1. Introduction
The planetary boundary layer (PBL) is the turbulent layer of the atmosphere nearest to Earth’s surface in which exchanges of energy and momentum between the surface and the atmosphere take place. An important parameter that measures the depth of this layer and is a key indicator in the mixing that takes place within the PBL is the PBL height (PBLH). The PBLH is used in many predictive and diagnostic methods and/or models to assess pollutant concentrations, and it is an important parameter in some atmospheric models (e.g., as a key term in some PBL mixing parameterizations).
The complicated interactions between the land/ocean surface and the PBL lead to large uncertainties in modeling the PBL, especially with regard to parameterization schemes, making it challenging to accurately simulate PBL thermodynamic structure and PBLH. These difficulties are perpetuated by the lack of globally comprehensive PBL observations, as well as the lack of both strategy and infrastructure to utilize all PBL-related observations cohesively. The importance of PBL dynamics was underscored by the 2017 Decadal Survey (National Academies of Sciences, Engineering, and Medicine 2018), which designated the PBL as an incubation class observable, and selected as priorities better observations of PBL temperature and water vapor profiles, and of PBLH. Following this designation, the PBL Incubation Study Team Report (Texeira et al. 2021) made clear that “a future global PBL observing system requires modeling and data assimilation as essential components.”
Data assimilation plays a critical role in combining a large volume of disparate observations with dynamical models to produce an optimal estimate of the atmospheric state, for use in research applications or to initialize model forecasts. The near-real-time NASA Goddard Earth Observing System (GEOS) data assimilation system uses approximately 4.5 million observations in each 6-h assimilation window, and overall, the observations have a significant impact on reducing model forecast errors. However, an assessment conducted on the effectiveness of the use of existing observing systems in the lower troposphere found that the data impacts are not satisfactory (Zhu et al. 2022). For example, satellite radiances have limited impact in the PBL, and model physics tends to return to its original free forecast mechanisms at 1-day forecast, retaining little observation information within the PBL. While several known factors contribute to the unsatisfactory data impact on the PBL, such as underutilization of GNSS-RO data and cloud-affected infrared radiances, our objective in the present study is to develop new PBL data assimilation capabilities in GEOS, focusing on the use of PBLH from multiple observing systems.
We aim to address the following key issues related to PBLH and its assimilation:
-
The PBLH itself is a key parameter in many applications, e.g., forecasting near-surface meteorology and air quality. Therefore, it is desirable to provide a global PBLH analysis and monitoring capability in GEOS. However, such an analysis requires utilizing PBLH derived from multiple sources with potentially disparate observables and retrieval algorithms. Key issues that need addressing in this context relate to the physical meanings and potential inconsistencies of the disparate PBLHs and the need for an optimal strategy for using these data together.
-
The second issue in need of addressing is related to the unique feature of a capping inversion. Traditional data assimilation methodologies tend to focus on amplitude error, whereas error in the PBL is often positional, such as a misplaced inversion height. As a consequence, current data assimilation algorithms tend to distort inversion structures (Fowler et al. 2012). In the present assimilation framework study, the use of PBLH data to capture the capping inversion and benefit all the other observations is explored.
-
A third issue is the retention of observation information by the forecast model. In the situation where PBLH is a prognostic variable, PBLH analysis increments as a result of PBLH data assimilation can be fed back to the forecast model in the same way as the increments of other control variables. However, PBLH is a diagnosed variable in GEOS. In addition, short process time scales within the PBL allow model parameterizations to rapidly compensate for analysis increments, limiting the impact of observations. One potential solution is model parameter estimation, frequently done along with analysis increment estimation in data assimilation (Zhu and Navon 1999). Methods to utilize the global PBLH analysis within the model parameterizations to better retain observation information will be discussed in detail in a follow-up paper.
To provide a comprehensive global estimate of PBLH and serve as the reference for PBLH data generated from both existing and new sensors, our overarching goal is to explore a global data synergy strategy for the assimilation of PBLH data from multiple observing systems and produce a global PBLH analysis in GEOS. Many groups have investigated and derived PBLH data from various observing systems (e.g., Lewis et al. 2013; McGrath-Spangler and Denning 2013; Molod et al. 2015; Ganeshan and Wu 2015; Palm et al. 2021a; Salmun et al. 2023). In this NASA PBL Incubation project, PBLH data from radiosondes, Global Navigation Satellite System-Radio Occultation (GNSS-RO), space-based lidars including the Cloud–Aerosol Lidar with Orthogonal Polarization (CALIOP) instrument aboard the Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) and the Cloud–Aerosol Transport System (CATS), ground-based Micro-Pulse Lidar Network (MPLNET), and global radar wind profiler networks are used. However, only PBLH data from radiosondes and GNSS-RO are included in this paper for methodology development. Other PBLH data and statistical evaluations will be discussed in a follow-up article, and the developed methodology is expected to be applicable to these PBLH data with the additional consideration that they are also sensitive to hydrometeors and aerosols.
Model PBLH from GEOS or the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2; Gelaro et al. 2017), has been used in previous studies to compare to PBLH data estimated from various observing systems (Molod et al. 2015; Ganeshan and Yang 2019). Although these studies found comparable patterns between PBLH data and model PBLH, differences were also noticed. One of the major reasons for the differences lies in the mismatch of PBLH definitions (Seidel et al. 2010; McGrath-Spangler and Molod 2014). PBLH products are derived using observing systems and retrieval algorithms that may be sensitive to different key variables, and as a result, the derived PBLH data may represent different physical heights. For example, radiosonde-based PBLH is often based on thermodynamic structure and gradients, GNSS-RO PBLH data can be based on refractivity gradient, while lidar PBLH data often use aerosols as a tracer to identify the top of the surface-attached mixed layer (Lewis et al. 2013; McGrath-Spangler and Denning 2013; Palm et al. 2021b). In this study, consistent model PBLH definitions will be used to compare PBLH data from different sources and to compute departures of first guess from observations (OmF).
