1. Introduction and background
Improving the representation of coupled processes in models is expected to enhance predictability on weather and climate scales (Dirmeyer et al. 2015). Previous studies have demonstrated that land surface characteristics, such as soil moisture and vegetation condition, can influence the intensity of heat waves (Hirsch et al. 2019; Miralles et al. 2014), formation of drought (Roundy et al. 2013; Wang et al. 2015), and precipitation patterns and intensity (Findell and Eltahir 1997; Koster et al. 2004; Seneviratne et al. 2010; Yang et al. 2018). At the same time, large-scale coordinated projects such as the Global Land Atmosphere Coupling Experiment (GLACE; Koster et al. 2004) have worked to quantify the influence of land initialization on forecast skill, finding significant contributions to temperature forecast skill at the subseasonal time scale (Koster et al. 2010).
Although emphasis is often placed on LA coupling at large time or space scales (e.g., global and seasonal predictability), the impact of the land surface on the water and energy cycle is modulated by its coupling to the PBL and begins at the local scale (Stull 1988; Sorbjan 1995; Peters-Lidard and Davis 2000; Cleugh et al. 2004; Santanello et al. 2005, 2007a, 2018). Atmospheric properties such as PBL evolution, atmospheric stability, and mixed-layer temperature and humidity are tightly linked to land surface variables such as soil moisture and surface fluxes on diurnal time scales (Stull 1988; Santanello et al. 2007a, 2013a). To better conceptualize these complex interactions, the local LA coupling (LoCo) project under the Global Energy and Water Exchanges (GEWEX) project describes a “process chain” that represents these interactions and feedbacks as “links” in a chain connecting soil moisture to PBL thermodynamics that ultimately influence clouds and precipitation (Santanello et al. 2018).
Thus, in order to improve numerical weather prediction (NWP), a greater understanding and integrated diagnosis of coupled model components and physics must be developed. Coupled and land-only (i.e., offline or uncoupled) models, whether operating at local, regional, or global scales suffer from uncertainty in surface hydrological estimates (e.g., soil moisture and evapotranspiration) and other essential water and energy cycle terms (Mueller et al. 2013). These limitations can largely be attributed to imperfect physics and associated parameters that are unable to capture the complexities of the governing processes and feedbacks of the LA system (Betts and Barr 1996; Entekhabi et al. 1999; Gu et al. 2006).
In particular, the physics of LSMs depend strongly on a large number of parameter values representing soil, vegetation, and other surface conditions. These parameters are typically assigned according to soil type, as defined by a soil texture map, and are critical in defining the movement of water within the soil. As a result, the parameters can impact the timing and duration of extreme events (e.g., droughts/floods) by controlling how easily water is drained and/or evaporated from the soil. Although this approach persists due to its ease of use, previous studies have shown that soil texture is a poor predictor of soil properties (Feddes et al. 1993; Gutmann and Small 2005; Santanello et al. 2007b; Harrison et al. 2012; De Lannoy et al. 2014; Dai et al. 2019). The classification of the soil type itself can often be erroneous, coarse, and/or discontinuous as global, high-resolution soil classification maps cannot be obtained from satellite like those for land cover or vegetation fraction. Approaches to address these model limitations fall into one of two groups, focused on either 1) the terrestrial leg of coupling [soil moisture to evaporative fraction (EF); Dirmeyer 2011], in which improvements are made to the land and expected to propagate to atmospheric components “downstream,” or 2) the atmospheric leg of coupling (surface flux to PBL; Dirmeyer et al. 2014), in which atmospheric properties are used to infer land states and fluxes “upstream.”
The most common method under the terrestrial approach is calibration of LSM parameters in land-only mode (i.e., offline) to observations of soil moisture or surface fluxes (Hess 2001; Hogue et al. 2005; Santanello et al. 2007b, 2013b; Peters-Lidard et al. 2008; Peng et al. 2020). For example, Shellito et al. (2016) calibrated soil parameters using both in situ and Soil Moisture Ocean Salinity (SMOS) soil moisture observations, showing comparable performance among the two datasets when SMOS had small biases. Other studies have found success in improving LSM biases when calibrating soil parameters to surface fluxes (Santanello et al. 2007b) or to a combination of observations (Gupta et al. 1999). Calibration has also been extended to semicoupled systems by assessing the impacts of LSM physics on uncoupled and single column models (Liu et al. 2003, 2004, 2005). Taken together, these studies imply that the use of a calibrated, land-only spinup has the potential to provide more accurate initial conditions consistent with the physics and parameters of the chosen model. However, even if the terrestrial leg of coupling is improved, there is no guarantee that the improvement will extend to the atmospheric leg (EF–PBL) in coupled mode. In fact, model deficiencies can result in new observations improving certain variables, but degrading others. For example, assimilation of brightness temperature from satellite can improve soil moisture but degrade fluxes (Crow et al. 2020; Munoz-Sabater et al. 2019).
The second approach includes efforts that constrain model error using atmospheric data, such as PBL profiles or screen-level temperature and humidity to infer surface fluxes. In particular, the operational weather community often uses near-surface temperature and humidity observations to improve fluxes by adjusting, or “nudging,” soil moisture (Drusch et al. 2009; de Rosnay et al. 2013; Hu et al. 1999). Other studies use PBL profiles or ambient temperature and humidity to estimate surface fluxes (Salvucci and Gentine 2013; Gentine et al. 2016; Denissen et al. 2021). Although these approaches can inform on the atmospheric leg of coupling (EF–PBL), they overlook the accuracy of the connections within the land (i.e., the terrestrial leg). Thus, it is critical to understand how an observed component of the process chain influences other components and the connections between them, in both the terrestrial and atmospheric legs of coupling. This approach inherently requires a coupled system and is particularly important when it comes to leveraging new observations (e.g., new satellite platforms) for improving NWP. The integration of new observations into models via calibration and assimilation, especially in LSMs, is often done once the observations become available (i.e., does not necessarily occur with the full range of other dependent variables).
In this study, a LA interaction and feedback investigation is conducted that allows for a quantitative understanding of the sensitivities of each link of the LoCo process chain using a coupled calibration approach in a single column modeling (SCM) framework. The use of the SCM enables an interactive PBL and land surface (and hence full process chain representation) without the full complexity and computational expense of a three-dimensional model. In this manner, the links in the chain that are upstream and/or downstream of the observations used in the calibration can be assessed. Soil hydraulic parameters (SHPs) and/or initial soil moisture (ISM) are calibrated to observations of surface fluxes, 2-m temperature (T2), 2-m humidity (Q2), and PBL height (PBLH). We focus on calibrating SHPs and ISM (i.e., LSM states and fluxes) due to their inherent importance in LA interactions as well as uncertainty in current LSMs.
The main questions this work seeks to answer are: Can observations of land and PBL variables be used to improve representation of the LA process chain, and what are the upstream and downstream consequences? As such, the goals of this work differ from traditional calibration studies in that we do not seek to produce the most accurate model using all available observations. Rather, we use calibration as a tool to understand how physical connections are made in the model, including how constraining one component of the process chain impacts the other components.
The paper is organized as follows: section 2 describes the LoCo process chain framework, as well as the observations, models, and the experimental design employed. Section 3 presents the results of the case study simulations. Section 4 offers a discussion of the context of the key results and limitations, and section 5 presents conclusions.
2. Methods
a. LA interactions and local LA coupling (LoCo)
(a) Schematic diagram of the Local Land Atmosphere Coupling (LoCo) process chain describing the interactions and feedbacks that link soil moisture variability to fluxes, PBL growth and entrainment, ambient weather, and ultimately clouds and precipitation. Adapted from Santanello et al. (2018, their Fig. 2). (b) Illustration of the suite of calibration experiments performed. The land parameters are calibrated [i.e., soil moisture (SM), soil parameters (SP) and both SM and SP] to atmospheric observations of Bowen ratio, 2-m temperature (T2), 2-m humidity (Q2), and PBL height (PBLH). The line style and color convention assigned to each experiment here is used throughout the analysis (e.g., the simulation where SM is calibrated to Bowen ratio is shown as a red dashed line in Figs. 3, 4, 7, and 8).
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
(a) Schematic diagram of the Local Land Atmosphere Coupling (LoCo) process chain describing the interactions and feedbacks that link soil moisture variability to fluxes, PBL growth and entrainment, ambient weather, and ultimately clouds and precipitation. Adapted from Santanello et al. (2018, their Fig. 2). (b) Illustration of the suite of calibration experiments performed. The land parameters are calibrated [i.e., soil moisture (SM), soil parameters (SP) and both SM and SP] to atmospheric observations of Bowen ratio, 2-m temperature (T2), 2-m humidity (Q2), and PBL height (PBLH). The line style and color convention assigned to each experiment here is used throughout the analysis (e.g., the simulation where SM is calibrated to Bowen ratio is shown as a red dashed line in Figs. 3, 4, 7, and 8).
