1. Introduction
The U.S. National Oceanic and Atmospheric Administration (NOAA) has acquired 20+ years of global multichannel infrared (IR) and solar-visible spectrum radiance data from the Advanced Very High Resolution Radiometer (AVHRR) flown on board NOAA sun-synchronous environmental satellites. Recognizing the climatological value in these remotely sensed global data, NOAA's National Environmental Satellite, Data and Information Service (NESDIS) has sought to generate long-term geophysical datasets through careful reprocessing of the archived AVHRR Level 1B radiometric data. In collaboration with the U.S. National Aeronautics and Space Administration (NASA), the NOAA–NASA Pathfinder Program was established to support initial (i.e., “pathfinder”) reprocessing efforts, culminating in the Pathfinder Atmospheres (PATMOS) (Stowe et al. 2002; Jacobowitz et al. 2003) and Oceans Pathfinder (Kilpatrick et al. 2001) datasets. Among other things, the AVHRR archive is valuable for producing global climate data records of sea surface temperature (SST) and aerosol optical depth (AOD), two important parameters for climate change studies that are the focus of this paper.
Although conventional SST climate data records are based upon in situ observations (e.g., from oceanographic buoys and/or ships), satellite data provide more complete global coverage, with higher spatiotemporal resolution. To date, the most comprehensive effort to derive accurate satellite SST climatology has been the European Remote Sensing Satellite (ERS) Along Track Scanning Radiometer (ATSR), a multichannel instrument with dual-view capabilities (e.g., Merchant et al. 1999). The NOAA AVHRR, on the other hand, provides the longest continuous data record (all freely accessible in the public domain) and is, thus, well suited for developing and distributing SST climate datasets (e.g., Kilpatrick et al. 2001; Armstrong and Vazquez-Cuervo 2001; Nalli 2003, 2004).
Satellite remote observing systems, such as AVHRR, provide distinctly different measurements from those obtained in situ. Satellite sensors observe the spatial mean radiance from the sensor field of view (FOV), typically spanning a spatial area on the order of 1–100 km2 at the surface. Geophysical parameters such as SST or AOD are derived from radiance observations using retrieval algorithms. In the case of IR surface emission, from which SST is derived, the radiance signal originates from the electromagnetic skin of the surface. In situ instruments, on the other hand, are in direct contact with the fluid medium, measuring the mean bulk properties at a particular point. These differences in observing systems can lead to various systematic and random differences between otherwise “coincident” observations. Because the majority of the SST record is from in situ ship and buoy observations, this poses a problem of “splicing” satellite data with the presatellite era in situ record. There are three error types that occur with any type of analysis. The type of error most often discussed is random error, which is the observational error caused by the instrument and/or the observer. Sampling error is the analysis error that occurs when the distribution of observations is not uniform. This error may be large for in situ–only analyses but is usually much smaller for satellite analyses because of the spatial density of satellite observations. The remaining source of error is bias error that is due to a systematic difference between one instrument or set of instruments and another. This is of particular concern for satellite data because of the high density of observations coming from only a small number of satellite instruments. Satellite biases can arise from limitations in the detector calibration and/or retrieval algorithms. Note that for climate change detection, satellite SST products must retain a high degree of absolute accuracy, namely ≲0.1 K decade−1 system stability (e.g., Allen et al. 1994).
Unlike random errors, systematic retrieval errors cannot, in general, be reduced by spatial and/or temporal averaging. To address such problems, NOAA has developed the Reynolds Optimum Interpolated SST (OISST) analysis product (Reynolds et al. 2002). The Reynolds OISST, derived from the “blending” of AVHRR with concurrent in situ SST observations, has been very well received in the geoscience community. Its success may arguably be attributed to the fact that, by effectively combining the strengths of IR satellite SST retrievals with an unbiased ensemble of in situ observations (Reynolds 1988; Reynolds and Smith 1994), a reliable, high-resolution (≃1°), global SST analysis can be attained. While the OISST with in situ bias correction is relatively free of systematic errors (relative to buoy data), there remains a small residual global bias (about −0.03 K over 60°S–60°N) that is difficult to remove completely using biased satellite SST data (Reynolds et al. 2002). Furthermore, in regions of sparse in situ data (e.g., poleward of 40°S), it is very difficult to reduce bias. Therefore, prior bias correction of satellite data is desirable to minimize any persistent residual biases in the analyses (Reynolds et al. 2002). Ideally, we seek to correct or eliminate the satellite biases using only satellite data and, thus, potentially obviate the need for an in situ correction.
Apart from gas absorption (namely, H2O), aerosols and clouds are the two primary atmospheric sources of radiance attenuation that, when unaccounted for, ultimately lead to biases in satellite IR SST retrievals. However, clouds, unlike aerosols, are generally opaque, meaning that they have zero transmittance.1 Because of this, the problem of cloud detection is usually treated as a problem distinct from retrievals, and cloud detection algorithms are thus developed and refined by dedicated specialists. We therefore direct our attention exclusively to the remaining problem of accounting for noncloudy tropospheric and/or stratospheric aerosol (e.g., Griggs 1985; Walton 1985; Xu and Smith 1986; Rao 1992; May et al. 1992; Reynolds 1993; Merchant et al. 1999; Diaz et al. 2001; Nalli and Stowe 2002; Vázquez-Cuervo et al. 2004). In the marine troposphere, seasonally persistent elevated levels of aerosol are found to occur in regions immediately downwind of continental sources. The primary sources of these aerosol are mineral dust from deserts (e.g., the Sahara), smoke from biomass burning sites (e.g., savanna grasslands in sub-Saharan Africa), and industrial emissions from large urban areas (e.g., the U.S. east coast urban corridor). In the stratosphere, the sulfate aerosol layer is perturbed substantially above background levels following the injection of SO2 from major volcanic eruptions (e.g., El Chichón and Mount Pinatubo). Although such eruptions occur relatively infrequently, the residence time of the enhanced sulfate layer can last a year or longer. Because the stratosphere is considerably colder than the surface, the impact on IR SSTs is also greater.
This paper presents optimum interpolated (OI), daytime AVHRR SST analyses derived from PATMOS. Two satellite bias correction methodologies are used: the first being a simple aerosol correction using information contained in AOD estimates derived from the solar reflectance channel 1 (0.63 μm) of AVHRR (following Nalli and Stowe 2002; Nalli 2003), and the second being an in situ correction of satellite biases based upon the classical solution of Poisson's equation (following Reynolds 1988). We consider only daytime AVHRR data because the AOD (visible channel) estimates are available only during daytime. In section 2, the PATMOS matchup data, multichannel SST algorithm, and bias-correction algorithms are described in detail. In section 3, four separate OI analyses are presented and intercompared based upon different combinations of PATMOS input data. Statistical analyses of residuals are computed relative to the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI) OI SST (e.g., Wentz et al. 2000) and in situ Extended Reconstruction SST version 2 (ERSST.v2) (Smith and Reynolds 2004). The section concludes with a brief demonstration of the 1985–2000 PATMOS OISST data product.