The traditional approach of the variational method to assimilate PBLH data is to construct an observation operator, which links control variables (or model prognostic variables) to PBLH data from each observing system and its tangent linear and adjoint operators. Amerault et al. (2024) use this approach in their variational assimilation of the PBLH data study and show clear sensitivities of PBLH with respect to potential temperature, water vapor mixing ratio, and wind. However, the observation operator can be very complicated and involves discontinuities, e.g., the lidar-based PBLH observation operator includes the total attenuated backscatter (ATB) lidar forward observation operator, the algorithm to derive PBLH from ATB, interpolation, and any calculations to handle the mismatch between observed and model scales. An alternative approach is to augment PBLH to the control variable vector. This approach technically moves part of the complicated forward observation operator (e.g., ATB calculation and PBLH retrieval algorithm) to the model, and as a result, it allows PBLH data to be assimilated with a straightforward observation operator, which consists of only interpolation and calculations handling the mismatch between observed and model scales, and generates an hourly global PBLH analysis. The correlations between PBLH and other control variables, which are embedded in the flow-dependent ensemble background error covariance, provide a mechanism for the PBLH data to impact other variables. Since the PBLH is a 2D variable, the requirement of extra computational resources will be minimal. This approach has been used in previous studies, e.g., the assimilation algorithm for radiosonde and aircraft PBLH data developed in the 2D variational Real-Time Mesoscale Analysis (RTMA) system for a dispersion modeling study (Tassone et al. 2012), and the PBLH assimilation study using lidar PBLH data from a field campaign in Greensburg, Kansas, which however employed a comparison of PBLH data with inconsistent model PBLH (Tangborn et al. 2021). Analogous to these applications, this approach of control variable augmentation was also used in the assimilation of radar reflectivity without tangent linear and adjoint of the nonlinear observation operator (Wang and Wang 2017). Because of its relative simplicity, this approach is adopted in the present study, and the original PBLH data assimilation algorithm used in the RTMA is adapted into the GEOS hybrid 4D ensemble-variational (EnVar) framework. In addition, with the PBLH data from different observing systems, special attention is paid to the consistency of PBLH data and model PBLH simulation. For example, a bulk Richardson number–based PBLH definition is used to compare the model and radiosonde data, and a refractivity gradient–based PBLH definition is used to compare the model and GNSS-RO data. Corresponding model PBLH definitions are augmented to the control variable vector.
The second issue we address here is how to capture capping inversions in the PBL data assimilation. A recent study conducted by Hilton et al. (2020) employed multiple existing retrieval and one-dimensional variational (1D-Var) algorithms and found that, in general, only where 1D-Var retrievals used numerical weather prediction (NWP) as input and already had an inversion could the retrieval have an inversion too. Algorithms to account for positional error explicitly have been investigated (e.g., Ravela et al. 2007; Fowler et al. 2012); however, these algorithms tend to be complicated and sensitive to the factors including choice of displacement function, observation vertical resolution, and whether a capping inversion is present. In this study, we have explored new approaches to better capture the locations of capping inversions, through improving control variables’ ensemble background error covariances and vertical localization length scales based on the PBLH data information. This will benefit the assimilation of all other PBL-related observations such as direct observations of temperature, specific humidity, and radiances. Therefore, PBLH data can affect the PBL thermodynamic structure through improving both ensemble background error covariances and the aforementioned correlations between PBLH and other control variables prescribed by ensemble background error covariances. The separate and combined PBLH data impacts are investigated in this study.
The present paper constitutes a first step toward developing and testing a comprehensive PBLH data assimilation methodology. It includes discussion of issues related to PBLH data, corresponding model PBLH definitions, and methods of using PBLH data in the data assimilation followed by preliminary results. The optimization of our methods and impact of PBLH data from multiple observing systems and methods to utilize the PBLH analysis in the model parameterization schemes will be presented in subsequent papers. Here, following this introduction, section 2 gives a brief overview of the global GEOS data assimilation system, and section 3 introduces the methodologies for PBL data assimilation, including PBLH data from radiosonde and GNSS-RO observations and their corresponding model PBLH definitions and the methods to help capture capping inversions and correlations between PBLH and other control variables in the background error covariance. Section 4 describes preliminary results of assimilation studies to demonstrate the PBLH assimilation capabilities. Conclusions and future plans are discussed in section 5.
2. Overview of the global GEOS data assimilation system
The global GEOS data assimilation (DA) system is used routinely to support NASA space missions and field campaigns. While the system includes a coupled atmosphere–ocean configuration used for seasonal prediction, our study focuses on the atmosphere-only configuration composed of the GEOS atmospheric general circulation model (AGCM) and the atmospheric data assimilation system (ADAS). The GEOS AGCM is built around the finite-volume cubed-sphere dynamical core (Putman and Lin 2007) and has 72 vertical levels with 12 levels below about 1.7 km. The PBL parameterizations include the “Lock” K-profile scheme driven by surface and cloud-top buoyancy fluxes (Lock et al. 2000) and the “Louis” local scheme for stable conditions based on the Richardson number (Louis 1979). Together, these elements determine a profile of diffusivity coefficients that are used to calculate PBL tendencies. The analytical diffusivity profile shape in the Lock scheme is determined by parcels released upward from the surface and downward from stratocumulus cloud top. Above the mixed layer defined by the Lock surface plume, shallow cumulus convection is represented by the Park and Bretherton (2009) mass flux scheme. Additional parameterizations are summarized in Arnold et al. (2020). Aerosols are an integral part of the GEOS system, including estimates and forecasts from the Goddard Chemistry Aerosol Radiation and Transport (GOCART) model and assimilation of aerosol optical depth measurements.
