Impact of Initialized Land Surface Temperature and Snowpack on Subseasonal to Seasonal Prediction Project, Phase I (LS4P-I): Organization and Experimental design

Myung-Seo Koo10, Zhaohui Lin11, Yuhei Takaya12, Tomonori Sato13, Constantin Ardilouze3, 5 Subodh K. Saha14, Mei Zhao15, Xin-Zhong Liang5, Frederic Vitart16, Xin Li2, Ping Zhao17, David Neelin1, Weidong Guo18, Miao Yu19, Yun Qian20, Samuel S. P. Shen21, Yang Zhang18, Kun Yang22, Ruby Leung20, Jing Yang23, Yuan Qiu11, Michael A. Brunke4, Sin Chan Chou24, Michael Ek 25 , Tianyi Fan23, Hong Guan 26 , Hai Lin 27 , Shunlin Liang 28 , Stefano Materia 29 , Tetsu Nakamura13, Xin Qi23, Retish Senan16, Chunxiang Shi30, Hailan Wang26, Helin Wei26, Shaocheng 10 Xie7, Haoran Xu5, Hongliang Zhang31, Yanling Zhan11, Weiping Li32, Xueli Shi32, Paulo Nobre24, Yi Qin22, Jeff Dozier33, Craig R. Ferguson34, Gianpaolo Balsamo16, Qing Bao35, Jinming Feng11, Jinkyu Hong36, Songyou Hong10, Huilin Huang1, Duoying Ji23, Zhenming Ji37, Shichang Kang38, Yanluan Lin22, Weiguang Liu39,19, Ryan Muncaster27, Yan Pan18, Daniele Peano29, Patricia de Rosnay16, Hiroshi G. Takahashi 40, Jianping Tang18, Guiling Wang39, Shuyu Wang18, Weicai 15 Wang2, Xu Zhou2, Yuejian Zhu26


