simulated sensitivity to land surface schemes in ... - Wiley Online Library

PUBLICATIONS Journal of Advances in Modeling Earth Systems RESEARCH ARTICLE 10.1002/2015MS000440 Key Points: WRF weather simulations are sensitive to the LSSs and lead times Each LSS possesses systematic simulated features over different lead times LSS deficiencies in simulating fluxes and processes induce the SAT errors

Correspondence to: N. Wing, [email protected]; X.-M. Zeng, [email protected]

Citation: Zeng, X.-M., N. Wang, Y. Wang, Y. Zheng, Z. Zhou, G. Wang, C. Chen, and H. Liu (2015), WRF-simulated sensitivity to land surface schemes in short and medium ranges for a hightemperature event in East China: A comparative study, J. Adv. Model. Earth Syst., 7, 1305–1325, doi:10.1002/ 2015MS000440. Received 20 FEB 2015 Accepted 17 AUG 2015 Accepted article online 19 AUG 2015 Published online 6 SEP 2015

WRF-simulated sensitivity to land surface schemes in short and medium ranges for a high-temperature event in East China: A comparative study Xin-Min Zeng1,2, Ning Wang1,2, Yang Wang1, Yiqun Zheng1, Zugang Zhou1, Guiling Wang1,2, Chaohui Chen1, and Huaqiang Liu1,2 1 2

College of Meteorology and Oceanography, PLA University of Science and Technology, Nanjing, Jiangsu, China, Key Laboratory for Mesoscale Severe Weather of Ministry of Education, Nanjing University, Nanjing, Jiangsu, China

Abstract We designed simulations for the high-temperature event that occurred on 23 July 2003 in East China using a series of forecast lead times, from short-range to medium-range, and four land surface schemes (LSSs) (i.e., SLAB, NOAH, RUC, and PX) in the Weather Research and Forecasting Model (WRF), Version 3. The sensitivities of short and medium-range simulations to the LSSs systematically varied with the lead times. In general, the model reproduced short-range, high-temperature distributions. The simulated weather was sensitive to the LSSs, and the LSS-induced sensitivity was higher in the medium range than in the short-range. Furthermore, the LSS performances were complex, i.e., the PX errors apparently increased in the medium range (longer than 6 days), RUC produced the maximum errors, and SLAB and NOAH had approximately equivalent errors that slightly increased. Additional sensitivity simulations revealed that the WRF modeling system assigns relatively low initial soil moisture for RUC and that soil moisture initialization plays an important role that is comparable to the LSS choice in the simulations. LSS-induced negative feedback between surface air temperature (SAT) and atmospheric circulation in the lower atmosphere was found in the medium range. These sensitivities were mainly caused by the LSS-induced differences in surface sensible heat flux and by errors associated with the lead times. Using the SAT equation, further diagnostic analyses revealed LSS deficiencies in simulating surface fluxes and physical processes that modify the SAT and indicated the main reasons for these deficiencies. These results have implications for model improvement and application.

1. Introduction The world has undergone rapid warming during the last several decades [IPCC, 2013], and hazardous weather/climate events, such as heat waves, have received considerable attention. Indeed, heat waves have severe impacts on the environment, society, and economy [e.g., Trigo et al., 2005; Zhou and Shepherd, 2010; Zeng et al., 2014]. Based on station records and reanalysis data sets, previous heat wave studies have mainly focused on data homogenization or theoretical analyses to characterize events, for example, homogenization of daily maximum temperatures [Della-Marta and Wanner, 2006; Kuglitsch et al., 2009], variations in daily €nnen, 2003], and changes in mean monthly maximum temperatures [T€ temperatures [Klein Tank and Ko urkes¸ et al., 2002], diurnal temperature ranges [Liu et al., 2006], the number of heat waves, length and intensity [Kuglitsch et al., 2010], and factors or mechanisms affecting heat waves [Stott et al., 2004; Della-Marta et al., 2007]. C 2015. The Authors. V

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.

ZENG ET AL.

In addition to the above mentioned investigations based on observational or analysis data, relatively few investigations on model performances have been performed for heatwaves or the factors/mechanisms affecting heat waves using dynamically and physically based numerical models. Using a general circulation model (GCM), Meehl and Tebaldi [2004] showed the presence of distinct geographic patterns in future heat wave €r et al. [2004] performed two regional climate model (RCM) simulations and demonstrated that changes. Scha the European summer climate might experience pronounced annual variability in response to greenhousegas forcing. In addition, these authors indicated that such an increase in variability might explain the unusual summer of 2003 in Europe and would strongly affect the incidence of heatwaves and droughts in the future. Kharin et al. [2007] evaluated temperature and precipitation extremes and their potential changes using an

WRF SENSITIVITY TO LAND SURFACE SCHEMES

1305

Journal of Advances in Modeling Earth Systems

10.1002/2015MS000440

ensemble of global coupled climate models that are part of the IPCC diagnostic exercise for the Fourth Assessment Report (AR4). While using a GCM, Feudale and Shukla [2011] found that global SST anomalies can explain many major features of the European summer heat wave of 2003. In general, most of these studies emphasized the mechanisms responsible for heatwave events at the climate time scale. Among the numerical studies of heatwave mechanisms, several investigations have considered the landatmosphere interactions that affect heatwave events. For example, Wolfson et al. [1987] conducted a series of numerical forecasting experiments using a GCM and showed that the sea-surface temperature (SST) anomalies and soil moisture anomalies in medium-range (10 days) simulations had opposite effects on the maintenance of the 1980 summer heat wave. Thus, the lower-than-normal soil moisture across the U.S. resulted in reduced surface evaporation, increased sensible heat flux from the ground to air, higher surface temperatures, lower sea-level pressures, and greater 500 hPa heights over much of the Great Plains. Using a GCM, Ferranti and Pedro [2006] showed that the linear atmospheric response to large initial perturbations in the root zone lasts for up to 2 months and is longer in drier regions. Additionally, the response of large, initially dry soil anomalies greatly exceeds the impacts of ocean boundary forcing. Fischer et al. [2007a] performed sensitivity simulations and demonstrated that soil moisture anomalies substantially affected the strength of the 2003 heat wave and the extent of geopotential height anomalies. Using RCM simulations, Fischer et al. [2007b] indicated that land-atmosphere coupling plays an important role in the evolution of heat waves through local and remote effects. During all simulated events, the soil moisture-temperature interactions increased the heat-wave duration and accounted for 50–80% of the hot summer days. The greatest impact was found for daily maximum temperatures during heat wave episodes. Using observations and mesoscale meteorological sensitivity simulations for the summer of 1994, Vautard et al. [2007] showed that hot summers are preceded by winter rainfall deficits over southern Europe and that subsequent drought and heat spreads northward throughout Europe during the early summer. All of these numerical studies generally presented soil-wetness-related mechanisms for developing and maintaining heat waves. Although numerical models have been employed for heat wave modeling, few studies have used regional/mesoscale models for weather simulations with various land surface schemes (LSSs), for example, the LSSs in the WRF model. This type of investigation is important for the following reasons: (1) current WRF sensitivities to different physical parameterizations have mostly focused on convective-precipitation cases [e.g., Fan, 2009], yet different LSSs might have different impacts on heat wave features (e.g., the intensity and spatial temperature distribution). In fact, the LSS-induced differences imply the extent by which different land surface processes may influence the occurrence and development of heat waves. For example, Zeng et al. [2011] revealed that simulated high-temperature events are sensitive to LSS and found that an LSS-induced negative feedback existed between the surface air temperature (SAT) and the low-level atmospheric circulation during an East China heat wave event. (2) Although GCMs have commonly been used in the climate mode for heat wave investigations, mesoscale temperature forecasting models have been used as an operational weather service, particularly for extremes, such as heat waves [e.g., Cheng and Steenburgh, 2005, among others]. Mesoscale model simulations of heat waves are affected by different physical processes because of scientific problems associated with regional models (e.g., effects of lateral boundaries and the choice of physical schemes). (3) In general, model simulation/ forecast errors increase with integration time or lead time. Different LSSs could behave differently in forecasts/ simulations with short and medium ranges. In these forecasts/simulations, the LSS and lead-time-induced sensitivities would be helpful for both model development and application. Therefore, this study seeks to comparatively quantify the sensitivities of short and medium-range WRF weather simulations to LSSs for the summer of 2003 heatwave event over East China using a series of forecast lead times. The structure of this paper is organized as follows. In section 2, we describe the model and the experiments. Then, we compare the short-range and medium-range WRF-simulated temperature sensitivities to the LSSs in section 3. A discussion of the mechanisms associated with the sensitivities is provided in section 4. Finally, we summarize the results and conclusions.

2. Methodology 2.1. The Coupled Mesoscale Modeling System The Advanced Research WRF (ARW) model (WRF-ARW Version 3.3) [Skamarock et al., 2008] is employed in this study. As an advanced modeling system, the WRF model includes many key dynamic features and

ZENG ET AL.