The ADAS uses the Gridpoint Statistical Interpolation analysis system (GSI) (Kleist et al. 2009). In the hybrid 4D EnVar configuration (Todling and El Akkraoui 2018; Wang et al. 2007), two DA systems run concurrently: the ensemble ADAS (Whitaker et al. 2008) that provides the flow-dependent background error covariance information to the high-resolution deterministic ADAS, and the deterministic ADAS that provides its analysis to the recentering procedure of the ensemble ADAS and bias correction information.
The near-real-time GEOS system combines all operational observations coherently with the model physics to generate optimal analysis increments of the control variables for the subsequent model forecast. The resultant analysis increments are fed back to the forecast model through the 4D incremental analysis update (IAU) approach (Takacs et al. 2018), where the model is integrated for 6 h with the 4D IAU tendencies. This approach allows the model to ingest information from observations and adjust its states and related physical balances gradually.
In this study, a lower horizontal resolution version of the then-current operational GEOS configuration in 2022 is used, with the deterministic cycle at 25 km for model forecast and 50 km for analysis and the ensemble cycle at 100 km, with 72 vertical levels in all its components.
3. Strategy and methodologies for assimilation of PBLH data from multiple observing systems
The PBLH data products used in this study have different strengths and weaknesses. Satellite PBLH data can provide better global coverage and complement in situ PBLH data, while in general ground-based PBLH data provide a more complete temporal sampling of the PBL diurnal cycle. In this section, we first focus on the derivation of PBLH data and appropriate model PBLH definitions, which are used to assess PBLH data and calculate the departures of model PBLH from PBLH data, i.e., PBLH OmFs. Then, the strategies of effectively utilizing the PBLH data from multiple observing systems for the construction of the global PBLH data monitoring and analysis framework are presented. The methods to capture PBL capping inversions are explored, and the correlation relationships among PBLHs and other control variables are also examined to help improve the PBL thermodynamic structure in the analysis.
a. PBLH data and model PBLH definitions
A successful PBLH data assimilation requires that comparison between observations-derived PBLH data and model simulations be consistent. Due to the sensitivity of PBLH data to the observing method and choice of algorithm, it is important to use a model definition appropriate for each observation type to compute consistent OmFs of PBLH. In this paper, while PBLH is used to refer to PBLH definitions and data in general,
1) Bulk Richardson number–based radiosonde and model definition
The GEOS model uses two PBLH definitions as length scales in the Louis turbulence scheme: the bulk Ri–based PBLH described above which is used over land, and a second definition based on the profile of thermal diffusivity is used over ocean. The latter PBLH is defined as the height at which the diffusivity drops to a threshold of 10% of the column maximum KH. We use the bulk Ri–based model
(a),(b) Bulk Ri–based radiosonde
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
2) Refractivity gradient–based GNSS-RO and model PBLHRF definition
Space-based observing systems, such as GNSS-RO, offer greater global data coverage than radiosonde or sparse ground-based networks. GNSS-RO is based on an atmospheric limb sounding technique that yields information at relatively coarse horizontal but high vertical resolution, which makes it potentially suitable for PBLH retrievals over oceans and flat land surfaces. The GNSS-RO
In this study, we use a modified version of the algorithm of Ao et al. (2012) to calculate PBLHRF. In the model, the refractivity of moist air (Cucurull et al. 2013) is computed using model profiles of temperature, humidity, and pressure. The algorithm of Ao et al. (2012) identifies
Model (a) original
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
Close inspection of the original
(a) Ensemble spread of original
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
These findings led us to modify the refractivity gradient–based algorithm as follows. We first identify all local minima in the refractivity gradient exceeding a certain threshold. The threshold is a weighted average of the overall minimum (25%) and the average gradient below 6 km (75%). This effectively removes small wiggles in the profile from consideration, while retaining more significant minima. The two most negative minima are then identified and labeled PBLHRFLOW and PBLHRFHGH based on their relative heights. In cases with only one minimum exceeding the threshold, both PBLHRFLOW and PBLHRFHGH are set to the same value. In the example shown in Fig. 3b, the lower minimum at an altitude of approximately 1326 m is denoted as PBLHRFLOW, while the upper minimum at 2356 m is denoted as PBLHRFHGH. A flowchart is provided in Fig. 4 to illustrate the algorithm used for estimating PBLHRFLOW and PBLHRFHGH.
Flowchart to illustrate the derivation algorithm for refractivity gradient–based PBLHRFLOW and PBLHRFHGH.
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
A global snapshot of
The global snapshot of
In principle, both low and high values could be assimilated; however, in this study, we will focus on assimilating
Figure 2c shows
b. PBLH data synergy strategy and quality control (QC)
One of our primary objectives is to build a global PBLH analysis and monitoring capability in the GEOS system. We seek to explore a strategy for assimilating PBLH data from various sources by using multiple PBLH control variables, i.e., the model
To successfully assimilate PBLH data, QC is a necessary and important step. Generally, QC procedures are unique for each PBLH dataset, e.g., the QC procedures for GNSS-RO PBLH data are different from those for PBLH data from space-based lidars. In this paper, only the QC procedures for PBLH data from radiosonde and GNSS-RO are presented. For both radiosonde and GNSS-RO PBLH data, while larger OmF errors are usually seen with deeper PBL, no obvious dependence of OmF errors on local time is found, and any PBLH data exceeding 6 km are excluded. For radiosonde
Subgrid surface height STDV (m) used to exclude GNSS-RO PBLH data at locations where the subgrid surface height STDV > 200 m.
