AVHRR Global Area Coverage (GAC) data provide daily
global coverage of the Earth, which are widely used for global environmental
and climate studies. However, their geolocation accuracy has not been
comprehensively evaluated due to the difficulty caused by onboard resampling
and the resulting coarse resolution, which hampers their usefulness in
various applications. In this study, a correlation-based patch matching
method (CPMM) was proposed to characterize and quantify the geo-location
accuracy at the sub-pixel level for satellite data with coarse resolution,
such as the AVHRR GAC dataset. This method is neither limited to landmarks nor
suffers from errors caused by false detection due to the effect of mixed
pixels caused by a coarse spatial resolution, and it thus enables a more robust
and comprehensive geometric assessment than existing approaches. Data of
NOAA-17, MetOp-A and MetOp-B satellites were selected to test the geocoding
accuracy. The three satellites predominately present west shifts in the
across-track direction, with average values of
Advanced Very High Resolution Radiometer (AVHRR) data provide valuable data sources with a near-daily global coverage to support a broad range of environmental monitoring research, including weather forecasting, climate change, ocean dynamics, atmospheric soundings, land cover monitoring, search and rescue, forest fire detection, and many other applications (Van et al., 2008). The unique advantage of AVHRR sensors is their long history dating back to the 1980s and thus enabling long-term analyses at climate-relevant timescales that cannot be covered by other satellites. However, AVHRR data are rarely used at the full spatial resolution for global monitoring due to the limited data availability (Pouliot et al., 2009; Fontana et al., 2009). Instead, the Global Area Coverage (GAC) AVHRR dataset with a reduced spatial resolution is generally employed in long-term studies at a global or regional perspective (Hori et al., 2017; Delbart et al., 2006; Stöckli and Vidale, 2004; Moulin et al., 1997).
However, there are several known problems with the geo-location of AVHRR GAC data, which have a profound impact on their application. (1) The drift of the spacecraft clock results in errors in the along-track direction (Devasthale et al., 2016). Generally, an uncertainty of 1 s approximately induces an error of 8 km in this direction. (2) Satellite orientation and position uncertainties influence the projection of the satellite geometry to the ground, which leads to errors in both along-track and across-track directions. (3) Earth surface elevation aggravates distortions in the across-track direction (Fontana et al., 2009). Without navigation corrections, the spatial misplacement of the GAC scene caused by these factors can be up to 25–30 km occasionally (Devasthale et al., 2016).
For geocoding of AVHRR data, a two-step approach is usually used: (1) geocoding based on orbit model, ephemeris data and time of onboard clock (Van et al., 2008), achieving an accuracy within 3–5 km depending on the accuracy of orbit parameters and model (Khlopenkov et al., 2010), and (2) using any kind of ground control points (GCPs) (e.g., road or river intersections, coastal lines) to improve geocoding (Takagi, 2004; Van et al., 2008). Additionally, in order to eliminate the ortho-shift caused by elevations, an orthorectification would be needed (Aguilar et al., 2013; Khlopenkov et al., 2010). The dataset used in this study is from the ESA (European Space Agency) cloud CCI (Climate Change Initiative) project, which has corrected clock drift errors by coregistration of AVHRR GAC data with a reference dataset and showed improved navigation by fitting the data to coastal lines.