1306


10.1002/2015MS000440

Table 1. Overview of the Physical Parameterizations in the Designed Experimentsa Test

Land Surface Scheme

Boundary Layer Scheme

Radiation Scheme

Cumulus Convection Scheme

Microphysics Scheme

Modified Kain-Fritsch scheme [Kain, 2004]

WRF Single moment five class [Hong et al., 2004]

As SLAB

As SLAB

SLAB

Five-layer thermal diffusion scheme [Dudhia, 1996]

Yonsei University scheme [Hong et al., 2006]

NOAH/ NOAHsm RUC/RUCsm

Noah scheme [Chen and Dudhia, 2001; Ek et al., 2003] Rapid Update Cycle scheme [Smirnova et al., 1997; 2000] Pleim–Xiu scheme [Xiu and Pleim, 2001]

As SLAB

RRTM longwave [Mlawer et al., 1997]; MM5 shortwave [Dudhia, 1989] As SLAB

As SLAB

As SLAB

As SLAB

As SLAB

As SLAB

As SLAB

As SLAB

As SLAB

PX a

Compared with NOAH and RUC, the initial soil moisture contents of NOAHsm and RUCsm, respectively, are adjusted for the sensitivity study; see the text.

sophisticated physical schemes, for example, mass-based terrain-following coordinates, Arakawa C-grid staggering, and physical options for radiation, cumulus convection, microphysics, and the planetary boundary layer. The WRF model includes four coupled LSSs, five layer thermal diffusion, Noah, RUC, and Pleim-Xiu. Additional LSS implementations have been conducted using the latest version of the WRF model (see NCAR website). These LSSs use different formulations for surface energy and moisture balances (Table 1). Without including explicit vegetation effects, the five layer thermal diffusion scheme (SLAB hereafter) [Dudhia, 1996] employs the thermal diffusion equation and the force-restore method to calculate soil temperatures with fixed soillayer thicknesses of 1, 2, 4, 8, and 16 cm and soil moisture values fixed using a seasonally varying function of land use, i.e., it does not account for dynamic soil moisture in the short and medium-range weather. The Noah scheme (NOAH) [Chen and Dudhia, 2001; Ek et al., 2003] was developed from the Oregon State University LSS and is now a cooperative model of several institutions. NOAH forecasts the temperatures and moisture contents of four soil layers (with thicknesses of 10, 30, 60, and 100 cm) using the force-restore method and includes detailed descriptions of physical vegetation (with monthly vegetation fraction) and hydrological processes (i.e., evapotranspiration, soil drainage, and runoff). In addition, NOAH includes an improved urban treatment method in which surface emissivity properties are considered. In particular, the LSS has an advantage over the other LSSs because of its consistency with time-dependent soil moisture in some commonly used analysis data sets, such as the National Centers for Environmental Prediction (NCEP) Final Analysis (FNL) data set. The Rapid Update Cycle LSS (RUC) [Smirnova et al., 1997; Smirnova et al., 2000] was operationally used in the NCEP operational weather prediction system and accounts for soil, snow cover, and vegetation at the six default levels set at 0, 5, 20, 40, 160, and 300 cm. This LSS employs a layer approach for solving energy and moisture budgets, in which a layer includes half of the lowest atmospheric layer and half of the surface soil layer. The Pleim–Xiu (PX) scheme [Xiu and Pleim, 2001] solves five partial differential equations for two layer (1 and 99 cm thick) soil temperatures, soil moisture, and canopy moisture and features three pathways for moisture fluxes, evapotranspiration, soil evaporation, and evaporation from wet canopies. By dynamically adjusting soil moisture and deep soil temperature, the PX scheme employs two indirect nudging schemes to correct biases in 2 m air temperature and relative humidity. For more details regarding LSSs, please see the above references. To simulate real-data cases, the initial conditions are preprocessed using the WRF model Preprocessing System (WPS), which defines a physical grid, interpolates static fields (e.g., terrain elevation), and reads the land and atmospheric data to the prescribed domain. Next, the WPS output is passed to the ARW real-data preprocessor, the REAL program, which produces initial and lateral boundary conditions, including vertically and linearly interpolated soil state variables (i.e., soil temperature and moisture) from the incoming levels to the soil layers for the specific LSSs in the ARW model [Skamarock et al., 2008]. 2.2. Experimental Design Using two-way nesting for two domains, the large domain of the model is centered at 28.58 N, 1178 E, and includes 57 3 55 gridpoints with a resolution of 30 km. The small domain (‘‘DN’’) has 100 3 82 gridpoints with a resolution of 10 km (Figure 1). The model features 31 uneven vertical levels, with 50 hPa at the top. In the next sections, we focus on the simulation results of the DN domain. The summer climate over continental China is largely controlled by the western Pacific subtropical high (WPSH) and boreal cold air masses. In the summer of 2003, the WPSH displayed extreme anomalies and

ZENG ET AL.


1307


10.1002/2015MS000440

Figure 1. Model domains (the nested domain is denoted as ‘‘DN’’) and topography (unit: m).

extended much farther westward, while the boreal cold air mass was relatively weak; therefore, an exceptional heat wave peaked in late June over East China, where the daily high temperatures (i.e., daily maximum SATs) at some stations exceeded 408C, particularly in the south region of the middle to lower reaches of the Yangtze River [Lin et al., 2005; Zeng et al., 2011]. Hence, using different LSSs, we focus on the simulations for 23 July 2003, when the daily maximum SAT appeared extremely intense and was spatially aggregated. The end time of the simulations is 0600 UTC 23 July 2003. We use different initial fields so that each start time corresponds to an integration length or lead time of days (i.e., 1–14 days) to investigate the effects of different LSSs on short and medium-range simulations. Thus, a series of forecast lead times is used for the integrations, with a focus on the SAT of the last day of the simulations. For simplicity, the integrations are denoted according to the lengths of the simulations, for example, ‘‘DAY6’’ represents experiments with 6 day integrations over the period of 0600 UTC 17 July 2003 to 0600 UTC 23 July 2003. Compared with weather simulations conducted in previous studies [e.g., Trier et al., 2008; Fan, 2009; Zeng et al., 2011; Cheng et al., 2013], these types of simulations have the three following features: (1) both short and medium-range simulations (with lead times less than or equal to 2 and 14 days, respectively) are carried out. (2) The model performances are analyzed at two time scales using the same criteria, for example, the simulation for SAT at 0600 UTC on 23 July 2003. (3) Because of the systematic characteristics of different simulations with the series of forecast lead times, this research could provide insight into forecasting short and medium-range heat waves. Notably, when we examine the succession of the 24 h WRF simulations for late July 2003 provided by Zeng et al. [2011], the performance of the Version 2.2 WRF model was generally good for the 10 day SAT means but did not successfully produce the SAT patterns for all individual 24 h simulations. It is understandable that not all of the individual 24 h simulations agree with the observations. Therefore, when comparing integrations obtained for specific dates, the experimental design used in this paper is reasonable for assessing short and medium-range weather simulations. In general, daily high and low temperatures (i.e., daily maximum and minimum SATs, respectively) and hourly SATs are the main outputs of SAT forecasts by Chinese weather services. Because land surface and atmospheric conditions differ, the timing of the daily maximum SAT may vary regionally. However, the daily high temperature typically occurs at approximately 1400 Beijing Time (i.e., 0600 UTC) under persistent WPSH-controlled weather over East China, where no other weather systems penetrate (i.e., local processes dominate). Additionally, a ‘‘high temperature’’ event occurs when the 2 m surface air temperature (SAT) exceeds 358C in China. Therefore, in the following sections, we focus on the SAT results and related quantities at 0600 UTC to analyze the occurrence and development of the simulated high-temperature event. We use a suite of model physical schemes, including the four LSSs (Table 1), SLAB, NOAH, RUC, and PX, for DAY1–DAY14 short and medium-range simulations. Because previous studies have emphasized the importance of land initialization [e.g., Koster et al., 2009] and the crucial role of soil moisture in the development

ZENG ET AL.


1308


10.1002/2015MS000440

of heatwaves [e.g., Miralles et al., 2014] in addition to SLAB, NOAH, RUC, and PX simulations, two groups of simulations, NOAHsm and RUCsm simulations, were designed to study the sensitivities of simulations to initial soil moisture when using the NOAH and RUC model configurations, respectively (i.e., totally 14 3 6 simulations were performed). Because the WRF modeling system presents soil moisture contents that are too dry for RUC simulations when using the REAL program (see section 2.1), the initial RUCsm soil moisture field is enhanced (by 20%) and adjusted to the FNL data (or NOAH and PX, both of which have the same values for the top soil layers and are very similar to the FNL field) and the initial soil moisture content prescribed by NOAHsm is 20% higher than that prescribed by NOAH. Simulation differences are only induced by the LSSs and the integration lengths because the same physical parameterization options, except for the LSSs, are used in this study. These differences are particularly noted in the following sections. 2.3. The Data Sets To drive the initial and boundary conditions of the WRF simulations, the NCEP Final (FNL) Analysis data at a 18 3 18 resolution were used with the model initialization addressed in section 2.1. To validate the simulated SAT results, conventional meteorological station data are employed. In addition, this study validates the LSS-simulated evapotranspiration against observationally based fields from Miralles et al. [2013].