2. Methodology
In this section, the AVHRR PATMOS data and SST retrieval method are described, including the aerosol bias correction algorithm. The OI methodology used for computing SST climate analyses, including the in situ correction of satellite biases, is then reviewed.
a. Data: AVHRR Pathfinder Atmospheres (PATMOS)
PATMOS is a 20-yr (1981–2001) climate dataset derived from the reprocessing of AVHRR five-channel Level 1b, Global Area Coverage (GAC) subsampled radiometric data from the NOAA-7, -9, -11, and -14 “afternoon” satellites and subsequently mapped to a quasi–equal area grid (110 km)2 (Stowe et al. 2002; Jacobowitz et al. 2003). PATMOS-1 data consist of the grid cell means and standard deviations of AVHRR channels 1 and 2 (0.63 and 0.83 μm) normalized solar reflectances, and channels 3–5 (3.7, 11, 12 μm) thermal IR radiances. Cloud-cleared grid cell statistics (also called “clear sky”) were obtained for each channel using the Clouds from AVHRR (CLAVR-1) cloud detection algorithm (Stowe et al. 1999). In a second processing step, geophysical climate parameters, including daytime retrievals of channel 1 AOD, τa(λ1 = 0.63 μm), are derived from the grid cell mean radiance statistics (i.e., PATMOS-2 derived products). The PATMOS AOD is derived over oceans (in solar-illuminated areas from “ascending” orbital passes) from the clear-sky, channel 1 statistics based upon a single-channel retrieval algorithm described in Stowe et al. (1997). As shown in Fig. 1, these AOD data were obtained from multiyear, multisatellite observations under conditions of tropospheric haze, dust, and smoke outflows from continents, as well as during the entire residence time of the Mount Pinatubo (1991–93) stratospheric aerosol layer. Note that there is a data gap between NOAA-11 and -14 that occurred when NOAA-11 failed in September 1994 and its replacement, NOAA-14, did not become operational until February 1995 (Stowe et al. 2002).
The AVHRR PATMOS gridded data have been merged with in situ buoy matchups (1985–99) provided by the Oceans Pathfinder Matchup Database (Kilpatrick et al. 2001). The merged data, referred to as PATMOS-BUOY, provide a unique matchup database from which SST algorithms, including aerosol bias corrections, can be derived empirically. The PATMOS-BUOY data have been subjected to rigorous quality assurance (QA) procedures to reduce errors and enhance parameter correlations (for more details, see Nalli and Stowe 2002; Ignatov and Nalli 2002). An overview of the PATMOS SST algorithms is presented below.
b. Multichannel SST retrieval
The AVHRR has three IR window channels with peak responses located within spectral regions where the majority of cloud-free atmospheric attenuation is due to H2O continuum absorption. Fortunately, because the spectral variation of transmittance correlates strongly with water vapor loading, multispectral SST retrieval algorithms that correct for atmospheric H2O can be derived statistically from satellite–buoy matchup training samples (e.g., McMillin 1975; McClain et al. 1985; Walton et al. 1998).
Equation (1) provides a brightness temperature correction for atmospheric water vapor,2 but not elevated levels of atmospheric aerosol. The problem of aerosol correction is discussed below.
c. Aerosol bias correction
Although water vapor is the primary atmospheric IR absorber, aerosols also attenuate the surface-leaving radiance through absorption and scattering. Unfortunately, it is difficult, if not impossible, to separate the spectral signal caused by atmospheric aerosols from that caused by water vapor using only two or three IR window channels. The empirical coefficients, bi, in Eq. (1) do account for aerosol extinction in a mean sense (i.e., under typical marine background aerosol conditions), but they are not valid when elevated levels of aerosol are present. Under such conditions, equations such as (1) yield systematic errors that are biased negative (e.g., May et al. 1992; Reynolds 1993; Merchant et al. 1999; Nalli and Stowe 2002). The absolute magnitude of these systematic errors often exceeds 0.5 K (e.g., Nalli and Stowe 2002).
sampling differences between the in situ point measurements and PATMOS (110)2 grid cell mean SST observations, which includes both horizontal and vertical (namely, skin versus bulk layer) variations in SST that are independent of aerosols;
variation in aerosol extinction associated with global differences in aerosol radiative properties (resulting from different aerosol types);
random error (i.e., σ ≃ 0.07) in the PATMOS AOD product (Stowe et al. 2002);
precision loss in the PATMOS IR radiance data due to an inadvertent rounding of pixel-level brightness temperatures (from 0.1 to 1.0 K) that occurred during PATMOS processing (Stowe et al. 2002; Nalli and Stowe 2002), which is estimated to contribute 0.2 K in random error, but systematic error is essentially unaffected; and
residual cloud contamination in the GAC pixels used for determining clear-sky statistics.
Although we employ QA procedures designed to minimize residual cloud contamination in PATMOS grid cells (e.g., Nalli and Stowe 2002), it is impossible to identify and eliminate all cloud contamination. However, under such conditions Eq. (3) will nonetheless provide a partial correction for these residual clouds implicitly since the cloud signal will be manifested in the τa retrieval (e.g., Ignatov and Nalli 2002).
Validation analyses using independent PATMOS-BUOY matchup samples (i.e., matchups not used for training) for each of the the NOAA satellites (NOAA-9, -11, and -14) confirm that global aerosol biases in the PATMOS MCSST are significantly reduced using the correction methodology (results are not shown here in the interest of brevity). We note here, however, that there are some limitations in both the data and method that will contribute to errors in the SST product. These are summarized as follows:
There are residual calibration errors and/or trends in the solar reflectance channels 1 and 2 that vary from satellite to satellite.
Residual regional SST bias, both positive and negative, is expected from nonlinear variations in aerosol extinction associated with variations in global aerosol types.
Due to nighttime limitations in the PATMOS CLAVR-1 (Stowe et al. 1999) cloud mask, and the unavailability of AOD retrievals, the PATMOS SST are limited to daytime retrievals only. Daytime conditions pose an additional complication of skin-buoy SST decoupling [discussed in more detail under subsection 3a(1)].