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
A final gross error check is applied to both radiosonde and GNSS-RO PBLH OmFs, where PBLH data will be rejected if the absolute value of the ratio of OmF to observation error exceeds a specified threshold, 5 in this study. This gross error check is modest and excludes only extreme outliers. As depicted in Fig. 6b for the 426 GNSS-RO
Difference in number of points of GNSS-RO
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
OmF histograms of (a)
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
Observation errors of
c. Capturing capping inversion and correlation between PBLHs and thermodynamic states
In the GEOS hybrid 4D EnVar formulation, observation information OmFs are propagated by background error covariance and projected onto analysis variables. The background error covariance is composed of static and ensemble components, and the flow-dependent ensemble component is generally provided by the model forecast ensemble. The static part in this study is designed to be univariate for
The impact of PBLH data on the PBL thermodynamic structure is realized through the ensemble background error covariance and through improving the performance of model physical parameterization schemes by utilizing the PBLH analysis. The discussion of the latter is not included in this paper but will be described in a subsequent article. The ensemble forecast provides the horizontal and vertical correlation structure of background errors and relationships among control variables, and the localization length scales are used to reduce sampling errors of the background error covariance with undersized 32 ensemble members. The accurate ensemble background error covariance and vertical localization length scale are critical for the assimilation of all observation profiles in the lower troposphere when analyzing a boundary layer capping inversion. When a capping inversion is present, the ensemble background error covariance is characterized by a large ensemble spread around the inversion height and a decoupling between the errors in the PBL and the free atmosphere, as a strong inversion prevents the air masses from mixing. However, when there is a large discrepancy in inversion height between background and observation, the ensemble spread, which peaks around the background inversion height, will cause an incorrect mapping structure for the observation profile information OmFs. As a result, as shown in Fowler et al. (2012), traditional data assimilation algorithms tend to erroneously consider the large OmFs in the vicinity of the inversion as amplitude error, which in turn leads to the analysis smearing out and degrading the inversion structure.
Therefore, it is key to accurately represent the ensemble spread structure around the observed capping inversion height to increase the impact of PBLH observation, and more importantly, the impact of profile observations, on the state variables in the model levels closest to the observed PBLH and to confine the impacts of observations around the observed inversion height properly. In this study, we use PBLH data to better capture the location of a capping inversion directly through inflation of the ensemble spread and adjustment of the vertical localization length scale around the model levels closest to the PBLH data. This method is referred to as AdjEnsVloc hereafter. In this way, PBLH data are no longer just a single piece of information, and their utilization will effectively help to draw the analysis profile more closely to observations for capping inversions. In each 6-h assimilation window, we inflate the ensemble virtual temperature Tυ and relative humidity (RH) perturbations at the four neighboring ensemble analysis grids around each PBLH datum horizontally and on five model levels as discussed below. The inflation is applied to the two ensemble time levels adjacent to the time of each PBLH datum. The inflation coefficients, defined as a function of model vertical level k, are set as c = 1 + α × f(k) with Gaussian probability density function
(a) Probability density function f(k) and (b) inflation coefficient c that are used for ensemble virtual temperature or relative humidity perturbations at the model level closest to each PBLH datum and two model levels above and below.
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
While the horizontal localization length scale is kept unchanged, adjustment is also made to the vertical localization length scale for the ensemble around inversions to ensure limited interactions between PBL and above air. The existing length scale varies at a different model level to mitigate spurious vertical error correlations but is a constant for the globe at each model level. It is specified in logarithmic pressure units for each model level and then converted to grid units used in recursive filter by scaling the length scale of each model level with the logarithmic pressure spacing of each model layer. In this study, PBLH data information is introduced into the process to allow the length scales to vary globally depending on the locations of PBLH data. Using the model’s geopotential height information at each model level, and latitude, longitude, and PBLH values at each PBLH observation location, we can identify the model level closest to the PBL top and the four neighboring horizontal ensemble grids surrounding the PBLH observation. In our preliminary tests, for simplicity, the vertical localization length scale is reduced by multiplication of 0.1 for the model level closest to the PBL top and two levels above and below, respectively, at these neighboring grids around each PBLH observation. A multiplication by the Gaussian-type function will be explored in the future.
Moreover, the correlation relationships among PBLHs and other control variables determine how the PBLH data information is projected and how PBLHs and other control variables interact with each other. All the PBLHs are tightly coupled with the PBL thermodynamic variables (e.g., as shown in Fig. 9), and thereby, the PBLH data will provide additional information for PBL temperature and moisture fields. For example, perturbation correlations between
Ensemble perturbation correlation coefficients (left) between PBLHRF and virtual temperature and (right) between PBLHRF and specific humidity at model level 9 at 0000, 0600, 1200, and 1800 UTC 9 Sep 2015. Black arrows illustrate the diurnal cycle, and blue arrows denote the strong correlation coefficients in the marine stratocumulus regions.
