1. Introduction
The issue of land–atmosphere interaction is one that has persisted for many years (Betts et al. 1996; Betts and Dias 2010; Eltahir 1998; Wood 1991; Yeh et al. 1984), yet is still inadequately understood. It is well recognized that the land surface state (i.e., soil moisture, temperature, albedo, roughness, overlying vegetation state, and thickness) can impact climate at local-to-regional scales via complex physical controls on the partitioning of surface turbulent fluxes. However, the degree to which these land–atmosphere interactions can influence weather varies regionally, seasonally, and even interannually, owing to the effect of other factors that modulate the ability of climate signals to persist and propagate through the hydrologic branch of the coupled land–atmosphere system (Dirmeyer 2006; Douville 2003). The scale of these interactions (and of the complete land–atmosphere feedback cycle) can range (often, as a function of the scale of surface spatial heterogeneity) from many kilometers to many hundred kilometers, making these processes difficult to observe, and hence, quantify and model. Findings from a limited number of cases suggest that coupling can play a role in monsoon (Douville 2002; Ferranti et al. 1999; Webster 1983), heatwave (Fischer et al. 2007b), and drought intensification (Atlas et al. 1993; Fennessy and Shukla 1999; Sud et al. 2003). Improving our ability to recognize, understand, and properly parameterize the many mechanisms that constitute land–atmosphere interaction (coupling) is viewed by the World Climate Research Programme (WCRP) Global Energy and Water Cycle Experiment (GEWEX) as a key step toward improving short-to-medium-range weather forecast skill. A multiyear global categorization of coupling is unprecedented and only now is feasible with remote sensing.
In the past, coupling research has largely focused on the analysis of large-scale model outputs. Betts (2004) developed many of the coupling diagnostics using the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40). One such diagnostic, the correlation strength between soil moisture (SM) and lifting condensation level (LCL), evaluates the model’s ability to capture soil moisture effects on boundary layer (BL) cloud development and, therefore, soil moisture–precipitation feedback (rainfall → increased SM → increased evaporative fraction → increased relative humidity, more BL clouds, lowered LCL → reduced net surface radiation and/or rainfall). Koster et al. (2004) showed that coupling “hot spots” exist in select regions of the world, akin to El Niño 3.4 (Trenberth and Hoar 1997) for sea surface temperatures, which impart more significant influence on climate than others. The GEWEX Global Land–Atmosphere Coupling Experiment (GLACE; Dirmeyer et al. 2006; Guo et al. 2006; Koster et al. 2006) quantified for the first time the range and (global) distribution of coupling strength within 12 AGCMs. GLACE showed that the strength of land surface–atmosphere coupling and its representation can affect a model’s ability to simulate climate predictability. However, even in the hot spot locations, there was large variability in the coupling strength between the 12 models (Koster et al. 2004). With only limited observational evidence with which to evaluate the realism of AGCM performance, GLACE concluded that the absence of objective quantifications of coupling strength from observational data was as a “major obstacle to the evaluation of model performance” and arguably for the next generation of improvement to AGCMs (Koster et al. 2006). The following questions remain unanswered:
How representative of real-world coupling is the distribution of coupling represented by the global climate models?
What is the accuracy of the AGCMs relative to one another?
Can land data assimilation be done correctly without accurately representing the coupling between land surface and atmosphere?
Findell and Eltahir (2003a, hereafter F&E2003) introduced a framework (Fig. 1) for describing atmospheric controls on soil moisture–boundary layer interactions that hinged on two lower-atmosphere metrics: the convective triggering potential (CTP) and low-level humidity index (HIlow, or simply HI) computed from radiosonde-observed atmospheric profiles. The framework was developed using results from a simple single-column boundary layer model (SCM) experiment performed at Lincoln, Illinois (ILX; 40.2°N, 89.3°W) and later transferred to 89 sites across the conterminous United States (Findell and Eltahir 2003b). Given early morning measurements of HI and CTP, each station was classified daily into one of four categories: atmospherically controlled, wet advantage (positive feedback), dry advantage (negative feedback), and transitional. The framework’s application, however, was ultimately constrained by the limited spatial and temporal coverage of the global radiosonde network. Indeed, operational radiosonde launch times (0000 and 1200 UTC daily) fall outside the critical observation time [approximately 0600 local time (LT), or prior to the degradation of the nocturnal stable boundary layer] for most of the world (except western Europe, West Africa, and western United States and Canada).
The F&E2003 CTP–HIlow framework for categorizing atmospheric profiles into one of four convective regimes: atmospherically controlled, wet soil advantage, dry soil advantage, and transition (reproduced from F&E2003, their Fig. 15).
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
Since then, we have seen dramatic advancements in the temporal and spatial coverage and accuracy of satellite remote-sensing-based estimates of soil moisture (Owe et al. 2008), rainfall (Joyce et al. 2004), and atmospheric temperature and humidity (Susskind et al. 2011), due in large part to the National Aeronautics and Space Administration (NASA) Aqua platform (launched 4 May 2002), which carries the Atmospheric Infrared Sounder (AIRS) and Advanced Microwave Scanning Radiometer for Earth Observing System (EOS) (AMSR-E), among other sensors. Coincidentally, AIRS is best suited (most accurate) for retrievals at the time and condition of interest for coupling: early morning and clear sky, respectively (Ferguson and Wood 2010a; Susskind et al. 2003). Collectively, these new observations enable a much more thorough and spatially comprehensive analysis than was possible at the time of F&E2003. As of yet, they remain an untapped resource.
In this study, we revisit the work of F&E2003 using satellite-based observations of HI and CTP from AIRS, soil moisture from AMSR-E, and the U.S. Climate Prediction Center (CPC) merged satellite rainfall product (CMORPH) (Joyce et al. 2004). We are concerned that a study based only on remote sensing data may lead to conclusions that are uncertain because of the data sources. Thus, we duplicate the analysis using side-by-side intercomparisons with model output from the recently released (and largely unvalidated) NASA Modern Era Retrospective Analysis for Research and Applications (MERRA; Rienecker et al. 2008b). With these new data streams, we systematically reevaluate the framework of F&E2003, discuss modifications necessary for global application, and, ultimately, provide global maps of the predominant coupling feedback signal (or lack thereof) for the period 2002–09. The specific motivation of our work is threefold: to validate the underlying datasets, classify global land areas by their dominant coupling feedbacks within a CTP–HI framework, and test the consistency of these patterns with those from GLACE.
The paper is organized as follows. Section 2 describes the datasets and methods. Section 3 defines the study period and domain. The results are summarized in section 4. Specifically, section 4a–d includes summaries of both the original (F&E2003) and revised (this study) CTP–HI frameworks, along with the respective regime categorizations. The applicability of HI and CTP for estimating SM, rainfall occurrence, and rainfall depth is explored in section 4e. Section 5 provides an overview of the ways in which coupling manifests itself in the real world, with descriptions of each possible feedback pathway. Summary and conclusions are given in section 6, followed by recommendations for future research in section 7. The AIRS and MERRA dataset validation is contained in the appendix.
2. Data and methods
a. Remote sensing products
1) AIRS
AIRS is on board NASA Aqua, which offers a sun-synchronous orbit with equatorial crossings of ~1330 and ~0130 LT on ascending and descending orbital nodes, respectively, with global repeat on the order of 2–3 days. The instrument provides atmospheric radiance measurements at 13.5-km (nadir) resolution at 2378 channels across three windows in the infrared spectrum: 3.74–4.61, 6.20–8.22, and 8.8–15.4 μm. AIRS is accompanied on Aqua by the Advanced Microwave Sounding Unit (AMSU), which measures microwave radiances at 15 channels between 23 and 89 GHz for a 45-km field of view (FOV) containing the AIRS 3 × 3 footprint array. AMSU measurements play a large role in the cloud-clearing algorithm (Chahine et al. 2006) that enables successful soundings 47.5% and 72.3% of the time over land and ocean, respectively (Susskind et al. 2003). AIRS was designed to provide sounding data with vertical resolution and accuracies on par with point-based radiosonde observations (raobs; 1-K rmse in 1-km tropospheric layers and 20% rmse in 2-km tropospheric layers for water vapor) for the purpose of improved weather prediction through data assimilation (Susskind et al. 2006). Importantly, AIRS overcomes the following limitations of the current raob network: 1) spacing that is too large to isolate regions of probable convection and 2) nonideal sampling times (0000 and 1200 UTC, globally). The accuracy of the AIRS retrievals largely depends on the accuracy of the measured infrared radiances, prescribed atmospheric transmittance functions, cloud-clearing algorithm, and inversion algorithm (Divakarla et al. 2006). Clear correlations between retrieval uncertainty and bias with the presence of clouds have been reported (Ferguson and Wood 2010a; Gao et al. 2008). Nevertheless, with the exception of near-surface retrievals, which in a multiyear intercomparison study using observations from ~1500 National Climate Data Center (NCDC) stations over the continental United States were found to have an rmse of 3.8 K and 32.0% specific humidity (Ferguson and Wood 2010a), AIRS atmospheric column layer averages are close to meeting the expected uncertainty goals given above (Divakarla et al. 2006; Tobin et al. 2006).
The AIRS–AMSU Level 2 Support Product Dataset (AIRX2SUP; Susskind et al. 2011) version 5 (from 1 September 2002 to 30 September 2007) and version 5.2 (from 1 October 2007 to 31 December 2008) provides retrievals of column water vapor and temperature at the AMSU horizontal resolution (45 km at nadir) and 100-pressure-level vertical resolution (from 1100 to 0.016 hPa) with measurements at increments of ~26 and ~14 hPa from 1100 to 700 hPa and 700 to 100 hPa, respectively (Olsen 2007). The version 5 retrieval algorithm implements a radiative transfer algorithm (RTA) that accounts for nonlocal thermodynamic equilibrium (NLTE) effects on the shortwave channels and provides, for the first time, case-by-case product error estimates (Susskind et al. 2011). For this study, only retrievals over land (SurfClass = 1) that satisfied the following quality checks were admitted: 1) the temperature profile from 50 hPa above ground level (AGL) to top of atmosphere (TOA) is of quality “good” or “best” [i.e., PGood ≥ (PsurfStd−50 hPa)], and 2) the moisture profile is of quality good or best (i.e., QualH2O < 2). We note that because of increasing radiometric noise in AMSU channel 4, an additional AIRS spectra-only cloud-cleared product (AIRS2SUP) has been produced coincidentally along with AIRX2SUP since June 2007. At the time of writing, the use of AIRX2SUP was still advised for post-June 2007 research applications (E. Olsen 2011, personal communication).
(i) HI and CTP
For each metric (HI and CTP), we first produced an intermediate 0.125° gridded product, selecting the nearest neighbor AIRS retrievals from within a 50-km great circle search radius of a cell, and then resampled to 1.25° resolution using bin averaging. Since we are interested in early morning soundings only, only the descending overpass is considered. If multiple retrievals (i.e., from multiple descending overpasses) were available for a given day, as occurs at high latitudes, the HI–CTP pair with the lowest surface air temperature (TSurfAir) was selected. In most cases, this should correspond to the sounding that is closest in time to local sunrise.
