Sampling errors in satellite-derived infrared sea-surface temperatures. Part II: Sensitivity and parameterization
Introduction
Clouds and inter-swath gaps are the primary reasons for incomplete coverage of satellite infrared (IR) measurements of the Earth's surface, and result in sampling errors in averaged IR Sea-Surface Temperature (SST) fields. In a recent paper (Liu and Minnett, 2016; hereafter LM16) we found that the MODIS (Moderate Resolution Imaging Spectroradiometer (Esaias et al., 1998)) monthly SST sampling error referenced to MUR SSTs (Multi-scale Ultrahigh Resolution (Chin et al., 2010), see details in Section 2) is up to O (1 K), which far exceeds the error threshold needed for climate research. The largest sampling error (> 5 K in monthly SSTs) is found in the Arctic. The 30°N–30°S zonal band has the smallest errors, with a notable exception being the persistent negative errors found in the Tropical Instability Wave (TIW) region, where mesoscale ocean-atmosphere interaction leads to a more frequent satellite sampling above areas with lower SSTs; SST-cloud relationships at different time and space scales were proposed to be the causes for certain error characteristics, which could introduce misleading SST values and patterns to the final Level 4 (see Table 1 for SST processing Level definitions) SST fields and potentially adversely affect many applications. The statistics based on the studied periods show that the global mean sampling error is generally positive and increases approximately exponentially with missing data fraction (gap fraction) in a fixed averaging interval, while the error variability is mainly controlled by SST variability.
Two further questions are addressed here. First, since the MODIS sampling error was initially calculated based on the use of MUR SST fields as the reference, whether another SST reference field with presumably different embedded variability would result in different sampling error patterns. The international Group for High Resolution SST (GHRSST: https://www.ghrsst.org/) was set up to help coordinate efforts to improve the accuracy of satellite-derived SST fields at all processing levels and to standardize data formats to facilitate the analysis of different SST fields by the research and operational communities (Donlon et al., 2007). With the growing number of Level 4 SST fields that blend observations, often including model simulations, differences in the SST structure exist among the different data products, especially in many dynamic and rarely sampled regions. SST differences among sixteen daily Level 4 fields are monitored and discrepancies are revealed by the online tool, L4-SQUAM (SST Quality Monitor (Dash et al., 2012): http://www.star.nesdis.noaa.gov/sod/sst/squam/L4/). For example, compared with the GHRSST multi-product ensemble (GMPE; Martin et al., 2012), MUR frequently shows lower SST estimates in the Southeast Asian Maritime Continent region, Falkland Islands (Islas Malvinas), and the Pacific and Atlantic eastern equatorial upwelling areas, while higher estimates are found in the Northern Hemisphere high latitudes. Such non-negligible differences constitute a potential source of uncertainty in the previously quantified sampling errors and are examined in this paper.
The second important question is whether the error magnitudes and patterns change significantly in different years. Sampling errors may have interannual variability due to ocean-atmosphere interactions associated with long-duration climate events such as ENSO (El Niño–Southern Oscillation). It is recognized that the eastern equatorial Pacific TIW activity can be influenced by ENSO (Yu and Liu, 2003, An, 2008, Kug et al., 2010): stronger (weaker) activity due to the increased (decreased) eastern equatorial Pacific meridional SST gradient during La Niña (El Niño). Sampling errors found in LM16 were quantified using the data of year 2011, which was during the 2010–2011 moderate La Niña event. How the negative TIW sampling errors may evolve with ENSO requires a comparative error quantification using data from an El Niño event; this can yield an assessment of the interannual changes in the sampling error.
Another focus of this study is whether sampling errors caused by clouds and interswath gaps can be predicted. As the properties of the error characteristics become better known, can the sampling errors be estimated using, for example the local SST anomaly, cloud persistence (the number of consecutive days during which a location is detected to be cloudy), or season and region? It is widely known that the primary component in any time series of L4 SST fields is the annual cycle. Therefore, the seasonally induced error component can be explicitly quantified by using a seasonal climatology, assuming the additional sampling errors in the climatology are neglected.
