourdin, F., Polcher, J., & Dupont, J. C. (2013). Response of land surface fluxes and precipitation to
different soil bottom hydrological conditions in a general circulation model. Journal of Geophysical Research: Atmospheres,118(19), 10725–
10739. https://doi.org/10.1002/jgrd.50627
Chen, S., & Yuan, X. (2022). Quantifying the uncertainty of internal variability in future projections of seasonal soil moisture droughts over
China. The Science of the Total Environment,824, 153817. https://doi.org/10.1016/j.scitotenv.2022.153817
Compo, G. P., Whitaker, J. S., Sardeshmukh, P. D., Matsui, N., Allan, R. J., Yin, X., et al. (2011). The twentieth century reanalysis project.
Quarterly Journal of the Royal Meteorological Society,137(654), 1–28. https://doi.org/10.1002/qj.776
Dai, A. (2017). Dai and Trenberth global river flow and continental discharge dataset. Research Data Archive at the National Center for At-
mospheric Research, Computational and Information Systems Laboratory.https://doi.org/10.5065/D6V69H1T
Dai, A., Qian, T., Trenberth, K. E., & Milliman, J. D. (2009). Changes in continental freshwater discharge from 1948 to 2004. Journal of Climate,
22(10), 2773–2792. https://doi.org/10.1175/2008JCLI2592.1
Dai, A., & Trenberth, K. E. (2002). Estimates of freshwater discharge from continents: Latitudinal and seasonal variations. Journal of Hydro-
meteorology,3(6), 660–687. https://doi.org/10.1175/1525‐7541(2002)003<0660:EOFDFC>2.0.CO;2
Danabasoglu, G., Lamarque, J. F., Bacmeister, J., Bailey, D. A., DuVivier, A. K., Edwards, J., et al. (2020). The community earth system model
version 2 (CESM2). Journal of Advances in Modeling Earth Systems,12(2). https://doi.org/10.1029/2019MS001916
Di Sante, F., Coppola, E., & Giorgi, F. (2021). Projections of river floods in Europe using EURO‐CORDEX, CMIP5 and CMIP6 simulations.
International Journal of Climatology,41(5), 3203–3221. https://doi.org/10.1002/joc.7014
Döscher, R., Acosta, M., Alessandri, A., Anthoni, P., Arneth, A., Arsouze, T., et al. (2021). The EC‐earth3 Earth system model for the climate
model intercomparison project 6. Geoscientific Model Development Discussions,15(7), 2973–3020. https://doi.org/10.5194/gmd‐15‐2973‐
2022
Eyring, V., Bony, S., Meehl, G. A., Senior, C. A., Stevens, B., Stouffer, R. J., & Taylor, K. E. (2016). Overview of the coupled model inter-
comparison project phase 6 (CMIP6) experimental design and organization. Geoscientific Model Development,9(5), 1937–1958. https://doi.
org/10.5194/gmd‐9‐1937‐2016
Faramarzi, M., Abbaspour, K. C., Schulin, R., & Yang, H. (2009). Modelling blue and green water resources availability in Iran. Hydrological
Processes: International Journal,23(3), 486–501. https://doi.org/10.1002/hyp.7160
Fisher, R. A., & Koven, C. D. (2020). Perspectives on the future of land surface models and the challenges of representing complex terrestrial
systems. Journal of Advances in Modeling Earth Systems,12(4), e2018MS001453. https://doi.org/10.1029/2018ms001453
Freni, G., Mannina, G., & Viviani, G. (2009). Urban runoff modelling uncertainty: Comparison among Bayesian and pseudo‐Bayesian methods.