As mentioned earlier, there are random errors in the PATMOS AOD retrievals as well as in the PATMOS IR grid cell radiances.
d. Optimum interpolation (OI)
Reynolds (1988) was the first to introduce and demonstrate the utility of combining satellite and in situ SST for deriving global SST analyses. Drawing upon this seminal work, Reynolds and Smith (1994) later went on to develop fully what is now known as the NOAA Reynolds Optimum Interpolated SST (OISST) analysis product, based upon the optimum interpolation (OI) spatial analysis method of Gandin (1963). For details on the OISST analysis methodology and data product, the reader is referred to Reynolds et al. (2002), and Reynolds and Smith (1994). What follows here is a very brief description of the method5 in its application to the present work.
Note that the OI analysis based upon Eqs. (8) and (9) does not correct for any bias contained in the observations, ϕi. This is why Eqs. (6) and (7) are first used to reduce large-scale satellite biases in the conventional OISST product (Reynolds and Smith 1994). In the next section, results from the PATMOS OISST weekly daytime climate products, both with and without bias corrections, are presented.
3. Results
In this section, PATMOS OISST analyses are presented based upon the input satellite SST data sources described above as follows:
PATMOS MCSST, Eq. (1);
PATMOS ACSST (MCSST with aerosol satellite bias correction), Eq. (5);
PATMOS MCSST with in situ satellite bias correction, Eqs. (1) and (7); and
PATMOS ACSST with in situ satellite bias correction, Eqs. (5) and (7).
To conduct a quantitative assessment of each of the four OI runs, quasi-independent SST measurements are necessary. In this work, we will use two different published SST analysis datasets: one primarily radiometric and one completely in situ based. The first is the TRMM Microwave Imager (TMI) OI SST product developed by Remote Sensing Systems (RSS) (e.g., Wentz et al. 2000; Gentemann et al. 2004). The TMI OI SST data only date back to January 1998, and are limited to 40°S–40°N. Therefore, we also rely on a second dataset, namely the extended reconstruction SST version 2 (ERSST.v2) developed by Smith and Reynolds (2003, 2004), to provide comparisons prior to 1998, including the crucial period of 1991–93 when IR data were severely contaminated by the eruption of Mount Pinatubo.
It is noted here that the ERSST is based upon in situ measurements similar to ours. While the TMI OI SST product provides a greater degree of independence from our data, the retrieved TMI SSTs are adjusted prior to the OI based upon 5-day means of the daily mean differences between collocated TMI and moored buoys (C. Gentemann 2005, personal communication). Thus, the ERSST and TMI OI SST datasets may be considered independent with respect to analyses 1 and 2, but only quasi independent for IR–in situ blended analyses 3 and 4. In any event, these datasets can assess the presence of aerosol- or moisture-dependent biases in all four analyses because such biases originate from the IR satellite data and are independent of in situ and/or MW measurements. Reynolds et al. (2002), for example, also used in situ data as “the standard of comparison” under a section discussing biases in the OISST analysis method.
a. Comparison against TMI OI SST
The TMI instrument constitutes the first multichannel passive microwave (MW) sensor suitable for accurate retrieval of SST (Wentz et al. 2000; Gentemann et al. 2004). Because MW radiation, unlike IR, is not absorbed or scattered significantly by clouds and/or aerosol (at frequencies ≲12 GHz), SST retrieved from TMI provide a useful independent data source for assessing the corresponding errors known to occur in IR retrievals. The TMI SST is retrieved using a statistical retrieval algorithm derived from a training sample based upon radiative transfer model (RTM) calculations. For more information on the TMI instrument and SST retrieval, the reader is referred to Wentz et al. (2000) and Gentemann et al. (2004). It is important to note that for the TMI OI SST product, the retrieved SSTs are “normalized” to the diurnal minimum at 0800 local standard time (LST) using the empirical model developed by Gentemann et al. (2003).
Similar to our OI analyses, the TMI OI SST product is derived via an optimum interpolation methodology, but is capable of higher spatial and temporal resolutions (≃25 km, daily). For simplicity in this work, we compute weekly means of the TMI OI SST, then linearly interpolate in space to the 1° PATMOS OISST resolution. We recognize that it would be more rigorous to compare analysis data at the same space and time resolution, but such analyses were not readily available.
1) SST measurement considerations: IR, MW, and in situ
In using TMI as a quasi-independent radiometric dataset, it is important to recognize how its measurement of SST will differ from the IR and in situ–blended methods described in this work. Because the TMI algorithm is derived from RTM training, the retrieved SST, TMWS, is representative of the bottom of the skin layer on the order of 1-mm depth (i.e., the “subskin,” denoted here as z1); that is, TMWS ≈ Ts(z1) (e.g., Donlon et al. 2002; Gentemann et al. 2004). In contrast to this, our empirical–statistical algorithms, namely Eqs. (1) and (3), are derived from AVHRR training data regressed against bulk-layer SST point measurements from buoys at z2 ≃ 0.5–1.0 m depth, Ts(z2). Although the AVHRR IR channel radiances are sensitive to the skin-layer SST, Ts(z0), where z0 ≃ 1–10 μm, Eqs. (1) and (5) nevertheless predict TIRS ≈ Ts(z2) because they have been “trained” to do so.
To complicate matters further, the daytime data used throughout this work will be impacted by diurnal warming and stratification of the ocean surface layer (e.g., Donlon et al. 2002; Gentemann et al. 2003). The magnitude of this impact will depend on the amount of solar insolation and turbulent mixing caused by surface winds. Under conditions of high solar insolation (i.e., under cloud-free skies typical of IR retrievals) and low wind speeds, significant stratification can result in the subskin SST being warmer than SST at depth, that is, δzTs ≡ Ts (z2) – Ts (z1) < 0. Conversely, under conditions of low insolation (i.e., under cloudy skies) or high wind speeds, the surface layer tends to be more isothermal, with the subskin SST being cooler than the SST at depth, δzTs > 0 [see Donlon et al. (2002) for an illustration of this]. In our case, Eqs. (1) and (3) are derived under daytime conditions, which implies they should predict SST with a mean negative difference from the subskin SST, 
2) Aerosol-related biases
Figures 4 and 5 show the geographic seasonal distribution of mean residuals (relative to TMI) over the TMI–PATMOS overlap data period (1998–2000). The top and bottom two plots in each figure show the results obtained without and with the in situ satellite bias correction Eq. (7), respectively. Likewise, the left- and right-hand plots show results obtained without and with the aerosol correction Eq. (5).