The ultimate goal is to predict the sampling errors without relying on a specific reference field after the error characteristics are well understood. Thus, we first test the error sensitivity by comparing the sampling errors using two different reference SST fields, and explain the prevalently small error differences and the few exceptions when the magnitude of sampling errors could be affected by the reference field selection. As an exploratory test of sampling error interannual variability, we quantified the errors in the El Niño year of 2009, as opposed to the previously studied La Niña year of 2010–2011. We also examined the error component of the seasonal cycle, and by combining the previously derived error statistics and the error sensitivity to the annual cycle, a preliminary empirical model is suggested to estimate and predict the MODIS SST sampling errors.
Section snippets
Methods and data
The sampling error quantification framework is described in detail in LM16. Here we briefly review the definitions that are relevant to this paper. In LM16, to minimize the effect of existing errors in MUR, sampling errors are calculated as the difference between the means of the sampled (number of nR) and the gap-free (number of NR) MUR SSTs at base resolution R0 in a fixed spatial or temporal interval defined by a coarser resolution R. R is a set of resolutions: 0.25°, 0.5°, 1°, 2.5°, and 5°
HYCOM reanalysis vs MUR
HYCOM and MUR SST fields have their own peculiarities. The input SSTs and the generating methods are different. The HYCOM SSTs used here are the 00Z mean temperatures of the top meter of the ocean. As a result, a global pattern of diurnal warming with the afternoon maximum heating at around 150°W will likely exist in this field. On the contrary, MUR represents the foundation temperature that is intended to be without diurnal effects. Therefore, we expect HYCOM to be warmer than MUR in many
Parameterization approach
When assessing sampling errors in satellite-derived IR SST fields, situations may arise where we do not have an appropriate reference field at the various temporal and spatial averaging intervals. An example is extending sampling error estimation to periods before the availability of MUR 1 km SSTs. Another example might be the use of assimilation of a reduced resolution SST field into atmosphere or ocean forecast models running in real time. However, we do have access to a number of relevant
Discussion and conclusions
Extending the work of LM16, here we study sampling error sensitivity on the choice of reference field and the possible dependence on large-scale interannual variations, for which we compare error estimates using data from an El Niño year to compare with the La Niña year data used in LM16. Comparisons conducted in both time and frequency dimensions show that HYCOM and MUR exhibit significant differences in the SST fields. The reasons for the HYCOM-MUR differences remain unclear due to the
Acknowledgements
This research was supported by a NASA graduate fellowship to Y. Liu (NNX14AL28H).
The 1/12° global HYCOM + NCODA Ocean Reanalysis was funded by the U.S. Navy and the Modeling and Simulation Coordination Office. Computer time was made available by the DoD High Performance Computing Modernization Program. The output is publicly available at http://hycom.org.
References (29)
- et al.
Sampling uncertainty in gridded sea surface temperature products and Advanced Very High Resolution Radiometer (AVHRR) Global Area Coverage (GAC) data
Remote Sens. Environ.
(2016) - et al.
Group for High Resolution Sea Surface Temperature (GHRSST) analysis fields inter-comparisons—Part 2: Near real time web-based level 4 SST Quality Monitor (L4-SQUAM)
Deep-Sea Res. II Top. Stud. Oceanogr.
(2012) - et al.
Sampling errors in satellite-derived infrared sea-surface temperatures. Part I: Global and regional MODIS fields
Remote Sens. Environ.
(2016) - et al.
Group for High Resolution Sea Surface temperature (GHRSST) analysis fields inter-comparisons. Part 1: a GHRSST multi-product ensemble (GMPE)
Deep-Sea Res. II Top. Stud. Oceanogr.
(2012) - et al.
Impact of missing data on the estimation of ecological indicators from satellite ocean-colour time-series
Remote Sens. Environ.