Environmental Modelling & Software,24(9), 1100–1111. https://doi.org/10.1016/j.envsoft.2009.03.003
Gao, X., Wang, M., & Giorgi, F. (2013). Climate change over China in the 21st century as simulated by BCC_CSM1. 1‐RegCM4. 0. Atmospheric
and Oceanic Science Letters,6(5), 381–386. https://doi.org/10.3878/j.issn.1674‐2834.13.0029
Guo, H., Zhan, C., Ning, L., Li, Z., & Hu, S. (2022). Evaluation and comparison of CMIP6 and CMIP5 model performance in simulating the
runoff. Theoretical and Applied Climatology,149(3–4), 1451–1470. https://doi.org/10.1007/s00704‐022‐04118‐0
Hagemann, S., Chen, C., Haerter, J. O., Heinke, J., Gerten, D., & Piani, C. (2011). Impact of a statistical bias correction on the projected hy-
drological changes obtained from three GCMs and two hydrology models. Journal of Hydrometeorology,12(4), 556–578. https://doi.org/10.
1175/2011JHM1336.1
Hall, D. K., & Riggs, G. A. (2016). MODIS/Aqua snow cover monthly L3 global 0.05Deg CMG (version 6). NASA national snow and ice data
center distributed active archive center. https://doi.org/10.5067/MODIS/MYD10CM.006
Hawkins, E., & Sutton, R. (2009). The potential to narrow uncertainty in regional climate predictions. Bulletin of the American Meteorological
Society,90(8), 1095–1108. https://doi.org/10.1175/2009BAMS2607.1
Hawkins, E., & Sutton, R. (2011). The potential to narrow uncertainty in projections of regional precipitation change. Climate Dynamics,37(1–2),
407–418. https://doi.org/10.1007/s00382‐010‐0810‐6
Haydon, S., & Deletic, A. (2009). Model output uncertainty of a coupled pathogen indicator–hydrologic catchment model due to input data
uncertainty. Environmental Modelling & Software,24(3), 322–328. https://doi.org/10.1016/j.envsoft.2008.09.004
Held, I. M., & Soden, B. J. (2006). Robust responses of the hydrological cycle to global warming. Journal of Climate,19(21), 5686–5699. https://
doi.org/10.1175/JCLI3990.1
Hirsch, A. L., Ridder, N. N., Perkins‐Kirkpatrick, S. E., & Ukkola, A. (2021). CMIP6 MultiModel evaluation of present‐day heatwave attributes.
Geophysical Research Letters,48(22), e2021GL095161. https://doi.org/10.1029/2021gl095161
Huntington, T. G. (2006). Evidence for intensification of the global water cycle: Review and synthesis. Journal of Hydrology,319(1–4), 83–95.
https://doi.org/10.1016/j.jhydrol.2005.07.003
Jiang, D., Tian, Z., & Lang, X. (2016). Reliability of climate models for China through the IPCC third to fifth assessment reports. International
Journal of Climatology,36(3), 1114–1133. https://doi.org/10.1002/joc.4406
Kiehl, J. T. (2007). Twentieth century climate model response and climate sensitivity. Geophysical Research Letters,34(22). https://doi.org/10.
1029/2007GL031383
Koster, R. D., Guo, Z., Yang, R., Dirmeyer, P. A., Mitchell, K., & Puma, M. J. (2009). On the nature of soil moisture in land surface models.
Journal of Climate,22(16), 4322–4335. https://doi.org/10.1175/2009JCLI2832.1
Koster, R. D., Mahanama, S. P. P., Yamada, T. J., Balsamo, G., Berg, A. A., Boisserie, M., et al. (2010). Contribution of land surface initialization
to subseasonal forecast skill: First results from a multi‐model experiment. Geophysical Research Letters,37(2). https://doi.org/10.1029/
2009gl041677
Lawrence, D. M., Fisher, R. A., Koven, C. D., Oleson, K. W., Swenson, S., Vertenstein, M., et al. (2018). Technical description of version 5.0 of
the community land model (CLM). Retrieved from https://escomp.github.io/ctsm‐docs/versions/release‐clm5.0/html/tech_note/index.html
Lawrence, D. M., Thornton, P. E., Oleson, K. W., & Bonan, G. B. (2007). The partitioning of evapotranspiration into transpiration, soil evap-
oration, and canopy evaporation in a GCM: Impacts on land–atmosphere interaction. Journal of Hydrometeorology,8(4), 862–880. https://doi.