In these figures, we first turn our attention to the aerosol-related negative SST biases. Note that there are three boxes in each map that outline regions of persistent tropospheric aerosol known to be problematic for causing biases: 1) equatorial Atlantic Ocean west of sub-Saharan Africa (smoke from biomass burning), 2) Arabian Sea (mineral dust), and 3) North Atlantic Ocean west of the Sahara (mineral dust). In the top-left plots, significant regional aerosol biases are clearly evident. These are reduced globally in the PATMOS data from the aerosol satellite bias correction shown in the top-right plots. While the overall aerosol correction (analysis 2; top-right plots) provides an improvement in terms of bias, there are some situations where the algorithm may undercorrect somewhat, for example, the Saharan box in July (Fig. 5) as evidenced by comparing against the in situ satellite bias correction (Fig. 5, bottom left). There are also situations where positive bias is apparently “generated” by the algorithm (e.g., Saharan box in January, Fig. 4; biomass box in July, Fig. 5), but these may be attributed, at least in part, to seasonal diurnal warming, which is discussed in more detail in section 3a(3).
To examine more quantitatively the seasonal cycle in these aerosol-prone regions, what follows are statistical analyses of the PATMOS OISST spatial mean residuals (relative to TMI) for temporal trends using box and whisker plots. The box and whisker plots quantify graphically the robust central tendency and dispersion of residuals within the regions specified (see the Fig. 6 caption for definition). Figures 6 –8 show the results grouped by month, revealing seasonal trends, for the biomass burning, and the Arabian and Saharan dust boxes, respectively. The plots in these figures are arranged in the same order as the maps in the previous figures, with aerosol-corrected plots on the right and in situ satellite-bias-corrected figures on the bottom. The negative biases present in the uncorrected analyses (top-left plots) are effectively eliminated (in a yearly mean sense) in the aerosol-bias-corrected results (top-right plots), and the dispersion is also moderately reduced. However, some residual negative biases remain in the biomass burning box during the Northern Hemisphere (NH) spring (Fig. 6, top right), Arabian box during May–July (Fig. 7, top right), and Saharan box during NH summer and December (Fig. 8, top right).
As documented in the earlier work by Reynolds et al. (2002), the in situ satellite bias correction (analyses 3 and 4; bottom plots) is very effective at consistently eliminating aerosol bias. Note that the dispersion is also reduced. However, because of the blend with satellite data, regions with persistent aerosol bias are difficult to remove completely, and are evident in all three aerosol regions (Figs. 6 –8) in a pattern similar to analysis 2 (top right). In the Saharan box (Fig. 8), analysis 3 (bottom left) yields a very small residual negative bias throughout the entire year. The use of aerosol-bias-corrected SST data along in situ satellite bias correction (bottom right) helps to reduce this persistent bias provides a slight improvement.
Figures 6 –8 also reveal pronounced seasonal cycles of residuals with positive biases, especially analysis 2 (top right). These biases occur most notably during July–October in the biomass burning box (Fig. 6), during August–December and January–April in the Arabian box (Fig. 7), and during March in the Saharan box (Fig. 8). As already mentioned, these positive biases are not necessarily an overcorrection, but may be expected from seasonal diurnal warming and stratification of the surface layer; this is discussed in more detail below.
3) Nonaerosol biases and diurnal warming
We now consider the occurrences of zonal and regional biases in the corrected data (Figs. 4, 5 and 6 –8, excluding top-left plots) not necessarily associated with aerosol. First there is the pronounced positive zonal bias that occurs in each hemisphere during their respective summers, that is, the Southern Hemisphere (SH) during January (Fig. 4) and the NH during July (Fig. 5). To a lesser magnitude, there are also some negative biases in each hemisphere during their respective winters. Because of the different sampling of the PATMOS (daytime-only observations of Ts(z2), both from AVHRR and in situ) and the TMI (day and night observations of Ts(z1) normalized to 0800 LST) analyses mentioned above, these are indicative of the seasonal cycle of diurnal warming associated with corresponding variations in insolation (e.g., Gentemann et al. 2003). At least two prior studies, namely Stuart-Menteth et al. (2003) and Tanahashi et al. (2003), have reported a seasonal pattern of diurnal warming in the midlatitudes during summer as well as coastal waters in the Tropics. During certain seasons, regional biases are also seen in our data. For example, positive bias occurs along the equator to the west of Ecuador, apparently independent of season. Consistent with our arguments above, these biases are partially indicative of diurnal warming associated with low surface wind speeds.6 Stuart-Menteth et al. (2003) also found regional patterns of diurnal warming similar to those observed here (cf. Figs. 4 and 5 against Fig. 1, op. cit.). Our results agree reasonably with, and in fact corroborate, their findings on diurnal warming. The seasonal cycle of diurnal warming observed in the Arabian Sea (Fig. 7, top right), for example, has been specifically documented by Stuart-Menteth et al. (2003).7
Some of these regional and zonal biases are mitigated, both in magnitude and extent, from the in situ correction of satellite biases (bottom plots). While the diurnal warming signal is reduced (bottom plots), some biases remain because diurnal warming can, in fact, reach water depths of ≃1 m, especially during high isolation and low wind speeds (e.g., Stuart-Menteth et al. 2003; Gentemann et al. 2004). The fact that the seasonal cycle of positive biases remain in the in situ–corrected data (Figs. 6 –8, bottom plots) demonstrates that the pronounced positive biases found in the aerosol-corrected data (top-right plots) cannot be solely attributed to algorithm overcorrection. It must also be noted that because of significant differences in the spatial and temporal resolutions of the raw measurements (AVHRR, TMI, and in situ), biases also remain in all four analyses within regions of high variability (e.g., the Gulf Stream) (e.g., Reynolds et al. 2002; Gentemann et al. 2004).
Table 3 summarizes the statistics for residuals derived from all four PATMOS OISST analyses for the aerosol contamination boxes as well as zonal regions 40°–20°S, 20°S–20°N and 20°–40°N. For OI analysis 2 (aerosol-bias-corrected data), there are slight negative residuals in the SH midlatitudes (−0.06°C) and tropical zone (−0.09°C), whereas a small positive bias is apparent in the NH midlatitudes (+0.16°C). The SH bias is more pronounced with the in situ satellite bias correction (−0.16°C), although it is not perfectly clear why this is, and the NH positive bias is eliminated, possibly because of reduced diurnal warming at depth. Note that the SH and tropical zones encompass a significant fraction of the earth's total ocean surface area and, thus, carry a relatively high weight in calculating the 40°S–40°N mean residual biases.
b. Comparison against ERSST.v2
As mentioned at the beginning of section 3, the TMI OI SST is spatially limited to 40°S–40°N, and it does not include the period prior to 1998. To allow comparisons outside of these bounds, we use the in situ ERSST.v2 (Smith and Reynolds 2004) as an independent (quasi-independent) dataset for analyses 1 and 2 (analyses 3 and 4). Among other things, we examine the impact of the Mount Pinatubo (1991–93) stratospheric aerosol layer upon the PATMOS OISST products.