(2014) - et al.
The convolution of dynamics and moisture with the presence of shortwave absorbing aerosols over the Southeast Atlantic
J. Clim.
(2015) Interannual Variations of the tropical ocean instability wave and ENSO
J. Clim
(2008)- et al.
Use of WindSat to extend a microwave-based daily optimum interpolation sea surface temperature time series
J. Clim.
(2013) - et al.
A 1/4°-spatial-resolution daily sea surface temperature climatology based on a blended satellite and in situ analysis
J. Clim.
(2014) - et al.
Satellite microwave SST observations of transequatorial tropical instability waves
Geophys. Res. Lett.
(2000)
Basin-scale, high-wavenumber sea surface wind fields from a multiresolution analysis of scatterometer data
J. Atmos. Ocean. Technol.
Algorithm Theoretic Basis Document: Multi-scale, Motion-compensated Analysis of Sea Surface Temperature Version 1.1
Mind the gap: the impact of missing data on the calculation of phytoplankton phenology metrics
J. Geophys. Res. Oceans
Variational Data Assimilation for the Global Ocean. Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications
Cited by (15)
Surface-based thermal infrared spectrometers
2022, Field Measurements for Passive Environmental Remote Sensing: Instrumentation, Intensive Campaigns, and Satellite ApplicationsA split-window method to retrieving sea surface temperature from landsat 8 thermal infrared remote sensing data in offshore waters
2020, Estuarine, Coastal and Shelf ScienceCitation Excerpt :A microwave radiometer is often used in meteorological satellites. It cannot accurately obtain observations in coastal waters because it can be greatly affected by land (Y. Liu et al., 2017b; Othman et al., 2002). Thermal infrared remote sensing has high spatial resolution and can effectively reduce the effect of land on offshore SST retrieval (Wang et al., 2013; Li et al., 2013a,b; Mathieu et al., 2017).
Half a century of satellite remote sensing of sea-surface temperature
2019, Remote Sensing of EnvironmentCitation Excerpt :This approach has been widely used for several satellite missions including for the (A)ATSR ((Advanced) Along-Track Scanning Radiometer) series (Merchant et al., 2012), the NOAA GOES (Geostationary Operational Environmental Satellite; Maturi et al., 2008), and Himawari-8 satellite (Kurihara et al., 2016) of the Japan Meteorological Agency (JMA). To address the specific cases where the decision-tree approach to cloud screening resulted in erroneous classification (Liu and Minnett, 2016; Liu et al., 2017), a new algorithm has been developed, called the Alternating Decision Tree (ADT) (Freund and Mason, 1999; Kilpatrick et al., 2019). Unlike in the decision-tree approach where a pixel identified as being cloud-free has to pass every test, in the ADT the output of each test is considered in determining the likelihood of a pixel being identified as being clear or cloudy.
Improving satellite retrieved night-time infrared sea surface temperatures in aerosol contaminated regions
2019, Remote Sensing of EnvironmentCitation Excerpt :Existing satellite retrieval algorithms discard potentially aerosol contaminated data to preserve data quality, leading to low numbers of retrievals in these regions. Correcting aerosol-related errors in SSTs would improve data coverage and decrease sampling errors in CDRs (May et al., 1992; Bulgin et al., 2016; Liu and Minnett, 2016; Liu et al., 2017). A joint analysis of the SSTs from TRMM Microwave Imager (TMI) and MODIS Aerosol Optical Depths (AODs), supported by upper ocean modeling, showed reductions in the SSTskin of up to 0.2–0.4 K simultaneously with, or shortly after, strong dust outbreaks (Martínez Avellaneda et al., 2010).
Satellite Remote Sensing of Sea Surface Temperatures
2019, Encyclopedia of Ocean Sciences, Third Edition: Volume 1-5Global Sea Surface Temperature
2019, Taking the Temperature of the Earth: Steps towards Integrated Understanding of Variability and Change