(ii) Cloud fraction
The AIRX2SUP provides a thermal infrared (TIR) cloud fraction (TotCld_4_CCfinal) estimate valid for the AMSU FOV (45 km) that has previously been shown to agree well (τKendall = 0.41) during daytime (ascending overpass) with cloud albedo derived from Clouds and the Earth’s Radiant Energy System (CERES) shortwave downward radiation (Ferguson and Wood 2010a). We produced a daily early morning cloud fraction (fc) map at 1.25° spatial resolution by bin averaging all available descending retrievals.
2) AMSR-E
AMSR-E, also on board NASA Aqua, makes dual-polarized passive radiation measurements at six frequencies (6.9, 10.7, 18.7, 23.8, 36.5, and 89.0 GHz).
(i) Rain rate and type
The AMSR-E Level 2B version 10 swath product (AE_Rain; Adler et al. 2007) provides instantaneous estimates of rain rate and rain type (convective or stratiform) for all ice- and snow-free land and ocean between 70°S and 70°N at a resolution of 5.4 km. The retrieval applies the physically based Goddard Space Flight Center (GSFC) profiling algorithm (GPROF) methodology (McCollum and Ferraro 2003) over both land and ocean, but limits over land the potential hydrometeor profile to 36 of the several thousand GPROF profiles (Kummerow et al. 2001) that satisfy an empirically derived temperature depression–rain rate relationship (Wilheit et al. 2003). The convective percentage of rainfall ranges from 0 (completely stratiform) to 100 (completely convective) and is computed as a function of brightness temperatures at 89.0, 36.5, and 10.7 GHz using an empirically derived linear regression (McCollum and Ferraro 2003).
Using all available swaths (ascending and descending) from January 2003 to December 2009 we calculated on a 1.25° geographic grid for all land areas the monthly climatology of convective activity, defined by the frequency of rain events characterized by a convective rainfall percentage in excess of 45%. We applied this particular threshold because it was found over the continental United States to mark the local minimum between modes in the bimodal AERain convective percentage distribution. We then defined a “convective season” for each 1.25° cell composed of the top ranking (most convective) calendar months whose sum did not exceed 80% of the annual event total, but that individually accounted for greater than 10% of the annual event total. There was no case in which a single calendar month was found to exceed 80% of the annual event total. Furthermore, no requirement was imposed that the convective “season” be unimodal (i.e., persist over consecutive months).
The monthly global coverage and duration (in months per year) is illustrated in Figs. 2 and 3, respectively. Season lengths range from 1–4 months and are not necessarily continuous. For example, at Cotonou, Benin (6.4°N, 2.4°E), where the annual convective rainfall distribution is bimodal, the classification correctly identifies the season as May, June, July, and September. The season duration generally represents a conservative estimate, partly owing to the fact that the underlying unit is a month. For example, the convective season at the Department of Energy Atmospheric Radiation Measurement (ARM) Central Facility in Lamont, OK (36.6°N, 97.5°W) is June–August. The approach does break down for the deserts of the world (e.g., Sahara, Arabian, Turkestan, and Taklamakan–Gobi), where it is shown (Fig. 3) to yield spurious results.
Monthly global maps of convective season coverage (shaded) over land derived from the AMSR-E rain rate and rain type product.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
Length (months) of convective season over land following from Fig. 2. Note that the months inclusive in the season are not necessarily continuous.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
Alternatively, the convective season may be defined using the MERRA convective precipitation estimates (PRECCON), an approach that we found yielded similar results over the conterminous United States (not shown).
(ii) Surface soil moisture
Estimates of surface soil moisture, representative of the uppermost 2 cm, may be derived on a global basis from AMSR-E C- (6.9 GHz) or X-band (10.7 GHz) dual-polarized brightness temperatures (Tb). Here, we use 0.25° Level 3 surface soil moisture estimates derived using the Land Surface Parameter Model (LPRM; Owe et al. 2008). The LPRM is a microwave radiative transfer emission model with three primary parameters: soil moisture, vegetation water content, and soil/canopy temperature. Given that the attenuation of microwaves by vegetation increases with vegetation density and microwave frequency, retrievals from the lower-frequency C band serve as the default. It is well known that severe radio frequency interference (RFI) contaminates the C-band signal over certain (developed) regions of the world (Li et al. 2004). In these regions, C-band retrievals were replaced with their X-band-derived counterparts. The LPRM is not valid for regions with high (>5%) surface water coverage, frozen (T < 273 K) or snow-covered ground, or dense vegetation (vegetation optical thickness > 0.8), therefore pixels that satisfy these qualities have been culled. Only estimates of “nominal” or higher accuracy are used. Additionally, we apply the global porosity map used in the LPRM dielectric mixing module to screen out any cell that does not exhibit dry down (i.e., remains in a saturated state for 5 or more of the previous 7 days). Previously, the LPRM SM was shown to compare well with in situ measurements at 5-cm depth (Wagner et al. 2007).
Since for this study we are interested in the initial soil moisture state at sunrise, we only consider the descending (early morning: ~0130 LT) overpass. In the case that there are multiple descending overpasses available, as occurs at high latitudes, only the last successful retrieval for each UTC day is considered. At 70°N, the last descending overpass occurs at approximately 0700 LT. Ultimately, the quality-cleared 0.25° Level 3 product is resampled by bin averaging to 1.25°.
3) CMORPH
The U.S. Climate Prediction Center morphing (CMORPH; Joyce et al. 2004) technique produces multisensor rainfall analyses at very high spatial (12 × 15 km2) and temporal (30 min) resolution, with coverage from 60°S to 60°N. Specifically, this technique uses rainfall estimates from existing microwave rainfall algorithms, derived from low orbiter satellite microwave observations, transports these features in time (when they are unavailable) via spatial propagation information obtained from geostationary infrared data, and “morphs” these features into a shape and rainfall intensity. CMORPH was found to best depict spatial patterns and temporal variations of rainfall in separate intercomparison studies over the conterminous United States (Tian et al. 2010) and China (Shen et al. 2010), relative to the following satellite-based rainfall products: Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN; Hsu et al. 1997, 1999; Sorooshian et al. 2000), Naval Research Laboratory (NRL) blended (Turk and Miller 2005), and Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis (TMPA) 3B42 (Huffman et al. 2007). Only the Global Satellite Mapping of Precipitation (GSMaP: Aonashi et al. 2009; Kubota et al. 2007; Ushio et al. 2009) had similarly high probability of detection and comparable threat scores (Tian et al. 2010). Notably, CMORPH is not gauge-corrected and, therefore, does suffer from biases, particularly because of overestimation of rain rates in excess of 20 mm day−1 (Tian et al. 2010). CMORPH average daily rainfall totals are biased by +88% and +32% during the summer over the west and east United States, respectively (Tian et al. 2010).
In this study, we resample and bin average 3-hourly 0.25° CMORPH estimates to produce 3-hourly 1.25° estimates. We then compute an afternoon-plus-evening rainfall total by summing the rainfall depths of the +3, +4, +5, and +6 (3 hourly) time steps subsequent to the AIRS AM (descending) overpass. For most points, this corresponds to the periods 1030–2230 or 1330–0130 LT—the peak hours of convective activity over land (Zipser et al. 2006). The CMORPH data record extends from 7 December 2002 to present.
b. Atmospheric reanalysis
1) MERRA HI and CTP
The MERRA (Rienecker et al. 2008a, 2011) is a NASA global (except Greenland) atmospheric–land reanalysis for the satellite era (1979–present) using version 5 of the Goddard Earth Observing System Data Assimilation System (GEOS-5 DAS). We use instantaneous temperature and specific humidity estimates from the 3-hourly 42-level reduced spatial resolution (1.25°) product (MAI3CPASM), which incorporates the Incremental Analysis Updates (IAU; Bloom et al. 1996) of the GEOS-5 system, to calculate HI and CTP [as per Eqs. (1) and (2)]. Notably, the pressure level increments from 1000 to 700 hPa and 700 to 100 hPa are a mere 25 and 50 hPa, respectively. Global raobs and AIRS radiances constitute two of the many data streams assimilated by GEOS-5, the temporal, vertical, and spatial extent of which is unclear because of record keeping gaps and/or the lack of a readily accessible metadata product.
2) MERRA SM
We use the MERRA topsoil-layer wetness (GWETTOP, or MERRA SM) estimates to provide a secondary (to LPRM SM) assessment of the strength of correlation between HI, CTP, and SM. We obtain SM from the hourly, time-averaged land surface diagnostics datastream (MAT1NXLND), which is distributed at the model’s native horizontal resolution of 0.5° latitude by 0.67° longitude. We first average the hourly values to 3-hourly, and then resample to 1.25° resolution by bin averaging. MERRA SM is an output of the catchment land surface model (Koster et al. 2000). It was parameterized to match moisture dynamics of the uppermost 2-cm soil layer, similar to LPRM SM.
3. Study period, extent, and resolution
We limit our study to the first 7 yr of AIRS data availability (September 2002–December 2009) and the geographic coverage of AMSR-E rain (±70°N/S). Since our intent is to evaluate regionally the potential for land surface conditions to influence the development of convection, we confine our analysis to days of the convective season (see section 2, Figs. 2 and 3) with relatively clear-sky mornings, as defined by an AIRS early morning (~0130 LT, descending overpass) TIR cloud fraction (fc) ≤ 0.4. We emphasize here that this cloud screening serves as the only mechanism by which synoptic weather events and other large-scale forcings are screened from our analysis. Over the southern Great Plains of the United States, only fifty percent of AIRS early morning quality-cleared (see section 2) retrievals during the convective season meet this criterion. Investigations of the CTP–HI–SM–rainfall correlation space (section 4e) have additional constraints imposed on them because of the geographic coverage of CMORPH (±60°N/S). All analyses are conducted on a 1.25° geographic grid—that of the MERRA (MAI3CPASM) atmospheric fields.
4. Results
Correspondence between the AIRS satellite observations, MERRA reanalysis, and global radiosonde soundings is demonstrated and quantified in the appendix. Using raobs, we found that AIRS HI was better correlated in general, with lower rmse but with larger absolute biases than MERRA (Figs. A1 and A2; Table A1). Patterns of bias in AIRS HI over the conterminous United States are consistent with previously published patterns of bias in AIRS near-surface relative humidity (Ferguson and Wood 2010a). As noted in Ferguson and Wood (2010a), it is difficult to generalize the accuracy of AIRS because the retrieval skill is a complex function of multiple surface and atmospheric scene characteristics. Regionally, AIRS and MERRA HI datasets were better correlated with raobs where specific humidity was low (Fig. A5). On the other hand, MERRA CTP was better correlated, with lower rmse and lower absolute bias than AIRS. Relative to MERRA, AIRS CTP showed much greater atmospheric stability (lower CTP), which may or may not be an artifact of noise (temperature inversions) in the AIRS profile. Overall, AIRS–MERRA correlations were higher for HI than CTP in 16 out of 18 global regions considered. The coefficient of variation was lower (higher) in AIRS for HI (CTP).