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
(a) Schematic diagram of the Local Land Atmosphere Coupling (LoCo) process chain describing the interactions and feedbacks that link soil moisture variability to fluxes, PBL growth and entrainment, ambient weather, and ultimately clouds and precipitation. Adapted from Santanello et al. (2018, their Fig. 2). (b) Illustration of the suite of calibration experiments performed. The land parameters are calibrated [i.e., soil moisture (SM), soil parameters (SP) and both SM and SP] to atmospheric observations of Bowen ratio, 2-m temperature (T2), 2-m humidity (Q2), and PBL height (PBLH). The line style and color convention assigned to each experiment here is used throughout the analysis (e.g., the simulation where SM is calibrated to Bowen ratio is shown as a red dashed line in Figs. 3, 4, 7, and 8).
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
b. Observations
The Department of Energy’s Atmospheric Radiation Measurement (ARM) program in the Southern Great Plains (SGP) provides a reliable, long-term record of surface and atmospheric observations. The ARM-SGP network includes the Central Facility (CF) in Lamont, Oklahoma, United States as well as more than 30 extended stations across Oklahoma and Kansas. Due to this abundance of observational data, the CF location has long been the focus of LA coupling research and therefore is an ideal location for the SCM simulations in this study (Fig. 2). The following sets of observations at the CF were obtained from the ARM Data Discovery portal (https://adc.arm.gov/discovery/#/): 1) soil moisture from the Soil Water and Temperature System (SWATS), 2) sensible and latent heat fluxes from the Bulk Aerodynamics Energy Balance Bowen Ratio (BAEBBR) and Eddy Covariance (ECOR) stations, and 3) standard meteorological variables (i.e., 2-m temperature and relative humidity, surface pressure, etc.) from the ARM Best Estimate (ARMBE) data stream.
Google Earth images showing the state of Oklahoma and (inset) the location of the ARM Southern Great Plains Central Facility in Lamont. The yellow box indicates the domain setup for the single column model (SCM) experiments.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Google Earth images showing the state of Oklahoma and (inset) the location of the ARM Southern Great Plains Central Facility in Lamont. The yellow box indicates the domain setup for the single column model (SCM) experiments.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Google Earth images showing the state of Oklahoma and (inset) the location of the ARM Southern Great Plains Central Facility in Lamont. The yellow box indicates the domain setup for the single column model (SCM) experiments.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
The ARM-SGP regularly hosts field campaigns that supplement the suite of routine observations. One such effort, the Enhanced Soundings for Local Coupling Studies (ESLCS) campaign took place in the summer of 2015 (Ferguson et al. 2016). The ARM-SGP routinely makes radiosonde measurements four times a day at the CF, and on 12 select intensive observation period (IOP) days during the ESLCS additional radiosondes were launched hourly during the daytime. Two of these IOP days, 11 July 2015 and 16 August 2015 are chosen for the case study simulations in this work. These days feature quiescent synoptic conditions, differing soil moisture states (~0.1 m3 m−3), and show a contrast in model skill, discussed further in section 4. Hourly PBLH is derived from the radiosonde profiles on these days by visual inspection of the height of the inversion above the mixed layer in the potential temperature profiles. The derived PBLH falls in line with derived estimates from the ESLCS campaign made using a suite of PBLH estimation methods (Ferguson et al. 2016). The BAEBBR and ECOR fluxes were compared and the dataset with the fewer data gaps was used for the analysis (i.e., ECOR on 11 July and BAEBBR on 16 August). Therefore, the full suite of observations used in this study include soil moisture, temperature, humidity, latent and sensible heat fluxes, and PBL height.
c. Models
1) Single column modeling framework
NASA’s Land Information System (LIS; Kumar et al. 2006) is a flexible land surface modeling and data assimilation system that allows for high-resolution simulation of land surface states and fluxes. LIS also includes an optimization and uncertainty subsystem (i.e., LIS-OPT; Kumar et al. 2012; Harrison et al. 2012) that allows for model calibration to satellite and other observation datasets using common algorithms such as the Levenberg–Marquardt (LM; Levenberg 1944; Marquardt 1963) and the genetic algorithm (GA; Holland 1975). LIS has been coupled to the Weather Research and Forecasting (WRF; Skamarock et al. 2005) Model under the NASA Unified WRF framework (NU-WRF; Peters-Lidard et al. 2015) and has been used effectively for numerous studies in land–atmosphere interactions (Santanello et al. 2011a, 2019; Lawston et al. 2020, 2015). Recently, the WRF single column model (SCM) capabilities were coupled to LIS under the NU-WRF framework for the first time, enabling interactive PBL and atmosphere without the full complexity and computational expense of a three-dimensional model. This configuration (hereafter LIS-SCM) is used here to better understand the modeled sensitivities of each component of the LoCo process chain.
2) OSTRICH calibration software
The primary model calibration software used in this study is the Optimization Software Toolkit for Research Involving Computational Heuristics (OSTRICH; Matott 2017). OSTRICH is an open source, model independent software tool for automatic, multivariate calibration using a choice of several common optimization algorithms. OSTRICH has been employed most often for the calibration or optimization of hydrologic models, with success in such applications as improving simulations of streamflow (Haghnegahdar et al. 2014), urban runoff and drainage (Behrouz et al. 2020), or to optimize rain barrel siting (Macro et al. 2019). To our knowledge, this is the first study to apply the OSTRICH calibration software to the WRF SCM. OSTRICH performs calibration by automatically running many model iterations with varying input parameters until the algorithm converges on a solution that minimizes the objective function. The coupled model setup used for calibration in this study allows for each component of the process chain to be constrained to atmospheric observations, potentially revealing strengths or deficiencies in the model’s ability to simulate each component.
3) Experimental design
LIS-SCM is run for two case studies, 11 July 2015 and 16 August 2015, over the ARM SGP Central Facility location (ARM CF; Fig. 2). In each case, the model is first run with default LSM parameters for 24 h beginning at 1200 UTC, hereafter referred to as the DEFAULT simulation. The land surface initial conditions are generated from a 5-yr land-only LIS simulation using the Noah LSM (Ek et al. 2003) version 3.6 forced by the Global Data Assimilation System (GDAS) data from the National Centers for Environmental Prediction (NCEP). The soil type and land cover for the ARM CF latitude and longitude in the model is silt loam and cropland, as given by the STATSGO soil texture and MODIS IGBP land cover datasets, respectively.
In all simulations, the North American Regional Reanalysis (NARR; Mesinger et al. 2006) analysis fields are used as atmospheric initial and boundary conditions at 3-hourly intervals. Other relevant parameterization and physics options include the use of the Mellor–Yamada–Nakaniski–Niino 2.5 (MYNN2.5; Nakanishi and Niino 2006) PBL scheme and Goddard shortwave and longwave radiation schemes (Chou and Suarez 1999). There is no external advection or subsidence. The initial profile is extracted from the NARR reanalysis data. The simulations use periodic lateral boundary conditions. The microphysics scheme is turned off, as this study is focused on the core LoCo process chain components and does not address cloud or precipitation impacts, which is an area for future extension of this work.
4) Calibration approach
Next, using OSTRICH as a wrapper around the LIS-SCM (hereafter LIS-SCM-CAL), the ISM, SHPs, or both ISM and SHPs together, are calibrated to observations of Bowen ratio, T2, Q2, and PBLH (Fig. 1b). The LIS-SCM-CAL simulations run from 1200 to 2300 UTC when calibrating to Bowen ratio and PBLH and from 1200 to 0400 UTC when using T2 and Q2, due to differing data availability. The calibration is completed within OSTRICH using the Genetic Algorithm (GA) with a population size of 30 and a mutation rate of 0.005 (Kumar et al. 2012). The algorithm achieves convergence when the convergence value, defined as the relative difference between the current minimum and the median of the last generation, is less than or equal to 0.005. The objective function that OSTRICH seeks to minimize is the weighted sum of squared error (WSSE) between the model and observations. As the calibration period is short and only uses high-quality daytime observations, a weight of 1 is used for all observation pairs. Further details and diagrams describing the LIS-SCM-CAL implementation are available in the supplement (Figs. S1–S2 in the online supplemental material). Once the calibrated parameters are obtained from the LIS-SCM-CAL runs, a 24-h simulation with the calibrated parameters is completed for each case study date. All 10 SHPs used by the Noah land surface model are calibrated. The full list of these parameters can be found in the supplement (Table S1). The soil moisture is calibrated only at the initial time and then allowed to evolve with the simulation.
To provide a remote sensing based reference, an additional experiment is conducted using the default LIS-OPT capabilities to calibrate Noah soil and vegetation parameters to the Soil Moisture Active Passive (SMAP; Entekhabi et al. 2014) Level 3 Enhanced soil moisture retrievals. This marks the first time to our knowledge that SMAP has been used for calibration of Noah LSM SHPs. The GA method is used with a population size of 50 and single point cross over elitism enabled, meaning that the best solution from a generation is carried over to the next. The calibration stops when the average fitness of the population does not improve beyond 5% (Kumar et al. 2012). The calibration required 76 generations to converge.