Introduction
Subseasonal-to-seasonal (S2S) prediction, especially the prediction of extreme hydroclimatic events such as droughts and floods, is not only scientifically challenging but also has substantial 45 societal impacts since such phenomena can have serious agricultural, economic, and ecological consequences (Merryfield et al., 2020). However, the prediction skill for precipitation anomalies in spring and summer months, a significant component of extreme climate events, has remained stubbornly low for years. It is therefore important to understand the sources of such predictability and to develop more reliable monitoring and prediction capabilities. Various 50 mechanisms have been attributed to the S2S predictability. For instance, oceanic basin-wide tropical sea surface temperature (SST) anomalies are known to play a major role in causing extreme events. The connection between SST [e.g., El Niño Southern Oscillation (ENSO), Pacific Decadal Oscillation (PDO), Atlantic Multidecadal Oscillation (AMO), and Madden-Julian oscillation (MJO)] and the associated weather and climate predictability has been 55 extensively studied for decades (Trenberth et al., 1988;Ting and Wang, 1997;Barlow et al., 2001;Schubert et al., 2008;Jia and Yang, 2013;Seager et al., 2014). The linkage of extreme hydrological events to tropical ocean basin SST anomalies allows us to predict them with useful skill at long lead times, ranging from a few months to a few years. Despite significant correlations and demonstrated predictive value, numerous studies based on observational data 60 analyses and numerical simulations have consistently shown that SST alone only partially explains the phenomena of predictability (Rajagopalan et al., 2000;Schubert et al., 2004Schubert et al., , 2009Scaife et al., 2009;Mo et al., 2009;Rui and Wang, 2011;Pu et al., 2016;Xue et al., 2016aXue et al., , b, 2018. For instance, the 2015-2016 El Niño event, one of the strongest since 1950, was associated with an extraordinary Californian drought, while the 2016-2017 La Niña event has 65 been related to record rainfall that effectively ended the 5-year Californian drought, contrary to established canonical SST-Californian drought/flood relationships. In South America, there is also an example where the canonical association of thermally direct, SST-driven atmospheric circulation fails (Robertson and Mechoso, 2000;Nobre et al., 2012). Although an important role for random atmospheric internal variability in such extreme events has been proposed (Hoerling 70 et al., 2009), such exceptions in explaining vital hydroclimatic extreme events as well as low prediction skill underscore the need to seek explanations beyond current traditional approaches.
It is therefore imperative to explore other avenues to improve S2S prediction skill. Studies have demonstrated that the predictive ability of models may come from their capability to represent land surface features that have inertia, such as vegetation (evolving cover 75 and density), soil moisture, snow, among others (e.g., Xue et al., 1996aXue et al., , 2010bLu et al., 2001;Delire et al., 2004;Koster et al., 2004Koster et al., , 2006Gastineau et al., 2017). Most land/atmosphere interaction studies have focused on local effects, for instance, such as those in the previous Global Land Atmosphere System Study (GLASS) experiment (Koster et al., 2006). The possible remote (non-local) effects of large-scale spring land surface/subsurface temperature 80 (LST/SUBT) anomalies in geographical areas upstream of the areas which experience late spring-summer drought/flood, an underappreciated relation, have largely been ignored.
Observational data in the Tibetan Plateau and the Rocky Mountains have shown that land surface temperature anomalies can be sustained for entire seasons, and that they are accompanied by persistent subsurface temperature, snow and albedo anomalies . Since only 2-m 85 air temperature (T-2m) has global coverage, and because it is very close to LST in stations with measurements for both ; also see the discussion in Section 5.1), observed T-2m data have been used in diagnostic studies to identify spatial and temporal characteristics of land surface temperature variability and its relationship with other climate variables. Figure 1 exhibits the persistence of the monthly mean difference of T-2m between warm and cold Mays, 90 which are selected based on a threshold of one-half standard deviation during the period 1981-2010. Those anomalies can persist for several months, especially during the spring. Preliminary studies have been carried out to explore the relationship between spring LST/SUBT anomalies and summer precipitation anomalies in downstream regions in North America and East Asia (Xue et al., 2002(Xue et al., , 2012(Xue et al., , 2016b(Xue et al., , 2018Diallo et al., 2019). Data analyses identify significant 95 correlations between springtime T-2m cold (warm) anomalies in both the Rocky Mountains and Tibetan Plateau and respective downstream drought (flood) events in late spring/summer.
Modeling studies using the NCEP Global Forecast System (GFS, Xue et al., 2004) and the regional climate model version of Weather Research and Forecasting (WRF; Skamarock et al., 2008), both of which were coupled with a land model SSiB (Xue et al., 1991;Zhan et al., 2003) 100 using observed T-2m and reanalysis data as constraints, have also suggested that there is a remote effect of land temperature changes in the Rocky Mountains and the Tibetan Plateau on their respective downstream regions with a magnitude comparable to the more familiar effects of https://doi.org/10.5194/gmd-2020-329 Preprint. Discussion started: 7 January 2021 c Author(s) 2021. CC BY 4.0 License.

6
SST and atmospheric internal variability. Recent studies have further revealed the presence of LST/SUBT effects in other seasons and regions (Shukla et al., 2019). 105 The main hypothesis is that LST and SUBT anomalies in early spring carry information about the amount of water locked in frozen ground (i.e., the amount of snow/ice on the ground and in the frozen soil layer below) which is melted in late spring and early summer, as well as the thermal status from the preceding winter which has a long memory. The more snow/ice on the ground and in the frozen soil layer, the longer the seasonal transition from spring to summer. 110 The timing of such a seasonal transition over high elevation areas in the western part (upstream) of the land mass plays an important role in setting up the circulation pattern downstream over the lower elevation areas to the east. The strength as well as the duration of LST/SUBT interactions with downstream circulation patterns should affect the occurrence of droughts or floods in late spring/summer over the eastern part of the continents.