org/10.1175/JHM596.1
Lehning, M., & Katul, G. (2013). Snow‐atmosphere interactions and hydrological consequences. Advances in Water Resources,55, 1–3. https://
doi.org/10.1016/j.advwatres.2013.02.001
Lovato, T., Peano, D., Butenschön, M., Materia, S., Iovino, D., Scoccimarro, E., et al. (2022). CMIP6 simulations with the CMCC Earth system
model (CMCC‐ESM2). Journal of Advances in Modeling Earth Systems,14(3), e2021MS002814. https://doi.org/10.1029/2021ms002814
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 20 of 22

Lu, J., Wang, G., Li, S., Feng, A., Zhan, M., Jiang, T., et al. (2021). Projected land evaporation and its response to vegetation greening over China
under multiple scenarios in the CMIP6 models. Journal of Geophysical Research: Biogeosciences,126(9), e2021JG006327. https://doi.org/10.
1029/2021jg006327
Ma, X. (2023). Evaluation and uncertainty analysis of the land surface hydrology in LS3MIP models over China [Dataset]. Zenodo.https://doi.org/
10.5281/zenodo.10081625
Ma, X., & Wang, A. H. (2022). Systematic evaluation of a high‐resolution CLM5 simulation over continental China for 1979–2018. Journal of
Hydrometeorology,23(12), 1879–1897. https://doi.org/10.1175/JHM‐D‐22‐0051.1
Martenss, B., Miralles, D. G., Lievens, H., Van Der Schalie, R., De Jeu, R. A., Fernández‐Prieto, D., et al. (2017). GLEAM v3: Satellite‐based land
evaporation and root‐zone soil moisture. Geoscientific Model Development,10(5), 1903–1925. https://doi.org/10.5194/gmd‐10‐1903‐2017
Masud, B., Cui, Q., Ammar, M. E., Bonsal, B. R., Islam, Z., & Faramarzi, M. (2021). Means and extremes: Evaluation of a CMIP6 multi‐model
ensemble in reproducing historical climate characteristics across Alberta, Canada. Water,13(5), 737. https://doi.org/10.3390/w13050737
Materia, S., Ardilouze, C., Prodhomme, C., Donat, M. G., Benassi, M., Doblas‐Reyes, F. J., et al. (2022). Summer temperature response to
extreme soil water conditions in the Mediterranean transitional climate regime. Climate Dynamics,58(7–8), 1943–1963. https://doi.org/10.