This experiment represents the more traditional approach to model calibration—that is, using a long-term offline (uncoupled LSM) calibration of model parameters to a single land surface variable (i.e., soil moisture). The SMAP satellite employs a passive microwave radiometer to measure soil moisture in the top 0–5 cm of soil at about 40-km resolution. The Enhanced product exploits the oversampling of the SMAP orbit to resample the retrievals and is gridded at 9-km resolution. The calibration time period was March 2015–December 2019, which represented the entire SMAP data record at the time of calibration. The accuracy of the SMAP soil moisture meets the mission requirements of 0.04 cm3 cm−3 where the volumetric water content (VWC) is less than or equal to 5 kg m−2. SMAP data contain a retrieval quality flag, which is an assessment of the robustness of the retrieval. Data points with retrieval quality flag of 0 or 8, considered high quality, are used.
3. Results
a. 11 July 2015 case study
The 11 July case study captures the beginning of a soil moisture drydown sequence (Ferguson et al. 2016). Figure 3 shows the diurnal cycle of top layer (0–10 cm) SM, LH, SH, T2, Q2, and PBLH for the default, each calibrated run, and observations. Note that the line styles (i.e., solid, dotted, dashed) correspond to the parameter, or sets of parameters, being calibrated (i.e., SHPs, ISM, both SHPs and ISM, respectively). The line colors correspond to the observation to which each parameter set has been calibrated, where red, blue, green, and orange correspond to Bowen ratio, PBLH, T2, and Q2, respectively. This convention is illustrated in a schematic in Fig. 1b and is used throughout the figures. The observations (open circles) confirm that soil moisture was quite moist (0.345 m3 m−3) and flux observations show a brief drop in radiation at 1800 UTC, likely the result of local passing clouds. The temperature peaks at around 305 K (89°F; 32°C) and maximum PBL height is limited to about 1400 m given the relative wet surface and LH dominated regime.
Time series comparing the results of the default (black) and calibrated (colored) model simulations as compared to observations (open circle markers) for the 11 Jul 2015 case study. The following variables are shown: (a) top layer SM, (b) LH, (c) SH, (d) T2, (e) Q2, and (f) PBLH.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Time series comparing the results of the default (black) and calibrated (colored) model simulations as compared to observations (open circle markers) for the 11 Jul 2015 case study. The following variables are shown: (a) top layer SM, (b) LH, (c) SH, (d) T2, (e) Q2, and (f) PBLH.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Time series comparing the results of the default (black) and calibrated (colored) model simulations as compared to observations (open circle markers) for the 11 Jul 2015 case study. The following variables are shown: (a) top layer SM, (b) LH, (c) SH, (d) T2, (e) Q2, and (f) PBLH.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
The LIS-SCM-CAL results in Fig. 3 produce a wide range of impacts to SM, surface fluxes, ambient weather, and PBLH. For example, the calibrated runs produce ISM values ranging from ~0.2 to greater than 0.4 m3 m−3, maximum LH from ~300 to 600 W m−2 and maximum PBLH from ~1300 to ~2800 m. The default ISM (generated by the LIS-Noah spinup) is drier than observations and lies near the middle in the spread of the runs, meaning that the calibration in some case pushes the model closer to SM observations, but in other cases exacerbates the dry SM bias depending on which process chain variable is used as the observation.
Table 1 shows the root-mean-square error (RMSE) of each simulation (in Fig. 3) and Fig. 4 shows the normalized RMSE, color coded to identify the best and worst performing runs for each observed variable (i.e., blue best, red worst using four groupings of 0.25). Table 1 and Fig. 4 together suggest that the calibration, by attempting to improve the simulation error with respect to one part of the process chain highlights deficiencies in other parts of the process chain. The most obvious example can be seen in the runs that were calibrated to PBLH (i.e., blue in Fig. 3) and stand out as the primary outliers for this case in terms of impacts on other process chain variables. Because the default PBLH was much higher than observed, in order to minimize the error in PBLH, the calibration either maximizes ISM and/or alters the SHPs so that LH is dramatically increased. The result is reduced SH, a moister and cooler near surface atmosphere, and reduced boundary layer growth more in line with, but still exceeding observations of PBLH. However, the resulting errors in LH, T2, and Q2 are the worst of all simulations. In particular, the fact that improving the simulated PBLH produces significantly degraded fluxes indicates that EF–PBLH component of the process chain is not represented well in this model for these conditions. This result is complementary to those of Liu et al. (2005), which showed an apparent “trade-off” in the ability of a different SCM to accurately simulate both sensible heat flux and downstream variables, in their case precipitation.
Root-mean-square error of each model simulation for the 11 Jul case study. Bold numbers indicate the run with the smallest error for each variable.
Normalized root-mean-square error for each simulation in Fig. 3 with respect to each process chain variable and the total error using four groupings of 0.25. Blue represents lower error (i.e., better performance).
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Normalized root-mean-square error for each simulation in Fig. 3 with respect to each process chain variable and the total error using four groupings of 0.25. Blue represents lower error (i.e., better performance).
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Normalized root-mean-square error for each simulation in Fig. 3 with respect to each process chain variable and the total error using four groupings of 0.25. Blue represents lower error (i.e., better performance).
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
In contrast to the PBLH runs, the runs that calibrate to Q2 (i.e., orange lines) seek to decrease Q2 by reducing SM, which in turn, increases SH, T2, and PBLH. Although the Q2 error is improved, the total error when considering all variables is degraded as compared to the default, resulting in some of the worst performing runs for this case study. This highlights that calibrating to Q2 (or T2) as a method of overall model improvement via ISM or SHPs can prove difficult, as Q2 is far downstream and therefore can be influenced by fluxes, PBLH, and entrainment. It should also be noted that SM affects fluxes locally and more directly, while PBL characteristics (T2, Q2, and PBLH) are integrative of a larger SM footprint through both local and advective effects. As a result, the fluxes are more sensitive to the calibration of soil parameters than the other observations.
Another example of a process chain deficiency exposed by the calibration experiments can be seen in the relationship between SM and surface fluxes (particularly LH) in observations and the model. For example, some of the runs that perform very poorly in terms of SM error (i.e., SPSM_Q2, SP_Bowen) show some of the lowest errors in fluxes and vice versa (e.g., SPSM_PBLH). If the terrestrial leg (i.e., SM–EF) were realistic, then improvement to SM should result in improvement in fluxes. This is likely a function of the optimized parameters in the SHP simulations, which are calibrated to a particular process chain variable but ultimately are absorbing inherent model error and biases (i.e., effective parameters) elsewhere. Thus, the set of parameters ideal for capturing the diurnal cycle of Q2 are not reflective of those that govern the SM–LH relationship. This again supports that the model struggles to accurately represent the process chain, as proper SM should lead to proper fluxes, which then drive PBL growth, entrainment, and ambient humidity (Q2).
Figure 5a shows the available energy (i.e., LH + SH) of each model run and observations. The default run shows more available energy than observations at each hour, with the hourly positive bias ranging from >100 W m−2 in the morning to 43 W m−2 late in the day and as much as 290 W m−2 at 1800 UTC. This excess incoming radiation in the model affects the LA coupling in ways that are difficult for the calibration to mitigate. For example, the runs calibrating to Bowen ratio (i.e., red lines) are most directly challenged to address the available energy issue. As a result, these runs modify the ISM and SHPs in such a way as to bury additional energy in the soil (i.e., increase ground heat flux in Fig. 5b). However, this is not enough to alleviate the strong positive bias in available energy. This result echoes that of Hu et al. (1999), who found that soil moisture nudging in the ECMWF model produced good estimates of sensible heat flux, but poor estimates of latent heat flux when the surface radiation was poor.
As in Fig. 3, but for (a) total available energy (latent + sensible heat flux) and (b) ground heat flux.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 3, but for (a) total available energy (latent + sensible heat flux) and (b) ground heat flux.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 3, but for (a) total available energy (latent + sensible heat flux) and (b) ground heat flux.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Some of the behavior seen across the process chain in Fig. 3 and Table 1 can be attributed to this available energy bias in the model. For example, the fact that the PBLH runs still exhibit greater PBLH growth than the observations, even despite modifying soil moisture and fluxes to an extreme degree, is the result of the positive energy bias. Overall, none of the model parameters or calibration pairs on their own can fully compensate for the available energy bias, resulting in much deeper PBL growth across all runs as compared to observations. This bias is likely also a factor in the poor representation of the SM–EF sensitivity for this case, as the model struggles to produce the observed flux magnitudes regardless of the SM value.
Figures 6 and 7 present the principal LoCo diagnostics of mixing diagrams (MDs) and EF versus PBLH plots (Santanello et al. 2009, 2011a, 2018). MDs allow for the quantification of heat and moisture budgets in the PBL and quantify the integrated LA response to different calibration results. The solid lines on the MDs show the hourly values (1300–2300 UTC) of T2 (y axis) and Q2 (x axis) multiplied by the specific heat of water and the latent heat of vaporization, respectively, to represent the diurnal evolution of temperature and humidity in energy space. The dashed lines are vectors, where the magnitudes are proportional to the fluxes of heat and moisture from the surface and atmosphere and the slope is equal to the Bowen ratio of the surface (βsfc) and entrainment (βent) fluxes.