115
One factor that is closely related to the LST/SUBT anomaly is light absorbing particles (LAPs) in snow. In particular, the snow darkening effect by LAPs in snow due to deposition of aerosols, e.g. desert dust, black carbon and organic carbon from industrial pollution, biomass burning, and nearby wildfires, can reduce snow albedo which increases the absorption of solar radiation by the land surface. This enhanced energy absorption can alter the surface energy 120 balance, leading to anomalous T-2m and snowmelt during the boreal spring. Recent studies have shown that snow darkening effect can lead to large increases in surface temperature over the Tibetan Plateau in April-May, thereby strongly affecting the subsequent evolution of the jet stream and variability of summertime precipitation over India, East Asia and Eurasia , Rashimi et al. 2019, Sang et al. 2019. At present, the representation of snow amount, 125 coverage, and LAPs in snow are either absent or grossly inadequate in most climate models, especially in high mountain regions. This could be one of the major reasons for the large diversity in simulated T-2m conditions in current Earth System Models (ESMs).
A number of studies have also started to pursue the potential causes of the spring LST/SUBT anomaly in the Tibetan Plateau and the Rocky Mountains. Analyses based on 130 observational station data over the Tibetan Plateau show that the LST anomaly is highly correlated with anomalous snow, surface albedo and SUBT in the preceding months. Using data from an off-line model incorporating permafrost processes (Li et al., 2010) driven with observed meteorological data as forcing over the Tibetan Plateau, a regression model can predict a LST https://doi.org/10.5194/gmd-2020-329 Preprint. Discussion started: 7 January 2021 c Author(s) 2021. CC BY 4.0 License. 7 anomaly at the monthly and seasonal scales, with surface albedo and middle-layer (40-160 cm) 135 SUBT as predictors . Additional analyses using observational data show that spring LST in the Tibetan Plateau is significantly coupled with the regional snow cover in preceding months. The latter is also strongly coupled with February atmospheric circulation patterns and wave activity in mid-to-high latitudes . Moreover, a modeling study focusing on North America (Broxton et al., 2017) showed that snow water equivalent 140 (SWE) anomalies more strongly affect April-June temperature forecasts than SST anomalies.
It is likely that a temporary filtered response to snow anomalies may be preserved in the LST and SUBT anomalies, and this mechanism deserves further investigation. Additional research on the causes of LST/SUBT anomalies and the association with LAPs in the snow would likely help us to better understand the sources of S2S predictability.

145
In the following text, Section 2 introduces the historical development of "Impact of initialized Land Surface temperature and Snowpack on Subseasonal to Seasonal Prediction" (LS4P) and its research objectives. Section 3 presents the LS4P Phase I protocol (LS4P-I): its experimental design and model output requirements. Section 4 discusses causes of current LS4P-I models' deficiencies in preserving land memory and possible approaches for improvement.

150
Section 5 briefly presents some preliminary LS4P-I results and discusses the future plan and prospectives.

Prediction Programs
Although T-2m measurement has the longest meteorological observational record with global coverage and the best quality among various land surface variables, its application in S2S prediction has largely been overlooked. Preliminary experiments to test the impact of model initialization of LST/SUBT on the S2S prediction are encouraging, but the results were obtained 160 from only one ESM and one RCM, with North America and East Asia as the focus regions (Xue et al., 2016b(Xue et al., , 2018. Due to the existing shortcomings and uncertainties associated with models, it is imperative to have a multi-model approach in order to further test the LST-memory hypothesis and to explore predictability in more regions. Furthermore, since LS4P proposes a new approach, involving a decade-long effort to explore, test, and understand the concept, as Since the inception of the LS4P in December 2018, more than forty groups worldwide have participated in this effort, including twenty-one (21) ESM groups, many of which are from major climate research centers, nine (9) RCM groups, and seven (7) data groups. A description of the major components of each of the ESM and RCM models is summarized in Appendix A.

180
A complete listing of LS4P group information can be found at https://ls4p.geog.ucla.edu/.
Because LS4P takes a new approach in S2S prediction, GEWEX, the Third Pole Environment skill at the S2S timescale, between two weeks and a season (WMO, 2013, Vitart et al., 2017Merryfield et al., 2020). Their S2S project has the study of land initialization and configuration as one of its major activities. The LS4P research activities to address these scientific challenges are consistent with those of the WWRP/WCRP S2S project. The LS4P activity is also closely related to the TPE program. The TPE has closely worked with LS4P to provide and maintain a 195 data base to support this project. The first phase of LS4P will be a joint effort with the TPE https://doi.org/10.5194/gmd-2020-329 Preprint. Discussion started: 7 January 2021 c Author(s) 2021. CC BY 4.0 License.

9
Earth System Model Inter-comparison Project (TPEMIP), which focuses on regional-scale Earth system modeling over the high elevation Tibetan Plateau region. The LS4P initiative is also relevant to the GLASS Panel because estimating the contribution of land memory to atmospheric predictability from convective to seasonal timescales is one of its main themes. This requires an This LS4P project intends to address the following questions: • What is the impact of initializing large scale LST/SUBT and LAPs in snow in climate models on S2S prediction in different regions?
• What are the relative roles and uncertainties of the associated land processes compared to 210 those of SST in S2S prediction? How do they synergistically enhance S2S predictability?
LS4P focuses on process understanding and predictability, hence it is different from, and complements, other international projects that focus on the operational S2S prediction. The majority of the models participating in LS4P are ESMs, although, there is a good amount of RCMs involved. Some difficulties have been identified regarding how to apply RCMs for 215 studying the LST/SUBT effect (Xue et al., 2012). The main concern is that imposition of the same lateral boundary conditions (LBC) for RCM's control and anomaly runs may hamper the necessary modification of circulations at larger scales in the anomaly run. This issue will be more comprehensively studied in LS4P using a much larger RCM domain configuration to reduce the LBC control on the large-scale change.