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
Normalized ensemble covariance of thermodynamic quantities with PBLH in three boundary-layer regimes. (a),(c) θυ or (b),(d) q covariance with (a),(b)
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
4. Assimilation testing results
To demonstrate the capability of the newly developed PBLH data assimilation framework and methodology, and of the impact of PBLH data on the representation of the PBL thermodynamic structure, in section 4a, we randomly select an analysis cycle of 0000 UTC 9 September 2015 to investigate the PBLH data impacts on atmospheric state variables. A sensitivity study of the AdjEnsVloc mechanism is first conducted with global radiosonde data profiles and
Although PBLH data contain unique height location information of the gradient extrema in air properties across the inversion, which is challenging for the assimilation of profile data to capture in data assimilation systems, we intend to use PBLH data conservatively. If the profiles that are used to derive PBLH data are separately assimilated in the GEOS system, we will confine the impact of the PBLH data to the PBLH variable only, i.e., only PBLH univariate static background error covariance is used for the assimilation of this PBLH data type. This is the case for
a. A case study: 0000 UTC 9 September 2015
1) Assimilation of radiosonde data
First, we examine the capability to assimilate radiosonde
Experiments for the assimilation of radiosonde
The first guess and analysis of variables of the four experiments are interpolated to radiosonde station 91680 and then are compared with radiosonde data profiles and
Vertical profiles of radiosonde observation (black solid), first guess (red solid), and analyses (Control: orange solid; Control_AllOb: green dotted; Inf2T2Q_0.1VLoc: light blue solid; Inf5T10Q_0.1VLoc: violet solid) of (a) virtual potential temperature (K), (b) specific humidity (g kg−1), (c) full-resolution specific humidity observations from the University of Wyoming site and corresponding first guess, and (d) temperature subtracted by lapse rate (K) at the radiosonde station 91680 (lat: 17.8°S, lon: 177.5°E) for the analysis cycle of 0000 UTC 9 Sep 2015. Horizontal dashed lines indicate
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
In both of the control runs, the radiosonde data for temperature T and specific humidity q profiles are assimilated. While small improvements in T and q analyses are observed, the analyses of T and θυ become worse than the first guess at and above the observed PBL top, and the PBL height is hardly moved. This is because the original ensemble spread (Fig. 12, red line) from the ensemble perturbations is small with little variation below 900 hPa and exhibits a slight increase from 900 hPa till peaking around 860 hPa with a shallower PBL than the observed. Compared with the analyses of control using only radiosonde data, the use of all available observations in Control_Allob has little additional benefits at this station; in fact, it moves analyses of T and θυ even farther away from observations at and above the observed PBL top.
Vertical profiles of ensemble spread of (a) virtual temperature and (b) relative humidity before (red line) and after (blue line) ensemble spread inflation in the Inf2T2Q_0.1VLoc experiment at the radiosonde station 91680 at 2300 UTC 8 Sep 2015. Black lines indicate first guess of virtual temperature (K) in (a) and specific humidity (g kg−1) in (b) at the corresponding time.
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
In the other two experiments where
Small additional improvements are observed when further adjustment of vertical localization length scale is applied. Smaller vertical localization length scale can better represent the strong inversion at this station, and the analysis increments are more confined around the PBL top (figure not shown).
The 2D map analysis increments of virtual temperature and specific humidity below and above the PBL top, i.e., at model levels 11 and 13, are examined in Fig. 13. Compared with the control run, as expected, the adjustments to Tυ and RH ensemble spread and the vertical localization length scale make the increments larger in magnitude in the vicinity of this radiosonde station, and maximum location of specific humidity increments is also shifted to the west. In addition, compared to the control run, the interactions between model levels 11 and 13 are decreased in Inf2T2Q_0.1VLoc experiment due to the reduction of the vertical localization length scale. The large negative OmFs below the PBL top produce larger negative analysis Tυ increment at model level 11 but have smaller impact on model level 13, where small positive Tυ increments are now present in Inf2T2Q_0.1VLoc as a result of the positive OmFs above PBL top. Similarly, the specific humidity variable exhibits larger opposite signs of analysis increments at the two model levels based on the respective OmFs information below/above PBL top. The impact of the reduced vertical localization length scale can also be observed in the cross section of analysis increments in Fig. 14, for example around 18°S, there is less interaction between model levels 11 and 13, and opposite signs of analysis increments are increased at these two model levels.
The analysis increments at model levels 11 (left four panels) and 13 (right four panels) for (a),(c),(e),(g) virtual temperature and (b),(d),(f),(h) specific humidity in (top) Control and (bottom) Inf2T2Q_0.1VLoc experiments at 2300 UTC 8 Sep 2015. The star indicates the location of radiosonde 91680.
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
As in Fig. 13, but showing vertical cross section of analysis increments for (a),(c) virtual temperature and (b),(d) specific humidity in the (top) Control and (bottom) Inf2T2Q_0.1VLoc experiments along a longitudinal line at 177°E, which is shown by the red dashed line in the inset at the top-right corner of (a).
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
Overall, the PBLH assimilation algorithm is working, producing PBLH analysis and improving the representation of PBL inversion. Inf2T2Q_0.1VLoc AdjEnsVloc setup will be used, hereafter, for conservative adjustments. Further studies are required to obtain optimal ensemble spread inflation and adjustment of the vertical localization length scale. It is also noticed that little improvement is observed near the surface. One of the possible causes may be associated with the small ensemble spread in the lowest model levels, which may be assessed in future stochastic model physics studies.
2) Assimilation of GNSS-RO data
Unlike the use of radiosonde
A set of experiments are performed to evaluate the effects of the two mechanisms on how
GNSS-RO
It is shown in Fig. 15 that the impacts of AdjEnsVLoc (upper row) are different from the impacts of the multivariable relationships in original BEC (middle row). The differences of analysis increments between Exp2 and Exp1 are as expected, PBLHRF increments are the same for the two experiments, but the increments for Tυ and q are changed due to AdjEnsVLoc based on the observed
Differences of analysis increments of (left) PBLHRF (m), (middle) virtual temperature (K), and (right) specific humidity (g kg−1) at model level 9 (around 880–910 hPa over tropical oceans) between (a),(b),(c) Exp2 and Exp1; (d),(e),(f) Exp3 and Exp1; and (g),(h),(i) Exp4 and Exp1.
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
The combined impact of AdjEnsVLoc and hybrid BEC of PBLHRF on the analysis is also reflected in the profiles of the mean of the normalized observation minus analysis (OmA) and its root-mean-square difference (RMSD) for GNSS-RO bending angle data for Exp1 and Exp4 (Fig. 16). The maximum impact on the analysis of thermodynamic variables can be expected at where
(a) The number of assimilated GNSS-RO bending angles in the analysis cycle of 0000 UTC 9 Sep 2015, (b) the mean of normalized OmA of bending angles, and (c) RMSD of normalized OmA (black solid: Exp1; blue dashed: Exp4) of bending angles as a function of impact height. Red line indicates change in RMSD of Exp4 from Exp1 (%). Statistics are obtained for each 300-m impact height interval by aggregating the data within those intervals, and there are fewer used data near the surface.