Mixing diagrams for the 11 Jul case for each parameter set calibrated: (a) SHPs, (b) ISM, (c) ISM and SHPs.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Mixing diagrams for the 11 Jul case for each parameter set calibrated: (a) SHPs, (b) ISM, (c) ISM and SHPs.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Mixing diagrams for the 11 Jul case for each parameter set calibrated: (a) SHPs, (b) ISM, (c) ISM and SHPs.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Plot of the EF vs maximum PBLH for the 11 Jul 11 case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Plot of the EF vs maximum PBLH for the 11 Jul 11 case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Plot of the EF vs maximum PBLH for the 11 Jul 11 case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
The overall shape of the MDs for all runs except the PBLH runs indicates warming and steady to very gradual drying of the PBL throughout the day that accelerates due to entrainment feedbacks toward the end of the day. In contrast, the PBLH runs (i.e., blue lines) show a moistening of the PBL as a result of the extreme LH values produced by the calibration and extremely high ISM values and SHPs that prioritize evaporation in an effort to reduce PBLH. The observations start out drier and slightly cooler than the model, drying more slowly and overall exhibiting less entrainment than the non-PBLH runs, though the diurnal cycle range in both is quite similar to the majority of the simulations. Although most of the calibrated runs are able to tap into the PBL entrainment feedbacks, the outlier PBLH runs are not due to their slow and stunted PBL growth. It should also be noted that although the surface Bowen ratio in the non-PBLH runs ranges from 0.6 to 0.91, the resultant T2 and Q2 diurnal cycles do not vary much at all, supporting that this case was not strongly coupled in terms of surface fluxes to ambient weather.
The EF versus PBLH plot shows (Fig. 7) the clustering of simulations by color, i.e., the observation used for calibration. All of the model runs overestimate PBLH, despite mean EF in the Bowen and T2 calibrated runs being quite close to observations (~0.6). To approach the maximum PBLH in observations, the PBLH calibrated runs (i.e., blue markers) dramatically increased LH, degrading the performance of these runs in terms of EF as compared to observations. This plot again highlights the available energy problem in that even when the proportion of latent to total flux is appropriate in the calibrated simulations (Bowen and T2 runs), the excess available SH drives much larger PBL growth than observed.
b. 16 August 2015 case study
The 16 August 2015 IOP date exhibited the largest diurnal temperature range of the ESLCS field campaign (Ferguson et al. 2016). Figure 8 shows the top layer SM, LH, SH, T2, Q2, and PBLH for the default, each calibrated run, and observations. The observed SM is quite a bit drier than in the July case study (~0.25 m3 m−3), but is rather moist for the location and time of the year, as the strong August drydown common to the region did not occur in 2015 (Ferguson et al. 2016). As compared to the July case, the observations show that the day was warmer and drier, resulting in deeper and more rapid PBL growth, with the maximum PBLH reaching more than 2000 m. Once again, the calibration results in a wide range of impacts to each variable in the process chain. Although the default run partitions too much energy into SH at the expense of LH, overall Q2 and PBLH are much closer to observations than in the July case.
As in Fig. 3, but for the 16 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 3, but for the 16 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 3, but for the 16 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Figure 9 shows the total available energy and ground heat flux and Fig. 10 shows the normalized RMSE of each run for the 16 August case study. The absolute values of RMSE are found in Table 2. Figure 9 indicates that one reason for the improved skill of the August default run compared to the previous case is due in part to the available energy in the model closely matching observations. In other words, the August case represents a well-constrained system in contrast to the 11 July case. As a result, the terrestrial leg is much stronger than in the previous case, as shown by the low SM error runs (i.e., SPSM_Bowen, SM_Bowen, SM_PBLH) generally exhibiting some of the best error in fluxes and vice versa (i.e., SM_T2, SPSM_Q2, SPSM_T2). In addition, the SM_PBLH run was one of the worst performing runs overall for the 11 July case, but exhibits the lowest overall error for the 16 August case, implying that there is promise for improving model error using new PBLH observations, provided that the coupled system is well constrained overall.
As in Fig. 5, but for the 16 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 5, but for the 16 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 5, but for the 16 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 4, but for the 16 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 4, but for the 16 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 4, but for the 16 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Several of the runs, each in their own unique ways, point to a deficiency in the model’s ability to simulate the atmospheric leg of coupling (EF–PBLH in the process chain). For example, calibrating to Bowen ratio (i.e., red lines, Fig. 8) improves LH and SH (by design) by lowering SH and increasing LH, but worsens PBLH in the process by reducing it considerably. The SM_PBLH run (i.e., blue dashed line) shows the lowest SM error and the lowest PBLH error, but retains considerable error in the fluxes. Thus, compared to the previous case, the EF to SM relationship is closer to observed, but once again the sensitivity of PBLH to fluxes is not represented well.
Figure 8 also shows that a noteworthy outlier in its diurnal behavior is the SPSM_PBLH run (i.e., blue dotted line). The observed PBLH grows more slowly than the default model run early in the day (1300–1500 UTC), then rapidly increases and exceeds the model around 1400–1700 UTC. By having both ISM and SHP modifications at its disposal, the SPSM_PBLH run increases initial soil moisture from ~0.18 to ~0.3 m3 m−3. At the same time, the SHPs are modified so that evaporation is increased early in the day, resulting in a rapid dry down of the top layer soil moisture that then drives down LH and increases SH later in the day. These flux impacts should slow morning PBL growth and drive an increase in the PBLH later in the day, more in line with observations. However, although this run improves the PBLH error over default, the impacts are very small and the diurnal cycle of PBLH remains relatively unchanged. This provides further support to the idea that the model struggles to accurately represent the relationship between fluxes and PBLH in this case, as the PBLH cycle is not very sensitive to the dramatic flux changes given in the SPSM_PBLH run.
The variation across runs shown in Fig. 7 is a bit less than in the 11 July case (Fig. 3), with the spread in SM, fluxes, and PBLH around 0.23 m3 m−3, 200 W m−2, and <1000 m, respectively. In terms of overall error across the process chain, the default simulation performs so well that it is difficult for calibration to any one variable to improve the total skill of the simulation. Only one run (i.e., SM_PBLH) is able to perform better than default, driven mostly by a reduction in SM, fluxes, and PBLH errors.
Figures 11 and 12 show the MDs and EF versus PBLH plots, respectively. The MDs show that the observations again start out slightly cooler and drier than the model simulations. As discussed above, the Bowen ratio calibrated runs (i.e., red lines) are the outlier for this case, but in contrast to the previous case, still exhibit a similar shape to the other model simulations. The observations show a rapid warming and steady drying of the PBL and stronger entrainment than in the previous case due to greater PBL growth. The model simulations exhibit a slower warming and similarly steady drying of the PBLH. Interestingly, although the observed and modeled PBLs start at a different combination of temperature and humidity and evolve a bit differently (i.e., the shape of the curves), they end up at a very similar temperature and humidity by the end of the day (i.e., end of solid lines around x = 24 000 J kg−1, y = 306 000 J kg−1). In addition, the mixing diagrams depict a drying signature in the models (i.e., path of the solid lines in the x direction) that differs more from observations than would be suggested by the good agreement between models and observations given in Fig. 8e. The two black dots in Fig. 11a indicate the heat and moisture state of the PBLH at 1700 UTC (noon local time) in the observations (on the gray line) and default (on the black line grouped with the other simulations). These dots highlight the fact that the temperature and humidity has changed more dramatically between 1300 and 1700 UTC in observations than in the model (i.e., the length of the solid gray line from start to black dot is much longer than the model simulated start to black dot). This is due to the rapid PBL growth in the observations that was not captured by the model (Fig. 8f).
As in Fig. 6, but for the 16 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 6, but for the 16 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 6, but for the 16 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 7, but for the 17 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 7, but for the 17 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
As in Fig. 7, but for the 17 Aug case study.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
The EF versus PBLH plots are shown in Fig. 12 and add support to the assertion that the model struggles to represent the EF–PBLH component of the process chain. The maximum PBLH in the default simulation is close to observations, but the mean EF is almost half that observed (i.e., ~0.6 in observations, 0.35 in default). There are four calibrated runs (i.e., the three PBL runs and SP_T2) that are also close to the maximum PBLH observed, but they span EF values ranging from 0.25 to 0.4, all of which are significantly lower than the observed EF of 0.6. The Bowen simulations (i.e., red markers) produce EF values closest to the observations, but the resulting maximum PBLH that is ~400 m lower than observed.
c. Effects of a long-term, calibrated spinup
The parameters produced by the calibration of Noah3.6 soil and vegetation parameters to SMAP soil moisture using LIS-OPT are shown in the supplement (Table S1). These parameters are used to complete a long-term land-only calibrated spinup that is then used to initialize calibrated simulations for the July and August case studies (hereafter referred to as SMAP_SPIN). Figure 13 shows the default and SMAP calibrated top layer SM and fluxes for the 11 July and 16 August simulations. In both cases, the SMAP calibrated parameters result in ISM that is 0.02–0.03 m3 m−3 drier than the default. This is consistent with SMAP observations, particularly during dry periods, being drier than the Noah LSM climatology.