1007/s00382‐021‐05815‐8
Meresa, H., Tischbein, B., & Mekonnen, T. (2022). Climate change impact on extreme precipitation and peak flood magnitude and frequency:
Observations from CMIP6 and hydrological models. Natural Hazards,111(3), 2649–2679. https://doi.org/10.1007/s11069‐021‐05152‐3
Miao, Y., & Wang, A. (2020). A daily 0.25×0.25 hydrologically based land surface flux dataset for conterminous China, 1961–2017. Journal of
Hydrology,590, 125413. https://doi.org/10.1016/j.jhydrol.2020.125413
Miralles, D. G., Holmes, T. R. H., De Jeu, R. A. M., Gash, J. H., Meesters, A. G. C. A., & Dolman, A. J. (2011). Global land‐surface evaporation
estimated from satellite‐based observations. Hydrology and Earth System Sciences,15(2), 453–469. https://doi.org/10.5194/hess‐15‐453‐2011
Mudryk, L., Santolaria‐Otín, M., Krinner, G., Ménégoz, M., Derksen, C., Brutel‐Vuilmet, C., et al. (2020). Historical Northern Hemisphere snow
cover trends and projected changes in the CMIP6 multi‐model ensemble. The Cryosphere,14(7), 2495–2514. https://doi.org/10.5194/tc‐14‐
2495‐2020
Mueller, B., & Seneviratne, S. I. (2014). Systematic land climate and evapotranspiration biases in CMIP5 simulations. Geophysical Research
Letters,41(1), 128–134. https://doi.org/10.1002/2013GL058055
Müller, W. A., Jungclaus, J. H., Mauritsen, T., Baehr, J., Bittner, M., Budich, R., et al. (2018). A higher‐resolution version of the max planck
institute earth system model (MPI‐ESM1. 2‐HR). Journal of Advances in Modeling Earth Systems,10(7), 1383–1413. https://doi.org/10.1029/
2017ms001217
Papadimitriou, L. V., Koutroulis, A. G., Grillakis, M. G., & Tsanis, I. K. (2017). The effect of GCM biases on global runoff simulations of a land
surface model. Hydrology and Earth System Sciences,21(9), 4379–4401. https://doi.org/10.5194/hess‐21‐4379‐2017
Qiao, L., Zuo, Z., & Xiao, D. (2022). Evaluation of soil moisture in CMIP6 simulations. Journal of Climate,35(2), 779–800. https://doi.org/10.
1175/JCLI‐D‐20‐0827.1
Qu, X., & Hall, A. (2014). On the persistent spread in snow‐albedo feedback. Climate Dynamics,42(1–2), 69–81. https://doi.org/10.1007/s00382‐
013‐1774‐0
Reick, C. H., Gayler, V., Goll, D., Hagemann, S., Heidkamp, M., Nabel, J. E. M. S., et al. (2021). JSBACH 3 ‐ The land component of the MPI
earth system model: Documentation of version 3.2. Berichte zur Erdsystemforschung,240.https://doi.org/10.17617/2.3279802
Roeckner, E., Bäuml, G., Bonaventura, L., Brokopf, R., Esch, M., Giorgetta, M., et al. (2003). The atmospheric general circulation model
ECHAM 5. PART I: Model description. Report Max‐Planck‐Institut für Meteorologie,349. Retrieved from https://hdl.handle.net/11858/00‐
001M‐0000‐0012‐0144‐5
Seneviratne, S. I., Wilhelm, M., Stanelle, T., van den Hurk, B., Hagemann, S., Berg, A., et al. (2013). Impact of soil moisture‐climate feedbacks on
CMIP5 projections: First results from the GLACE‐CMIP5 experiment. Geophysical Research Letters,40(19), 5212–5217. https://doi.org/10.
1002/grl.50956
Sheffield, J., Goteti, G., & Wood, E. F. (2006). Development of a 50‐year high‐resolution global dataset of meteorological forcings for land
surface modeling. Journal of Climate,19(13), 3088–3111. https://doi.org/10.1175/jcli3790.1
Sillmann, J., Kharin, V. V., Zwiers, F. W., Zhang, X., & Bronaugh, D. (2013). Climate extremes indices in the CMIP5 multimodel ensemble: Part
2. Future climate projections. Journal of Geophysical Research: Atmospheres,118(6), 2473–2493. https://doi.org/10.1002/jgrd.50188
Srivastava, A., Grotjahn, R., & Ullrich, P. A. (2020). Evaluation of historical CMIP6 model simulations of extreme precipitation over contiguous
US regions. Weather and Climate Extremes,29, 100268. https://doi.org/10.1016/j.wace.2020.100268
Sun, P., Zou, Y., Yao, R., Ma, Z., Bian, Y., Ge, C., & Lv, Y. (2023). Compound and successive events of extreme precipitation and extreme runoff
under heatwaves based on CMIP6 models. The Science of the Total Environment,878, 162980. https://doi.org/10.1016/j.scitotenv.2023.