In a separate assessment, we calculated the MERRA error characteristics averaged over all cloud conditions and sampling times, regardless of AIRS overpass availability. It was shown (Fig. A4; Table A2) that the effect was to reduce MERRA HI and CTP bias and rmse, and to increase correlation. In fact, MERRA (both HI and CTP) was found to be nearly unbiased in some regions. The extended geographic coverage afforded by the 3-hourly global model revealed a clear zonal pattern: in the tropics, both MERRA HI and CTP were poorly correlated with raobs. While raob coverage is dense in some regions (e.g., western Europe) and could be used to bias correct AIRS and MERRA, coverage is inadequate and correlation is too low to do so globally. For this reason, we used AIRS and MERRA data “as is.”
Having confirmed the generally good (albeit biased) quality of AIRS data, and hence the capacity to carry out large-scale CTP–HI analyses, the next step is to test the suitability of the F&E2003 framework for global application. In this section, we include a brief overview of the original F&E2003 framework, its shortcomings, and details of a newly developed classification methodology that successfully overcomes these limitations.
a. F&E2003 CTP–HI framework revisited
The CTP–HI framework of F&E2003 (see also Fig. 1) defines first a region in which the surface soil moisture state is likely to influence the triggering of convection (CTP > 0 J kg−1, 5°C ≤ HI ≤ 15°C) and secondly, within that region, a set of distinct conditions in which convection is either (i) equally likely over wet or dry surface states (transition), (ii) favored over wet soils, or (iii) favored over dry soils. Beyond this region, or when CTP < 0 J kg−1, HI < 5°C (too wet), or HI > 15°C (too dry), convection is considered independent of the effects of energy partitioning at the surface. They developed and refined the framework using an SCM initialized with clear-sky early morning (~0600 LT) raob profiles from Lincoln, Illinois (station ILX: 40.2°N, 89.3°W). The SCM was run for three summers’ worth of data, twice for each profile: once with prescribed anomalous dry (SM = 20%) conditions and again with anomalous wet (SM = 100%) conditions.
In their follow-up manuscript, Findell and Eltahir (2003b), a station classification protocol was introduced and applied at 89 raob stations across the conterminous United States. The classification was conducted on a yearly basis using all available summertime (June–August) soundings, but no fewer than 10 yr per station, from 1957–98. If a station had more than 80% of samples per summer fall into atmospherically controlled regions of the CTP–HI space, it was classified accordingly. Otherwise, it was submitted for further evaluation. Specifically, a two-level scheme was applied in which stations with greater than 50% of the remaining nonatmospherically controlled days contained entirely in either wet-advantage, dry-advantage, or transition regions were given a “level-1” designation of the respective regime. Finally, using the percentage of years classified as level-1 wet or dry advantage at each station, rough boundaries were drawn (Findell and Eltahir 2003b, their Fig. 2), dividing the conterminous United States into geographic regions of varying feedback sign and strength.
Ferguson and Wood (2010b,c) recently demonstrated sensitivity of the Findell and Eltahir (2003b) protocol to the number of years, as well as the specific years considered, especially where interannual variability was substantial. Because Findell and Eltahir (2003b) did not limit their classification to a single period of overlap amongst the stations, their regional categorizations should be used with discretion. Moreover, given the sparse nature of the raob network, the suggested boundaries (Findell and Eltahir 2003b, their Fig. 2) are prone to imprecision and should be considered as being approximate.
b. Assessing the suitability of F&E2003 for global analyses
Spatially comprehensive satellite-based HI and CTP observations from AIRS afford us with an opportunity to overcome the data coverage constraints underlying Findell and Eltahir (2003b). Here we evaluate whether the F&E2003 framework is robust enough for application globally.
First, we revisit four of the stations singled out in the original F&E2003 study—Desert Rock, NV (36.6°N, 116.0°W); Tampa, FL (27.7°N, 82.4°W); Albuquerque, NM (35.0°N, 106.6°W); and Topeka, KS (39.1°N, 95.6°W)—each of which reside in a different convective regime. Figure 4 shows the CTP–HI space populated with AIRS data from the study period, with the F&E2003 (Fig. 1) framework overlain. An important improvement on F&E2003 is the addition of surface soil moisture and rainfall information, here from LPRM and CMORPH, respectively. Upon viewing Fig. 4 two features stand out: 1) there are a large number of rain samples that fall in CTP–HI subspace previously labeled “too dry” (HI > 15°C) or “too stable” (CTP < 0 J kg−1) for rainfall in F&E2003 and 2) there is a distinguishable wet-to-dry gradient in SM with increasing HI and CTP that verifies the alignment of “wet-soil-advantage” and “dry-soil-advantage” regions in F&E2003 (Fig. 1)—this issue will be explored further in section 4e. The precise number of samples with rain as determined from CMORPH—and for additional emphasis—the NOAA/CPC unified daily gauge analysis (CPC-UNI; Xie et al. 2007; Chen et al. 2008) are listed for each convective regime in Table 1. Finding (1) is found to hold true over a much greater extent than these points alone. Figure 5 shows that the median AIRS (MERRA) HI for all rainy days exceeds 15°C for latitudes between 20°–60°N (25°–60°N) and 15°–55°S (20°–55°S). [Additional (10th, 25th, 75th, and 90th) HI percentile values are provided along with these median values in Table A4.] This suggests that the F&E2003 framework may have been artificially influenced by either the local climate (Lincoln, Illinois) for which it was developed or the specific SCM employed, rendering it, therefore, unsuitable for global application.
Sample AIRS CTP–HI scatterplots at four radiosonde sites, each located in one of four distinct convective regimes, formatted to match the style of Findell and Eltahir (2003b, their Figs. 4 and 5). The site classifications according to the method of this study (Findell and Eltahir 2003b) are as follows: (a) dry advantage (atmospherically controlled), (b) wet advantage (wet advantage), (c) atmospherically controlled (dry advantage), and (d) transitional (transitional). Superimposed grid boxes demarcate the classification framework of F&E2003, as also shown in Fig. 1. These observations have been bias corrected using coincident radiosonde profiles. The corrections (AIRS HI and AIRS CTP) and corresponding convective seasons are as follows: Desert Rock (+15.3°C, +78 J kg−1, May, Jul, and Aug), Tampa (−1.4°C, +55 J kg−1, Jun–Aug), Albuquerque (−10.2°C, −145 J kg−1, Jun–Sep), and Topeka (−1.5°C, +88 J kg−1, May, Jun, Aug, and Sep). The coincident LPRM SM is indicated by shading only for days with afternoon-plus-evening rainfall. Note that the plotting scale on (a) is inconsistent with (b)–(d). Also, the SM color bar varies by station.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
An accompaniment of Fig. 4. For each radiosonde site, the rainfall frequency for each of the four convective regimes computed from CMORPH afternoon-plus-evening and CPC-UNI version 1.0 daily rainfall data. The CPC-UNI data were obtained at 0.25° resolution for the continental United States and resampled by box averaging to 1.25°. The daily aggregation ends at 1200 UTC. Accordingly, for these stations over the United States, we use the subsequent day’s data.
(a) Latitudinal median of AIRS (solid) and MERRA (dashed) HI and (b) CTP. The “R” and “NR” denote the medians of sample subsets conditioned on the occurrence or absence of subsequent afternoon-plus-evening rainfall, respectively. Specifically, a decision threshold of 0.001 mm was applied to the CMORPH 1.25° afternoon-plus-evening rainfall. Rainfall-screened subsets are limited in extent by the coverage of CMORPH, or ±60°N/S. The vertical gray line at HI = 15°C in (a) represents the previously suggested (F&E2003) upper threshold of surface influence on the triggering of convection.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
Further analyses revealed one additional inconsistency of the F&E2003 framework (Fig. 1). Specifically, the assumption of independence implied by the rectangular CTP–HI bins of the F&E2003 framework was not substantiated by the AIRS and MERRA datasets. To the contrary, Fig. 6 shows that HI and CTP are significantly correlated over much of the world. The correlation was strongest for areas characterized by median HI and CTP in excess of 20°C (dry) and 100 J kg−1 (atmospherically unstable), respectively. The correlations in both AIRS and MERRA (not shown) were found to increase linearly with median HI and CTP in the range of 0° to 20°C and −100 to 250 J kg−1, respectively. For example, for AIRS the slope was 0.02°C−1 (τKendall /HI) or 0.006 (J kg−1)−1 (τKendall /CTP).
CTP–HI correlation (τKendall) as computed from the (a) AIRS and (b) MERRA datasets, along with (c) the frequency histogram of these correlations globally. The absence of color (i.e., white) indicates cells with insufficient (n < 40) AIRS observations.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
c. Modified CTP–HI framework
To overcome the inconsistencies in the real-world representativeness of F&E2003 (discussed above), a new, globally applicable approach was developed. The first step in transferring the work of F&E2003 beyond the conterminous United States was to define a global convective season. This we accomplished using the AMSR-E rain rate and type product (see section 2). Next, given 1) the latitudinal variability in HI–rainfall characteristics (Fig. 5), 2) the latitudinal (and regional) variability in data error characteristics (Tables A1 and A2), and 3) the inability (due to the lack of raob coverage) to satisfactorily bias correct AIRS or MERRA consistently on a global basis, we needed to develop a framework that did not impose a fixed HI–CTP framework (as in F&E2003; Fig. 1) and could account inherently for the error structure of the data sources. Figure 7 includes a detailed flowchart of an approach that satisfies these requirements.
(a) CTP–HI convective regime classification protocol designed and implemented in this study. (b) Example (taken from 35° to 40°N) frequency distributions of AIRS HI and (c) CTP for each of the four convective regime classes (classified using AIRS data) relative to respective zonal distribution (30° to 45°N).
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
This “regime classification system” (Fig. 7) is based on the relative distribution of HI and CTP at the location (cell) being classified, compared to the distribution of HI and CTP in the bounding 15° latitudinal zone. First, the zonal HI–CTP distributions are calculated at intervals of 5° from 70°N to 60°S—the same zones used earlier in the HI–CTP validation (Tables A3 and A4). Then, cells within the central 5° band are classified using the cumulative distribution composed of all other cells in the encompassing 15° zone. The framework’s eight input requirements are the 25th, 50th, and 75th percentiles, and standard deviation of HI and CTP for the cell being classified and the encompassing zone. As in F&E2003, the classification does not explicitly account for advection and feedbacks to the mesoscale circulation. In other words, the time series of instantaneous AIRS observations for each cell are treated independently (uncorrelated) of adjacent cells.