3. Comparison of Simulation Results 3.1. Geographical SAT Distribution Because the same FNL data were employed for the initial and boundary conditions and because the physical parameterization options are nearly the same, the 0600 UTC SAT (SAT06) simulations over the same integration lengths are consistent compared with the corresponding FNL analysis results. Figures 2a, 2c, 2e, and 2g provide the DAY1 SAT06 distributions for 23 July simulated by the four LSSs. In addition, Figure 2l displays the corresponding analysis. For the 24 h short-range simulations, the four LSSs can typically reproduce the geographic SAT06 distribution, for example, high values in the central areas of the study domain and low values north or south of these areas. These LSSs can also simulate SAT patterns higher than 358C. However, large LSS-induced differences in SAT06 have also been observed. Overall, PX produces the best SAT06 distribution. As shown in Figure 2g, PX (DAY1) generally reproduces a detailed SAT06 distribution, for example, there are three observational centers with high values near 1138E, 1158E, and 1178E that correspond to the ‘‘western,’’ ‘‘middle,’’ and ‘‘eastern’’ centers, respectively (Figure 2l). PX successfully simulates the first two centers in terms of position and intensity and represents the SAT06 band of values greater than 358C. In addition, PX better depicts the SAT distribution in the upper portion of the model domain when compared with the other LSSs, which produce positive biases. NOAH (DAY1) represents the three centers and intensities of the ‘‘western’’ and ‘‘middle’’ centers well and simulates a lower ‘‘eastern’’ center. Furthermore, NOAH produces a small area of SATs higher than 388C, with high SATs in the upper portion of the model domain (Figure 2c). The RUC LSS successfully simulates the positions of the three SAT06 centers and produces much stronger values than the observations over most of the study area (Figure 2e). SLAB can reproduce the ‘‘western’’ and ‘‘middle’’ centers well in terms of position and intensity but fails to simulate the ‘‘eastern’’ center. Moreover, the scheme produces high SATs to the north (Figure 2a). In addition, large differences between the simulations appear to be induced by differences in the integration length or forecast lead time. As the lead time increases, the SLAB and NOAH SAT distributions slightly change in both short-range (i.e., DAY1–DAY2) and medium-range (i.e., DAY3–DAY14) forecasts. For example, Figures 2a, 2b and 2c, 2d present the DAY14 SAT distributions simulated by SLAB and NOAH, respectively. For a specific LSS, the 14 day simulation retains nearly the same distribution as the 1 day simulation. The difference field results between the DAY1 and other SLAB (or NOAH) experiments suggest that the SLAB (NOAH) SATs slightly change with the lead time. Figures 3a and 3b shows the DAY14–DAY1 SAT differences of the two LSSs. The absolute values are generally less than 18C over land. However, the RUC-simulated SAT changes with lead time are more apparent. In general, the maximum values and areas of high values increase with the integration time. RUC-DAY2 also reproduces the general SAT06 pattern when compared with DAY1, but the high-temperature area (i.e., >358C) expands and the intensity increases, which suggests a poorer performance (not shown). When the integration time increases to 3 days, the SATs are positively biased and the high-temperature area expands (Figure 3c). As the lead time increases to 14 days, RUC presents similar temperature distributions and intensities (e.g., Figure 3k).

ZENG ET AL.


1309


10.1002/2015MS000440

Figure 2. Distributions of SAT06 (unit: 8C) on 23 July 2003, where (a) SLAB (DAY1), (b) SLAB (DAY14), (c) NOAH (DAY1), (d) NOAH (DAY14), (e) RUC (DAY1), (f) RUC (DAY3), (g) PX (DAY1), (h) PX (DAY4), (i) PX (DAY11), (j) RUCsm (DAY1), and (k) NOAHsm (DAY1) represent the simulated results as they change with the LSSs (lead times), and (l) FNL denotes the NCEP Final Analysis data.

The PX SATs greatly change with lead time. For short-range simulations of DAY2 (not shown), the SAT06 distribution is very similar to that of DAY1. In general, PX reproduces the positions and intensities of the ‘‘western’’ and ‘‘middle’’ centers well but fails to capture the ‘‘eastern’’ center. In the medium-range simulations of DAY4 (Figure 2h), PX can simulate the three centers very well despite the smaller total area with high temperatures. Similar to RUC, PX also produces higher SATs, larger areas with high temperatures, and further increases in the lead time. For example, for DAY4 through DAY10 (not shown), PX can simulate the overall positions of the three centers despite the enhanced intensities. When the integration time increases to DAY11, the SATs have greater positive biases and larger areas of high temperature than the observations. However, when the integration time further increases, the large biases slightly decrease (Figure 3k). Compared with DAY1, the other experiments (DAY2–DAY11) simulate higher temperatures and larger areas of high temperature (e.g., Figure 3d). For experiments with lead times longer than 11 days (i.e., DAY12– DAY14), the simulated SATs do not change in intensity (Figure 3k). With enhanced initial soil moisture contents, both RUCsm and NOAHsm present SATs that are lower than those of the RUC and NOAH over land, respectively (e.g., Figure 2e versus 2j and Figure 2c versus 2k). In particular, RUCsm simulates much lower SAT06 values than RUC (Figure 3h), with SAT06 values that increase with lead time (Figure 3g). This result demonstrates the crucial role of initial soil moisture and the substantial effects of lead time in the weather simulations. 3.2. Area-Averaged SATs Here, the domain (DN)-averaged SAT06 is used to examine the differences induced by different LSSs and lead times. For example, Figure 3j provides the diurnal SAT variations for DAY1 and shows that all simulated

ZENG ET AL.


1310


10.1002/2015MS000440

Figure 3. Distributions of SAT06 difference fields (unit: 8C) for 23 July 2003 due to different LSSs and leads times (a–i) and the domain (DN)-averaged DAY1 diurnal SAT variations (j) and SAT06 (k).

ZENG ET AL.


1311


10.1002/2015MS000440

diurnal SAT variations agree well with the observed variations. The daily minimum SAT occurred at approximately 0500 Beijing Time (2100 UTC), and the daily maximum SAT occurred at 1400 Beijing Time (0600 UTC). In addition, large differences are induced when choosing the LSS, for example, the largest hourly differences are within 1.5–2.08C over 24 h. Notably, for PX, the SAT is the lowest during the first four integration hours (0700–1000 UTC) and reaches its maximum value after integration for 7 h from the start (1400– 2200 UTC). Next, the SAT decreases to a minimum after the 18 h integration period (2400 UTC–0600 UTC), which corresponds to the lowest SAT06. These results suggest that an LSS would present different characteristics at different time scales (e.g., daily versus subdaily in this case) because of the nonlinearity of the land-atmosphere interactions. Figure 3k shows the variations in the area-averaged SAT06 due to different lead times. With time, the NOAH and SLAB-produced SATs slightly change. Specifically, the SLAB (NOAH) SATs are slightly higher (lower) than 358C and are similar to the observed value at 34.98C. Therefore, both schemes are consistent for forecasting short and medium-range high temperatures. However, for RUC and PX, the SAT06 increased with time. The accumulated errors peaked at a particular time (DAY7 and DAY11 for RUC and PX, respectively) and then remain steady. Additionally, results show that RUCsm still has features that are similar to RUC (Figure 3k). Overall, the simulated SAT06 is lower in the short range than in the medium range, and the SAT06 values both generally increase with the lead times for DAY1–DAY8. For DAY9–DAY14, the RUCsm SAT06 values remain almost unchanged, while those of RUC slightly decrease. Overall, all of these results indicate that the RUC LSS is a consistent feature for modeling SAT06 with changing lead time. Due to the initial soil moisture contents, substantial differences occur between the RUCsm and RUC, for example, the simulated RUCsm SAT06 is more than 2.18C lower than that simulated by the RUC in all of the simulations (Figure 3k). This difference is very large relative to the NOAHsm-NOAH difference, where NOAHsm enhances the initial soil moisture by as much as 20% and its SAT06 is roughly 1.58C lower than that of NOAH. These results suggest that soil moisture initialization can lead to strong drift for simulated SAT in addition to the deficiencies of the WRF modeling system when assigning initial values to the RUC LSS, which was noted by Koster et al. [2009] for climate modeling. LSSs with ‘‘normal’’ initial soil moistures, including NOAH, PX, and RUCsm, can still have different large areaaveraged SAT06 values (Figure 3k). For instance, differences of more than 0.58C are observed for simulations with lead times of less than 6 days and can reach 2.78C between PX and the other LSSs. This result indicates the high sensitivity of the results to the LSS when the initial soil moisture effect was considered.

3.3. Assessment Using Measures 3.3.1. Threat Score A threat score (TS) is used by Chinese meteorological agencies to determine the quality of the weather forecast and is calculated as follows: TS5NA =ðNA 1NB 1NC Þ;

(1)

where NA is the number of stations for which high SATs (i.e., over 358C) are correctly simulated, NB is the number of stations with simulated high SATs but observed low SATs, and NC is the number of stations with simulated low SATs and observed high SATs. In the present study, 131 stations within DN, the central area of East China, were chosen for the assessment. Figure 4a exhibits the threat scores of SAT06 over 358C and shows that large TS differences can be induced by the LSSs and lead times. On average, for different LSSs with a given lead time, the TSs of PX, RUC, SLAB, and NOAH become increasingly smaller, while RUCsm and NOAHsm produce lower values compared with RUC and NOAH, respectively, under enhanced initial soil moisture conditions. All of these results indicate the systematic performances of the LSSs when simulating high-temperature events. Moreover, for a given LSS without enhanced soil moisture, the difference in the TS in the medium-range simulations is larger than that in the short-range simulations. For example, the largest TS difference exists between PX and NOAH, with an average (maximum) difference of approximately 0.10 (0.12) for the short-range simulations (DAY1– DAY2) and approximately 0.19 (0.26) for the medium-range simulations (DAY3–DAY14). In addition, this

ZENG ET AL.