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
b. Assessment for 23 August–17 September 2015
So far, we have investigated the two mechanisms through which PBLH data affect the PBL thermodynamic structure, AdjEnsVloc using
Three stand-alone analysis experiments shown in Table 3 are conducted to access the impact of our algorithms. While all experiments assimilate radiosonde
Radiosonde
The RMSDs of thermodynamic variable analysis departures from radiosonde observation data (Fig. 17) clearly demonstrate the direct impact of adjusting ensemble BEC of thermodynamic variables and vertical localization length scales on the analyses of thermodynamic variables at radiosonde observation locations. While the RMSDs of radiosonde temperature and specific humidity OmA below 700 hPa are displayed in Figs. 17a and 17b for all experiments, the corresponding OmA RMSD changes of T and q for experiments AdjStat (red solid line) and AdjHybr (blue dashed line) with respect to CNTL below 300 hPa are displayed in Figs. 17c and 17d. It is shown that the results of AdjStat and AdjHybr are basically the same at the radiosonde data locations because the multivariate aspect of BEC is lacking due to the fact that only univariate BEC is employed for
The profiles of RMSD of analysis departures OmA of (a) temperature and (b) specific humidity from CNTL (black solid), AdjStat (red dashed), and AdjHybr (blue dotted) experiments, respectively, against radiosonde observations below 700 hPa from 23 Aug to 17 Sep 2015. The profiles of percentage change in OmA RMSD of (c) temperature and (d) specific humidity of AdjStat (red solid) and AdjHybr (blue dashed) from CNTL below 300 hPa. Reduction of RMSD (negative) indicates improvement over CNTL experiment. The right y axis indicates the number of observations aggregated in each pressure level interval for statistics. The corresponding shading indicates the confidence intervals with the 95% significant level.
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
The normalized GNSS-RO bending angle OmA RMSDs and the percentage changes from CNTL are presented in Fig. 18. GNSS-RO data used in this period are aggregated into 300-m impact height intervals to calculate the bending angle OmA RMSD. Here, bending angle OmA is normalized by the corresponding bending angle observation. Although adjustments are made to ensemble BEC of thermodynamic variables in the AdjEnsVloc method, the analysis fits GNSS-RO bending angle data also show improvement mostly below 4000 m, and the largest improvement happens around 2400 m. There is a slight degradation near surface, but this corresponds to much fewer GNSS-RO data compared to available data counts at other heights. Moreover, it is noticed that a majority of the improvements has already been obtained in experiment AdjStat, and only small additional impact is introduced into GEOS by AdjHybr. This suggests that the impact through AdjEnsVloc is more effective than the multivariate correlation aspects of hybrid BEC. Nevertheless, when relying on the ensemble to constrain variables, the multivariate aspects are expected to play an important role to use PBLH observations to constrain the model states. Although we have noticed very small PBLHRF ensemble spread in Fig. 3c over many of the marine stratocumulus regions where high ensemble perturbation correlations are observed as shown in Fig. 9, the real cause of the smaller
As in Fig. 17, but for the normalized bending angle OmA RMSD as a function of impact height.
Citation: Monthly Weather Review 153, 3; 10.1175/MWR-D-24-0141.1
In addition, the confidence intervals with the 95% significant level are calculated from a bootstrap resampling method with a resampling size of 2000 and are shown as shading in Figs. 17 and 18.
5. Conclusions and future work
An accurate PBLH representation is essential for many applications such as forecasting near-surface meteorology, air quality, water and energy cycles, aerosol and cloud chemistry, and climate. This paper focuses on developing a methodology of utilizing PBLH data to improve the PBL thermodynamic representation in the GEOS data assimilation system and building the framework of a global PBLH analysis and monitoring capability for multiple observing systems to help support the assessment of program of record and future PBL observations as recommended in the 2017 Decadal Survey Report.
We first derived PBLH from radiosonde and GNSS-RO refractivity data and discussed the necessity of using consistent model PBLH definitions for comparison with each observing system, since PBLH is sensitive to the observing method and retrieval algorithm. To this end, a new refractivity gradient–based
The strategy and framework for assimilating PBLH data from multiple observing systems were then developed. Unlike the traditional data assimilation method to construct observation operators connecting control/state variables and PBLH variables, both
With the PBLH data assimilation approach adopted in this study, the impacts of PBLH data on the PBL thermodynamic structure can be realized through interaction with other variables via correlation relationships reflected in ensemble BEC. More importantly, PBLH data can help to capture capping inversions, which may present the biggest challenge in PBL profile assimilation. This challenge stems from the uncertainties of BEC, particularly the accuracy of ensemble BEC and the vertical localization length scale. These elements are key to a good representation of the PBL thermodynamic structure. Therefore, we developed the AdjEnsVloc method in this study to improve ensemble BEC by inflating ensemble spread of Tυ and RH and adjusting the vertical localization length scale based on observed PBLH data. Preliminary results are promising, demonstrating an improved representation of capping inversions and beneficial impact on the use of other observations in the lower troposphere. The results also indicate that, in the lower troposphere, the impacts of PBLH data on Tυ and q through AdjEnsVloc are different from their impacts through multivariate correlation relationships in the original BEC and are more important. The combined PBLH data impact through AdjEnsVloc and multivariate aspects of the improved BEC provides the largest benefits to PBL data assimilation.