The classification proceeds as follows. First, the “atmospherically controlled” region is determined, defined by cells with standard deviations of HI and CTP both larger than the respective zonal median values. Here, their standard deviations are taken to imply synoptic variability. While it is feasible for the land surface to play an important role in convection within regions of strong atmospheric variability, differentiating these cases would require the use of an additional diagnostic—one that provided an indication of the flux partitioning at the surface (e.g., evaporative fraction). Hence, the method of determination of atmospherically controlled regions excludes areas with strong synoptic controls on precipitation from having a land component. If the cell’s HI histogram is particularly tight, as determined by whether the 25th and 75th percentiles are greater than and smaller than the zonal values, respectively, then the pixel is reclassified as “transitional.” All remaining cells are classified into “wet advantage,” “dry advantage,” or “transition,” depending on whether the histograms of HI–CTP have a tight wet/low bias against the zonal values, a tight dry/high bias, or tight/nonsubstantial bias. Refer to Fig. 7 for detailed classification criteria. If the pixel remains unclassified, it is classified as transitional. In fact, classification by default accounts for 55% (64%) of the total global transitional regime coverage derived from AIRS (MERRA) data. We note in our classification scheme that cells initially classified as atmospherically controlled may later be reclassified as transitional (refer to Fig. 7). Perhaps a more conservative convention would have been to prioritize atmospherically controlled classification, as in F&E2003. Implementation of this classification loophole affects only a limited area, including West Africa, southern India, central Brazil, Tibet, and the Sierra Madre Oriental mountain range of Mexico. Ultimately, the fact that West Africa would have otherwise been classified as atmospherically controlled motivated us to implement the current protocol. Previous studies have shown both positive (Van den Hurk and Van Meijgaard 2010) and negative (Taylor and Ellis 2006) feedback signals in West Africa, as well as strong synoptic controls (e.g., Nicholson 2009). By latitude, the loophole results in a transitional regime that is 3%–5% drier (SM) than would have otherwise been the case.
The decision tree and criteria (Fig. 7a) are the product of a result-based approach, using the F&E2003 regime boundaries over the conterminous United States (Findell and Eltahir 2003b, their Fig. 2) as our training target. Throughout the development process, we remained mindful of the inexact nature (as discussed earlier) of the F&E2003 boundaries, due in part to the sparse nature of the raob network employed in that study.
d. Global maps of coupling
Figures 8a,b show the global maps of derived convective regimes generated by applying the classification method of this study, using AIRS and MERRA, respectively. For comparison, Figs. 8c,d provide the global maps attainable via a strict application of the original F&E2003 framework. The global convective regime composition for each of the data–classification method pairs (or panel of Fig. 8) is listed in Table 2. Figure 9 translates the spatial maps of Figs. 8a,b, as well as their differences, into latitudinal terms. Regional differences in the AIRS- and MERRA-based classification results (using the method of this study) are provided in Fig. 10 for most regions of the world, as well as select hydrological basins. Unless specified, reference to AIRS- and MERRA-based regime classification maps shall imply those derived using the method of this study.
(a) AIRS- and (b) MERRA-derived global regime classification for 2002–09 convective seasons, following the classification protocol of this study (Fig. 7). (c),(d) As in (a)–(b), but using the unmodified classification scheme of F&E2003. Only AIRS-available days were included. The absence of color (i.e., white) indicates an insufficient number (n < 40) of AIRS retrievals. On a global basis, 200 ± 118 days contributed to the classification, with a median of 166.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
Breakdown of the global coverage (%) of the four regime classes for each of the data source–classification method pairs (panels of Fig. 8).
Zonal composition of convective regimes derived from (a) AIRS and (b) MERRA, as well as (c) their differences (AIRS minus MERRA). Note that the fractional coverage of land declines approximately from north to south, leaving fewer than 1000 and 500 (1.25°) cells at 15°N and 40°S to constitute this figure, respectively.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
(a) AIRS- and (b) MERRA-derived regime compositions for select regions and hydrologic basins. Acronyms for the regions are defined in Table A1. For the basins, the following abbreviations were applied: AMU = Amur, CHA = Changjiang (Yangtze), CON = Congo (Zaire), DAN = Danube, GAN = Ganges, MEK = Mekong, M–D = Murray–Darling, and VOL = Volga.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
The AIRS-based classification yielded greater transitional and less atmospherically controlled coverage than MERRA (Fig. 9). This tendency is most pronounced between 25°–40°N and 20°–45°S with ~16% and ~27% more transitional classifications, respectively, and for 25°N–20°S, with ~13% fewer atmospherically controlled classifications. Relative to AIRS, MERRA yielded a more stable ratio between regime classes pole to pole (Fig. 9). Typically, the MERRA-based regime composition was as follows: atmospherically controlled: 25%–30%, wet advantage: 37%–43%, dry advantage: 12%–17%, and transitional: 11%–16%.
While the differences between AIRS- and MERRA-based classifications (Figs. 8a,b and 9) were substantial (i.e., 46% of pixels differed), there were sizeable regions of coherence. The largest contiguous regions by class are as follows:
Atmospherically controlled: southern Australia, western United States, and southern tier of Canada from western Quebec to Alberta.
Wet advantage: Great Lakes; Northeast United States; eastern Quebec, Newfoundland, and New Brunswick, Canada; United States Gulf Coast; Central America (CAM); Amazon (AMZ); northern Europe (NEU); India (Ganges and IND^); East Asia (EAS, including Changjiang); and Southeast Asia (SEA, including Mekong).
Dry advantage: Arizona, United States; northern Mexico; Middle East; central Asia (CAS); and Tibet (TIB).
Transition: West Africa (WAF); northern Alberta and Saskatchewan, Canada; and Botswana.
Previous modeling studies also showed a positive (wet advantage) feedback between summertime SM and rainfall distribution over much of Europe (Fischer et al. 2007a; Schär et al. 1999; Seneviratne et al. 2006). Consistent with Fischer et al. (2007a), who noted that “coupling is much weaker in maritime regions near the French and Portuguese west coasts,” France, Spain, and Portugal were found to be atmospherically controlled. The fact that AIRS- and MERRA-based results agree on the regime classification (wet advantage) over AMZ is particularly encouraging because independently the two data products (both HI and CTP) are poorly correlated (Figs. A5e,f) in this region. Interestingly, the Congolese Rainforest (Congo basin) was not included in the wet-advantage regime, unlike the other tropical rain forests of the world.
Of the three GLACE hot spots considered in this study—central Great Plains (CGP^), West Africa (WAF^), and India (IND^) (refer to Table A1 for domain extents)—WAF^ was found to be predominantly dry advantage (49%) or transition (21%) based on AIRS and predominantly atmospherically controlled (50%) or transition (26%) according to MERRA. The largest fraction of CGP^ was transition (39%) as classified by AIRS, but MERRA yielded fairly equal (all < 29%) class coverage. For IND^, both AIRS- and MERRA-based classifications showed atmospheric (AIRS: 44%; MERRA: 36%) and wet-advantage conditions (AIRS: 34%; MERRA: 39%).
For all data–classification method pairs (Figs. 8a–d) considered, the dry- and wet-advantage regimes comprised the smallest and largest fraction of the globe (Table 2). The only exception being when the F&E2003 framework was applied using AIRS (Fig. 8c), in which case 71% of the cells fell into the atmospherically controlled class. The transition regime constituted less than one-quarter of the globe.
The results from the F&E2003 approach (Figs. 8c,d) highlight the sensitivity of that approach to the properties of the underlying input dataset. Specifically, when AIRS was used, 71% of the pixels were classified as atmospherically controlled, relative to only 31% when inputs were taken from MERRA (Table 2). We attribute this result in part to the dry (stable) HI (CTP) bias of AIRS relative to MERRA, as quantified in the appendix. When the AIRS-derived classification maps (Figs. 8a,c) are contrasted with those derived from MERRA (Figs. 8b,d), greater consistency is found between the MERRA-derived maps (50% pixel agreement), relative to the AIRS-derived maps (40% pixel agreement). This finding suggests that the F&E2003 framework, despite its shortcomings, may yield reasonable results given inputs from MERRA.
e. Applicability of HI and CTP for soil moisture models and rainfall forecasts
The premise of F&E2003 is that initial (early AM) soil moisture anomalies can influence the triggering of convection and, hence, precipitation. HI and CTP are put forth as diagnostics that correlate well with the feedback signal and we propose, by extension, the initial SM anomaly itself. Unlike in 2003, we now do have access to accurate, globally available independent estimates of SM and rainfall. We employ them here to test further the explanatory power of HI and CTP. It was F&E2003 that suggested “even when the occurrence of rainfall is atmospherically controlled, the land surface moisture condition can indeed impact the depth of rain.” Eltahir and Pal (1996) had previously found linear correlation between CAPE and rainfall depth over AMZ. But AMZ is predominantly a region of positive feedbacks (i.e., wet advantage), as we have shown (Fig. 8).
Figure 11 shows the conditional expectation of SM (LPRM), rainfall (CMORPH) occurrence, and rain event depth, given AIRS HI and CTP for each of the four regime classes, globally. The figure reiterates earlier findings (section 4b) that 1) there are an abundance of atmospheric samples less humid than HI = 15°C with rain and 2) a clear linear correlation structure persists between SM and HI and to a lesser extent between SM and CTP. Furthermore, Fig. 11 shows that there are distinct ranges in HI and CTP with either elevated rainfall frequency and/or depth. The SM–HI correlation is summarized in Fig. 12—a bin-averaged plot of SM at 1°C intervals of HI. Across regimes, the SM–HI slope is approximately constant [atmospherically controlled: −0.55% (°C−1), wet advantage: −0.43% (°C−1), dry advantage: −0.50% (°C−1), and transition: −0.64% (°C−1)]. Best-fit linear equations for SM (both LPRM and MERRA) were calculated on a regime basis using AIRS and MERRA—the intercepts and coefficients of which are listed in Table 3. The explained variances of LPRM and MERRA SM are greatest within dry-advantage regimes. For the wet-advantage regime, the explanatory power of AIRS HI for MERRA SM is considerably higher than for LPRM. A plausible explanation is that for higher SM, the C- or X-band microwave signal used in the LPRM retrieval becomes saturated (less sensitive). Not inconsequential is the fact that while the frequency distribution (mean and dynamic range) of LPRM and MERRA SM are substantially different, as shown in Fig. 13, the SM–HI correlation structure (linear) is very similar (Table 3). The relation between HI, CTP, and SM to rainfall frequency and depth is more nonlinear (not shown) for all variables and data source combinations. Since rainfall depth is largely driven by convergence, an important missing variable from this study is likely the low-level wind.