1312


10.1002/2015MS000440

Figure 4. Assessment using the following measures: (a) TS, (b) BIAS, and (c) RMSE.

result suggests that different LSSs can substantially influence the precision of the forecasted high SAT06 values for the short and medium-range forecasts. The simulated sensitivities to different lead times with a given LSS are shown in Figure 4a. The maximum difference of the total TS was 0.05 for SLAB (i.e., DAY8 minus DAY2), 0.03 for NOAH (DAY1 minus DAY3), and 0.04 RUC (DAY8 minus DAY1). The PX TS differences are generally the largest among the LSSs, with a maximum of 0.16 between DAY14 and DAY2. From DAY2 to DAY6, the PX TS becomes increasingly larger and changes in a complex manner with the two maxima on DAY6 and DAY14. For each LSS (especially for PX), the TS of the simulated high temperature is highly sensitive to the lead time. 3.3.2. Simulation Errors We also apply the model bias (BIAS; Figure 4b) and root-mean-squared error (RMSE; Figure 4c) for assessment, as described by Zeng et al. [2014]. The BIAS values are identical to the above modeled and observed SAT06 values (Figure 3k), showing the overall estimations of all of the LSSs, except NOAH. In addition, large BIAS values are consistent with high TS values for high-temperature events. For any particular simulation using the four LSSs, except for DAY1, the RUC RMSE is typically greater for short and medium forecasts of SAT06. Apart from the simulations with adjusted soil moisture contents (i.e., RUCsm and NOAHsm), large differences occur between the RUC RMSE and the lowest RMSE, for example, the total RUC-PX difference for the DAY2 simulations is 0.78C, which is 26% of the PX RMSE. The total RUC-PX DAY6 difference is 2.18C, which is 78% of the PX RMSE. For the DAY1 through DAY7 simulations without enhanced soil moisture, the PX RMSE is the lowest among the LSSs. However, the PX RMSE increases starting on DAY6 such that the DAY8 RMSE surpasses that of the SLAB and NOAH, both of which are very similar with very small error increases. For a given LSS, the RMSE values vary due to lead times. Overall, the SLAB and NOAH errors both slightly change and have time-induced RMSE maxima of 0.238C and 0.348C, respectively. Although the errors only slightly increase with the lead time, the RUC and PX errors change substantially. The RUC error rapidly increases and produces a large difference of 2.08C (DAY7 minus DAY1) or 69% compared with DAY1. Compared with DAY7, the RUC RMSEs in the simulations with longer lead times are slightly lower and become

ZENG ET AL.


1313


10.1002/2015MS000440

stabile over 4.38C. The PX RMSE is complicated, for example, it slightly decreases from DAY1 to DAY3 and slightly increases from DAY3 to DAY6. The error rapidly increases from DAY6 to DAY11 and then slowly decreases from DAY11 to DAY14. Because of the differences among these TS and RMSE results and the spatial SAT06 distributions, no LSS is superior in any aspect. Regardless of the short or medium-range predictions, none of the LSSs provide the most reasonable spatial SAT06 pattern with the lowest error. Notably, TS and RMSE are different methods for assessing the SAT06 simulations, and neither measure necessarily coincides. For example, although the RUC RMSE is generally much larger than that of NOAH, the NOAH TS is generally much lower than that of the RUC TS. Specifically, the RUC typically simulates the number of stations with high temperatures (over 358C), and the simulated domain-averaged SAT06 is higher than the NOAH and the observed SAT06, which have greater and more consistent initial soil moisture. In addition, RUCsm results in BAIS and RMSE values that are lower than RUC and a worse SAT06 pattern (e.g., shown in TS values). Moreover, despite the very small difference in the RMSE between SLAB and NOAH, a large difference in the TS exists between the two schemes (approximately 0.03; Figure 4a). Thus, the simulated SAT06 is highly sensitive to the LSSs. In the 24 h simulations using WRF Version 2.2 [Zeng et al., 2011], the SLAB LSS presents the highest TS, RUC produces smaller values than SLAB, and NOAH produces the lowest values, with RMSE values between 4 and 58C. In contrast, in the 24 h (DAY1) simulations using the WRF Version 3 model (except for PX in the four LSSs), RUC, SLAB, and NOAH produce increasingly smaller TS values, with RMSE values between 2.6 and 3.18C. Therefore, the behavior of a given LSS can differ when different versions of the model are used, and the overall performance improves due to the updated model.

4. Interpretation of Physical Mechanisms 4.1. Preliminary Analysis 4.1.1. Sensible and Latent Heat Fluxes The above results show that the simulated SAT06 can be greatly modified by different LSSs for a given lead time and by using different lead times for a given LSS. In terms of physical mechanisms, all of the modifications directly result from different LSS-induced land surface heat fluxes and from the integration-lengthinduced error growth. Figures 5a and 5b presents the DAY1 variations in the area-averaged sensible and latent heat fluxes. Because of the different LSSs, the sensible heat flux generally decreases, while the latent heat flux increases, and vice versa, at a given time. Specifically, the net radiation remains nearly unchanged under given conditions and is partitioned between sensible and latent heat fluxes, which have also been reported in previous studies [e.g., Zeng et al., 2014]. Compared with diurnal variations in the SAT and sensible heat flux (e.g., DAY1 in Figures 3j and 5a), the two types of diurnal variations usually agree, for example, high values of sensible heat flux correspond with high SATs during the daytime and correspond very well with high SAT06 values. Regarding the 0600 UTC simulation results with longer lead times, the area-averaged sensible heat flux has a very high, positive correlation with SAT06 (Figures 5c versus 3k). For example, some LSS-induced temperature features, such as the small NOAH-SLAB SAT06 differences, the small but steadily increasing SAT06 RMSE values that change with the lead time for both LSSs, the rapidly increasing PX RMSE values, and the rapidly increasing RUC RMSE values over DAY1–DAY7, are strongly consistent with the simulated sensible heat flux at 0600 UTC. Notably, the thermodynamic variables are most affected by changes in the land surface conditions, particularly during the intensive daytime heating, as previously reported [e.g., Trier et al., 2008]. Figure 6 shows the spatial distribution of the differences of the 0600 UTC sensible heat flux. For a given LSS, the spatial distribution of the lead-time-induced differences in sensible heat flux correspond with the temperature difference distribution (e.g., the DAY3–DAY1 differences for RUC in Figures 6a and 3c and the DAY7–DAY1 differences for PX in Figures 6b and 3d), which indicates that the SAT06 warming is closely related to the higher sensible heat flux and vice versa. Theoretically, a higher (lower) sensible heat flux indicates stronger (weaker) direct heating in the lower atmosphere and greater simulated SATs, as confirmed by the simulation results of the different schemes (e.g., the SLAB-NOAH results in Figures 6c and 3e and the RUC-PX results in Figures 6d and 3f).

ZENG ET AL.


1314


10.1002/2015MS000440

Figure 5. Domain (DN)-averaged DAY1 diurnal SAT variations in land surface sensible and latent heat fluxes (a and b, respectively), the simulated sensible heat fluxes at 0600 UTC on 23 July and derived from hourly values (c and d, respectively), latent heat fluxes (e) with observation-based data from daily evaporations by Miralles et al. [2013], and simulated 850 hPa geopotential heights (f) at 0600 UTC on 23 July according to the lead times and LSSs.

The 0600 UTC sensible heat flux generally increases with time for all LSSs (Figure 5c), for which RUC and PX present relatively rapid sensible heat flux increases, particularly over DAY1–DAY7, and become stabile over longer periods (i.e., DAY8–DAY14). In contrast, the NOAH and SLAB produce slowly sensible heat fluxes that slowly increase with lead time. Substantial differences occur in the 0600 UTC area-averaged sensible heat flux (e.g., up to 185 W m22 between RUC and NOAH on DAY7). These trends of the sensible heat flux variations are consistent with those of the temperature variations (Figure 3k). Overall, each LSS poorly simulates medium-range and high-temperature events when compared with short-range and high-temperature events. One exception is NOAH, which consistently underestimates SAT06. Although the instantaneous 0600 UTC fields of the sensible heat flux agree well with with those of SAT06, the simulated SAT06 fields are actually produced over the entire integration period. Figure 5d presents the average sensible heat fluxes derived from the hourly values over the entire period, which are very similar to the 0600 UTC values of sensible heat flux (Figure 5c) and the SAT06 (Figure 3k). This result indicates that the simulated SAT06 described above responds very well to cumulative surface sensible heating.

ZENG ET AL.


1315


10.1002/2015MS000440

Figure 6. Distributions of sensible heat flux difference fields at 0600 UTC on 23 July due to the different LSSs and leads times (e.g. ‘‘DAY3– DAY1 (RUC)’’ denotes RUC-simulated DAY3 minus DAY1 results; units: W m22).

Because land surface latent heat flux can be obtained using direct moisture (evaporation) observations, it appears to be more physically based than the sensible heat flux in the assimilation data sets. To test how well these LSSs simulate surface fluxes, Figure 5e presents the LSS-simulated area-averaged evapotranspiration and corresponding observation-based evapotranspiration derived from Miralles et al. [2013]. All four LSSs overestimate the evapotranspiration. The RUC evapotranspiration is closest to the observations, while the PX evapotranspiration, which generally decreases and varies greatly over the integration periods, appears unreasonable compared with the observations. In addition, according to the direct relationship between the SAT06 (Figure 3k) and the sensible heat flux (Figure 5d), SLAB and NOAH provide the least and most reasonable sensible heat fluxes overall. For medium-range evapotranspiration overestimations (Figure 5e), it is suggested that the four LSSs overestimate the net surface radiation (sum of sensible and latent heat fluxes), which plays a fundamental role in land-atmosphere interactions [Betts, 2009] and influences atmospheric energies. 4.1.2. Soil Moisture Previous investigations have revealed the importance of soil moisture in weather/climate simulations. For instance, Koster et al. [2009] explained the general problem of LSS initialization that leads to climate drift and suggested the importance of LSS consistency when determining initial soil moisture contents. However, for the mega-heatwave event, Miralles et al. [2014] indicated that the crucial effect of soil desiccation also depends on atmospheric heat accumulation. As discussed in section 2, NOAH, RUCsm, and PX have ‘‘consistent’’ initial soil moisture contents (the soil moisture values are nearly the same in the top soil layer of each LSS, which rapidly respond to atmospheric forcing in the short and medium ranges), while RUCsm (NOAHsm) has a 17% (20%) enhancement relative to RUC (NOAH). Figure 7 presents the area-averaged initial soil moisture contents and the soil moisture variations for the LSSs with changing lead time. Regardless of the inconsistencies in the LSS soil layering, the initial soil moisture content is relatively consistent across NOAH, RUCsm, and PX (Figures 7a, 7c, and 7g) in the top layer and in the top 1 m layer. However, soil moisture depletion is produced very differently. The NOAH

ZENG ET AL.