The ERSST is a global, 2° gridded analysis based upon the Comprehensive Ocean–Atmosphere Data Set (COADS), a well-established long-term historical archive of daytime and nighttime in situ SST observations. For simplicity in this work, the 2° monthly ERSST is linearly interpolated in space and time to the 1° weekly PATMOS OISST resolution. OISST residuals for each run, defined by ΔTS = TIRS − TERS, are then computed for each week and 1° grid cell within the zonal region bounded by 50°S and 60°N. Reynolds et al. (2002) constrained their statistical analyses to the region 60°S–60°N in an effort to avoid contamination by sea ice. We too sought initially to conduct our statistical assessment over this region for consistency, but subsequently found the satellite SST data to exhibit unacceptable negative bias poleward of 45° in the SH. One cause of this bias may be the misclassification of sea ice and/or clouds in PATMOS (e.g., the PATMOS cloud fraction greatly increases poleward of 50°S) (Stowe et al. 2002). Note also that the daytime geographic coverage from AVHRR substantially diminishes as one progresses toward the poles (e.g., see Fig. 1). Some bias may be due to the ERSST analysis itself because in situ data are very sparse south of 40°S. Finally, the presence of high wind speeds over these latitudes (cf. Gentemann et al. 2003, Fig. 4b) will suppress both diurnal warming and thermal stratification of the surface layer. This would yield buoy measurements close to, or warmer than, the skin SST throughout the day, whereas our Eqs. (1) and (5) will predict SST with a mean negative difference from the skin SST, thus possibly contributing to the negative bias.
1) Aerosol-related biases
Figure 9 shows the geographic distribution of mean residuals over the ERSST–PATMOS overlap data period, 1998–2000 including Mount Pinatubo, with maps arranged in the same order as previous figures. We again first consider aerosol-related biases only. In the analysis using uncorrected data (top left), the impacts of Saharan dust and Mount Pinatubo aerosol attenuation are clearly apparent as cool bias throughout the tropical latitudes, with values reaching −1°C in the Saharan and biomass aerosol regions. These biases are greatly reduced in the aerosol-bias-corrected data (top right), as well as in the data with in situ satellite bias correction (bottom plots).
The long-term stability of the analyses over the spatial domain and sampling period including Mount Pinatubo is demonstrated in the box and whisker plots in Fig. 10. In the uncorrected data (top left), the analysis is severely contaminated by the Mount Pinatubo eruption as indicated by the large dispersion and negative biases (< −0.6°C) during 1991–92. There is also persistent negative bias associated with tropospheric aerosol throughout 1985–2000. The negative biases are practically eliminated in the aerosol-bias-corrected data (top right), and there are no discernable trends over the time period spanning three satellites. The in situ satellite bias corrections (bottom plots) demonstrate significantly less dispersion, but retain a very small residual negative bias, particularly during the Mount Pinatubo period, for reasons discussed in section 1.
2) Nonaerosol biases/diurnal effects
There are once again certain biases evident in Fig. 9 that are unrelated to aerosols. First, there are the negative biases in the SH poleward of ≃45°S as already discussed above. These cannot be readily corrected because of the sparsity of in situ (and satellite) data, as well as the diurnal sampling issues already mentioned and, thus, are manifested in all four analyses. Second, there are large positive biases to the south of the Gulf Stream gradient and cold biases to the north of the gradient. This results from the interpolation of the courser spatial and temporal resolution (2 km, monthly) of the ERSST.v2 to the higher resolution of the PATMOS OISST (1 km, weekly) (e.g., Reynolds et al. 2002). Finally, we note what appears to be enhanced positive bias in the aerosol-corrected analysis 2 (top-right map). Notably, these include the Arabian Sea, the Bay of Bengal, the Yellow Sea, the eastern North Pacific Ocean along the 30°N axis, the west coasts of North and South America, the Gulf of Mexico, the North Atlantic Ocean, and the west coasts of the Sahara and southern Africa. Although some of these biases may result in part from overcorrection of aerosols, we argue that they are indicative of diurnal warming (cf. Gentemann et al. 2003, Figs. 4a,c) using arguments similar to those in sections 3a(1) and 3a(3).
c. Discussion
The above statistical analyses demonstrate that the PATMOS 1°, daytime weekly OISST analyses developed in this work compare favorably against radiometric MW TMI and in situ–based ERSST.v2 analyses. As would be expected, the least reliable OI results are obtained from satellite data without any bias correction (OI analysis 1). The aerosol bias correction (OI analysis 2), derived from Eq. (3), provides a global mean correction for tropospheric dust and smoke, as well as volcanic aerosol from Mount Pinatubo. The in situ correction for satellite biases (OI analysis 3), derived from Eq. (7), corrects both negative and positive satellite biases, and also reduces the magnitude of the diurnal warming signal. The combined correction (i.e., analysis 4) yields a very slight reduction in aerosol-related bias, particularly in regions subject to persistent tropospheric aerosol bias (namely, Saharan dust). Both analyses 3 and 4 also reduce the dispersion found in analyses 1 and 2. Analysis 4 minimizes the diurnal effects and provides the best continuity with respect to prior in situ records. It is emphasized here, however, that aerosol contamination is significantly reduced in analysis 2, which is derived strictly from daytime satellite imager data without depending upon an in situ–based correction. It is also of critical importance to note that in situ data can only be used to correct satellite biases in regions with in situ data. Because in situ data are very sparse south of 40°S, it is difficult to correct satellite biases there. Thus, aerosol-bias-corrected SST may be the only correction available.
Figure 11 summarizes the 1985–2000 daytime AVHRR PATMOS OISST record (analysis 4) in terms of weekly zonal-mean (50°S–60°N) anomalies with respect to ERSST.v2 climatological mean computed for the 1960–89 base period. This figure reveals seasonal, interannual, and decadal trends in SST detectable by our analysis, and the data are consistent over the NOAA-9, -11, and -14 mission lifetimes. A general warming trend in SST is readily observable from tropical to midlatitudes over the data period, although the trend is dampened from 1992 to mid-1994 because of surface cooling from Mount Pinatubo sulfate aerosol. The zonal evolution in the phases of the El Niño–Southern Oscillation (ENSO) cycle are observable, with El Niño occurring during late 1986–1988, 1991–mid-1992, 1993, late 1994, and 1997–mid-1998, and La Niña occurring during 1985, 1988–89, and late 1998–2000 (e.g., McPhaden 2004). Warm anomalies are seen propagate from the Tropics into the midlatitudes (especially in the NH) for several years following major El Niño events.
4. Summary and future directions
Sixteen years (1985–2000) of reprocessed daytime (ascending node) AVHRR PATMOS gridded data from the NOAA sun-synchronous afternoon satellites NOAA-9, -11, and -14 were used to derive optimum interpolated SST (OISST) analyses (after Reynolds and Smith 1994). The PATMOS data included normalized solar reflected and IR emitted radiances, along with AOD retrieved from channel 1 (0.63 μm), all mapped to a global (110 km)2 equal-area grid.