Two-dimensional gridded conditional probability plot of mean (a) LPRM SM, (b) CMORPH afternoon-plus-evening rainfall frequency, and (c) CMORPH afternoon-plus-evening rainfall depth in the AIRS CTP–HI space for each of the four convective regimes (as classified by the method of this study using AIRS). The grid values are computed at 20 J kg−1 × 1°C grid spacing. Coverage is limited to the CMORPH extent, or ±60°N/S. There was a minimum requirement of 40 observations per grid cell in order for a mean to be calculated.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
Global bin average of LPRM SM for each of the four convective regimes (as classified by the method of this study using AIRS) computed at 1°C intervals of AIRS HI. Refer to Table 3 for the complete mathematical expressions.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
Explained variance of global near-surface soil moisture (both LPRM and MERRA) obtainable via multiple linear regression using HI (°C) and CTP (J kg−1) (both AIRS and MERRA) for each of the four convective regime classes (classified by the method of this paper using AIRS). Contributing data pairs are limited to ±60°N/S. LPRM SM is given in units of percent volumetric, whereas MERRA SM is a percent saturation.
The (a) LPRM and (b) MERRA SM distribution per convective regime class (as classified following the method of this study using AIRS). Note that LPRM is percent volumetric; MERRA is percent saturation.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
Figures 11–13 show globally averaged results. In addition to particular CTP–HI conditions, it makes sense that for certain geographic regions, periods, or events the explanatory power of HI and CTP for SM, and rainfall frequency and/or depth would be comparably higher (or lower). While this is a topic of future, more detailed analyses that fall beyond the scope of the current work, we point briefly to the southern Great Plains 2006/07 dry–wet convective seasons. Figures 14a,b shows the U.S. Drought Monitor issued by the National Drought Mitigation Center in early August of both years. Figures 14c,d show the effects of drought (conversely, drought recovery), captured in the form of distinct shifts in the distribution of HI, CTP, and SM. Such dipole years offer unique (perhaps ideal) opportunities for detailed exploration and understanding of land–atmosphere interactions through coupled observation–modeling studies.
The National Drought Mitigation Center (NDMC; http://www.drought.unl.edu/) drought severity index map issued for the conterminous United States on (a) 8 Aug 2006 and (b) 7 Aug 2007. The frequency histograms of AIRS (c) HI, (d) CTP, and (e) LPRM SM for the 2006 and 2007 convective seasons over the outlined domain (32.5°–38.75°N, 101.25°–92.5°W). NDMC uses the following abbreviations: D0 = abnormally dry, D1 = moderate drought, D2 = severe drought, D3 = extreme drought, and D4 = exceptional drought. D3 conditions are characterized by a PDSI of −4.0 to −4.9. D4 denotes a 1-in-50-yr event and Palmer drought severity index (PDSI) of −5.0 or less.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
5. Discussion
Thus far, we have highlighted our current capacity to overcome the limits imposed by the raob network through the use of satellite-based remote sensing. In the previous section, global maps (Fig. 8) were produced that indicate the predominant regional convective season land–atmosphere feedback signal (wet, dry, or transition), or lack thereof (atmospherically controlled). Here we discuss how these signals would manifest themselves in observable phenomena.
After sunrise, the net surface radiation (Rnet) balance becomes positive, marking the initiation of the energy transmission into the BL. In the absence of large-scale synoptic (background) forcing (i.e., atmospheric circulation, monsoon, topography, and sea surface temperatures), the structure and depth of the daytime BL is driven primarily by the large fluxes of heat and moisture at the surface. Whether convection is favored depends fundamentally on two variables: the Bowen ratio, or ratio between sensible (SH) and latent heat (LH), and the vertical structure (temperature, humidity, and total depth) of the BL (Betts 2009; Betts and Ball 1995; Haiden 1997). Both positive and negative feedbacks are possible over either wet or dry soil conditions, leaving us with the following four scenarios:
Wet positive (rainfall on already-wet soil). Areas of elevated SM (low vegetative resistance) exhibit higher LH (lower SH) that moistens (a possibly already humid) BL and, in turn, reduces surface Rnet, SH, and BL growth (and magnitude of entrainment). Overall, restricted SH limits the development of a turbulent layer, which inhibits the efficiency of mixing with subsiding dry air. Convection is triggered by lowering the level of free convection (LFC) down to the BL top through a strong increase in moist static energy (MSE; F&E2003). Relative to the dry-positive case (#3 below), the input of water vapor is larger and distributed over a thinner (~100 hPa) and cooler BL (Ferguson and Wood 2008, unpublished manuscript; Schär et al. 1999).
Wet negative (rainfall suppression over wet soil). As above, but the BL remains too shallow. Lower Bowen ratios, cooler air temperatures, and increased pressure at the surface contribute to increasingly stable atmospheric conditions, which impedes the vertical movement of air parcels, and can limit the occurrence of rainfall by preventing moist air from reaching the LFC (Cook et al. 2006).
Dry positive (rainfall suppression over already-dry soil). When SM (or vegetative resistance) limits LH (high Bowen ratio; SH ≫ LH), a near-surface atmospheric vapor pressure deficit and increased air temperature results, which, in turn, leads to intensified growth of the BL and enhanced entrainment (typically, warm and dry air). From this point, the process feeds back on itself: less LH → enhanced SH → further warming of the air and reduction of relative humidity → drying (and enhanced vegetative resistance). The result is a deep [~200 hPa: Ferguson and Wood (2008, unpublished manuscript) and ~150 hPa: Schär et al. (1999)], well-mixed BL, driven by a large Bowen ratio at the surface. The only escape from this cycle occurs whenever large-scale and synoptic factors are strong enough to overcome the strength of the feedback (Entekhabi et al. 1996). A deeper BL is associated with more vigorous entrainment of above BL air with low MSE, which acts to decrease the MSE per unit of BL air and thereby reduce potential for convective development.
Dry negative (rainfall on dry soils). Areas with SM dry anomalies are characterized by higher Bowen ratios (similar to dry positive) and deeper BLs that require greater instability (CTP) for convection to occur. In such an environment, sheets of moist and cold advected air are susceptible to being “triggered” up to the LFC. According to F&E2003, convection is favored if the temperature profile between 100 and 300 hPa above the surface is close to dry adiabatic (CTP > 200 J kg−1). Note that this is the opposite of the wet-positive case in that the BL grows up to the LFC, rather than the LFC dropping down to the BL. In the Sahel (West Africa), where dry Saharan air aloft overlies a low-level moist adiabatic monsoon layer (Parker et al. 2005), this dry-negative feedback can be observed (Taylor and Ellis 2006).
In this study (and in F&E2003), the wet- and dry-advantage regimes correspond to regions typified by wet-positive and dry-negative tendencies, respectively. The transition zone applies to areas where the observed signals are mixed. Atmospherically controlled areas correspond to those regions with strong temporal variability in HI. The most basic example of atmospheric control is early morning cloud cover, which provides radiative cooling and stabilization of the mixed layer. In some cases, the predominant land–atmosphere feedback may serve to reinforce the impacts of synoptic forcing. For example, northeasterly winds carry large quantities of water vapor into AMZ from the Atlantic Ocean. Once convection is initiated over the basin (via wet-positive feedbacks), the resulting LH release deepens the BL, which must then be supported by additional inflow of moist air from the Atlantic at low levels. Figures 8a,b show that 40% (AIRS based) to 46% (MERRA based) of the tropics (23.5°N–23.5°S) were indeed found to sustain positive feedbacks.
Importantly, the scale at which these feedbacks are resolved has been shown to impact not only the strength but also the sign of the signal detected (Hohenegger et al. 2009; Taylor and Ellis 2006; Van den Hurk and Van Meijgaard 2010). Over the Alpine region of Europe, Hohenegger et al. (2009) showed that while 25-km simulations yielded a strong positive feedback, those at 2.2 km yielded a negative feedback. Similarly, over West Africa, the ~50-km simulations of Van den Hurk and Van Meijgaard (2010) showed a dominant positive feedback, while the ~37-km satellite-based study of Taylor and Ellis (2006) showed a negative feedback. In both cases described, positive feedbacks dominated at larger scales. However, there is no evidence to suggest that this pattern always holds true. Based on these limited studies, as well as our observation that patches of dry and moist soils (or stressed and unstressed vegetation) create sharp gradients in SH at local scales (10–15 km), we believe that scale dependency exists and is observable in the real world. We are unaware of examples following our methodology that fully address the scale dependency issue and this is the potential focus of future research.
6. Summary and conclusions
In this study, we have revisited the work of Findell and Eltahir (2003a,b) with satellite-based data from AIRS, AMSR-E, and the merged CMORPH rainfall product and subsequently redeveloped their CTP–HI classification framework for global application. The classification framework (Fig. 7) developed herein contributes to the first step in enhancing conceptual understanding of coupling, which is to identify where and when (both regionally and temporally) coupling persists, and thereby provides an intelligent guidance tool for future focused process and observational (on the ground, model, and satellite based) studies. There is clear value in narrowing the study area from one that is global in scale to one that is both spatially and temporally well defined.
Throughout the study, HI, CTP, and GWETTOP from the MERRA reanalysis were used in parallel to provide both results for intercomparison as well as a measure of the sensitivity of the classification to its inputs. The validation of both AIRS and MERRA HI and CTP presented in the appendix was a necessary prerequisite given that neither dataset had previously been evaluated in the manner they were used here. In Fig. 5 (and Tables A3 and A4) we characterized the zonal distribution of AIRS and MERRA HI and CTP for “all,” “rain,” and “no-rain” sample sets. The median and select percentile values provided in Fig. 5 (and Tables A3 and A4) should serve as a stepping-off point for zone- or condition-specific data validation studies. Users should consider applying the 10th and 90th zonal percentiles of HI from the rain sample (Table A4) as an improvement on the 5°C ≤ HI ≥ 15°C bounding threshold for rain previously suggested by F&E2003.
In section 4a, we began our critique of the F&E2003 framework by demonstrating its application at four sites in the United States (Fig. 4). We concluded that because of the following two findings, which contradict adopted conventions of F&E2003, the approach is unsuitable for global application: 1) median values and, hence, thresholds of HI and CTP for rain days vary as a function of latitude (climate; Fig. 5); and 2) HI and CTP exhibit substantial correlation in many regions (Fig. 6). In response, we outlined a new distribution-based technique that shifts with the zonal properties (Fig. 7). An important feature of this scheme, which improves upon F&E2003, is that it accounts for the inherent biases of the underlying input dataset. Also not inconsequential is the new convective season identification methodology (see section 2), which enabled us to properly handle regions where the typical annual cycle of succession of dry and wet periods does not correspond with conventional meteorological seasons (e.g., December–February, March–May, June–August, and September–November).
Using our new CTP–HI scheme, we produced global maps of dominant feedback signals (Fig. 8), from which we extracted the zonal (Fig. 9), regional (Fig. 10), and global (Table 2) compositional makeup. Using inputs from AIRS and MERRA, we compared and contrasted the results from not only the classification framework put forth in this study, but also that of F&E2003. It was encouraging to find multiple regions of concurrence between the pair (AIRS and MERRA) of global maps produced using the method of this study, while much fewer existed between those based on F&E2003. The strong variation between the F&E2003–based maps (Figs. 8c,d) highlights the sensitivity of their fixed-threshold technique to the choice of dataset (and its inherent biases). On the other hand, many similarities existed between the MERRA F&E2003-based map (Fig. 8d) and the pair of maps (both AIRS and MERRA) produced using the method of this study (Figs. 8a,b). This implies that forced with an unbiased dataset of which MERRA is our best approximation, the two frameworks actually yield similar categorizations.