1316


10.1002/2015MS000440

Figure 7. (left) Domain-averaged initial soil moisture contents and (right) soil moisture variations for the LSSs have changed with the lead times, where soil moisture variations are from hourly averages for individual simulations with the following soil layers: NOAH (Layer 1: 0– 10 cm; Layer 2: 10–40 cm; Layer 3: 40–100 cm; Layer 4: 100–200 cm), PX (Layer 1: 0–1 cm; Layer 2: 1–99 cm), RUC (Level 1: 0 cm; Level 2: 5 cm; Level 3: 20 cm; Level 4: 40 cm; Level 5: 160 cm; Level 6: 300 cm), and FNL (Layer 1: 0–10 cm; Layer 2: 10–200 cm). The values for the top 1 m Layer (Layer 1 m) are calculated using linear interpolation of the soil moisture content in the LSS layers. In addition, some lines may overlap due to very close values (e.g., lines of Layer 2 and Layer 1 m in Figure 7d).

ZENG ET AL.


1317


10.1002/2015MS000440

successfully simulates soil moisture variations (Figures 7b versus 7l), while the RUCsm and PX present more soil moisture depletion than NOAH. Because PX, RUCsm, and NOAH simulate evapotranspiration that is approximately equivalent for simulations DAY1–DAY6 (shown in Figure 5e in latent heat flux), the large PX soil moisture depletion is suggested to result from the large internal flow (gravity drainage) across the soil layers, which is further attributed to a deficiency in the soil hydraulic parameterization. For example, PX has the fewest soil layers (two layers) among the LSSs and is less detailed than a multilayer LSS based on the solution of the vertical one-dimensional moisture diffusion equations [e.g., Garratt, 1993]. This outcome would cause soil moisture and SAT biases. As shown in Figure 4c, PX exhibits the largest overall SAT06 BIAS among the LSSs. In addition, it appears that evaporation is less important than soil water drainage for affecting soil moisture depletion/variation and SAT (Figures 5 and 7). For instance, consistent large surface soil moisture depletion can be seen in the PX for all DAY1–DAY14 simulations, whereas high evaporation can only be seen for DAY1–DAY7. In the RUC, high soil moisture depletion and low evaporation can be seen for DAY2–DAY14. However, surface soil moisture is more negatively correlated with the 0600 sensible heat flux and SAT06. Although NOAH and NOAHsm simulate small soil moisture sensitivity (Figures 7a, 7b versus 7i, 7j), both RUC and RUCsm simulate apparently large surface soil moisture depletion (Figures 7e, 7f versus 7g, 7h). The NOAH-NOAHsm differences in surface soil moisture depletion appear smaller than the RUC-RUCsm differences. When considering the approximately equivalent differences in the initial soil moisture contents, this result suggests that the RUC LSS is more sensitive to soil moisture initialization than the NOAH LSS. Additionally, the SAT06 differences due to soil moisture initialization are comparable to those induced by the LSSs (e.g., the NOAHsm-NOAH differences versus the PX-NOAH differences; Figure 3k). Even with consistent initial soil moisture contents, it can be concluded that the LSSs produce very different soil moisture values. RUC and especially PX generate large soil moisture depletions that largely deviate from the FNL data, whereas NOAH shows good performance in the soil moisture simulations. In addition, different LSSs have different sensitivities to soil moisture initialization in the soil moisture simulations, and the SAT sensitivities induced by soil moisture initialization are comparable with those induced by LSSs. All of these LSS behaviors in the soil moisture simulations are consistent with corresponding SAT simulations because the simulation of soil moisture can only affect surface heat fluxes in the SAT simulations. 4.1.3. Geopotential Height Fields Although the simulated surface fluxes directly affect the SAT results, changes in the surface fluxes would also modify the atmospheric circulation. For example, Walker and Rowntree [1977] showed the substantial effects of soil moisture on regional atmospheric circulation. Zhou et al. [2009] indicated that the westerly extension of the WPSH is partially caused by surface heating in the India Ocean and western Pacific Ocean. The modified atmospheric circulation can also result in SAT changes. As indicated by Lin et al. [2005], the 2003 summer heat wave in southern China primarily resulted from the persistent strong anomalies of the WPSH. As further discussed by Zeng et al. [2011], the weather in the simulation region was controlled by the WPSH in July 2003, and the effect induced by moist convective activities was relatively small. Therefore, the effect of the LSS on the high-temperature short-range (24 h) simulations was dominated by differences in the land surface’s direct sensible heating, while the effect of the LSS-induced changes in atmospheric circulation on the surface temperature is secondary. Figures 3k, 5c, and 5f present the variations in the regional mean surface temperature, sensible heat fluxes, and 850 hPa geopotential heights, respectively. Negative feedback is apparent in the simulations with consistent initial soil moisture contents (e.g., NOAH, PX, and RUCsm) and in the simulations with adjusted initial soil moisture contents (i.e., NOAH versus NOAHsm, and RUC versus RUCsm). Thus, as the sensible heat flux increases, the SAT06 increases and the geopotential heights decrease in the lower atmosphere and vice versa. According to Wei et al. [2010], who separated land effects and feedback in a study on landatmosphere interactions, the feedbacks between temperature, sensible heat flux, and geopotential heights in the atmosphere can be explained clearly. Enhanced sensible heat flux (regardless of whether it is induced by low soil moisture or by land parameterization) generally leads to higher SAT with a stronger thermal low pressure regime that is accompanied by a weak updraft in the lower atmosphere and intensified anticyclone circulation in the upper troposphere [Pal and Eltahir, 2003; Fischer et al., 2007a]. In addition, when considering the WPSH that controlled the simulation area (Figures 8a and 8b), a mean subsidence occurred that corresponded to the strong, dynamical anti-cyclone pressure structure throughout the troposphere,

ZENG ET AL.


1318


10.1002/2015MS000440

Figure 8. Distributions of geopotential height fields at 0600 UTC on 23 July due to the different LSSs and lead times (unit: gpm): (a) SLAB-DAY14 geopotential heights at 500 hPa derived from hourly values; (b) same as Figure 8a but for 850 hPa; (c) SLAB minus NOAH differences in the 850 hPa geopotential heights from DAY7; (d) same as Figure 8c but for RUC minus PX differences; (e) same as Figure 8c but for DAY1 RUCsm minus NOAH differences; and (f) same as Figure 8c but for DAY14 NOAHsm minus NOAH differences.

which would act on the land surface-induced changes and is less affected by land surface processes than thermal variables [Trier et al., 2008]. Therefore, the strong anti-cyclone with subsidence would typically be reduced by the weak cyclone in the lower troposphere and enhanced in the upper troposphere due to the increased SAT. These results indicate an LSS-induced negative (positive) feedback between the temperature and atmospheric circulation in the lower (upper) atmosphere and are consistent with the results presented by Zeng et al. [2011, 2014], who performed 24 h WRF simulations using different LSSs and initial moisture contents. In addition, these results are inconsistent with (but do not theoretically contradict) the low-level positive temperature-circulation feedback proposed by Fischer et al. [2007a], who investigated the sensitivity of the regional climate to initial soil moisture.

ZENG ET AL.


1319


10.1002/2015MS000440

Moreover, in the medium-range simulations, the LSS-induced differences in the SAT correspond better with the differences in the sensible heat flux (e.g., Figures 6c, 6d versus 3e, 3f) than with the differences in the geopotential heights at low pressure levels (e.g., Figures 6c, 6d versus 8c, 8d). This finding highlights the dominant (secondary) role of the surface sensible heat flux (circulation) in the medium-range surface temperature forecasts and further suggests that for the LSS-induced temperature errors occur within a reasonable range. In addition, the errors induced by the surface fluxes would typically be larger than those induced by the mean subsidence in the WPSH in both the short and medium ranges, as discussed below. 4.2. Detailed Explanation 4.2.1. Diagnostic Equation for Temperature As qualitatively addressed in section 4.1, land surface sensible heat flux plays an important role in LSSinduced SAT changes when compared with the mean subsidence in the WPSH. However, except for the atmospheric dynamic effect, the SAT is directly modified by the net heat flux in the atmospheric surface layer rather than at the land surface, and the land surface (skin) sensible heat flux only contributes to the SAT change as a component of the net heat flux. These results can be explained using equations for temperature change. Derived from the thermodynamic equation [Zeng et al., 2014], the prognostic equation of air temperature is given as follows: @T 52V rT2wðcd 2cÞ1Ht ; @t