Results of the run initialized with a SMAP calibrated spinup (SMAP_SPIN) as compared to default (black) and observations for the (left) 11 Jul and (right) 16 Aug cases. The variables analyzed are (a),(d) top layer soil moisture, (b),(e) latent heat flux, and (c),(f) sensible heat flux.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Results of the run initialized with a SMAP calibrated spinup (SMAP_SPIN) as compared to default (black) and observations for the (left) 11 Jul and (right) 16 Aug cases. The variables analyzed are (a),(d) top layer soil moisture, (b),(e) latent heat flux, and (c),(f) sensible heat flux.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Results of the run initialized with a SMAP calibrated spinup (SMAP_SPIN) as compared to default (black) and observations for the (left) 11 Jul and (right) 16 Aug cases. The variables analyzed are (a),(d) top layer soil moisture, (b),(e) latent heat flux, and (c),(f) sensible heat flux.
Citation: Journal of Hydrometeorology 22, 9; 10.1175/JHM-D-20-0263.1
Because observed SM is much wetter than the default simulation, the SMAP calibration leads to an increase in SM error relative to in situ soil moisture measurements. Despite the drier ISM, the calibrated LH is greater and SH is less than the default simulation in both cases, as a result of calibrated SHPs that have modified the SM saturation, critical threshold, and wilting point values in the Noah LSM. These new parameter values support larger LH throughout the range of middry SM that is produced. This results in reduced T2, increased Q2, and a lower PBLH in both cases (Fig. S3), which for some variables represents an improvement in error and for others a degradation as compared to the default simulation. Overall, the normalized error across all variables shows that the SMAP calibrated spinup run performs in the top third of runs for the 16 August case, but performs in the bottom of half runs for the 11 July case.
The better relative performance for the August case is likely due to the SM being drier in August. SMAP is typically wetter than the Noah LSM during and immediately following precipitation events, but it dries out more quickly than the model during drydowns (Shellito et al. 2020; Santanello et al. 2019). This is due in part to the mismatch in the nominal sensing depth of the SMAP radiometer (0–5 cm) and the Noah definition of top layer depth (0–10 cm). As there are more drydown events than precipitation events over the course of the calibration for the study area, on the whole SMAP is generally drier. As a result, the calibration generates parameters that promote maximum evaporation in order to dry out the soil to better match SMAP. These parameters when applied to a wetter case (July) are mixed, but perform better when applied to a drier case.
4. Discussion
This study employs a coupled calibration approach to constrain individual components of the LoCo process chain and to evaluate the impacts on the remaining links and overall LA coupling. The additional SMAP experiment serves as a reference point for what is more traditionally done in the LSM calibration community—that is, calibrating surface parameters to another single surface variable in a long-term calibration. This standard approach was included to better understand how improving the land surface (with respect to SMAP) affects the rest of the process chain. For example, the parameters given by SMAP calibration may be appropriate for soil moisture, but not necessarily for the rest of the process chain. This is true for the July case where LH was already overestimated by the default simulation and the SMAP calibrated parameters exacerbate that bias by increasing evaporation. This highlights the need for coupled calibration to better constrain the atmospheric response to land surface changes (and ultimately, vice versa as well).
The goal of calibrating model parameters to SMAP, or other soil moisture observations, is that accurate soil moisture will yield more accurate fluxes, and so on down the process chain, leading to improvement in LA coupling, clouds, precipitation, and overall NWP. However, this requires a consistency between models and observations, referred to as “observability,” that remains a challenge when reconciling modeled, in situ, and satellite soil moisture (Shellito et al. 2020; Koster et al. 2009). For example, the SMAP calibration produced parameters that dried the modeled soil moisture to levels more in line with SMAP observations. However, when evaluating the SMAP calibrated run against in situ observations, the SMAP calibrated run increased the SM error because the in situ observations were wetter than SMAP. This also presents challenges in evaluating the terrestrial leg of the process chain. Nonetheless, the relative strength of the SM–EF relationship in August as compared to July is still apparent and supports tighter coupling in that case.
The results from both case studies show a model deficiency in accurately representing the sensitivity of PBLH to EF at this location and for these dates, and the inability of the calibration to improve it. One could envision a theoretical, continuous curve of the EF versus PBLH relationship that is governed by the model. In the default simulation, it will have a certain shape reflecting the default SHPs and ISM. In the calibrated runs, the curve will shift in slope to reflect the new parameters. Ideally, this would move toward a “true” EF versus PBLH relationship as observed in nature. In practice, however, the new curves will reflect inherent model biases (e.g., too much available energy) and/or insufficiencies in the physics such that it will reflect an “effective” relationship. The goal of this study is to first examine whether observations of different links of the process chain can be used to better constrain relationships such as EF versus PBLH, and if not, identify model errors and biases in the process.
In both case studies, the dominant factor in the spread of atmospheric impacts across calibration runs is a result of the observation (i.e., Bowen, T2, etc.) that is used for calibration. This is evident in Figs. 3 and 8, where most of the time series exhibit stronger grouping by color (i.e., observation) than by line style. The parameters being calibrated (ISM versus SHPs) serve as a secondary delineator among the runs. The exception to this behavior is SM, which shows more grouping based on the parameter calibrated because the parameters most directly impact SM itself. Taken together, the grouping among the soil and atmospheric variables indicates that there are multiple ways in which the parameters can be modified to better match the observations. Whether any of these parameter solution sets represents a realistic (i.e., physically consistent) soil is not explored in this study. In addition, these two cases sample a certain range within the SM–EF spectrum that skews toward the wetter end, implying both case studies are likely in atmosphere-limited regimes. The sensitivities may vary for cases that lie in other parts of the SM–EF spectrum. As such, future work is needed in different locations and moisture regimes, but the number of sites with the required collocated measurements, particularly hourly PBL observations is limited. New initiatives that prioritize the collection of coupled land and atmosphere observations, such as the GEWEX Land–Atmosphere Feedback Observatory (GLAFO; Wulfmeyer et al. 2020) should be fully leveraged to explore other potential coupling and moisture regimes.
Another factor affecting the results in this calibration approach is that the model simulates (i.e., imposes) a diurnal cycle in the atmospheric variables, driven by the model physics (e.g., radiation, PBL schemes, etc.) that itself shows low sensitivity to calibration. That is, the calibration is unable to significantly modify the shape of the imposed diurnal cycle, instead primarily adjusting the magnitude of the cycle with an occasional slight nudge in the phase. This is apparent in all of the atmospheric variables shown across the two cases, but is most evident in T2 on 16 August. Although the default run simulates well the available energy, Q2, and PBLH, the diurnal cycle of T2 exhibits a slow linear increase in temperature rather than the expected, observed rapid increase in T2 after sunrise that is supported by rapid PBL growth. This behavior is likely a function of PBL initial conditions or physics. As T2 is merely a diagnostic variable, its diurnal shape does not influence the integration of the model and instead offers an opportunity to more clearly illustrate the limitations of calibration. Even when the parameters are calibrated to T2 observations, little improvement is made in the T2 error because the model cannot change the imposed diurnal cycle.
This study focused on the core LoCo process chain components and did not address cloud or precipitation impacts, which is an area for future extension of this work. Although previous studies have shown the effectiveness of calibrating both land and atmospheric parameters together to improve model skill (Liu et al. 2004, 2005), no atmospheric parameters are calibrated in this work. This is due to the short time and space scales in the experimental design that allow for the dominant atmospheric processes to be resolved rather than parameterized. In addition, calibrating only the LSM parameters in a coupled framework reveals how the modeled atmosphere responds to changes in the land, allowing for the identification of model deficiencies in individual process chain components that would otherwise be obscured if land and atmosphere parameters were calibrated together.
It should also be noted that the evaluation methods undertaken in this study are different than traditional calibration approaches. This study does not seek universal, effective parameters for model improvement, to make overarching statistical conclusions about LA coupling, or to identify the most effective calibration approach for traditional metrics of NWP (e.g., 500-hPa height, T2, relative humidity, precipitation). Rather, we use calibration as a tool to understand how physical connections are made in the model for these case studies. As such, in order to understand the implications in constraining one component of the process chain and not the others, we evaluate the simulations (Tables 1 and 2 and Figs. 4 and 10) with the same observations used to calibrate the model runs. This differs, by design, from traditional evaluation that typically uses one set of observations to improve the model when tested against independent observations.