The feedback signals were mixed over the coupling hot spots (Koster et al. 2004), with no more than 50% coverage by any one regime. Having said this, the dominant regimes within CGP^, WAF^, and IND^ were transition, dry advantage, and atmospherically controlled (dry advantage, atmospherically controlled, and wet advantage), respectively, according to AIRS (MERRA)-based classifications. We point out that the AGCM ensemble members of GLACE-1 were initialized with SST boundary conditions from 1994, run at resolutions varying from 1.875° to 3.75° for the period of 1 June–31 August, and analyzed at a 6-day temporal scale (Koster et al. 2006). Accordingly, any differences could be attributable to inconsistencies in period, temporal, and/or spatial resolution. We hypothesize that real-world surface heterogeneity not parameterized in the AGCMs, including the effects of large-scale irrigation prevalent in CGP^ and IND^, plays a key role in model–observation differences.
We ended our analysis with a brief look at the explanatory (for SM) and predictive (for rainfall frequency and depth) power of HI and CTP (Fig. 11). Globally, a strong linear relationship was found between HI and SM (Fig. 12), which we modeled using regression (Table 3). Despite large differences in the dynamic range of LPRM and MERRA SM (Fig. 13), their variance was similarly explained by the models, except for the wet-advantage regime, where models with MERRA SM performed better. We conclude that LPRM may have less sensitivity (more noise) near saturation.
From a conditional probability perspective, Fig. 11 reveals the potential for HI and CTP to be used for predicting initial SM or subsequent rain occurrence and depth. It implies that there may be specific geographic regions and/or events (i.e., certain HI–CTP subspace) where the explanatory or predictive power (value) of the diagnostics is particularly strong. As an example and motivator for future applications, we demonstrated the effects of drought on the frequency histograms of HI, CTP, and SM using the southern Great Plains 2006/07 dry–wet dipole years (Fig. 14).
7. Future work
Further analyses that fall beyond the scope of this work are warranted to further refine the classification methodology and redefine the role of HI and CTP in future modeling and/or forecast applications. For example, could these diagnostics be used to improve methodologies for separating stratiform and convective rain (e.g., the modified GSFC profiling algorithm used in the generation of the AMSR-E rain rate and rain type over land)? Can HI and rainfall be merged to correct retrieval biases and spread? The rate of change of HI and CTP as measured from subsequent overpasses may hold additional value for confirming model-based diurnal rainfall cycles and mean SM, especially at high latitudes. For example, our interpolated AIRS fields provide up to five observations daily for Fairbanks, Alaska (64.8°N, 147.9°W). Together HI and CTP could provide estimates, constraints, or validation of satellite-based SM and evapotranspiration, ideally, where those retrievals are most uncertain: optically dense canopies and dry regimes, respectively. Currently, the LandFlux community is in search of additional ways to evaluate their multimodel results (Mueller et al. 2011). Using HI and CTP in hydrologic consistency checks is not an entirely new idea (McCabe et al. 2008).
As we have discussed in section 5, the feedbacks may have inherent scale dependencies. This brings up the issue of advection. Advection plays a large role in averaging over the landscape, as winds of 2 (weak) and 10 m s−1 (strong) traverse 86.4 and 432 km in 12 h, respectively. Findell and Eltahir (2003c) showed that wind fields serve as a crucial third component in their CTP–HI framework (F&E2003). However, because satellite-based estimates of wind fields over land are not currently available, wind was not addressed in our study. Future research should consider the inclusion of wind from readily available reanalysis or else rely on geostationary satellite IR cloud vectors as a surrogate.
Our climate is not static. The concept of land–atmosphere interaction, or the ability of land surface–related processes to directly influence the state of the boundary layer and, hence, rainfall, feeds directly into the concept of hydrologic intensification in a changing climate. For example, coupling suggests that negative feedback regions (i.e., dry negative) can have drought that is unconnected with higher mean temperatures, while positive feedback regions, such as the southwest United States, experience drought that tends to be associated with warming, which reinforces the state (i.e., dry positive). It is also likely that land–atmosphere interactions themselves will be modified as a part of climate change, with marked shifts both geographically (in location and extent) and in amplitude. The current range of Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4) generation model predictions for temperature and precipitation has been linked to intermodel disparities in coupling strength (Seneviratne et al. 2006). Accordingly, improvements to our observing, modeling, and understanding of land–atmosphere coupling and the dominant underlying mechanisms and feedbacks that constitute it are critical to forecasting climate into the future. We foresee future application of the CTP–HI–SM space (Fig. 11a) as a potential “key-hole” criterion for deciding which AGCM models are included in ensemble runs.
Future research will focus on deciphering whether the hydrologic cycle is changing and, if so, quantifying the role that coupling plays. While this study focused on the mean feedback signal over a 7-yr period, it would be meaningful to analyze event-scale processes, as well as interannual variability of the signal strength and sign. The GEWEX Local Coupled Project (LoCo) is leading efforts to develop new diagnostics—beyond HI and CTP—that will enable easier recognition of conditions and geographic regions where coupling significantly impacts climate (Santanello et al. 2009; Van den Hurk and Blyth 2008).
Acknowledgments
This research was funded through the lead author’s NASA Earth and Space Science Fellowship NNX08AU28H: Understanding Hydrologic Sensitivity and Land–Atmosphere Coupling through Space-Based Remote Sensing. GEWEX GLASS/WATCH provided travel funding to attend the LoCo Workshop held in De Bilt, the Netherlands, in June 2008. This manuscript benefitted from the helpful suggestions of attendees at that workshop, as well as Ming Pan and Justin Sheffield. We also acknowledge Edward T. Olsen for his relentless efforts through AskAIRS to resolve our AIRS algorithmic and product inquiries.
APPENDIX
Dataset Validation
AIRS- and MERRA-derived HI and CTP have not previously been evaluated. In this section, we use raob data to evaluate the suitability (accuracy and uncertainty) of these datasets for use in a global atmospheric regime classification. While AIRS provides only instantaneous observations on overpass, hence limiting the number of potential raob matchups and qualifying stations, MERRA is produced 3 hourly, enabling matchups 100% of the time for all stations.
a. Raobs
We use raobs from 452 stations globally, each offering more than 50 quality-cleared (see below) profiles during the September 2002–December 2009 study period. Soundings are performed on an operational basis twice daily at 0000 and 1200 UTC, globally, but select stations do routinely release radiosondes at 3- or 6-hourly intervals. The stations are disproportionately located in the Northern Hemisphere (United States and Europe), with few stations in Latin America and Africa. This study benefits from recent activities of the African Monsoon Multidisciplinary Analysis (AMMA; Redelsperger et al. 2006) experiment, which reactivated, renovated, and expanded the West Africa radiosonde network (Parker et al. 2008).
Radiosondes typically record measurements at 2-s intervals; however, the data are not transmitted or archived at full resolution. For example, the Met Office (UKMO) standard-resolution radiosonde data contain measurements at only the following standard pressure levels: 1000, 925, 850, 700, 500, 400, 300, 250, 200, 150, 100, 70, 50, 30, 20, and 10 hPa, with additional pressure levels included if they are deemed significant (e.g., turning points) with respect to temperature, relative humidity, and wind according to the criteria outlined in the World Meteorological Organization’s (WMO) Manual on Codes (WMO 2009). With regard to temperature and relative humidity, the manual stipulates that the accuracy of the linearly interpolated value (in log pressure) between significant levels should agree with the observed values within 1 K and 15% (here, % refers to the amount of relative humidity, not % of observed value), respectively. First priority is given to representing the temperature profile, such that the additional levels are the actual levels at which “prominent changes in the lapse rate of air temperature” occur. The effect is a net reduction of the water vapor information content of the operational sondes. Therefore, we use the University of Wyoming standard-resolution radiosonde data (http://weather.uwyo.edu/upperair/sounding.html), which comprise 6–7 times the number of pressure levels included in the UKMO archive (Li et al. 2003).
In order for a radiosonde profile to be included, we required that the profile pass the following quality checks: 1) the first valid report occurs within 20 m of the surface, 2) the profile includes at least 12 valid reports between the ground and 500 hPa AGL, and 3) at least five of these must occur between 100 and 300 hPa AGL. For intercomparison with AIRS, we require that the raobs fall within a collocation window of ±3 h and a radius of 100 km, consistent with convention applied in previous studies (Divakarla et al. 2006; McMillin et al. 2007). For intercomparisons with MERRA, we use the 3-hourly file within which the raob timestamp falls. In both cases, if more than one AIRS or MERRA grid satisfies the distance requirement, the matchup with the lowest absolute difference in HI is selected.
We accept that radiosonde sensors have their own error characteristics. Currently, a wide (>14) variety of sensors are used across the raob network, each with unique error characteristics (Wang and Zhang 2008). Radiosonde accuracy has been linked to observed temperature and relative humidity, age of sensor, and pressure. Different radiosonde types exhibit different strengths and weaknesses in different realms of the temperature and relative humidity space (Miloshevich et al. 2006). Much work has been done to develop and validate (bias) correction techniques for these individual sensors (i.e., Agustí-Panareda et al. 2009; Miloshevich et al. 2004, 2001; Vömel et al. 2007; Wang et al. 2002; Yoneyama et al. 2008). Generally, accuracy degrades with increasing cloud thickness and altitude (Miloshevich et al. 2006). However, operationally not all the additional metadata required to apply the existing humidity bias correction schemes are available (e.g., prelaunch reference measurements such as relative humidity, temperature and pressure, specific production batches, and age of radiosonde). Neither the UKMO nor University of Wyoming datasets include a record of the sensor type used. Accordingly, we use the raobs record as is, without correction.
b. Data validation
Figures A1 and A2 show the HI and CTP error statistics computed from all AIRS-available matchups at 253 stations globally. Figure A3 shows the AIRS–MERRA HI and CTP joint τKendall correlation. Figure A4 shows the MERRA error characteristics for 444 sites computed from all available (0000, 0300, … 2100 UTC) raobs. Note that this latter comparison is for a distinctly different temporal period than the first. Firstly, sampling times and frequency varies by station. Secondly, and perhaps more importantly, no distinction is made between clear or cloudy conditions. Figure A4 is included because it provides 1) an accurate representation of the geographic distribution of MERRA–raob agreement and 2) an estimate of the all-sky diurnally averaged uncertainty of MERRA. Figures A1–A4 are summarized regionally in Tables A1 and A2. Note that limited matchups preclude the inclusion of Sahara (SAH) in Table A1 and eastern Africa (EAS) in both Tables A1 and A2. There are three supplemental regions defined by the coupling hot spots of Koster et al. (2004)—they are central Great Plains (CGP^), West Africa (WAF^), and India (IND^). There was insufficient coverage over WAF^ for the AIRS-available case. Figures A5 and 5 show the results of AIRS–MERRA intercomparisons conducted at full spatial coverage, the zonal statistics of which are summarized in Table A3.