(2)

where V and w denote the horizontal wind vector and vertical velocity, respectively, c and cd are the atmospheric temperature and dry adiabatic lapse rates, respectively, and Ht is the diabatic heating rate. Because of the lack of sufficient output for the 2 m level, as described by Zeng et al. [2014], we use equation (2) for the lowest model level (30 m) to quantify the relative importance of near-surface air temperature change processes. In addition, the weather was controlled by strong and persistent WPSHs in July 2003 when no weather systems penetrated the study region. Thus, the process of advection was less important. Referring to section 4.1, we focus on the two terms in equation (2), i.e., Ht (the diabatic term that mainly represents the convergence of sensible heat flux [Zeng et al., 2014]) and CN (the adiabatic convection term that represents the mean subsidence effect in the WPSH), where CN 52wðcd 2cÞ:

(3)

Therefore, according to equation (2), which uses the output for the lowest model level (30 m), the former three terms that include CN in their equations can be calculated directly. The Ht term, which is difficult to directly compute, is obtained using the three terms of the equation [Zeng et al., 2014]. 4.2.2. Comparison Between the Adiabatic and Diabatic Effects Figure 9 displays the area-averaged CN and Ht terms in equation (2) as affected by the LSSs and lead times. The most striking feature of the CN values is that only small differences (generally lower than 2 3 10258C s21) occur among SLAB, NOAH, RUC, and RUCsm, while the PX values are much lower than those of other LSSs (Figure 9a). Meanwhile, RUCsm, SLAB, NOAH, RUC, and PX present increasingly higher Ht values (Figure 9b), and large Ht differences exist among the LSSs. The NOAH-RUCsm differences are even larger than the RUCsm magnitudes. These findings clearly suggest large LSS uncertainties in simulating the net heat flux. Regarding the effects of the initial soil moisture (Figures 9a and 9b), the NOAH-NOAHsm CN (Ht) differences are roughly 1/2 (1/3) of the RUC-RUCsm differences, confirming their high sensitivity to initial soil moisture. Lower soil moisture contents lead to lower CN (higher Ht) values, which is consistent with the results of Zeng et al. [2014], who suggested the presence of negative SAT-circulation feedback due to soil moisture in the lower layer (low soil moisture enhances adiabatic heating (Figure 9b) and the SAT but reduces the WPSH circulation, which results in reduced subsidence (Figure 9c) and reduced adiabatic heating (Figure 9a)). In addition, a comparison between the surface area-averaged sensible heat flux (from hourly values; Figure 5d) and the convergence of the sensible heat flux (i.e., Ht; Figure 9b) indicates that low soil moisture leads to high land (skin) surface sensible heat flux and consistently high flux convergence (i.e., NOAH versus NOAHsm; RUC versus RUCsm). However, when the initial soil moisture content is consistent, the LSSproduced surface sensible heat flux and flux convergence may be inconsistent, for example, the NOAH

ZENG ET AL.


1320


10.1002/2015MS000440

Figure 9. Domain-averaged rate of change of SAT in the CN (a) and Ht (b) terms of equation (2), and the vertical velocity (c) and air temperature lapse rate (d) derived from hourly values according to the LSSs and lead times.

differences are very small for the surface sensible heat flux and very large for the flux convergence. This inconsistency occurs because LSSs lead to inconsistent changes in the lower atmosphere, which further affect the heat transfer near the surface. To explain the large CN differences, Figures 9c and 9d provides values for the vertical velocity and air temperature lapse rates that affect the CN calculation shown in equation (3). PX produces subsidence that is substantially weaker than that of the other LSSs, in contrast with Trier et al. [2008], who indicated that thermodynamic quantities are most often affected by perturbations of land conditions. Moreover, the c calculation indicates that PX simulates the largest c values (close to 0.00988C m21, the cd value), which clearly suggests that PX induces unrealistic conditions during most integration periods (i.e., the super-adiabatic state is too long) (Figure 9d). When tested against observations, all of the LSSs overestimate the mean temperature lapse rates, which explain why the SAT06 values are generally higher than the observations (Figure 4b). By combining the cd minus c results with the w values, the weakest adiabatic subsidence effect of PX is determined. Notably, daytime intensive heating and areas with abnormally large c values dominate the area-averaged c calculation for PX. These results reflect the deficiencies of the LSS for simulating the SAT with unrealistically weak adiabatic heating (CN) and an intensive diabatic effect (Ht) in high-temperature weather.

5. Conclusions and Discussion To investigate the sensitivities of short and medium-range high-temperature simulations to different LSSs, we applied four LSSs (i.e., SLAB, NOAH, RUC, and PX) in the mesoscale model Advanced WRF, Version 3.3. We designed a series of simulations, including sensitivity simulations induced by soil moisture initialization (i.e., RUCsm and NOAH) with forecast lead times (i.e., integration lengths) that varied from 1 to 14 days, to simulate the East China high-temperature event on 23 July 2003, which is used as the end time in the

ZENG ET AL.


1321


10.1002/2015MS000440

simulations (DAY1–DAY14). We focused on the 0600 UTC (1400 Local Time) surface air temperature to evaluate the occurrence and development of high temperatures. This type of experimental design facilitates a comparison of the simulations under a given criterion. The major conclusions were as follows: 1. The four LSSs can generally simulate surface air temperature distribution patterns for the short and medium ranges, whereas different schemes lead to substantially different SAT performances for the medium range, suggesting that the simulated high temperature is sensitive to different LSSs. For shortrange simulations (i.e., DAY1–DAY2), the overall PX produces a spatial distribution of SAT06 that is more consistent with the observations. Followed by NOAH and RUC, the performance of SLAB is the worst. Regarding the TS of SAT06 values greater than 358C, PX produces the highest value, followed by RUC and SLAB. NOAH presents the lowest values, with a maximum difference of approximately 0.15 between NOAH and PX. All of the LSSs, except for the NOAH LSS, overestimate the SAT06. PX provides the lowest RMSE value, SLAB has a value slightly higher than NOAH, while RUC presents the fastest RMSE increase, which is much higher than that of the other schemes (DAY3–DAY1 difference of 1.18C). In the mediumrange simulations (i.e., DAY3–DAY14), each LSS has an above-358C SAT06 TS that is approximately consistent. However, the sensitivity to the LSSs is greater in the medium-range simulations than in the short-range simulations (e.g., a maximum DAY6 TS difference of approximately 0.26 between PX and NOAH). Similarly, the error analysis indicates a higher sensitivity compared with the short range. PX presents substantial growth in the SAT06 error with lead times longer than 6 days and produces errors that exceed those of SLAB and NOAH in the DAY8 and longer simulations. RUC maintains the largest error (e.g., in the DAY6 simulations, the area-averaged RUC-PX difference amounts to 2.18C), while SLAB and NOAH produce similar small errors. The SLAB-NOAH error differences steadily increase with the lead time in all of the simulations (DAY1–DAY14), i.e., in simulations longer than DAY8, SLAB, NOAH, PX, and RUC produce increasingly larger errors. 2. The high-temperature event simulated by a given LSS is also sensitive to the length of the lead time. Although NOAH and SLAB have stable error magnitudes in all of the simulations, PX presents significantly increasing errors in the simulations longer than DAY6, with RMSE values of 2.6 and 3.78C in the short and medium ranges, respectively. The RUC errors continue to increase rapidly for lead times of less than 7 days, with RMSE values of 3.0 and 4.78C in the short and medium ranges, respectively. 3. Although the sensitivity simulations focus on the high-temperature event of 23 July 2003, the simulated results reflect the systematic effects of the LSSs. For example, a high-temperature overestimation is apparent in all of the RUC simulations with varying lead times, while PX produces reasonable simulations for lead times shorter than or equal to 8 days (both in the short and medium ranges). The SLAB LSS also produces accurate and steady simulations in the medium ranges longer than 7 days. These results elucidate the model performance for choosing suitable LSS options for simulating regional high-temperature events. 4. The variations in the simulated SAT06 fields include the cumulative (or time-averaged) results of the simulated sensible heat fluxes. Validated against observation-based data, the simulated latent heat fluxes are generally overestimated. Because the surface net radiation is partitioned between sensible and latent heat fluxes, the net radiation may also be overestimated by all four LSSs. Thus, although the LSS simulations are complex, at least in terms of surface fluxes, the latent heat flux, and net radiation might appear unreasonable (e.g., NOAH in this case), even when an LSS produces a reasonable sensible heat flux. This result would further affect the soil temperature and SAT. Because of differences in the LSSs and lead times, errors in the simulated land surface fluxes can be amplified, which would induce higher errors when simulating higher temperatures in the medium-range simulations than in the short-range simulations. In general, all four LSSs, except the NOAH LSS, overestimate the SAT. The PX, NOAH, SLAB, and RUC LSSs produce increasingly higher surface sensible heat fluxes (mean hourly values) in the shortrange simulations. Thus, the LSS-produced SAT06 becomes increasingly higher. In the medium range, with lead times of more than 7 days, NOAH and SLAB produce nearly equivalent sensible heat fluxes along with higher PX and higher RUC sensible heat fluxes, which are consistent with the SAT06 values. 5. The LSS SAT simulations are sensitive to soil moisture initialization, even if consistent soil moisture is applied in the LSSs. PX generates the largest soil moisture depletions, while NOAH shows good performance in the soil moisture simulations. In addition, different LSSs have different sensitivities to soil moisture initialization, and the SAT sensitivities to soil moisture initialization are comparable with the sensitivities to the LSS choice. The parameterization of soil moisture affects soil moisture depletion and is

ZENG ET AL.