220
LS4P will organize inter-comparison and validation studies, using satellite, and available ground-based observations, among participating model subgroups in order to pursue a better understanding of the relationship among LAPs in snow in high mountain regions, their deposition, snowmelt/albedo reduction, and the corresponding LST/SUBT anomaly, as well as how snow darkening affects the S2S predictability.

225
The project will ultimately consist of several phases, and each of which will focus on a particular high mountain region on one continent as a focal point. The LS4P-I will investigate the LST/SUBT effect in Tibetan Plateau. Testing the effect of LAPs in snow is still in the preparation stage and will not be conducted in the Phase I experiments. The second phase of LS4P will focus on the Rocky Mountains of North America. It is intended that this project will 230 also provide motivation for examining additional high mountains in other continents with similar geographic structure, such as those in South America, for the potential of the LST/SUBT effect to provide added-value to S2S prediction and understanding of the pertinent physical principles.

235
The Tibetan plateau region provides an ideal geographic location for the LS4P-I test owing to its relatively high elevation and large-scale (areal extent) as well as the presence of persistent LST anomalies. The Tibetan Plateau provides thermal and dynamic forcings which drive the Asian monsoon through a huge, elevated heat source in the middle troposphere, and this has been reported in the literature for decades (e.g., Ye, 1981;Yanai et al., 1992;Wu et al., 2007;Wang et 240 al., 2008;Yao et al., 2019). A large impact of the Tibetan Plateau LST/SUBT anomaly effect should be expected and has been demonstrated in a preliminary test (Xue et al., 2018).

Observational data for LS4P Phase I (LS4P-I)
There are large amounts of observational data available in the Tibetan Plateau area. The TPE 245 has conducted comprehensive measurements over Tibetan Plateau for more than a decade and has integrated the observational data into the National Tibetan Plateau Data Center (Li et al., 2020), which has more than 2400 different data sets for scientific research focused on the  (Peng and Zhu, 2017); soil temperature and moisture observations (Su et al., 2011;Yang et al., 2013); multi-scale observations of the Heihe River Basin (Li et al., 2017;Liu et al., 2018;Che et al., 2019;Li et al., 2019); and multiple datasets from the coordinated Asia-European long-term observing system for the Tibetan Plateau (Ma et al., 2009).

255
The Third Tibetan Plateau Atmospheric Scientific Experiment (TIPEX-III, Zhao et al., 2018) also provides field measurement data for the LS4P project. The Chinese Meteorological Administration (CMA) provides some field measurements with long term records. The observed https://doi.org/10.5194/gmd-2020-329 Preprint. Discussion started: 7 January 2021 c Author(s) 2021. CC BY 4.0 License.

11
CMA monthly mean precipitation and T-2m, and topography data, with a 0.5-degree resolution based on station measurements (Han et al., 2019), are used in LS4P to evaluate the LS4P models' measurements for shallow soil layers, e.g., only reaching down to 101.6 cm (Hu and Feng, 2004). Because of the lack of subsurface measurements, there has been some speculation as to whether the LST/SUBT anomaly and memory, as well as the hypothesized relationship between T-2m/LST/SUBT truly exist in the real world.
In addition to the ground measurements, satellite products from 1981 to 2018 from the 270 Global LAnd Surface Satellite (GLASS, Liang et al., 2013Liang et al., , 2020) data set will also be employed for this project. This dataset consists of surface skin temperature, albedo, emissivity, surface radiation components, and vegetation conditions (www.glass.umd.edu). anomaly to the south (north) of the Yangtze River. Meanwhile, a preliminary modeling study revealed the causal relationship between the May T-2m/LST/SUBT anomaly over the Tibetan Plateau and the June drought/flood in East Asia (Xue et al., 2018). LS4P intends to further test 290 and confirm such causal relationships with multiple state-of-the-art ESMs in order to assess the uncertainty, and to compare the T-2m/LST/SUBT effect with that of the SST.