Both case studies showed that calibrating to Q2 degraded the overall model performance. Simply trying to reduce Q2 (or T2) error via ISM or SHPs is difficult as Q2 is far downstream and can be influenced by fluxes, PBLH, entrainment, and the limitations of the associated physics schemes. This result has implications for operational centers that rely on screen-level variable (T2, Q2) assimilation for SM nudging (e.g., ECMWF; Drusch et al. 2009; Drusch and Viterbo 2007; de Rosnay et al. 2013; Hu et al. 1999). Although such an approach can improve Q2 and surface fluxes, it may ultimately bury the errors elsewhere in the coupled system, rather than improving the system as a whole. Such a result is somewhat unsurprising as NWP development to date has largely overlooked LA processes and by extension has not been designed to fully leverage quality land information. In addition, approaches that infer surface fluxes from PBL variables (e.g., Denissen et al. 2021) assume that radiation and other atmospheric model parameters are appropriate (e.g., entrainment ratio). The SM_PBLH run conducted here represents a similar approach and shows that when these assumptions are not met (e.g., radiation bias), the error will appear in the fluxes and negatively impact other components of the process chain.
5. Conclusions
This study used the SCM capability of the NU-WRF modeling system to assess model coupling at the local scale in two summer case studies at the ARM program site in Lamont, OK. The NU-WRF SCM was coupled to NASA’s Land Information System (LIS-SCM) enabling a fully interactive PBL and atmosphere without the complexity of a three-dimensional model. A series of experiments were performed in which SHPs, ISM, or both, were calibrated to atmospheric observations of the LoCo process chain including T2, Q2, Bowen ratio, and PBLH. To do so, the OSTRICH model calibration software was deployed as a wrapper around LIS-SCM (and inherently to WRF and NU-WRF for the first time). An additional calibration experiment was also completed in which soil and vegetation parameters were calibrated to SMAP SM, representing the traditional approach to calibration in the LSM community (i.e., calibrating one land parameter to one land observation over a long period of record). The results of the SMAP calibration experiments underscore the need for coupled calibration, as it cannot be assumed that parameters that improve soil moisture also positively impact the remainder of the LA process chain.
The LIS-SCM calibration produced a wide range of impacts to SM, fluxes, ambient weather, and PBLH in both case studies explored. Results show that optimizing the coupled model to predict PBLH produces vastly different SHP and ISM solutions than if optimized for surface fluxes or ambient temperature or humidity. This is due to the varying uncertainties in model parameterizations that connect these quantities in the process chain. The range in solutions also provides insight into the various ways land surface parameters and ISM can be combined to achieve a desired impact on the observed downstream (atmospheric) variables, highlighting that improvement in either SHP or ISM when not in tandem with the other can provide undesirable results. For example, improvement of ISM via soil moisture data assimilation, with improper soil type specification, will not necessarily improve SM–EF or the atmosphere downstream. It should be noted that the boundary conditions, initial conditions, and the model parameterizations have some level of inherent background uncertainty, as in all models. The case studies were chosen in part because they are quiescent weather days where synoptic-scale forcing is weak, limiting the impacts of more complicated atmosphere processes (e.g., deep convection, etc.).
Overall, this work demonstrates how the LIS-SCM with OSTRICH can be an effective tool for understanding model connections, feedbacks, strengths, and deficiencies, and where the most positive impact will be felt in incorporating observations via assimilation or calibration approaches. In the near term, we will see improvements in observations of all the components of the process chain, which will be integrated into coupled models via calibration or assimilation. The approach presented here allows us to parse out the unique impact of each of these observations on what should be a tightly coupled system (LA process chain), but as demonstrated here is often far from that. The August case shows that when the system is well constrained, the calibration of ISM and/or SHPs has more positive impacts across the process chain. This implies that the utility of new observations, such as PBLH, will depend on underlying biases in the coupled model including those outside of the process chain.
Overall, this work demonstrates how the LIS-SCM with OSTRICH can be an effective tool for understanding model connections, feedbacks, strengths, and deficiencies, and where the most positive impact will be felt in incorporating observations via assimilation or calibration approaches. The ability to select from both a suite of LSMs within LIS as well as numerous advanced physics option in WRF SCM, provides unique flexibility of the LIS-SCM-CAL system that can aid future model development. In particular, this system can facilitate a “model development hierarchy” approach, utilized by the land team at NCEP to systematically evaluate models in hierarchical steps starting with process-level components and ending with fully coupled Earth system models (Ek et al. 2016).
This study demonstrated the utility of the LIS-SCM-CAL system using one site and two case studies. To fully leverage this system to improve NWP and predictions of extremes, future work is required to comprehensively assess coupling across regions with differing climate and surface characteristics, but such a task remains a challenge. Although global networks of certain process chain variables exist, few sites feature all needed variables at the sufficient hourly temporal resolution. For example, FLUXNET and AmeriFlux networks provide fluxes, but not PBL properties. Most meteorological sites provide T2 and Q2, but not SM, fluxes, or PBLH. The global radiosonde network provides PBL profiles twice a day, but does not capture the necessary diurnal cycle features of PBL development. In short, a lack of PBL observation is the primary factor limiting the immediate extension of this work to many other regions. Advances in the collection of PBL observations, for example through the proliferation of supersites like GLAFO (Wulfmeyer et al. 2020) and the push for PBL measurements from space (Teixeira et al. 2021) would go a long way toward improving NWP using many approaches, including the one presented here.
This study also focused specifically on land parameters that control LSM states (soil moisture) and fluxes (via SHPs), but future work should be performed that also addresses multicriteria atmospheric parameters. This approach could also be used to generate composites over seasons (e.g., one date over many years) to better understand uncertainty in the calibration results stemming from interannual variability in weather conditions. In addition, future studies improving the efficiency and accessibility of coupled calibration methods will aid in further improvements to NWP.
Acknowledgments
The authors declare no conflicts of interest. This work was supported by the NASA Science Utilization of SMAP (SUSMAP) program and Jared Entin under GSFC Grant 15-SUSMAP15-1047.
Data availability statement
The in-situ data used in this study are publicly available via the ARM Data Discovery portal via https://adc.arm.gov/discovery/#/. SMAP soil moisture used for calibration is publicly available via the National Snow and Ice Data Center via https://nsidc.org/data/SPL3SMAP/versions/3.
REFERENCES
Behrouz, M. S., Z. Zhu, L. S. Matott, and A. Rabideau, 2020: A new tool for automatic calibration of The Storm Water Management Model (SWMM). J. Hydrol., 581, 124436, https://doi.org/10.1016/j.jhydrol.2019.124436.
Betts, A. K., and A. G. Barr, 1996: First International Satellite Land Surface Climatology Field Experiment 1987 sonde budget revisited. J. Geophys. Res., 101, 23 285–23 288, https://doi.org/10.1029/96JD02247.
Chou, M.-D., and M. J. Suarez, 1999: A solar radiation parameterization for atmospheric studies. NASA Tech. Rep. NASA/TM-1999-10460, Vol. 15, 38 pp., https://ntrs.nasa.gov/api/citations/19990060930/downloads/19990060930.pdf.
Cleugh, H. A., M. R. Raupach, P. R. Briggs, and P. A. Coppin, 2004: Regional-scale heat and water vapour fluxes in an agricultural landscape: An evaluation of CBL budget methods at OASIS. Bound.-Layer Meteor., 110, 99–137, https://doi.org/10.1023/A:1026096711297.
Crow, W. T., and Coauthors, 2020: Soil moisture/evapotranspiration over-coupling and L-band brightness temperature assimilation: sources and forecast implications. J. Hydrometeor., 21, 2359–2374, https://doi.org/10.1175/JHM-D-20-0088.1.
Dai, Y., and Coauthors, 2019: A global high-resolution data set of soil hydraulic and thermal properties for land surface modeling. J. Adv. Model. Earth Syst., 11, 2996–3023, https://doi.org/10.1029/2019MS001784.
De Lannoy, G. J., R. D. Koster, R. H. Reichle, S. P. Mahanama, and Q. Liu, 2014: An updated treatment of soil texture and associated hydraulic properties in a global land modeling system. J. Adv. Model. Earth Syst., 6, 957–979, https://doi.org/10.1002/2014MS000330.
de Rosnay, P., M. Drusch, D. Vasiljevic, G. Balsamo, C. Albergel, and L. Isaksen, 2013: A simplified Extended Kalman Filter for the global operational soil moisture analysis at ECMWF. Quart. J. Roy. Meteor. Soc., 139, 1199–1213, https://doi.org/10.1002/qj.2023.
Denissen, J. M. C., R. Orth, H. Wouters, D. G. Miralles, C. C. van Heerwaarden, J. Vila-Guerau de Arellano, and A. J. Teuling, 2021: Soil moisture signature in global weather balloon soundings. npj Climate Atmos. Sci., 4, 13, https://doi.org/10.1038/s41612-021-00167-w.
Dirmeyer, P. A., 2011: The terrestrial segment of soil moisture–climate coupling. Geophys. Res. Lett., 38, L16702, https://doi.org/10.1029/2011GL048268.
Dirmeyer, P. A., Z. Wang, M. J. Mbuh, and H. E. Norton, 2014: Intensified land surface control on boundary layer growth in a changing climate. Geophys. Res. Lett., 41, 1290–1294, https://doi.org/10.1002/2013GL058826.
Dirmeyer, P. A., C. Peters-Lidard, and G. Balsamo, 2015: Land–atmosphere interactions and the water cycle. Seamless Prediction of the Earth System: From Minutes to Months, G. Brunet, S. Jones, and P.M. Ruti, Eds., WMO-1156, World Meteorological Organization, 145–154, https://epic.awi.de/id/eprint/38291/1/WMO_Book_WWOSC.pdf.