(a),(b) Bias and (c),(d) nonparametric correlation (τKendall) of (a),(c) AIRS and (b),(d) MERRA HI computed relative to coregistered radiosonde profiles. There are 253 stations displayed, with a mean of 155 ± 96 data pairs available for each.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
As in Fig. A1, but for CTP.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
The joint τKendall correlation of AIRS and MERRA (a) HI and (b) CTP computed using the same data sample as in Figs. A1 and A2.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
Bias and nonparametric correlation (τKendall) of the 3-hourly MERRA (a),(b) HI and (c),(d) CTP at 444 radiosonde sites globally. Using all available matchups (i.e., 0000, 0300, … 2100 UTC) afforded 810 ± 500 matchups per station.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
Regional summary statistics of MERRA–raob, AIRS–raob, and AIRS–MERRA matchups. The supplemental zones (A–C) are included because they were previously identified as hot spots of land–atmosphere interaction (Koster et al. 2004).
Median values of (a) AIRS HI and (b) CTP, (c),(d) bias relative to respective, coregistered MERRA data, and (e) AIRS–MERRA HI and (f) CTP τKendall correlation. The absence of color (i.e., white) denotes cells with fewer than 40 data pairs, due to retrieval limitations of AIRS.
Citation: Journal of Hydrometeorology 12, 6; 10.1175/2011JHM1380.1
Statistical summary of AIRS and MERRA HI and CTP retrievals for 15° zones. For each zone, the table includes (from left to right) the mean and standard deviation of the mean, 25th percentile (PCT), median, 75th percentile, and the median of the standard deviation of all cells. Note that while the first four columns consider the lump distribution comprising all retrievals for all cells within the zone, the last column is obtained by first calculating the σ from the time series at each cell and then finding the areal median over the zone.
1) Point based
(i) AIRS available
Table A1 shows that AIRS and MERRA HI bias share the same sign in 14 of 18 regions. However, in 10 of 18 regions the absolute bias of AIRS is larger. AIRS and MERRA share particularly strong wet (−) and dry (+) biases over Australia (AUS) and central Asia (CAS), respectively. MERRA has a remarkable wet bias relative to the single raob station in southern Africa (SAF). Interestingly, MERRA exhibited a strong dry bias over the western United States (WUS) relative to AIRS, but over the central United States (CUS) is very similarly biased. As measured by rmse, the AIRS HI has an overall equal or better agreement than MERRA for 15 of 18 regions. While the AIRS rmse is less than 10°C for all but one region (CAS), MERRA rmse exceeds 10°C in five regions: WUS, SAF, SAH, CAS, and Tibet. AIRS–raob correlation was also found to be stronger than that of MERRA–raob at 15 of 18 sites. MERRA correlations are substantially weaker in CAS, North Asia (NAS), and the Amazon (AMZ). The strongest correlations in the case of both datasets are found in the low specific humidity regions of the world, including AUS, WUS, Mediterranean (MED), and Alaska and Canada (AK–CA) . Correlations between the two datasets are strongest and weakest over AUS and SAF, respectively.
In terms of CTP, Table A1 shows that it was MERRA that yielded the strongest correlations in 15 of 18 (other than WAF, SAS, and NAS) regions. AIRS is biased in the same direction as MERRA for 11 of 18 regions, but its absolute bias is larger in 13 of 18 regions. AIRS is negatively biased (atmosphere is more stable) in 13 of 18 regions, as opposed to only 6 of 18 for MERRA. Figure A2a shows the pervasiveness of AIRS more-stable-than-raob soundings. Overall, MERRA provides better agreement with raobs (16 of 18 stations), as determined by rmse. AIRS–MERRA correlations are strongest and weakest over NAS and SAF, respectively. Relative to HI, AIRS–MERRA CTP correlations are lower for 17 of 18 regions.
(ii) All available (MERRA only)
Table A2 shows that MERRA HI intercomparisons that included all available raob matchups (regardless of the time of day or cloudiness) yielded reduced absolute biases in 13 of 18 regions relative to the MERRA HI statistics computed from AIRS-available, clear-sky (fc ≤ 0.4) mornings only (Table A1). In all but three regions—AMZ, AUS, and SAF—the mean HI is depressed (more humid) relative to the clear-sky morning sample, as would be expected. Furthermore, rmse is reduced in all regions, except EAS. Areas of relatively strong and weak correlation remain to be in the semiarid and humid regions, respectively. A comparison between Figs. A1d and A4b reveals that correlations are enhanced most over North America, eastern Europe, and central Asia.
MERRA CTP absolute bias is reduced in 12 of 18 regions relative to the AIRS-available sample set (Table A2). Southern South America (SSA), CAM, and EAS are the only regions where rmse was not reduced. In all but the following four regions, the absolute bias was less than 50 J kg−1: CAM, WUS, WAF, and CAS. The sign of the shift in bias between the AIRS-available and all-available samples is split between those with relatively stronger positive (8 of 18) and stronger negative (10 of 18) biases. Areas of relatively strong and weak correlation include CUS, eastern United States (EUS), AK–CA, and northern Europe (NEU), and WAF and SAS, respectively.
Figures A4b and A4d show a dramatic gradient in both MERRA–raob HI and CTP correlation as a function of latitude, with the lowest values in the tropics. Hints of this general zonal pattern are also evident in Figs. A2d and A4d. Because of data coverage limitations, we cannot confirm whether this pattern also holds true for AIRS.
2) AIRS–MERRA intercomparisons at global coverage
(i) Regional comparisons
Figure A5c shows that, on average, AIRS HI is larger (less humid) than MERRA HI in the following areas: the Rocky Mountain range, Great Lakes, Brazilian Highlands, Ethiopian Highlands, Tibet, and northern Russia. Conversely, AIRS HI is smaller (more humid) than MERRA HI in the western United States (CA, NV, AZ, and UT), central Argentina, central and southern Australia, Sahel, Congo, and central Asia.
Globally, the median AIRS CTP evokes a more stable atmosphere than MERRA CTP (Fig. A5d). AIRS CTP is, on average, more than 200 J kg−1 less than MERRA CTP over the Rocky Mountain range, central South America, Tibet, and Congo. There are only a few scattered exceptions where the AIRS median CTP exceeds that of MERRA by 50 J kg−1 or more. They are southern Texas, Sahara, and the eastern coast of Africa. Figure A5b supports the earlier findings of Riemann-Campe et al. (2009) that suggest specific humidity, not air temperature, has a more pronounced effect on CTP.
It is the case for both HI and CTP that the strongest and weakest AIRS–MERRA correlations occur in the mid- to high latitudes and tropics, respectively. On a global basis, variability in the respective datasets, as measured by the coefficient of variation (CV), is relatively lower for AIRS HI than MERRA HI, but higher for AIRS CTP than MERRA CTP (not shown).
(ii) Latitudinal comparisons
Figure 5 illustrates three zonal profiles each for AIRS and MERRA median HI and CTP—one calculated from all available data pairs, and the other two conditioned on the occurrence (rain or no-rain) of afternoon-plus-evening (CMORPH) rainfall. Overall, both AIRS and MERRA HI are sensitive to rain, except at high latitudes, whereas the sensitivity of CTP to rain was limited to Northern Hemisphere midlatitudes—specifically, 15°–30°N. The largest spread (in absolute terms) in HI between the rain and no-rain samples is found within the 10°–35°N and 15°–30°S zonal windows. The largest AIRS–MERRA discrepancies in median HI occur between 20° and 35°N for both the all- and rain-only cases. In terms of CTP, median values of AIRS and MERRA differ substantially, especially outside of the tropics (Fig. 5).
The minimum and maximum AIRS (MERRA) zonal medians occur between 5°N–10°S and 25°–40°N (5°N–10°S; 25°–40°S) for HI and between 45°–60°S and 15°–30°S (same) for CTP. These and other key zonal statistics calculated from the “all-available” sample are provided in Table A3. Table A4 provides the zonal 10th, 25th, 50th, 75th, and 90th percentiles of HI from the “rain-only” subset. Together, Tables A3 and A4 provide a potential stepping-off point for more detailed bias (AIRS) or process (MERRA) studies.
As in Table A3 but only for HI and with the added condition that there was afternoon-plus-evening rainfall, as defined by CMORPH. Also, the mean ±σ and median σCELL columns have been replaced with the 10th and 90th percentiles.
REFERENCES
Adler, R., Wilheit T. , Kummerow C. , and Ferraro R. , 2007: AMSR-E/Aqua L2B Global Swath Rain Rate/Type GSFC Profiling Algorithm. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://nsidc.org/data/ae_rain.html.]
Agustí-Panareda, A., and Coauthors, 2009: Radiosonde humidity bias correction over the West African region for the special AMMA reanalysis at ECMWF. Quart. J. Roy. Meteor. Soc., 135, 595–617.
Aonashi, K., and Coauthors, 2009: GSMaP passive microwave precipitation retrieval algorithm: Algorithm description and validation. J. Meteor. Soc. Japan, 87, 119–136.
Atlas, R., Wolfson N. , and Terry J. , 1993: The effect of SST and soil moisture anomalies on GLA model simulations of the 1988 U.S. summer drought. J. Climate, 6, 2034–2048.
Betts, A. K., 2004: Understanding hydrometeorology using global models. Bull. Amer. Meteor. Soc., 85, 1673–1688.
Betts, A. K., 2009: Land-surface-atmosphere coupling in observations and models. J. Adv. Model. Earth Syst., 1, 4, doi:10.3894/JAMES.2009.1.4.
Betts, A. K., and Ball J. H. , 1995: The FIFE surface diurnal cycle climate. J. Geophys. Res., 100D, 25 679–25 693.
Betts, A. K., and Dias M. A. F. S. , 2010: Progress in understanding land-surface-atmosphere coupling from LBA research. J. Adv. Model. Earth Syst., 2, 6, doi:10.3894/JAMES.2010.2.6.
Betts, A. K., Ball J. H. , Beljaars A. C. M. , Miller M. J. , and Viterbo P. A. , 1996: The land surface-atmosphere interaction: A review based on observational and global modeling perspectives. J. Geophys. Res., 101D, 7209–7225.
Bloom, S. C., Takacs L. L. , DaSilva A. M. , and Ledvina D. , 1996: Data assimilation using incremental analysis updates. Mon. Wea. Rev., 124, 1256–1271.
Chahine, M. T., and Coauthors, 2006: AIRS: Improving weather forecasting and providing new data on greenhouse gases. Bull. Amer. Meteor. Soc., 87, 911–926.