Experimental Design: Baseline and Sensitivity Experiments
(1). Task 1. In Task  Before calculating the model bias, the model-simulated T-2m data must be adjusted with a 320 proper lapse rate to the elevation height of the observational data (Xue et al., 1996a;Gao et al., 2017).
The major goals of Task 1 are to check whether most ESMs have large biases in simulating these two variables and whether they are able to produce the observed T-2m and precipitation anomalies. The relationship between these two biases are evaluated to see whether anomaly, it may also be able to produce the observed June precipitation anomaly.
The discoveries from Task 1 will provide crucial information for the LS4P project as it pursues its objectives as discussed in Section 2. If the LS4P ESMs produce no large bias in precipitation and T-2m and/or they are able to simulate the observed anomaly well over Tibetan Plateau and eastern China, the justification for LS4P would be questionable. Should the model 335 bias relationship between the May T-2m and the June precipitation be the opposite of the observed anomaly relationship of these two variables, it would also be difficult, if not impossible, to pursue the LS4P approach further for these models. The preliminary assessments, however, are encouraging and strongly support the need for LS4P to further pursue its goals, and they will be briefly demonstrated in Section 5. It should be pointed out that the evaluation of the 340 bias relationship between May T-2m in the Tibetan Plateau and June precipitation in eastern China is just a necessary condition for LS4P to pursue its approach. It is not sufficient to guarantee the model can improve the June precipitation prediction by using improved May T-2m initial conditions. Only Task 3, as discussed below, will serve this purpose.
(2). Task 2. A number of LS4P modeling groups are from big climate modeling centers, observational data have shown a high correlation between LST and SUBT, and the memory in the soil subsurface is one of the major factors for producing soil surface temperature anomalies (Hu and Feng, 2004;.

375
To improve the LST/SUBT initialization, a surface temperature mask for each grid point,

15
: the averaged observed anomaly and model bias, respectively, over the entire area where the mask is intended to be applied, such as the Tibetan Plateau. Because all current land surface models are unable to maintain the soil temperature anomaly, a tuning parameter "n" (e.g., 1, 2, 3, etc.) is introduced. Through trial and error, each model selects a proper "n" intending to produce the T-2m anomaly close to observation. For the subsurface, the "n" may be different from that for LST 395 depending on the ESM's land surface scheme. But currently, most modeling groups use the same "n" for every soil layer. This approach can be improved after more deep soil layer measurements are conducted such that we have a better idea about the LST and SUBT relationships. Figure 2 shows schematic diagrams for imposed masks for surface temperature initialization under different conditions, which shows the concept for the mask formulation. In   atmospheric sounding data over the Tibetan Plateau for data assimilation. That said, lower atmosphere temperature is also subject to model bias. Since there are no observed near surface layer observations, we compare the reanalysis surface and near surface temperature anomalies with their own climatology. These 490 anomalies are very close (not shown), which means even if we impose a mask to overcome the LST/SUBT bias, the bias in the lower troposphere is still there. This bias in the reanalysis data has an important implication in affecting the LST initialization and its simulation, which will be discussed further in section 4.2.
In addition to the surface temperature, subsurface temperature initialization is also challenging in  The deficiencies in the reanalysis products pose a challenge for properly producing the observed T-2m anomalies since the reanalyses are used to provide the basis for the surface initial condition for most

ESMs. Since every LS4P ESM showed a large bias in simulating the May 2003 T-2m anomaly over the
Tibetan Plateau, we have addressed how to take the bias into account in producing the initial condition mask in Section 3.2. In the next section, the efforts from different modeling groups to generate the 515 observed T-2m anomaly will be presented further.