Two satellite bias correction methods were employed: namely, the statistical aerosol correction of Nalli and Stowe (2002) and the in situ correction of satellite biases of Reynolds (1988). Global aerosol bias correction equations were derived from observed statistical relationships in PATMOS–BUOY matchup data between slant-path AOD and MCSST depressions for elevated levels of tropospheric (e.g., from continental mineral dust) and stratospheric (e.g., Mount Pinatubo sulfate) aerosol.
Four weekly 1° equal-angle OISST analyses were generated based on four combinations of PATMOS input data: 1) MCSST without any satellite bias correction, 2) MCSST with aerosol satellite bias correction, 3) MCSST with in situ correction of satellite biases, and 4) MCSST with both aerosol and in situ correction of satellite biases. These analyses were then compared statistically against the MW-based TMI OI SST and the ERSST.v2 in situ–derived analysis products. Optimum interpolation analysis 1 was found to exhibit large negative and positive biases as would be expected, while OI analysis 2 was found to reduce, in a global sense, the negative bias associated with elevated atmospheric aerosol. Analysis 2, which was derived solely from satellite data, revealed variations in diurnal warming consistent with the findings of Donlon et al. (2002), Stuart-Menteth et al. (2003), Tanahashi et al. (2003), and Gentemann et al. (2003, 2004). The aerosol bias correction algorithm used in analysis 2 was particularly effective at removing bias associated with Mount Pinatubo stratospheric aerosol. OI analyses 3 and 4, derived using the in situ satellite bias correction, greatly mitigated all biases, including those resulting from diurnal effects, and also reduced the random error. Both of these analyses retained a small residual negative bias during the Mount Pinatubo residence time. Analysis 4 performed slightly better than analysis 3 under conditions of persistent aerosol (namely, the Saharan dust box and Mount Pinatubo). However, south of 40°S, in situ data are very sparse and the corrections in analyses 3 and 4 become unreliable. Thus, improvement of satellite data using the correction methods such as the methods that we have presented here are essential to produce a reliable global SST product.
The PATMOS OISST 1985–2000 daytime climate analyses derived in this work provide a relatively high-resolution (l° weekly) global database for studying seasonal and interannual climate processes. New AVHRR reprocessing efforts currently under way at NOAA/NESDIS (i.e., PATMOS-x) will improve upon the known limitations in PATMOS (e.g., Heidinger et al. 2005) from which successive iterations can be undertaken toward deriving optimal SST climatologies (Nalli 2004). Retrospective bias corrections that can potentially improve upon those presented here should be explored.
We advocate that satellite biases should be corrected and eliminated for individual retrievals, as was done in analysis 2, rather than for averages, as was done by the in situ Poisson correction in the OI (analyses 3 and 4). This paper is the first step to correct explicitly for daytime AVHRR retrieval biases due to aerosol contamination. Other procedures are needed to continue the effort to correct satellite biases. Diurnal signals over the daytime, for example, are not resolved in the OI and thus may appear as biases, and nighttime data have not yet been corrected. Subsequent work might include computation of daily analyzed fields of AOD. These fields could be interpolated to allow aerosol corrections of the nighttime AVHRR retrievals. Another important step would be to utilize MW satellite SST data as an additional data source (e.g., Reynolds et al. 2004; Gentemann et al. 2004). These data have different error characteristics that are not dependent on clouds and aerosols and thus provide an independent method of detecting satellite biases.
Acknowledgments
This research was supported by the NESDIS/ORA Satellite Meteorology and Climatology Division (M. D. Goldberg) under ORA contract, and by the NESDIS Ocean Remote Sensing Program (L. Dantzler and E. Bayler) under CIRA grant (T. Vonder Haar). We gratefully acknowledge L. L. Stowe (retired) for his foresight and leadership in AVHRR PATMOS reprocessing and aerosol–SST algorithms at NOAA/NESDIS. A. Kidd (NOAA Satellite Active Archive) and K. Knapp (NESDIS/NCDC) assisted us in accessing the PATMOS daily archive. Microwave OI SST data are produced by Remote Sensing Systems and sponsored by the NASA Earth Science Physical Oceanography Program and the NASA REASoN DISCOVER Project. Data are available online (at www.remss.com).
The views, opinions, and findings contained in this report are those of the authors and should not be construed as an official National Oceanic and Atmospheric Administration or U.S. government position, policy, or decision.
REFERENCES
Allen, M. R., C. T. Mutlow, G. M. C. Blumberg, J. R. Christy, R. T. McNider, and D. T. Llewellyn-Jones, 1994: Global change detection. Nature, 370 , 24–25.
Armstrong, E. M., and J. Vazquez-Cuervo, 2001: A new global satellite-based sea surface temperature climatology. Geophys. Res. Lett., 28 , 4199–4202.
Diaz, J. P., M. Arbelo, F. J. Expósito, G. Podestá, J. M. Prospero, and R. Evans, 2001: Relationship between errors in AVHRR-derived sea surface temperature and the TOMS Aerosol Index. Geophys. Res. Lett., 28 , 1989–1992.
Donlon, C. J., P. J. Minnett, C. Gentemann, T. J. Nightingale, I. J. Barton, B. Ward, and M. J. Murray, 2002: Toward improved validation of satellite sea surface temperature measurements for climate research. J. Climate, 15 , 353–369.
Gandin, L. S., 1963: Objective Analysis of Meteorological Fields. Gidrometeorirzdar, 238 pp. (Translated from Russian by Israeli Program for Scientific Translations in 1965.).
Gentemann, C. L., C. J. Donlon, A. Stuart-Menteth, and F. J. Wentz, 2003: Diurnal signals in satellite sea surface temperature measurements. Geophys. Res. Lett., 30 .1140, doi:10.1029/2002GL016291.
Gentemann, C. L., F. J. Wentz, C. A. Mears, and D. K. Smith, 2004: In situ validation of Tropical Rainfall Measuring Mission microwave sea surface temperatures. J. Geophys. Res., 109 .C04021, doi:10.1029/2003JC002092.
Griggs, M., 1985: A method to correct satellite measurements of sea surface temperature for the effects of atmospheric aerosols. J. Geophys. Res., 90 , 12951–12959.
Heidinger, A. K., M. D. Goldberg, D. Tarpley, A. Jelenak, and M. J. Pavolonis, 2005: A new AVHRR cloud climatology. Proc. SPIE, 5658 , 197–205. doi:10.1117/12.579047.