The present paper is a first step toward developing and testing a comprehensive PBLH data assimilation methodology. One specific follow-on step is to continue to explore the optimal adjustments of ensemble spread and the vertical localization length scale, either from climatology or a physical perspective. A follow-up study will focus on including PBLH data derived from space- and ground-based lidar observations (McGrath-Spangler and Denning 2012; Palm et al. 2021b) as well as from global radar wind profiler measurements (Molod et al. 2015). In addition, methods to retain observation information through using PBLH analysis to adjust model PBL parameterization schemes will be addressed in the follow-up studies.
Acknowledgments.
This study was supported by the National Aeronautics and Space Administration, NNH21ZDA001N-DSI, Decadal Survey Incubation program. Resources supporting this work were also provided by the NASA High-End Computing (HEC) Program through the NASA Center for Climate Simulation (NCCS) at the Goddard Space Flight Center. The authors would also like to thank Dr. Daryl T. Kleist and the three reviewers for their valuable comments and suggestions.
Data availability statement.
The GEOS model and data assimilation system and supporting software are maintained by the Global Modeling and Assimilation Office and publicly available through the NASA Open Source Agreement (NOSA) on GitHub (https://github.com/GEOS-ESM). Changes to GEOS software resulting from this project and used in publications are also available in the GitHub PBLH branch. The data generated in this study include PBL height data and output from the GEOS model and data assimilation system. All the unprocessed output and data are archived and can be accessed on the NCCS Discover Centralized Storage System.
REFERENCES
Amerault, C., A. Castellanos-Morales, B. Blaylock, D. Tyndall, P. Pauley, C. Harten, and A. Sims, 2024: Variational assimilation of boundary layer height measurements. 12th Workshop on Sensitivity Analysis and Data Assimilation, Lake George, NY, U. S. Naval Research Laboratory, 1–14, https://www.adjoint-workshop.org/presentations.
Ao, C. O., D. E. Waliser, S. K. Chan, J.-L. Li, B. Tian, F. Xie, and A. J. Mannucci, 2012: Planetary boundary layer heights from GPS radio occultation refractivity and humidity profiles. J. Geophys. Res., 117, D16117, https://doi.org/10.1029/2012JD017598.
Arnold, N. P., W. M. Putman, and S. R. Freitas, 2020: Impact of resolution and parameterized convection on the diurnal cycle of precipitation in a global nonhydrostatic model. J. Meteor. Soc. Japan, 98, 1279–1304, https://doi.org/10.2151/jmsj.2020-066.
Cucurull, L., J. C. Derber, and R. J. Purser, 2013: A bending angle forward operator for global positioning system radio occultation measurements. J. Geophys. Res. Atmos., 118, 14–28, https://doi.org/10.1029/2012JD017782.
Fowler, A., R. Bannister, and J. Eyre, 2012: A new floating model level scheme for the assimilation of boundary-layer top inversions: The univariate assimilation of temperature. Quart. J. Roy. Meteor. Soc., 138, 682–698, https://doi.org/10.1002/qj.955.
Ganeshan, M., and D. L. Wu, 2015: An investigation of the Arctic inversion using COSMIC RO observations. J. Geophys. Res. Atmos., 120, 9338–9351, https://doi.org/10.1002/2015JD023058.
Ganeshan, M., and Y. Yang, 2019: Evaluation of the Antarctic boundary layer thermodynamic structure in MERRA2 using dropsonde observations from the Concordiasi campaign. Earth Space Sci., 6, 2397–2409, https://doi.org/10.1029/2019EA000890.
Gelaro, R., and Coauthors, 2017: The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2). J. Climate, 30, 5419–5454, https://doi.org/10.1175/JCLI-D-16-0758.1.
Guo, P., Y.-H. Kuo, S. V. Sokolovskiy, and D. H. Lenschow, 2011: Estimating atmospheric boundary layer depth using COSMIC radio occultation data. J. Atmos. Sci., 68, 1703–1713, https://doi.org/10.1175/2011JAS3612.1.
Hilton, F., and Coauthors, 2020: An investigation into the performance of retrievals of temperature and humidity from IASI. 8 pp., https://www-cdn.eumetsat.int/files/2020-04/pdf_conf_p55_s7_37_hilton_p.pdf.
Kleist, D. T., D. F. Parrish, J. C. Derber, R. Treadon, W.-S. Wu, and S. Lord, 2009: Introduction of the GSI into the NCEP Global Data Assimilation System. Wea. Forecasting, 24, 1691–1705, https://doi.org/10.1175/2009WAF2222201.1.
Lewis, J. R., E. J. Welton, A. M. Molod, and E. Joseph, 2013: Improved boundary layer depth retrievals from MPLNET. J. Geophys. Res. Atmos., 118, 9870–9879, https://doi.org/10.1002/jgrd.50570.
Lock, A. P., A. R. Brown, M. R. Bush, G. M. Martin, and R. N. B. Smith, 2000: A new boundary layer mixing scheme. Part I: Scheme description and single-column model tests. Mon. Wea. Rev., 128, 3187–3199, https://doi.org/10.1175/1520-0493(2000)128<3187:ANBLMS>2.0.CO;2.
Louis, J.-F., 1979: A parametric model of vertical eddy fluxes in the atmosphere. Bound.-Layer Meteor., 17, 187–202, https://doi.org/10.1007/BF00117978.
McGrath-Spangler, E. L., and A. S. Denning, 2012: Estimates of North American summertime planetary boundary layer depths derived from space-borne lidar. J. Geophys. Res., 117, D15101, https://doi.org/10.1029/2012JD017615.
McGrath-Spangler, E. L., and A. S. Denning, 2013: Global seasonal variations of midday planetary boundary layer depth from CALIPSO space-borne LIDAR. J. Geophys. Res. Atmos., 118, 1226–1233, https://doi.org/10.1002/jgrd.50198.