Approaches in Improving the LST/SUBT Initialization and T-2m Anomaly Simulation
In addition to the data that are used for LST/SUBT initial conditions, land models also have deficiencies in maintaining the anomalies that are imposed using an initial mask as discussed in Section 3.2. In the LS4P-I 520 experiment, most models are only able to partially produce the observed T-2m anomaly in May despite the fact that the imposed initial masks normally contain much larger anomalies than those observed. This section highlights some specific approaches undertaken by a few groups during their application of the LS4P-I protocol to improve the T-2m anomaly simulation.
The surface soil (20-30 cm) in the central and eastern Tibetan Plateau contains a large amount of 525 organic matter which greatly reduces the soil thermal conductivity and increases the soil heat capacity (Chen et al., 2012;. However, this factor is not taken into account in the LS4P ESMs, except CNRM-CM6-1. That said, the soil thermal conductivity/heat capacity over the Tibetan Plateau in the ESMs is too high/too low. In addition, some ESMs overestimate the precipitation over the Tibetan Plateau, making the soil water content higher than in reality (Su et al., 2013), which also leads to higher soil 530 thermal conductivity. Less soil organic matter and high soil moisture both accelerate the heat exchange rate between the soil and the atmosphere, which causes the rapid loss of soil thermal anomalies in the models.
The soil layer depth in the ESM also affects the model's ability to generate the observed T-2m anomaly. The long memory in deeper soil helps to preserve the soil temperature anomaly in shallower 535 layers. In a sensitivity study that changed the soil depth from 6 m to 3 m, it was found that with reduced total soil column depth, a similar magnitude anomalous soil temperature can only be kept for about 20 days, then it disappears much faster thereafter compared with the 6-m soil layer model . In a number of LS4P land models, the total soil column depth is less than 3 m. To overcome these shortcomings in current ESMs and to reproduce the observed T-2m anomaly, a tuning parameter "n" is 540 introduced (Eq. 1) when setting up the surface mask since it is not a simple task to increase the soil layer depth for all of the ESMs.
One of the intentions of the initialization of LST/SUBT is to influence the lower atmosphere since the corresponding initial condition from reanalysis also has inherent errors as discussed in section 5.1, and for some models they can be quite large. A number of modeling groups have started the model simulation 545 earlier, for instance on April 01, in order to have sufficient time for the lower atmosphere to spin-up and to be consistent with the within-mask imposed soil surface conditions. In some models, such as ACCESS-S2 and KIM, the models make an adjustment after reading in the initial condition, usually referred to as shock adjustment, in order to avoid an imbalance between the atmosphere, land, and ocean initial conditions. This shock adjustment has become a more popular practice in a number of modeling groups. The idea 550 behind the shock adjustment arises from the potential inconsistency among different sources for initial conditions, and the belief that the atmospheric components are considered to be relatively the most reliable.
With such an approach, within the first week or 10 days, the atmospheric forcing plays a dominant role in adjusting the other components' initial conditions. As such, the imposed initial soil temperature from the mask at the top soil layers could be compromised very dramatically toward the lower atmospheric 555 conditions, which, unfortunately, also have large errors over theTibetan Plateau as previously discussed.
Although the imposed deep soil temperatures eventually start to affect the air temperature, this process generally takes more than 20 days. For the model with such a shock adjustment, the mask needs to be imposed when the shock adjustment becomes weak, such as at the second day in ACCESS-S2 or half a month after the initial simulation date, as done in KIM. As such, the models may have to start their 560 integrations much earlier. A couple of models tried to impose the mask more than once to produce the T-2m anomaly. For instance, the FGOALS-f2 model imposed the LST/SUBT anomaly on both May 1 and May 2 to better produce the observed T-2m anomalies. It should be pointed out that if a mask is imposed too many times, the ΔT in the mask may add up every time when it is imposed to become quite large sink/heat source. Furthermore, enforcing the LST/SUBT perturbation too many times during the model 565 simulation with accumulated large ΔT may distort the atmospheric conditions. Precautions must be taken in this type of approach, probably with ΔT imposed no more than twice with a well-designed scheme to avoid the excessive accumulation of heating/cooling.
For the E3SM and CESM2, which are mainly used in long-term climate research (e.g., centurylong simulations), real time initialization for S2S prediction is not very closely related to the research 570 objective the model centers intend to pursue. To conduct LS4P type research, the modeling groups have to develop an approach in nudging the reanalysis data for a real time initialization. Nudging is one of the simplest data assimilation methods (Hoke and Anthes, 1976) and has been widely used in climate model evaluation and sensitivity studies (e.g., Xie et al., 2008;Sun et al., 2019;Tang et al., 2019) to constrain the simulations towards a predefined reference (the reanalysis data in this case) and hence to facilitate time-575 specific comparisons between model and observations. For the LS4P simulations, E3SM and CESM2 used 1-month worth of nudging of the horizontal wind components (U & V) with a 6-hour relaxation time scale before the land mask for the initial LST perturbation was applied. A study (Ma et al., 2015) has shown that nudging only horizontal winds produces better results compared with those with nudging of more variables, such as temperature, specific humidity, etc.