Ignatov, A., and N. R. Nalli, 2002: Aerosol retrievals from the multiyear multisatellite AVHRR Pathfinder Atmosphere (PATMOS) dataset for correcting remotely sensed sea surface temperatures. J. Atmos. Oceanic Technol., 19 , 1986–2008.
Jacobowitz, H., L. L. Stowe, G. Ohring, A. Heidinger, K. Knapp, and N. R. Nalli, 2003: The Advanced Very High Resolution Radiometer Pathfinder Atmosphere (PATMOS) climate dataset: A resource for climate research. Bull. Amer. Meteor. Soc., 84 , 785–793.
Kilpatrick, K. A., G. P. Podestá, and R. Evans, 2001: Overview of the NOAA/NASA Advanced Very High Resolution Radiometer Pathfinder algorithm for sea surface temperature and associated matchup database. J. Geophys. Res., 106 , 9179–9197.
May, D. A., L. L. Stowe, J. D. Hawkins, and E. P. McClain, 1992: A correction for Saharan dust effects on satellite sea surface temperature measurements. J. Geophys. Res., 97 , 3611–3619.
McClain, E. P., W. G. Pichel, and C. C. Walton, 1985: Comparative performance of AVHRR-based multichannel sea surface temperatures. J. Geophys. Res., 90 , 11587–11601.
McMillin, L. M., 1975: Estimation of sea surface temperatures from two infrared window measurements with different absorption. J. Geophys. Res., 80 , 5113–5117.
McPhaden, M. J., 2004: Evolution of the 2002/03 El Niño. Bull. Amer. Meteor. Soc., 85 , 677–695.
Merchant, C. J., A. R. Harris, M. J. Murray, and A. M. Závody, 1999: Toward the elimination of bias in satellite retrievals of sea surface temperature, 1, Theory, modeling and interalgorithm comparison. J. Geophys. Res., 104 , 23565–23578.
Nalli, N. R., 2003: Aerosol corrected sea surface temperature climatology (1985–2000) derived from the AVHRR Pathfinder Atmospheres (PATMOS) dataset. Proc. 2003 EUMETSAT Meteorological Satellite Conf., Darmstadt, Germany, EUMETSAT EUM P 39, 520–525.
Nalli, N. R., 2004: Satellite remote sensing and the NOAA/NESDIS sea-surface temperature science team. Backscatter, 15 , 24–28.
Nalli, N. R., and L. L. Stowe, 2002: Aerosol correction for remotely sensed sea surface temperatures from the National Oceanic and Atmospheric Administration Advanced Very High Resolution Radiometer. J. Geophys. Res., 107 .3172, doi:10.1029/2001JC001162.
Rao, C. R. N., 1992: Aerosol radiative corrections to the retrieval of sea surface temperatures from infrared radiances measured by the Advanced Very High Resolution Radiometer (AVHRR). Int. J. Remote Sens., 13 , 1757–1769.
Reynolds, R. W., 1988: A real-time global sea surface temperature analysis. J. Climate, 1 , 75–76.
Reynolds, R. W., 1993: Impact of Mount Pinatubo aerosols on satellite-derived sea surface temperatures. J. Climate, 6 , 768–774.
Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surface temperature analyses using optimum interpolation. J. Climate, 7 , 929–948.
Reynolds, R. W., N. A. Rayner, T. M. Smith, D. C. Stokes, and W. Wang, 2002: An improved in situ and satellite SST analysis for climate. J. Climate, 15 , 1609–1625.
Reynolds, R. W., C. L. Gentemann, and F. Wentz, 2004: Impact of TRMM SSTs on a climate-scale SST analysis. J. Climate, 17 , 2938–2952.
Smith, T. M., and R. W. Reynolds, 2003: Extended reconstruction of global sea surface temperatures based on COADS data (1854–1997). J. Climate, 16 , 1495–1510.
Smith, T. M., and R. W. Reynolds, 2004: Improved extended reconstruction of SST (1854–1997). J. Climate, 17 , 2466–2477.
Stowe, L. L., A. M. Ignatov, and R. R. Singh, 1997: Development, validation, and potential enhancements to the second-generation operational aerosol product at the National Environmental Satellite, Data and Information Service of the National Oceanic and Atmospheric Administration. J. Geophys. Res., 102 , 16923–16934.
Stowe, L. L., P. A. Davis, and E. P. McClain, 1999: Scientific basis and initial evaluation of the CLAVR-1 global clear/cloud classification algorithm for the Advanced Very High Resolution Radiometer. J. Atmos. Oceanic Technol., 16 , 656–681.
Stowe, L. L., H. Jacobowitz, G. Ohring, K. Knapp, and N. R. Nalli, 2002: The Advanced Very High Resolution Radiometer Pathfinder Atmosphere (PATMOS) climate dataset: Initial analyses and evaluations. J. Climate, 15 , 1243–1260.
Stuart-Menteth, A. C., I. S. Robinson, and P. G. Challenor, 2003: A global study of diurnal warming using satellite-derived sea surface temperature. J. Geophys. Res., 108 .3155, doi:10.1029/2002JC001534.
Tanahashi, S., H. Kawamura, T. Takahashi, and H. Yusa, 2003: Diurnal variations of sea surface temperature over the wide-ranging ocean using VISSR on board GMS. J. Geophys. Res., 108 .3216, doi:10.1029/2002JC001313.
Thiébaux, H. J., and M. A. Pedder, 1987: Spatial Objective Analysis: With Applications in Atmospheric Science. Academic Press, 299 pp.
Vázquez-Cuervo, J., E. M. Armstrong, and A. Harris, 2004: The effect of aerosols and clouds on the retrieval of infrared sea surface temperatures. J. Climate, 17 , 3921–3933.
Walton, C. C., 1985: Satellite measurement of sea surface temperature in the presence of volcanic aerosols. J. Climate Appl. Meteor., 24 , 501–507.
Walton, C. C., W. G. Pichel, J. F. Sapper, and D. A. May, 1998: The development and operational application of nonlinear algorithms for the measurement of sea surface temperatures with the NOAA Polar-Orbiting Environmental Satellites. J. Geophys. Res., 103 , 27999–28012.
Wentz, F. J., C. Gentemann, D. Smith, and D. Chelton, 2000: Satellite measurements of sea surface temperature through clouds. Science, 288 , 847–850.
Xu, L., and W. L. Smith, 1986: Numerical simulation of the influence of volcanic aerosols on VAS-derived SST determinations. Appl. Opt., 25 , 1137–1144.
PATMOS weekly zonal mean AOD, τa, derived from “clear sky,” grid cell mean AVHRR channel 1 (λ1 = 0.63 μm) normalized reflectances. Seasonal variations in tropospheric aerosol are evident, as is the evolution of the stratospheric sulfate layer following the major eruption of Mount Pinatubo in 1991. Note that the decreasing coverage in the extratropical latitudes during the lifetime of each satellite results from the drift in the satellite orbit to later afternoon equator crossing times (e.g., Jacobowitz et al. 2003).