1322


10.1002/2015MS000440

very important, with major deficiencies regarding soil water drainage due to gravity. However, evaporation is less important. All of these soil moisture behaviors are simulated by the LSSs, with adjusted soil moisture contents that are consistent with the SAT simulations. 6. The sensitivity of the simulated temperature is directly caused by different land surface fluxes (in particular, sensible heat flux) and is affected by different LSS and lead times, and soil moisture, including soil moisture initialization and soil moisture depletion, which are affected by land parameterizations and are the main causes of differences in the sensible heat flux. In addition, the corresponding temperature changes induced by atmospheric circulation are of secondary importance in the high-temperature medium range weather that occurs over East China. LSS-induced negative feedback also occurs between the SAT and low-level circulation in the medium-range simulation (rather than the positive feedback that is commonly found in land-atmosphere interactions). This result was previously shown in the short-range simulation and was explained by the combined effects of surface sensible heating and the WPSH [Zeng et al., 2011]. 7. Using the SAT change equation, further theoretical analysis of the physical processes that affect the SAT shows that the choice of the LSS and lead times, within a reasonable range, affects the error growth mainly due to the diabatic (mainly sensible net heat flux) term rather than the adiabatic convective (subsidence) term. Large uncertainty in the simulation of these heating terms is observed. Additionally, an inconsistency between the surface sensible heat flux and the flux convergence is produced due to the LSS-induced changes at the land surface and in the lower atmosphere. When tested against observational temperature data, the PX simulations show large errors. The simulated surface air superadiabatic state is too strong (with an averaged temperature lapse rate that is greater than the adiabatic temperature lapse rate, i.e., c>cd,) and covers a long duration with different lead times. Consequently, large errors are induced when calculating the adiabatic (CN) and diabatic (Ht) terms (i.e., adiabatic heating is too weak and the diabatic effect is too intense at the surface). In short, the LSS-induced difference in sensible heat flux (i.e., the direct quantity at the land surface with the largest contribution to net heat flux), the LSS-induced difference in the convergence of the sensible heat flux (i.e., the most direct quantity for the SAT changes in the atmosphere), and the parameterization differences of soil moisture initialization and depletion (i.e., the most important factor for soil layers) are the main three causes for the SAT differences. Investigations of all of these causes are valuable and result from different angles. However, the parameterization differences in soil moisture would be chosen as the most important cause because it addresses the deepest LSS issues. Overall, this study highlights the importance of LSS choice and soil moisture initialization when simulating high-temperature weather events with various lead times. In addition, several systematic features or deficiencies of the WRF modeling system and the LSSs are found in the SAT simulations. Regarding the drift of the LSSs in the SAT simulations, the drift is strong for RUC and PX and weak for NOAH and SLAB (Figure 3k). This difference is likely associated with soil moisture initialization due to the following reasons. (i) Although the SLAB LSS does not include dynamic soil moisture, the prescribed seasonal averages likely prevent soil moisture values from becoming overly biased. (ii) The NOAH LSS has been widely used in various applications, including its operational application in the NCEP Global Forecasting System (GFS). In addition, the FNL data are derived from the GFS, which provide consistent data for use in this study. (iii) The RUC LSS does not have initial soil moisture contents that are consistent with the FNL, NOAH, or PX LSSs, which results from the WRF modeling system in land initialization and deviates from FNL by as much as 20%. Additionally, both RUC and RUCsm produce soil moisture contents that are much lower than those of the FNL data (Figure 7), resulting in large, positive SAT06 biases (Figure 3k). (iv) Although the PX LSS has consistent soil moisture, the simulated surface soil moisture is low relative to the FNL data, potentially due to PX soil layering (i.e., it has the fewest soil layers (two layers) among the LSSs and is less detailed than a multilayer LSS based on the solutions of the vertical one-dimensional moisture diffusion equations [e.g., Garratt, 1993]). This finding indicates large soil moisture depletion and strong gravity drainage, which induce large positive SAT06 biases. Because each LSS presents a systematic simulation error growth with lead time, the simulation results have implications for the predictabilities of WRF’s for medium-range weather, which can help us understand the model performance and practically apply the model. The results and conclusions presented here were obtained using WRF models for a high-temperature case. Because the performances of regional models are

ZENG ET AL.


1323


10.1002/2015MS000440

closely associated with model settings, such as domain size, buffer zone, and initial and boundary conditions [e.g., Leung and Qian, 2003; Zeng et al., 2014], further simulations are needed (e.g., for examining sensitivities under different settings).

Acknowledgments The authors would like to thank the anonymous reviewers for their helpful comments on the manuscript. This study was supported by the National Natural Science Foundation of China (grant 41275012 and 41205073). The conventional meteorological station data employed in this paper are freely available upon request at the China Meteorological Administration. In addition, we wish to thank Diego Miralles at University of Bristol for providing the evaporation data. The initial and boundary conditions of the WRF were obtained from Final (FNL) Analysis data, which were prepared by the National Centers for Environmental Prediction (NCEP) and are available from the Computational and Information Systems Laboratory (CISL) at the National Center for Atmospheric Research (NCAR) under data set number ds083.0 (http://dss.ucar.edu). NCAR is financially supported by the National Science Foundation.

ZENG ET AL.

References Betts, A. K. (2009). Land-surface-atmosphere coupling in observations and models, J. Adv. Model. Earth Syst., 1, 4, doi:10.3894/ JAMES.2009.1.4. Chen, F., and J. Dudhia (2001), Coupling an advanced land-surface hydrology model with the Penn State-NCAR MM5 modeling system. Part I: Model description and implementation, Mon. Weather Rev., 129, 569–585, doi:10.1175/15200493(2001)129 2.0.CO;2. Cheng, W. Y. Y., and W. J. Steenburgh (2005), Evaluation of surface sensible weather forecasts by the WRF and the Eta models over the western United States, Weather Forecast., 20, 812–821, doi:10.1175/WAF885.1. Cheng, F. Y., Y. C. Hsu, P. L. Lin, and T. H. Lin (2013), Investigation of the effects of different land use and land cover patterns on mesoscale meteorological simulations in the Taiwan area, J. Appl. Meteorol. Climatol., 52, 570–587, doi:10.1175/JAMC-D-12-0109.1. Della-Marta, P. M., and H. Wanner (2006), A method of homogenizing the extremes and mean of mean of daily temperature measurements, J. Clim., 19, 4179–4197, doi:10.1175/JCLI3855.1. Della-Marta, P. M., M. R. Haylock, J. Luterbacher, and H. Wanner (2007), Doubled length of Western European summer heat waves since 1880, J. Geophys. Res., 112, D15103, doi:10.1029/2007JD008510. Dudhia, J. (1989), Numerical study of convection observed during the winter monsoon experiment using a mesoscale two-dimensional model, J. Atmos. Sci., 46, 3077–3107, doi:10.1175/1520-0469(1989)0462.0.CO;2. Dudhia, J. (1996), A multi-layer soil temperature model for MM5, in 6th PSU/NCAR Mesoscale Model Users’ Workshop, pp. 49–50, Boulder, Colo. Ek, M. B., K. E. Mitchell, Y. Lin, E. Rogers, P. Grunmann, V. Koren, G. Gayno, and J. D. Tarpley (2003), Implementation of Noah land surface model advances in the national centers for environmental prediction operational mesoscale Eta model, J. Geophys. Res., 108(D22), 8851, doi:10.1029/2002JD003296. Fan, X. (2009), Impacts of soil heating condition on precipitation simulations in the Weather Research and Forecasting Model, Mon. Weather Rev., 137, 2263–2285, doi:http://dx.doi.org/10.1175/2009MWR2684.1. Ferranti, L., and P. Viterbo (2006), European summer of 2003: Sensitivity to soil water initial conditions, J. Clim., 19, 3659–3680, doi:10.1175/ JCLI3810.1. Feudale, L., and J. Shukla (2011), Influence of sea surface temperature on the European heat wave of 2003 summer. Part II: A modeling study, Clim. Dyn., 36, 1705–1715, doi:10.1007/s00382-010-0789-z. Fischer, E. M., S. I. Seneviratne, P. L. Vidale, D. L€ uthi, and C. Sch€ar (2007a), Soil moisture-atmosphere interactions during the 2003 European summer heat wave, J. Clim., 20, 5081–5099, doi:10.1175/JCLI4288.1. Fischer, E. M., S. I. Seneviratne, D. L€ uthi, and C. Sch€ar (2007b), Contribution of land-atmosphere coupling to recent European summer heat waves, Geophys. Res. Lett., 34, L06707, doi:10.1029/2006GL029068. Garratt, J. R. (1993), Sensitivity of climate simulations to land-surface and atmospheric boundary-layer treatments: A review, J. Clim., 6, 419–448, doi:10.1175/1520-0442(1993)0062.0.CO;2. Hong, S.-Y., J. Dudhia, and S.-H. Chen (2004), A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation, Mon. Weather Rev., 132, 103–120, doi:10.1175/1520-0493(2004)1322.0.CO;2. Hong, S.-Y., Y. Noh, and J. Dudhia (2006), A new vertical diffusion package with an explicit treatment of entrainment processes, Mon. Weather Rev., 134, 2318–2341, doi:10.1175/MWR3199.1. IPCC (2013), Summary for policymakers, in Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, edited by T. F. Stocker et al., Cambridge Univ. Press, Cambridge, U. K. Kain, J. S. (2004), The Kain-Fritsch convective parameterization: An update, J. Appl. Meteorol., 43, 170–181, doi:10.1175/1520-0450(2004) 0432.0.CO;2. Kharin, V. V., F. W. Zwiers, X. Zhang, and G. C. Hegerl (2007), Changes in temperature and precipitation extremes in the IPCC ensemble of global coupled model simulations, J. Clim., 20, 1419–1444, doi:10.1175/JCLI4066.1. Klein Tank, A. M. G., and G. P. K€ onnen (2003), Trends in indices of daily temperature and precipitation extremes in Europe, 1946–99, J. Clim., 16, 3665–3680, doi:10.1175/1520-0442(2003)0162.0.CO;2. Koster, R. D., Z. Guo, R. Yang, P. A. Dirmeyer, K. Mitchell, and M. J. Puma (2009) On the nature of soil moisture in land surface models, J. Clim., 22, 4322–4335, doi:10.1175/2009JCLI2832.1. Kuglitsch, F. G., A. Toreti, E. Xoplaki, P. M. Della-Marta, J. Luterbacher, and H. Wanner (2009), Homogenization of daily maximum temperature series in the Mediterranean, J. Geophys. Res., 114, D15108, doi:10.1029/2008JD011606. Kuglitsch, F. G., A. Toreti, E. Xoplaki, P. M. Della-Marta, C. S. Zerefos, M. T€ urkes¸, and J. Luterbacher (2010), Heat wave changes in the eastern Mediterranean since 1960, Geophys. Res. Lett., 37, L04802, doi:10.1029/2009GL041841. Leung, L. R., and Y. Qian (2003), The sensitivity of precipitation and snowpack simulations to model resolution via nesting in regions of complex terrain, J. Hydrometeorol., 4, 1025–1043, doi:10.1175/1525-7541(2003)0042.0.CO;2. Lin, J., B. G. Bi, and J. H. He (2005), Physical mechanism responsible for western Pacific subtropical high variation and hot wave in Southern China in July 2003, Chin. J. Atmos. Sci., 29(4), 594–599, doi:10.3878/j.issn.1006-9895.2005.04.10. Liu, X., Z.-Y. Yin, X. Shao, and N. Qin (2006), Temporal trends and variability of daily maximum and minimum, extreme temperature events, and growing season length over the eastern and central Tibetan Plateau during 1961–2003, J. Geophys. Res., 111, D19109, doi:10.1029/ 2005JD006915. Meehl, G. A., and C. Tebaldi (2004), More intense, more frequent and longer lasting heat waves in the 21st century, Science, 305, 994–997, doi:10.1126/science.1098704. Miralles, D. G., et al. (2013), El Ni~ no–La Ni~ na cycle and recent trends in continental evaporation, Nat. Clim. Change, 4, 122–126, doi:10.1038/ nclimate2068. Miralles, D. G., A. J. Teuling, C. C. van Heerwaarden, and J. V. de Arellano (2014), Mega-heatwave temperatures due to combined soil desiccation and atmospheric heat accumulation, Nat. Geosci., 7, 345–349, doi:10.1038/ngeo2141.