Drusch, M. K., and R. Viterbo, 2007: Assimilation of screen-level variables in ECMWF’s integrated system: A study on the impact of the forecast quality and analyzed soil moisture. Mon. Wea. Rev., 135, 300–314, https://doi.org/10.1175/MWR3309.1.
Drusch, M. K., K. Scipal, P. de Rosnay, G. Balsamo, E. Andersson, P. Bougeault, and P. Viterbo, 2009: Towards a Kalman filter based soil moisture analysis system for the operational ECMWF Integrated Forecast System. Geophys. Res. Lett., 36, L10401, https://doi.org/10.1029/2009GL037716.
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, 8851, https://doi.org/10.1029/2002JD003296.
Ek, M. B., Y. Xia, J. Meng, R. Shresha, H. Wei, J. Dong, Y. Wu, and W. Zheng, 2016: Land prediction in NCEP modeling systems: Current status and future plans (NGGPS Land Team). Next Generation Global Prediction System (NGGPS) Meeting, College Park, MD, NOAA, 29 pp., https://www.weather.gov/media/sti/nggps/10_Ek%20EMC-land-NGGPS-Feb2016.pdf.
Entekhabi, D., and Coauthors, 1999: An agenda for land-surface hydrology research and a call for the second International Hydrological Decade. Bull. Amer. Meteor. Soc., 80, 2043–2058, https://doi.org/10.1175/1520-0477(1999)080<2043:AAFLSH>2.0.CO;2.
Entekhabi, D., and Coauthors, 2014: SMAP Handbook: Mapping Soil Moisture Freeze/Thaw from Space, NASA, 182 pp. https://smap.jpl.nasa.gov/system/internal_resources/details/original/178_SMAP_Handbook_FINAL_1_JULY_2014_Web.pdf.
Feddes, R. A., M. Menenti, P. Kabat, and W. G. M. Bastiaanssen, 1993: Is large-scale inverse modeling of unsaturated flow with areal average evaporation and surface soil moisture as estimated from remote-sensing feasible. J. Hydrol., 143, 125–152, https://doi.org/10.1016/0022-1694(93)90092-N.
Ferguson, C.R., J. A. Santanello, and P. Gentine, 2016: Enhanced soundings for local coupling studies: 2015 ARM Climate Research Facility field campaign. DOE Rep., DOE/SC-ARM-15-035, 15 pp., https://www.osti.gov/servlets/purl/1232666.
Findell, K. L., and E. A. B. Eltahir, 1997: An analysis of the soil moisture-rainfall feedback, based on direct observations from Illinois. Water Resour. Res., 33, 725–735, https://doi.org/10.1029/96WR03756.
Gentine, P., A. Chhang, A. Rigden, and G. Salvucci, 2016: Evaporation estimates using weather station data and boundary layer theory. Geophys. Res. Lett., 43, 11 661–11 670, https://doi.org/10.1002/2016GL070819.
Gu, L., and Coauthors, 2006: Direct and indirect effects of atmospheric conditions and soil moisture on surface energy partitioning revealed by a prolonged drought at a temperate forest site. J. Geophys. Res., 111, D16102, https://doi.org/10.1029/2006JD007161.
Gupta, H. V., L. A. Bastidas, S. Sorooshian, W. J. Shuttleworth, and Z. L. Yang, 1999: Parameter estimation of a land-surface scheme using multicriteria methods. J. Geophys. Res., 104, 19 491–19 503, https://doi.org/10.1029/1999JD900154.
Gutmann, E. D., and E. E. Small, 2005: The effect of soil hydraulic properties vs. soil texture in land surface models. Geophys. Res. Lett., 32, L02402, https://doi.org/10.1029/2004GL021843.
Haghnegahdar, A., B. Tolson, B. Davison, F. Seglenieks, E. Klyszejko, E. D. Soulis, V. Fortin, and L. S. Matott, 2014: Calibrating environment Canada’s MESH modelling system over the Great Lakes Basin. Atmos.–Ocean, 52, 281–293, https://doi.org/10.1080/07055900.2014.939131.
Harrison, K. W., S. V. Kumar, C. D. Peters-Lidard, and J. A. Santanello, 2012: Quantifying the change in soil moisture modeling uncertainty from remote sensing observations using Bayesian inference techniques. Water Resour. Res., 48, W11514, https://doi.org/10.1029/2012WR012337.
Hess, R., 2001: Assimilation of screen level observations by variational soil moisture analysis. Meteor. Atmos. Phys., 77, 145–154, https://doi.org/10.1007/s007030170023.
Hirsch, A. L., and Coauthors, 2019: Amplification of Australian heatwaves via local land-atmosphere coupling. J. Geophys. Res., 124, 13 625–13 647, https://doi.org/10.1029/2019JD030665.
Hogue, T. S., L. Bastidas, H. Gupta, S. Sorooshian, K. Mitchell, and W. Emmerich, 2005: Evaluation and transferability of the Noah land surface model in semiarid environments. J. Hydrometeor., 6, 68–84, https://doi.org/10.1175/JHM-402.1.
Holland, J., 1975: Adaptation in Natural and Artificial Systems. University of Michigan Press, 232 pp.
Hu, Y., X. Gao, W. J. Shuttleworth, H. Gupta, J.-F. Mahfouf, and P. Viterbo, 1999: Soil- moisture nudging experiments with a single-column version of the ECMWF model. Quart. J. Roy. Meteor. Soc., 125, 1879–1902, https://doi.org/10.1002/qj.49712555719.
Koster, R. D., and Coauthors, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305, 1138–1140, https://doi.org/10.1126/science.1100217.
Koster, R. D., Z. Guo, R. Yang, P. Dirmeyer, K. Mitchell, and M. J. Puma, 2009: On the nature of soil moisture in land surface models. J. Climate, 22, 4322–4335, https://doi.org/10.1175/2009JCLI2832.1.
Koster, R. D., and Coauthors, 2010: Contribution of land surface initialization to subseasonal forecast skill: First results from a multi-model experiment. Geophys. Res. Lett., 37, L02402, https://doi.org/10.1029/2009GL041677.
Kumar, S. V., and Coauthors, 2006: Land information system: An interoperable framework for high resolution land surface modeling. Environ. Modell. Software, 21, 1402–1415, https://doi.org/10.1016/j.envsoft.2005.07.004.
Kumar, S. V., R. H. Reichle, K. W. Harrison, C. D. Peters-Lidard, S. Yatheendradas, and J. A. Santanello, 2012: A comparison of methods for a priori bias correction in soil moisture data assimilation. Water Resour. Res., 48, W03515, https://doi.org/10.1029/2010WR010261.
Lawston, P. M., J. A. Santanello, B. F. Zaitchik, and M. Rodell, 2015: Impact of irrigation methods on land surface model spinup and initialization of WRF forecasts. J. Hydrometeor., 16, 1135–1154, https://doi.org/10.1175/JHM-D-14-0203.1.
Lawston, P. M., J. A. Santanello, B. Hanson, and K. Arsenault, 2020: Impacts of irrigation on summertime temperatures in the Pacific Northwest. Earth Interact., 24, 1–26, https://doi.org/10.1175/EI-D-19-0015.1.
Levenberg, K., 1944: A method for the solution of certain non-linear problems in least squares. Q. Appl. Math., 2, 164–168, https://doi.org/10.1090/qam/10666.
Liu, Y., L. A. Bastidas, H. V. Gupta, and S. Sorooshian, 2003: Impacts of a parameterization deficiency on offline and coupled land surface model simulations. J. Hydrometeor., 4, 901–914, https://doi.org/10.1175/1525-7541(2003)004<0901:IOAPDO>2.0.CO;2.
Liu, Y., H. V. Gupta, S. Sorooshian, L. A. Bastidas, and W. J. Shuttleworth, 2004: Exploring parameter sensitivities of the land surface using a locally coupled land-atmosphere model. J. Geophys. Res., 109, D21101, https://doi.org/10.1029/2004JD004730.
Liu, Y., H. V. Gupta, S. Sorooshian, L. A. Bastidas, and W. J. Shuttleworth, 2005: Constraining land surface and atmospheric parameters of a locally coupled model using observational data. J. Hydrometeor., 6, 156–172, https://doi.org/10.1175/JHM407.1.
Macro, K., L. S. Matott, A. J. Rabideau, S. H. Ghodsi, and Z. Zhu, 2019: 2019: OSTRICH-SWMM: A new multi-objective optimization tool for green infrastructure planning with SWMM. Environ. Modell. Software, 113, 42–47, https://doi.org/10.1016/j.envsoft.2018.12.004.
Marquardt, D., 1963: An algorithm for least-squares estimation of nonlinear parameters. SIAM J. Appl. Math., 11, 431–441, https://doi.org/10.1137/0111030.