Discussion: Prospective and Impact of LS4P
LS4P is the first international grass-root effort focused on introducing spring LST/SUBT anomalies over high mountain areas as a factor to improve S2S precipitation prediction through the remote effects of land/atmosphere interactions. Although the original idea of starting LS4P was more limited and only 585 aimed at evaluating whether the results from preliminary tests with one ESM and one RCM (Xue et al., 2016b(Xue et al., , 2018 could be reproduced by more modeling groups, multi-model participation has quickly lead to the recognition that the Tibetan Plateau's spring LST/SUBT effect on the precipitation anomaly to the south and north of the Yangtze River was only a small part of broader aspects.

Furthermore, the T-2m cold bias over the Tibetan Plateau is associated with a cold bias in the Iranian
Highlands and a warm-cold-warm wave train over the Eurasian continent, which is also generally consistent with the observed T-2m anomalies. Moreover, the consistencies suggest a possibly much larger scale remote effect of the Tibetan Plateau LST/SUBT on summer precipitation over many parts of the world and support the LS4P's approach in its experimental design as discussed in Section 3.2. As a result, 610 the diagnostic analyses from the tasks in Experiment 1 will cover the entire globe. Comprehensive analyses and discussion will be presented in subsequent papers after the LS4P groups have completed their experiments.
Although the T-2m anomaly covers large areas, our previous study has shown that only the LST/SUBT anomaly over high mountains had a substantial impact on the subsequent drought (Xue et al.,615 2012). One of the LS4P groups, KIM, also tested the effect of the LST anomaly in other parts of East Asia, but found their effects are incompatible with the Tibetan Plateau LST/SUBT effect. In addition to year 2003, we also checked the May T-2m and June precipitation bias in the climatologies of the different models. The thirteen ESMs shown in Figure 6 have also provided their climatological data sets. Figure 7 shows the climatological biases for May T-2m and June precipitation from these ESMs. The patterns 620 between the bias in the 2003 simulation and the bias in the model climatologies are generally consistent, which suggests that the findings from the 2003 case may have a broader implication.
In Phase I, through the LS4P RCM efforts in incorporating the TPE and TIPEX-III data, we also intend to adequately simulate water and energy cycle and atmospheric conditions in the Tibetan Plateau and their variability. These simulations will provide the data for better atmospheric and surface initialization, along with obtaining an improved understanding of the atmospheric circulation and water cycle in Tibetan Water Tower.
Thus far, our discussion has been focused on the modeling approach. A recent statistical study has shown that spring soil temperature in central Asia could be a predictor of summer heat waves over northwestern China . In addition, surface temperatures from five Northern European 630 observing stations have been used as a predictor for long-range forecasting of southwest monsoon rainfall over India (Rajeevan, et al., 2007). Moreover, spring (April-May) precipitation and 2m air temperature over northwestern India, Pakistan, Afghanistan, and Iran have been found to have a strong link with the first phase (June-July) of summer monsoon rainfall over India (Rai et al., 2015). We will extend the data analyses for different major mountains and different seasons and to identify hot spots over the globe where 635 LST has significant impacts. Preliminary statistical forecasts will also be explored, using methods such as the Canonical-Correlation Analysis (CCA) and Joint Empirical Orthogonal Analysis (JEOF) (Smith et al., 2016). Based on the statistical analyses, a Tibetan Plateau Oscillation Index (TPO) and a Rocky Mountain Oscillation Index (RMO) will be proposed for predictions of the hydroclimatic extreme events, and a relationship between the TPO and the RMO indexes will also be investigated. As discussed in Section 3, 640 the Rocky Mountain LST/SUBT effect will be the focus of LS4P Phase II (LS4P-II).
The LS4P research has revealed some severe deficiencies in current land models in preserving the land memory. In many models, the force-restore method (Deardorff, 1978;Dickinson, 1988;Xue et al., 1996b) is used to represent subsurface heat transfer and soil thermal status. This simple method produces adequate diurnal and seasonal cycles of surface temperature 645 and thus has been widely used by many land models for decades. However, its severe deficiency in keeping the soil memory is apparent in the LS4P studies. We have found that excessively shallow soil depths along with simplified parameterizations of subsurface heat transfer are acting to limit the soil memory effect in many models, especially in cold regions. An innovative approach has been developed for the land model initialization that can help maintain the monthly 650 LST/SUBT anomaly. The LS4P's finding on why ESMs have difficulty to maintain the LST anomaly, and its proposed approach to help solving the issue should be a significant contribution from the LS4P project to improve the S2S prediction.
One issue that hampers the application of the LST/SUBT approach for S2S prediction is data availability. The TPE has conducted comprehensive measurements over the high mountain planning for more measurements that are related to land/atmosphere interactions (Wulfmeyer et al., 2020;Schneider and van Oevelen, 2020). We hope that the results from LS4P will demonstrate the substantial role of high mountain surface conditions on global climate and atmospheric circulation, and therefore stimulate more initiatives to increase land/atmosphere interaction measurements over high mountain regions. Although land has lower heat capacity and less moisture compared to the oceans, the land surface has a much stronger response to changes in surface net radiation at diurnal, sub-seasonal, and seasonal scales compared to oceans. This is particularly true in high elevation areas, which 675 could provide a useful source for predictability at these scales. LS4P intends to improve the S2S precipitation prediction through a better representation of land surface processes in the current generation of ESMs and aims to make a fundamental contribution in advancing S2S prediction through proper initialization of LST/SUBT in high mountain regions. The LS4P approach proposes a new front in S2S prediction to complement other existing approaches. We hope activities and results   Mellor-Yamada-Janjic (Schaefer, 1990) Noah (Ek et al. 2003) CBMZ (Zaveri and Peters, 1999 LS4P data archive and data formats, including the common naming system, is provided in Appendix C. The data portal is available at http://data.tpdc.ac.cn/en/. The login is "LS4P_group". National Tibetan Plateau Data Center has an online data submission system which is similar to paper submission system. For instance, folders can be uploaded, but not needed to be tarred in one file. It is recommended that each modeling group to create its own folder using the following naming: InstituteName_ESMModelName (example: UCLA_CFS-SSiB2). SSiB2/Task1/daily/…).