162980
Swenson, S. C., & Lawrence, D. M. (2014). Assessing a dry surface layer‐based soil resistance parameterization for the Community Land Model
using GRACE and FLUXNET‐MTE data. Journal of Geophysical Research: Atmospheres,119(17), 10–299. https://doi.org/10.1002/
2014JD022314
Tatebe, H., Ogura, T., Nitta, T., Komuro, Y., Ogochi, K., Takemura, T., et al. (2019). Description and basic evaluation of simulated mean state,
internal variability, and climate sensitivity in MIROC6. Geoscientific Model Development,12(7), 2727–2765. https://doi.org/10.5194/gmd‐12‐
2727‐2019
Taylor, K. E. (2001). Summarizing multiple aspects of model performance in a single diagram. Journal of Geophysical Research,106(D7), 7183–
7192. https://doi.org/10.1029/2000JD900719
Trenberth, K. E., & Guillemot, C. J. (1998). Evaluation of the atmospheric moisture and hydrological cycle in the NCEP/NCAR reanalyses.
Climate Dynamics,14(3), 213–231. https://doi.org/10.1007/s003820050219
van den Hurk, B., Kim, H., Krinner, G., Seneviratne, S. I., Derksen, C., Oki, T., et al. (2016). LS3MIP (v1.0) contribution to CMIP6: The land
surface, snow and soil moisture model intercomparison project – Aims, setup and expected outcome. Geoscientific Model Development,9(8),
2809–2832. https://doi.org/10.5194/gmd‐9‐2809‐2016
Vano, J. A., Das, T., & Lettenmaier, D. P. (2012). Hydrologic sensitivities of Colorado River runoff to changes in precipitation and temperature.
Journal of Hydrometeorology,13(3), 932–949. https://doi.org/10.1175/JHM‐D‐11‐069.1
Viovy, N., & Ciais, P. (2009). A combined dataset for ecosystem modelling, CRUNCEP. Atmospheric forcing data for the community land model.
Research Data Archive at the National Center for Atmospheric Research, Computational and Information Systems Laboratory.https://doi.org/
10.5065/PZ8F‐F017
Wang, A., Kong, X., Chen, Y., & Ma, X. (2022). Evaluation of soil moisture in CMIP6 multimodel simulations over conterminous China. Journal
of Geophysical Research: Atmospheres,127(19). https://doi.org/10.1029/2022JD037072
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 21 of 22

Wang, A., Miao, Y., Kong, X., & Wu, H. (2022). Future changes in global runoff and runoff coefficient from CMIP6 multi‐model simulation
under SSP1‐2.6 and SSP5‐8.5 scenarios. Earth's Future,10(12), e2022EF002910. https://doi.org/10.1029/2022ef002910
Wang, A., & Shi, X. (2019). A multilayer soil moisture dataset based on the gravimetric method in China and its characteristics. Journal of
Hydrometeorology,20(8), 1721–1736. https://doi.org/10.1175/JHM‐D‐19‐0035.1
Wang, A., & Zeng, X. (2012). Evaluation of multireanalysis products with in situ observations over the Tibetan Plateau. Journal of Geophysical
Research,117(D5). https://doi.org/10.1029/2011JD016553
Wang, A., Zeng, X., & Guo, D. (2016). Estimates of global surface hydrology and heat fluxes from the Community Land Model (CLM4. 5) with
four atmospheric forcing datasets. Journal of Hydrometeorology,17(9), 2493–2510. https://doi.org/10.1175/JHM‐D‐16‐0041.1
Weedon, G. P., Balsamo, G., Bellouin, N., Gomes, S., Best, M. J., & Viterbo, P. (2014). The WFDEI meteorological forcing data set: WATCH
forcing data methodology applied to ERA‐interim reanalysis data. Water Resources Research,50(9), 7505–7514. https://doi.org/10.1002/
2014WR015638
Weedon, G. P., Gomes, S., Viterbo, P., Shuttleworth, W. J., Blyth, E., Österle, H., et al. (2011). Creation of the WATCH forcing data and its use to
assess global and regional reference crop evaporation over land during the twentieth century. Journal of Hydrometeorology,12(5), 823–848.