Citation: Journal of Climate 19, 3; 10.1175/JCLI3644.1
PATMOS–BUOY empirical relationships between quality-controlled MCSST residuals (relative to buoy matchups), T ′S − TS0, where TS0 is the buoy SST, vs slant-path AOD for elevated aerosol conditions: (top) tropospheric (trop) aerosols (NOAA-14, 1995–99) and (bottom) stratospheric/tropospheric (strat/trop) aerosols following the eruption of Mount Pinatubo (NOAA-11, July 1991–mid-March 1993).
Citation: Journal of Climate 19, 3; 10.1175/JCLI3644.1
Geographic histograms of PATMOS–BUOY matchups: (top) for elevated trop aerosols over the period 1995–99, and (bottom) for elevated strat/trop aerosols over the period July 1991–mid-March 1993. Map projection is equal area.
Citation: Journal of Climate 19, 3; 10.1175/JCLI3644.1
Geographic distribution (40°S–40°N) equal-area map projection) of January mean PATMOS OISST residual bias relative to the TMI OI SST over the period 1998–2000, based upon four separate satellite SST inputs: (top left) MCSST from Eq. (1), (top right) ACSST from Eq. (5), (bottom left) MCSST with in situ satellite bias correction from Eqs. (1) and (7), and (bottom right) ACSST with in situ satellite bias correction, Eqs. (5) and (7). The three boxes outline regions of persistent tropospheric aerosol known to be problematic for causing IR satellite SST biases: Saharan dust, bounded by 5°, 20°N latitude and 60°, 10°W longitude; subSaharan biomass burning, bounded by 8°S, 5°N and 30°W, 10°E; and Arabian dust, bounded by 5°, 20°N and 60°, 70°E.
Citation: Journal of Climate 19, 3; 10.1175/JCLI3644.1
Box and whisker plots showing the robust central tendency and dispersion of PATMOS OISST spatial-mean residuals for the biomass burning region (relative to the TMI OI SST over the period 1998–2000, as in Figs. 4 and 5), as grouped by month to reveal seasonal trends: (top left) MCSST, (top right) ACSST, (bottom left) MCSST with in situ correction for satellite bias, and (bottom right) ACSST with in situ correction for satellite bias. The boxes span the interquartile range (bounded by the 25th and 75th percentiles), the horizontal lines within the boxes denote the medians (the 50th percentile), and the whiskers (dashed lines extending from the boxes) show the spread in the remaining data (out to 1.5 times the interquartile range to exclude outliers).
Citation: Journal of Climate 19, 3; 10.1175/JCLI3644.1
Same as in Fig. 6 except restricted to the Arabian dust region.
Citation: Journal of Climate 19, 3; 10.1175/JCLI3644.1
Same as in Fig. 6 except restricted to the Saharan dust region.
Citation: Journal of Climate 19, 3; 10.1175/JCLI3644.1
Similar to Figs. 4 and 5 except showing mean residual bias relative to the ERSST.v2 over the spatial domain 50°S–60°N and the period 1985–2000, regardless of month, and including the residence time of Mount Pinatubo stratospheric aerosol. Note the color scale is the same as before.
Citation: Journal of Climate 19, 3; 10.1175/JCLI3644.1
Similar to Fig. 6 except showing residual bias relative to the ERSST.v2 over the spatial domain 50°S–60°N and the period 1985–2000, grouped by year to reveal interannual trends and including the residence time of Mount Pinatubo stratospheric aerosol.
Citation: Journal of Climate 19, 3; 10.1175/JCLI3644.1
Time series of weekly daytime 1985–2000 PATMOS OISST analysis 4 (aerosol and in situ corrections for satellite bias) zonal mean anomalies (50°S–60°N) relative to the ERSST.v2 climatology (computed over the 1960–89 base period). The operational NOAA-9, -11, and -14 satellite lifetimes are also shown for reference.
Citation: Journal of Climate 19, 3; 10.1175/JCLI3644.1
AVHRR PATMOS daytime split-window MCSST algorithm statistics. MCSST algorithm is defined by Eq. (1). The training sample size and root-mean-square error (rmse) of the least squares fit are also given.
PATMOS SST aerosol correction regression statistics. Coefficients, bi, along with their estimated standard error, σb, are defined for aerosol bias correction as in Eq. (3). The test for significance of regression is given by the Student's t statistic for H0: βi = 0, and p is the probability that H0 is true. The strat/trop coefficients are used during the Mount Pinatubo residence time (NOAA-11, July 1991–mid-March 1993). The trop coefficients are used for the remaining time periods. For NOAA-9, the constant term b0 was empirically adjusted by −0.12 to compensate for an apparent offset observed in a limited buoy-matchup sample available for validation.
PATMOS daytime OISST residual statistics relative to TMI OI SST. Here, SD is the standard deviation. Units are degrees Celsius. Statistics are valid for the PATMOS-TMI overlap period of January 1998–December 2000. PATMOS daytime OI analyses are defined at the beginning of section 3. TMI OI SST is “normalized” to the daily minimum (0800 LST) using the diurnal warming model of Gentemann et al. (2003).
With the exception of certain high-level ice clouds, which have IR extinction properties similar to aerosols.
Note that Eq. (1) also provides an implicit correction for nonunity surface emissivity and reflectance under mean sea-state conditions. The last term on the right side roughly accounts for variations in both atmospheric pathlength and surface emissivity as functions of θ.
Note that it is minimally necessary to distinguish between these two aerosol categories because the majority of extinction in the IR is due to absorption and reemission. The mean emission temperature of absorbing layers confined to the lower troposphere will be significantly warmer than those distributed throughout the stratosphere and troposphere. Note also that the definition of τa as a column-integrated quantity, indicated by Eq. (2), precludes the ability to separate tropospheric from stratospheric aerosol using AVHRR.
Nalli and Stowe (2002) also found that an additional predictor, namely the reflectance ratio of channels 1 and 2, provided slightly better results for a given satellite, although it was strongly correlated with slant-path AOD. Because of remaining uncertainties in the PATMOS retrospective vicarious calibration of channels 1 and 2 between satellites, it was necessary to exclude this predictor in the present work.
Mathematical notation has been here adapted in a couple places.
For example, compare the bias patterns in Figs. 4 and 5 against Fig. 7 in Donlon et al. (2002) and Figs. 4a–c in Gentemann et al. (2003).
Compare Fig. 7, top right, against Fig. 2 in Stuart-Menteth et al. (2003). The magnitudes and trends in bias (i.e., monthly mean ΔT) agree reasonably well.