A) Uploading Data into National Tibetan Plateau Data Center using Filezilla
To upload data into the National Tibetan Plateau Data Center, we recommend to use "Filezilla". With Filezilla, the host, username and password are generated automatically for the Filezilla when the data are uploaded. The following procedure is based on "Filezilla".

760
The procedure will utilize the following steps.
3). You will see the webpage "CREATE METADATA". Please fill in your data information, such as i) Overview (Title, abstract, data file naming, file size, time range,…), ii) Reference, iii) Keyword(s),…etc. After complete, click "Save" to save the information.  the ''root directory". If you have created the folder before, you will find it, when you log back.

7)
. Then from your Filezilla window, you can drag your data from your local site to the 785 newly created folder/subfolder, such as Task1.

8).
Send an email to Duo at panxd@itpcas.ac.cn. Then she will synchronize the data for you directly! 790 9). Click "submit" to submit the online data in the window which appeared in step 3.

10).
Duo will send you a confirmation email to confirm/acknowledge the proper submission.
By that time, you should be able to see your data.

B) Acquiring LS4P-I Project Data
a) Log in to the online National Tibetan Plateau Data Center (http://data.tpdc.ac.cn/en/), using the aforementioned login details (see Section II).
b) Go to "LS4P_group" / "Personal Center" 800 c) Select "My Data", and then select "Review" or "My Draft" d) You will see all the metadata belonging to LS4P group. e) Under the metadata, click "edit" button, and move to "Data Files" item, you will find the host, port, username and passport for the specific group data you selected.
f) Open Filezilla using the information's from e), 805 g) Now from Filezilla you can manage the LS4P directory and see what has been uploaded, along with the current directories/sub-directories.

Competing Interest
The authors declare that they have no conflict of interest.   (2) The North American Regional Reanalysis (NARR, Mesinger et al., 2006) assimilated the observed T-2m and is viewed as having an accurate representation of the observed surface air temperature.

Figure 2. Schematic Diagram for an Imposed Mask for Surface Temperature Initialization Corresponding to (a) a Cold Anomaly Year; (b) a Warm Anomaly Year
Notes: 1). T0 is the original model initial condition and 0 � is the initial condition after imposing the mask. 2). The +/-sign in the parentheses indicate that the value is positive/negative, respectively.  Note: The CMA climatology is used as reference for the anomalies. Because each T-2m data set has its own elevation, all the data have been adjusted to the CMA elevation for comparison.