Chen, M., Shi W. , Xie P. , Silva V. B. S. , Kousky V. E. , Higgins R. W. , and Janowiak J. E. , 2008: Assessing objective techniques for gauge-based analyses of global daily precipitation. J. Geophys. Res., 113, D04110, doi:10.1029/2007JD009132.
Cook, B. I., Bonan G. B. , and Levis S. , 2006: Soil moisture feedbacks to precipitation in southern Africa. J. Climate, 19, 4198–4206.
Dirmeyer, P. A., 2006: The hydrologic feedback pathway for land–climate coupling. J. Hydrometeor., 7, 857–867.
Dirmeyer, P. A., Koster R. D. , and Guo Z. C. , 2006: Do global models properly represent the feedback between land and atmosphere? J. Hydrometeor., 7, 1177–1198.
Divakarla, M. G., Barnet C. D. , Goldberg M. D. , McMillin L. M. , Maddy E. , Wolf W. , Zhou L. H. , and Liu X. P. , 2006: Validation of Atmospheric Infrared Sounder temperature and water vapor retrievals with matched radiosonde measurements and forecasts. J. Geophys. Res., 111, D09S15, doi:10.1029/2005JD006116.
Douville, H., 2002: Influence of soil moisture on the Asian and African monsoons. Part II: Interannual variability. J. Climate, 15, 701–720.
Douville, H., 2003: Assessing the influence of soil moisture on seasonal climate variability with AGCMs. J. Hydrometeor., 4, 1044–1066.
Eltahir, E. A. B., 1998: A soil moisture–rainfall feedback mechanism: 1. Theory and observations. Water Resour. Res., 34, 765–776.
Eltahir, E. A. B., and Pal J. S. , 1996: Relationship between surface conditions and subsequent rainfall in convective storms. J. Geophys. Res., 101D, 26 237–26 245.
Entekhabi, D., Rodriguez-Iturbe I. , and Castelli F. , 1996: Mutual interaction of soil moisture state and atmospheric processes. J. Hydrol., 184, 3–17.
Fennessy, M. J., and Shukla J. , 1999: Impact of initial soil wetness on seasonal atmospheric prediction. J. Climate, 12, 3167–3180.
Ferguson, C. R., and Wood E. F. , 2010a: An evaluation of satellite remote sensing data products for land surface hydrology: Atmospheric Infrared Sounder. J. Hydrometeor., 11, 1234–1262.
Ferguson, C. R., and Wood E. F. , 2010b: Observing land-atmosphere interaction globally with satellite remote sensing. Proc. Earth Observation and Water Cycle Science Symp., Frascati, Italy, ESA, 8.
Ferguson, C. R., and Wood E. F. , 2010c: Quantifying land–atmosphere interaction with satellite remote sensing: Current capabilities, findings, and limits. Extended Abstracts, 24th Conf. on Hydrology, Atlanta, GA, Amer. Meteor. Soc., J13.4. [Available online at http://ams.confex.com/ams/90annual/techprogram/paper_163465.htm.]
Ferranti, L., Slingo J. M. , Palmer T. N. , and Hoskins B. J. , 1999: The effect of land-surface feedbacks on the monsoon circulation. Quart. J. Roy. Meteor. Soc., 125, 1527–1550.
Findell, K. L., and Eltahir E. A. B. , 2003a: Atmospheric controls on soil moisture–boundary layer interactions. Part I: Framework development. J. Hydrometeor., 4, 552–569.
Findell, K. L., and Eltahir E. A. B. , 2003b: Atmospheric controls on soil moisture–boundary layer interactions. Part II: Feedbacks within the continental United States. J. Hydrometeor., 4, 570–583.
Findell, K. L., and Eltahir E. A. B. , 2003c: Atmospheric controls on soil moisture-boundary layer interactions: Three-dimensional wind effects. J. Geophys. Res., 108, 8385, doi:10.1029/2001JD001515.
Fischer, E. M., Seneviratne S. I. , Lüthi D. , and Schär C. , 2007a: Contribution of land-atmosphere coupling to recent European summer heat waves. Geophys. Res. Lett., 34, L06707, doi:10.1029/2006GL029068.
Fischer, E. M., Seneviratne S. I. , Vidale P. L. , Lüthi D. , and Schär C. , 2007b: Soil moisture–atmosphere interactions during the 2003 European summer heat wave. J. Climate, 20, 5081–5099.
Gao, W. H., Zhao F. S. , Xu Y. F. , and Feng X. , 2008: Validation of the surface air temperature products retrieved from the atmospheric infrared sounder over China. IEEE Trans. Geosci. Remote Sens., 46, 1783–1789.
Guo, Z. C., and Coauthors, 2006: GLACE: The Global Land–Atmosphere Coupling Experiment. Part II: Analysis. J. Hydrometeor., 7, 611–625.
Haiden, T., 1997: An analytical study of cumulus onset. Quart. J. Roy. Meteor. Soc., 123, 1945–1960.
Hohenegger, C., Brockhaus P. , Bretherton C. S. , and Schär C. , 2009: The soil moisture–precipitation feedback in simulations with explicit and parameterized convection. J. Climate, 22, 5003–5020.
Hsu, K.-L., Gao X. , Sorooshian S. , and Gupta H. V. , 1997: Precipitation estimation from remotely sensed information using artificial neural networks. J. Appl. Meteor., 36, 1176–1190.
Hsu, K.-L., Gupta H. V. , Gao X. , and Sorooshian S. , 1999: Estimation of physical variables from multichannel remotely sensed imagery using a neural network: Application to rainfall estimation. Water Resour. Res., 35, 1605–1618.
Huffman, G. J., and Coauthors, 2007: The TRMM Multisatellite Precipitation Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeor., 8, 38–55.
Joyce, R. J., Janowiak J. E. , Arkin P. A. , and Xie P. , 2004: CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydrometeor., 5, 487–503.
Koster, R. D., Suarez M. J. , Ducharne A. , Stieglitz M. , and Kumar P. , 2000: A catchment-based approach to modeling land surface processes in a general circulation model 1. Model structure. J. Geophys. Res., 105D, 24 809–24 822.
Koster, R. D., and Coauthors, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305, 1138–1140.
Koster, R. D., and Coauthors, 2006: GLACE: The Global Land–Atmosphere Coupling Experiment. Part I: Overview. J. Hydrometeor., 7, 590–610.
Kubota, T., and Coauthors, 2007: Global precipitation map using satellite-borne microwave radiometers by the GSMaP Project: Production and validation. IEEE Trans. Geosci. Remote Sens., 45, 2259–2275.
Kummerow, C., and Coauthors, 2001: The evolution of the Goddard profiling algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 1801–1820.
Li, L., Njoku E. G. , Im E. , Chang P. S. , and Germain K. S. , 2004: A preliminary survey of radio-frequency interference over the US in Aqua AMSR-E data. IEEE Trans. Geosci. Remote Sens., 42, 380–390.
Li, Z., Muller J.-P. , and Cross P. , 2003: Comparison of precipitable water vapor derived from radiosonde, GPS, and Moderate-Resolution Imaging Spectroradiometer measurements. J. Geophys. Res., 108, 4651, doi:10.1029/2003JD003372.
McCabe, M. F., Wood E. F. , Wójcik R. , Pan M. , Sheffield J. , Gao H. , and Su H. , 2008: Hydrological consistency using multi-sensor remote sensing data for water and energy cycle studies. Remote Sens. Environ., 112, 430–444.
McCollum, J. R., and Ferraro R. R. , 2003: Next generation of NOAA/NESDIS TMI, SSM/I, and AMSR-E microwave land rainfall algorithms. J. Geophys. Res., 108, 8382, doi:10.1029/2001JD001512.
McMillin, L. M., Zhao J. , Raja M. K. R. V. , Gutman S. I. , and Yoe J. G. , 2007: Radiosonde humidity corrections and potential Atmospheric Infrared Sounder moisture accuracy. J. Geophys. Res., 112, D13S90, doi:10.1029/2005JD006109.
Miloshevich, L. M., Vömel H. , Paukkunen A. , Heymsfield A. J. , and Oltmans S. J. , 2001: Characterization and correction of relative humidity measurements from Vaisala RS80-A radiosondes at cold temperatures. J. Atmos. Oceanic Technol., 18, 135–156.
Miloshevich, L. M., Paukkunen A. , Vömel H. , and Oltmans S. J. , 2004: Development and validation of a time-lag correction for Vaisala radiosonde humidity measurements. J. Atmos. Oceanic Technol., 21, 1305–1327.
Miloshevich, L. M., Vömel H. , Whiteman D. N. , Lesht B. M. , Schmidlin F. J. , and Russo F. , 2006: Absolute accuracy of water vapor measurements from six operational radiosonde types launched during AWEX-G and implications for AIRS validation. J. Geophys. Res., 111, D09S10, doi:10.1029/2005JD006083.
Mueller, B., and Coauthors, 2011: Evaluation of global observations-based evapotranspiration datasets and IPCC AR4 simulations. Geophys. Res. Lett., 38, L06402, doi:10.1029/2010GL046230.
Nicholson, S. E., 2009: A revised picture of the structure of the “monsoon” and land ITCZ over West Africa. Climate Dyn., 32, 1155–1171.
Olsen, E. T., Ed., 2007: AIRS/AMSU/HSB version 5 L2 support pressure levels. NASA JPL Tech. Rep., 3 pp. [Available online at http://disc.sci.gsfc.nasa.gov/AIRS/documentation/v5_docs/AIRS_V5_Release_User_Docs/V5_L2_Support_Pressure_Levels.pdf.]
Owe, M., de Jeu R. , and Holmes T. , 2008: Multisensor historical climatology of satellite-derived global land surface moisture. J. Geophys. Res., 113, F01002, doi:10.1029/2007JF000769.
Parker, D. J., Thorncroft C. D. , Burton R. R. , and Diongue-Niang A. , 2005: Analysis of the African easterly jet, using aircraft observations from the JET2000 experiment. Quart. J. Roy. Meteor. Soc., 131, 1461–1482.
Parker, D. J., and Coauthors, 2008: The AMMA radiosonde program and its implications for the future of atmospheric monitoring over Africa. Bull. Amer. Meteor. Soc., 89, 1015–1027.
Redelsperger, J. L., Thorncroft C. D. , Diedhiou A. , Lebel T. , Parker D. J. , and Polcher J. , 2006: African Monsoon Multidisciplinary Analysis: An international research project and field campaign. Bull. Amer. Meteor. Soc., 87, 1739–1746.
Riemann-Campe, K., Fraedrich K. , and Lunkeit F. , 2009: Global climatology of Convective Available Potential Energy (CAPE) and Convective Inhibition (CIN) in ERA-40 reanalysis. Atmos. Res., 93, 534–545.
Rienecker, M. M., and Coauthors, 2008a: The GEOS-5 data assimilation system—Documentation of versions 5.0.1 and 5.1.0. NASA GSFC Tech. Rep. NASA/TM-2007-104606, 92 pp.
Rienecker, M. M.