1324


10.1002/2015MS000440

Mlawer, E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, and S. A. Clough (1997), Radiative transfer for inhomogeneous atmosphere: RRTM, a validated correlated-k model for the long-wave, J. Geophys. Res., 102(D14), 16,663–16,682, doi:10.1029/97JD00237. Pal, J. S., and E. A. B. Eltahir (2003), A feedback mechanism between soil moisture distribution and storm tracks, Q. J. R. Meteorol. Soc., 129, 2279–2297, doi:10.1256/qj.01.201. Sch€ar, C., P. L. Vidale, D. L€ uthi, C. Frei, C. H€aberli, M. A. Liniger, and C. Appenzeller (2004), The role of increasing temperature variability in European summer heatwaves, Nature, 427, 332–336, doi:10.1038/nature02300. Skamarock, W. C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker, M. G. Duda, X.-Y. Huang, W. Wang, and J. G. Powers (2008), A description of the Advanced Research WRF Version 3, NCAR Tech. Note TN–4751STR, 125 pp., Natl. Cent. for Atmos. Res., Boulder, Colo. Smirnova, T. G., J. M. Brown, and S. G. Benjamin (1997), Performance of different soil model configurations in simulating ground surface temperature and surface fluxes, Mon. Weather Rev., 125, 1870–1884, doi:10.1175/1520-0493(1997)1252.0.CO;2. Smirnova, T. G., J. M. Brown, S. G. Benjamin, and D. Kim (2000), Parameterization of cold-season processes in the MAPS land surface scheme, J. Geophys. Res., 105(D3), 4077–4086, doi:10.1029/1999JD901047. Stott, P., D. A. Stone, and M. Allen (2004), Human contribution to the European heatwave of 200, 3, Nature, 432, 610–614, doi:10.1038/ nature03089. Trier, S. B., F. Chen, K. W. Manning, et al. (2008), Sensitivity of the PBL and precipitation in 12-Day simulations of warm-season convection using different land surface models and soil wetness conditions, Mon. Weather Rev., 136, 2321–2343, doi:10.1175/2007MWR2289.1. Trigo, R. M., R. Garcıa–Herrera, J. Dıaz, I. F. Trigo, and M. A. Valente (2005), How exceptional was the early August 2003 heatwave in France?, Geophys. Res. Lett., 32, L10701, doi:10.1029/2005GL022410.issn:0094-8276. T€ urkes¸, M., U. M. S€ umer, and I. Demir (2002), Re-evaluation of trends and changes in mean, maximum and minimum temperatures of Turkey for the period 1929–1999, Int. J. Climatol., 22, 947–977, doi:10.1002/joc.777. Vautard, R., P. Yiou, F. D. Andrea, N. de Noblet, N. Viovy, C. Cassou, J. Polcher, P. Ciais, M. Kageyama, and Y. Fan (2007), Summertime European heat and drought waves induced by wintertime Mediterranean rainfall deficit, Geophys. Res. Lett., 34, L07711, doi:10.1029/ 2006GL028001. Walker, J., and P. R. Rowntree (1977), The effect of soil moisture on circulation and rainfall in a tropical model, Q. J. R. Meteorol, Soc., 103, 29–46, doi:10.1002/qj.49710343503. Wei, J., P. A. Dirmeyer, Z. Guo, L. Zhang, and V. Misra (2010), How much do different land models matter for climate simulation? Part I: Climatology and variability, J. Clim., 23, 3120–3134, doi:10.1175/2010JCLI3177.1. Wolfson, N., R. Atlas, and Y. Sud (1987), Numerical experiments related to the summer 1980 U.S. heat wave, Mon. Weather Rev., 115, 1345– 1357, doi:10.1175/1520-0493(1987)1152.0.CO;2. Xiu, A., and J. E. Pleim (2001), Development of a land surface model. Part I: Application in a mesoscale meteorological model, J. Appl. Meteorol., 40, 192–209, doi:10.1175/1520-0450(2001)0402.0.CO;2. Zeng, X.-M., Z. Wu, S. Xiong, S. Song, Y. Zheng, and H. Liu (2011), Sensitivity of simulated short-range high-temperature weather to land surface schemes by WRF, Sci. China Earth Sci., 54, 581–590, doi:10.1007/s11430-011-4181-6. Zeng, X.-M., B. Wang, Y. Zhang, S. Song, X. Huang, Y. Zheng, C. Chen, and G. Wang (2014), Sensitivity of high-temperature weather to initial soil moisture: A case study using the WRF model, Atmos. Chem. Phys., 14, 9623–9639, doi:10.5194/acp-14-9623-2014. Zhou, Y., and J. M. Shepherd (2010), Atlanta’s urban heat island under extreme heat conditions and potential mitigation strategies, Nat. Hazards, 52(3), 639–668, doi:10.1007/s11069-009-9406-z. Zhou, T., et al. (2009), Why the western Pacific subtropical high has extended westward since the late 1970s, J. Clim., 22, 2199–2215, doi: 10.1175/2008JCLI2527.1.

ZENG ET AL.


1325

simulated sensitivity to land surface schemes in ... - Wiley Online Library

simulated sensitivity to land surface schemes in ... - Wiley Online Library

Suggest Documents

A Multialgorithm Approach to Land Surface ... - Wiley Online Library

Noah land surface model modifications to ... - Wiley Online Library

Effects of Land Surface Schemes on WRF-Simulated Geopotential ...

coding schemes - Wiley Online Library

Diagnosing errors in a land surface model - Wiley Online Library

biennial oscillation simulated in WACCM to the ... - Wiley Online Library

Response to simulated climatic change in an ... - Wiley Online Library

Salt sensitivity in chickpea - Wiley Online Library

Land Commodification - Wiley Online Library

In vitro sensitivity to dasatinib in lymphoblasts ... - Wiley Online Library

Land cover, climate, and the summer surface ... - Wiley Online Library

Land Surface Temperature over Global Deserts ... - Wiley Online Library

On the applicability of current land surface ... - Wiley Online Library

Scaling land surface parameters for globalscale ... - Wiley Online Library

Use of remotely-sensed land surface ... - Wiley Online Library

Microwave land surface emissivities estimated ... - Wiley Online Library

ECOCLIMAP: a global database of land surface ... - Wiley Online Library

Hyperspectral retrieval of land surface ... - Wiley Online Library

Remote sensing land surface temperature for ... - Wiley Online Library

Impact of Shanghai urban land surface forcing ... - Wiley Online Library

Disaggregation of remotely sensed land surface ... - Wiley Online Library

Accelerated simulated annealing algorithm ... - Wiley Online Library

Geophysical Monitoring of Simulated ... - Wiley Online Library

Surface polysaccharide involvement in ... - Wiley Online Library