McGrath-Spangler, E. L., and A. Molod, 2014: Comparison of GEOS-5 AGCM planetary boundary layer depths computed with various definitions. Atmos. Chem. Phys., 14, 6717–6727, https://doi.org/10.5194/acp-14-6717-2014.
Molod, A., H. Salmun, and M. Dempsey, 2015: Estimating planetary boundary layer heights from NOAA Profiler Network wind profiler data. J. Atmos. Oceanic Technol., 32, 1545–1561, https://doi.org/10.1175/JTECH-D-14-00155.1.
National Academies of Sciences, Engineering, and Medicine, 2018: Thriving on Our Changing Planet: A Decadal Strategy for Earth Observation from Space. The National Academies Press, 716 pp., https://doi.org/10.17226/24938.
Palm, S. P., Y. Yang, U. Herzfeld, D. Hancock, A. Hayes, P. Selmer, W. Hart, and D. Hlavka, 2021a: ICESat-2 atmospheric channel description, data processing and first results. Earth Space Sci., 8, e2020EA001470, https://doi.org/10.1029/2020EA001470.
Palm, S. P., P. Selmer, J. Yorks, S. Nicholls, and E. Nowottnick, 2021b: Planetary boundary layer height estimates from ICESat-2 and CATS backscatter measurements. Front. Remote Sens., 2, 716951, https://doi.org/10.3389/frsen.2021.716951.
Park, S., and C. S. Bretherton, 2009: The University of Washington shallow convection and moist turbulence schemes and their impact on climate simulations with the Community Atmosphere Model. J. Climate, 22, 3449–3469, https://doi.org/10.1175/2008JCLI2557.1.
Putman, W. M., and S.-J. Lin, 2007: Finite-volume transport on various cubed-sphere grids. J. Comput. Phys., 227, 55–78, https://doi.org/10.1016/j.jcp.2007.07.022.
Ravela, S., K. Emanuel, and D. McLaughlin, 2007: Data assimilation by field alignment. Physica D, 230, 127–145, https://doi.org/10.1016/j.physd.2006.09.035.
Salmun, H., H. Josephs, and A. Molod, 2023: GRWP-PBLH: Global radar wind profiler planetary boundary layer height data. Bull. Amer. Meteor. Soc., 104, E1044–E1057, https://doi.org/10.1175/BAMS-D-22-0002.1.
Seidel, D. J., C. O. Ao, and K. Li, 2010: Estimating climatological planetary boundary layer heights from radiosonde observations: Comparison of methods and uncertainty analysis. J. Geophys. Res., 115, D16113, https://doi.org/10.1029/2009JD013680.
Seidel, D. J., Y. Zhang, A. Beljaars, J.-C. Golaz, A. R. Jacobson, and B. Medeiros, 2012: Climatology of the planetary boundary layer over the continental United States and Europe. J. Geophys. Res., 117, D17106, https://doi.org/10.1029/2012JD018143.
Takacs, L. L., M. J. Suarez, and R. Todling, 2018: The stability of incremental analysis update. Mon. Wea. Rev., 146, 3259–3275, https://doi.org/10.1175/MWR-D-18-0117.1.
Tangborn, A., B. Demoz, B. J. Carroll, J. Santanello, and J. L. Anderson, 2021: Assimilation of lidar planetary boundary layer height observations. Atmos. Meas. Tech., 14, 1099–1110, https://doi.org/10.5194/amt-14-1099-2021.
Tassone, C., M. Tsidulko, Y. Zhu, L. Cucurull, G. Manikin, J. McQueen, and G. DiMego, 2012: An assessment of a real-time analysis and its impact on dispersion modeling. Air Pollution Modeling and Its Application XXI, D. Steyn and S. Trini Castelli, Eds., Springer, 391–396.
Texeira, J., and Coauthors, 2021: Toward a global planetary boundary layer observing system: The NASA PBL incubation study team report. NASA PBL Incubation Study Team Rep., 134 pp., https://ntrs.nasa.gov/api/citations/20230001633/downloads/AFridlindPBLTowardsReport.pdf.
Todling, R., and A. El Akkraoui, 2018: The GMAO hybrid ensemble-variational atmospheric data assimilation system: Version 2.0. NASA Tech. Memo. NASA/TM-2018-104606/Vol. 50, 185 pp., https://gmao.gsfc.nasa.gov/pubs/docs/Todling1019.pdf.
Wang, X., C. Snyder, and T. M. Hamill, 2007: On the theoretical equivalence of differently proposed ensemble–3DVAR hybrid analysis schemes. Mon. Wea. Rev., 135, 222–227, https://doi.org/10.1175/MWR3282.1.
Wang, Y., and X. Wang, 2017: Direct assimilation of radar reflectivity without tangent linear and adjoint of the nonlinear observation operator in the GSI-based EnVar system: Methodology and experiment with the 8 May 2003 Oklahoma City tornadic supercell. Mon. Wea. Rev., 145, 1447–1471, https://doi.org/10.1175/MWR-D-16-0231.1.
Whitaker, J. S., T. M. Hamill, X. Wei, Y. Song, and Z. Toth, 2008: Ensemble data assimilation with the NCEP Global Forecast System. Mon. Wea. Rev., 136, 463–482, https://doi.org/10.1175/2007MWR2018.1.
Zhu, Y., and I. Navon, 1999: Impact of parameter estimation on the performance of the FSU global spectral model using its full-physics adjoint. Mon. Wea. Rev., 127, 1497–1517, https://doi.org/10.1175/1520-0493(1999)127<1497:IOPEOT>2.0.CO;2.
Zhu, Y., R. Todling, and N. Arnold, 2022: Observation impact and information retention in the lower troposphere of the GMAO GEOS data assimilation system. Mon. Wea. Rev., 150, 2187–2205, https://doi.org/10.1175/MWR-D-21-0334.1.