Matott, L. S., 2017: OSTRICH: An Optimization Software Tool, Documentation and User's Guide, Version 17.12.19, University at Buffalo Center for Computational Research, 79 pp., www.eng.buffalo.edu/~lsmatott/Ostrich/OstrichMain.html.
Mesinger, F., and Coauthors, 2006: North American Regional Reanalysis. Bull. Amer. Meteor. Soc., 87, 343–360, https://doi.org/10.1175/BAMS-87-3-343.
Miralles, D. G., A. J. Teuling, C. C. van Heerwaarden, and J. V.-G. de Arellano, 2014: Mega-heatwave temperatures due to combined soil desiccation and atmospheric heat accumulation. Nat. Geosci., 7, 345–349, https://doi.org/10.1038/ngeo2141.
Mueller, B., and Coauthors, 2013: Benchmark products for land evapotranspiration: LandFlux-EVAL multi-dataset synthesis. Hydrol. Earth Syst. Sci., 13, 3707–3720, https://doi.org/10.5194/hess-17-3707-2013.
Muñoz-Sabater, J., H. Lawrence, C. Albergel, P. Rosnay, L. Isaksen, S. Mechklenburg, Y. Kerr, and M. Drusch, 2019: Assimilation of SMOS brightness temperatures in the ECMWF Integrated Forecasting System. Quart. J. Roy. Meteor. Soc., 145, 2524–2548, https://doi.org/10.1002/qj.3577.
Nakanishi, M., and H. Niino, 2006: An improved Mellor–Yamada Level-3 model: Its numerical stability and application to a regional prediction of advection fog. Bound.-Layer Meteor., 119, 397–407, https://doi.org/10.1007/s10546-005-9030-8.
Peng, F., M. Mu, and G. Sun, 2020: Evaluations of uncertainty and sensitivity in soil moisture modeling on the Tibetan Plateau. Tellus, 72A, 1–16, https://doi.org/10.1080/16000870.2019.1704963.
Peters-Lidard, C. D., and L. H. Davis, 2000: Regional flux estimation in a convective boundary layer using a conservation approach. J. Hydrometeor., 1, 170–182, https://doi.org/10.1175/1525-7541(2000)001<0170:RFEIAC>2.0.CO;2.
Peters-Lidard, C. D., D. M. Mocko, J. A. Santanello, M. Tischler, M. S. Moran, M. Garcia, and Y. Wu, 2008: Role of precipitation uncertainty in the estimation of hydrologic soil properties using remotely sensed soil moisture in a semiarid environment. Water Resour. Res., 44, W05S18, https://doi.org/10.1029/2007WR005884.
Peters-Lidard, C. D., and Coauthors, 2015: Integrated modeling of aerosol, cloud, precipitation and land processes at satellite resolved scales. Environ. Modell. Software, 67, 149–159, https://doi.org/10.1016/j.envsoft.2015.01.007.
Roundy, J. K., C. R. Ferguson, and E. Wood, 2013: Temporal variability of land–atmosphere coupling and its implications for drought over the Southeast United States. J. Hydrometeor., 14, 622–635, https://doi.org/10.1175/JHM-D-12-090.1.
Salvucci, G. D., and P. Gentine, 2013: Emergent relation between surface vapor conductance and relative humidity profiles yields evaporation rates from weather data. Proc. Natl. Acad. Sci. USA, 110, 6287–6291, https://doi.org/10.1073/pnas.1215844110.
Santanello, J. A., Jr., M. A. Friedl, and W. Kustas, 2005: Empirical investigation of convective planetary boundary layer evolution and its relationship with the land surface. J. Appl. Meteor., 44, 917–932, https://doi.org/10.1175/JAM2240.1.
Santanello, J. A., Jr., M. A. Friedl, and M. B. Ek, 2007a: Convective planetary boundary layer interactions with the land surface at diurnal time scales: Diagnostics and feedbacks. J. Hydrometeor., 8, 1082–1097, https://doi.org/10.1175/JHM614.1.
Santanello, J. A., Jr., C. D. Peters-Lidard, M. Garcia, D. Mocko, M. Tischler, M. S. Moran, and D. P. Thoma, 2007b: Using remotely-sensed estimates of soil moisture to infer soil texture and hydraulic properties across a semi-arid watershed. Remote Sens. Environ., 110, 79–97, https://doi.org/10.1016/j.rse.2007.02.007.
Santanello, J. A., Jr., C. D. Peters-Lidard, S. V. Kumar, C. Alonge, and W. K. Tao, 2009: A modeling and observational framework for diagnosing local land–atmosphere coupling on diurnal time scales. J. Hydrometeor., 10, 577–599, https://doi.org/10.1175/2009JHM1066.1.
Santanello, J. A., Jr., C. D. Peters-Lidard, and S. V. Kumar, 2011a: Diagnosing the sensitivity of local land–atmosphere coupling via the soil moisture–boundary layer interaction. J. Hydrometeor., 12, 766–786, https://doi.org/10.1175/JHM-D-10-05014.1.
Santanello, J. A., Jr., and Coauthors, 2011b: Results from Local Land-Atmosphere Coupling (LoCo) Project. GEWEX News, Vol. 21, No. 4, International GEWEX Project Office, Silver Spring, MD, 7–9, www.gewex.org/gewex-content/files_mf/1432209597Nov2011.pdf.
Santanello, J. A., Jr., C. D. Peters-Lidard, A. Kennedy, and S. V. Kumar, 2013a: Diagnosing the nature of land–atmosphere coupling: A case study of dry/wet extremes in the U.S. Southern Great Plains. J. Hydrometeor., 14, 3–24, https://doi.org/10.1175/JHM-D-12-023.1.
Santanello, J. A., Jr., S. V. Kumar, C. D. Peters-Lidard, K. Harrison, and S. Zhou, 2013b: Impact of land model calibration on coupled land-atmosphere prediction. J. Hydrometeor., 14, 1373–1400, https://doi.org/10.1175/JHM-D-12-0127.1.
Santanello, J. A., Jr., and Coauthors, 2018: Land–atmosphere interactions: The LoCo perspective. Bull. Amer. Meteor. Soc., 99, 1253–1272, https://doi.org/10.1175/BAMS-D-17-0001.1.
Santanello, J. A., Jr., P. Lawston, S. Kumar, and E. Dennis, 2019: Understanding the impacts of soil moisture initial conditions on NWP in the context of land–atmosphere coupling. J. Hydrometeor., 20, 793–819, https://doi.org/10.1175/JHM-D-18-0186.1.
Seneviratne, S. I., T. Corti, E. L. Davin, M. Hirschi, E. B. Jaeger, I. Lehner, B. Orlowsky, and A. J. Teuling, 2010: Investigating soil moisture–climate interactions in a changing climate: A review. Earth-Sci. Rev., 99, 125–161, https://doi.org/10.1016/j.earscirev.2010.02.004.
Shellito, P. J., E. E. Small, and M. H. Cosh, 2016: Calibration of Noah soil hydraulic property parameters using surface soil moisture from SMOS and basinwide in situ observations. J. Hydrometeor., 17, 2275–2292, https://doi.org/10.1175/JHM-D-15-0153.1.
Shellito, P. J., and Coauthors, 2020: Assessing the impact of soil layer depth specification on the observability of modeled soil moisture and brightness temperature. J. Hydrometeor., 21, 2041–2060, https://doi.org/10.1175/JHM-D-19-0280.1.
Skamarock, W. C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker, W. Wang, and J. G. Powers, 2005: A description of the Advanced Research WRF version 2. NCAR Tech. Note NCAR/TN-4681STR, 88 pp., https://doi.org/10.5065/D6DZ069T.
Sorbjan, Z., 1995: Toward evaluation of heat fluxes in the convective boundary layer. J. Appl. Meteor., 34, 1092–1098, https://doi.org/10.1175/1520-0450(1995)034<1092:TEOHFI>2.0.CO;2.
Stull, R. B., 1988: An Introduction to Boundary Layer Meteorology. Kluwer Academic Publishers, 670 pp.
Teixeira, J., and Coauthors, 2021: Toward a Global Planetary Boundary Layer Observing System: The NASA PBL Incubation Study Team Report. NASA PBL Incubation Study Team, 134 pp., https://science.nasa.gov/science-pink/s3fs-public/atoms/files/NASAPBLIncubationFinalReport.pdf.
Wang, S.-Y. S., and Coauthors, 2015: An intensified seasonal transition in the central U.S. that enhances summer drought. J. Geophys. Res. Atmos., 120, 8804–8816, https://doi.org/10.1002/2014JD023013.
Wulfmeyer, V., and Coauthors, 2020: The GEWEX Land-Atmosphere Feedback Observatory (GLAFO). GEWEX Quarterly, Vol. 30, No. 1, International GEWEX Project Office, Silver Spring, MD, 6–10, https://www.gewex.org/gewex-content/files_mf/1583952472Feb2020.pdf.
Yang, L., G. Sun, L. Zhi, and J. Zhao, 2018: Negative soil moisture-precipitation feedback in dry and wet regions. Sci. Rep., 8, 4026, https://doi.org/10.1038/s41598-018-22394-7.