https://doi.org/10.1175/2011JHM1369.1
Wu, J., Gao, X., Giorgi, F., & Chen, D. (2017). Changes of effective temperature and cold/hot days in late decades over China based on a high
resolution gridded observation dataset. International Journal of Climatology,37(S1), 788–800. https://doi.org/10.1002/joc.5038
Wu, J., & Gao, X. J. (2013). A gridded daily observation dataset over China region and comparison with other datasets. Chinese Journal of
Geophysics,56(4), 1102–1111. (In Chinese with English abstract). https://doi.org/10.6038/cjg20130406
Wu, T., Lu, Y., Fang, Y., Xin, X., Li, L., Li, W., et al. (2019). The Beijing climate center climate system model (BCC‐CSM): The main progress
from CMIP5 to CMIP6. Geoscientific Model Development,12(4), 1573–1600. https://doi.org/10.5194/gmd‐12‐1573‐2019
Xie, W., Zhou, B., Han, Z., & Xu, Y. (2022). Substantial increase in daytime‐nighttime compound heat waves and associated population exposure
in China projected by the CMIP6 multimodel ensemble. Environmental Research Letters,17(4), 045007. https://doi.org/10.1088/1748‐9326/
ac592d
Xin, X., Wu, T., Zhang, J., Yao, J., & Fang, Y. (2020). Comparison of CMIP6 and CMIP5 simulations of precipitation in China and the East Asian
summer monsoon. International Journal of Climatology,40(15), 6423–6440. https://doi.org/10.1002/joc.6590
Xu, Y., Zhang, X., Hao, Z., Hao, F., & Li, C. (2021). Projections of future meteorological droughts in China under CMIP6 from a three‐
dimensional perspective. Agricultural Water Management,252, 106849. https://doi.org/10.1016/j.agwat.2021.106849
Yang, X., Yong, B., Ren, L., Zhang, Y., & Long, D. (2017). Multi‐scale validation of GLEAM evapotranspiration products over China via
ChinaFLUX ET measurements. International Journal of Remote Sensing,38(20), 5688–5709. https://doi.org/10.1080/01431161.2017.
1346400
Yip, S., Ferro, C. A., Stephenson, D. B., & Hawkins, E. (2011). A simple, coherent framework for partitioning uncertainty in climate predictions.
Journal of Climate,24(17), 4634–4643. https://doi.org/10.1175/2011JCLI4085.1
Yuan, S., Quiring, S. M., & Leasor, Z. T. (2021). Historical changes in surface soil moisture over the contiguous United States: An assessment of
CMIP6. Geophysical Research Letters,48(1), e2020GL089991. https://doi.org/10.1029/2020gl089991
Zeng, J., Yuan, X., Ji, P., & Shi, C. (2021). Effects of meteorological forcings and land surface model on soil moisture simulation over China.
Journal of Hydrology,603, 126978. https://doi.org/10.1016/j.jhydrol.2021.126978
Zhang, H., Zhang, F., Zhang, G., & Yan, W. (2022). Why do CMIP6 models fail to simulate snow depth in terms of temporal change and high
mountain snow of China skillfully? Geophysical Research Letters,49(15), e2022GL098888. https://doi.org/10.1029/2022gl098888
Zhu, Y. Y., & Yang, S. (2020). Evaluation of CMIP6 for historical temperature and precipitation over the Tibetan Plateau and its comparison with
CMIP5. Advances in Climate Change Research,11(3), 239–251. https://doi.org/10.1016/j.accre.2020.08.001
Earth and Space Science
10.1029/2023EA003391
MA AND WANG 22 of 22