Unlike the Local Area Coverage (LAC) data with a full spatial resolution of
AVHRR, GAC data are sampled on board the satellite in real time to generate
coarser-resolution data (Kidwell, 1998). This is achieved by averaging
values from four out of five pixel samples along a scan line and eliminating
two out of three scan lines, resulting in a spatial resolution of
There are generally three approaches to assess the non-systematic geometric errors of satellite images: (1) the coastline crossing method (CCM) which detects the coastline in the along-track and across-track directions through a cubic polynomial fitting (Hoffman et al., 1987); (2) the land–sea fraction method (LFM) which develops a linear radiance model as a function of land–sea fraction and land and sea radiance and then finds the minimum difference between model-simulated and instrument-observed radiance by shifting the pixels in the along-track and across-track directions (Bennartz, 1999); and (3) the coregistration method which computes the difference or similarity relative to a reference image (Khlopenkov et al., 2010). The abilities of these three methods in characterizing the geometric errors are limited and dependent on different, method-dependent factors. The CCM is subject to the structure of the coastline, and the LFM depends on the accuracy of the land–sea model but shows advantages on complex coastlines (Han et al., 2016). The coregistration method is usually applied to high-resolution visible and infrared images (Wang et al., 2013; Wolfe et al., 2013) as it relies on individual objects/landmarks in both datasets. However, when it comes to coarse-resolution data with several kilometers' pixel size, the main difficulties arise from false detection due to the effect of mixed pixels, which hampers the application of the existing methods. An approach assessing the geolocation accuracy of coarse-resolution satellite data is thus strongly needed. The geometric accuracy is important as even small geometric errors can lead to significant noises on the retrieval of surface parameters, such as normalized difference vegetation index (NDVI), leaf area index (LAI) and albedo, which mask the reality or bias the final results and conclusions (Khlopenkov et al., 2010; Arnold et al., 2010). For instance, anomalous NDVI dynamics during the regeneration phase of forest-fire-burnt areas can be explained by the imprecise geolocation of the dataset used (Alcaraz-Segura et al., 2010). Therefore, it is critical to develop a rigorous geometric accuracy assessment method in order to ensure the effectiveness of AVHRR GAC data in the generation of climate data records (CDRs) (Khlopenkov et al., 2010; Van et al., 2008).
Based on the idea of the coregistration method, this study proposes a method
named correlation-based patch matching method (CPMM), which is capable of
quantifying the geometric accuracy of coarse-resolution satellite data
available as fundamental climate data records (FCDRs) for global applications
(Hollmann et al., 2013). We show the procedure based on AVHRR GAC data,
which are compiled for the ESA CCI cloud project (Stengel et al., 2017) and
are now also used for the ESA CCI
AVHRR is a multipurpose imaging instrument aboard the NOAA satellite series since 1978 and the Meteorological Operational Satellites (MetOp) operated by EUMETSAT since 2006, delivering daily information of the Earth in the visible, near-infrared and thermal wavelengths. They provide observations from four to six spectral bands, depending on the generation of AVHRR sensors. This study only focuses on the AVHRR GAC data observed by NOAA-17 (AVHRR-3 generation), MetOp-A and MetOp-B. The spectral characteristics of the AVHRR sensors on board these three platforms are the same and summarized in Table 1. Since the spatial resolution of AVHRR GAC data is often considered to be 4 km (Fontana et al., 2009), the analysis in this study was conducted at the 4 km level using the data acquired on 13 August 2003 for NOAA-17 and 12 March 2017 for MetOp-A and MetOp-B.
Spectral characteristics of AVHRR sensors.
From a standpoint of geometric accuracy assessment, the reflectances in bands
1 and 2 were employed in this study. However, these two bands are not only
affected by the atmosphere but also by the earth surface anisotropy
characterized by the bidirectional reflectance distribution function (BRDF)
(Cihlar et al., 2004). Given the fact that BRDF effects can be reduced
through the calculation of vegetation indices such as NDVI (Lee and
Kaufman, 1986), the NDVI is employed in this study, which is derived from
the reflectance in bands 1 and 2 according to Eq. (1).
Ideally, the referenced data in geometric quality assessment should meet the
required accuracy of a one-third field of view (FOV) (WMO and UNEP, 2006) and also
satisfy the accuracy requirement of an order of magnitude better than
The purpose of this study is not only to assess the geolocation accuracy of 4 km AVHRR GAC data, but also to explore the potential impact factors related to geolocation accuracy. Therefore, the investigations were made at different latitudes and longitudes, at different locations with different SatZs, for different land covers, as well as different topographies. The swaths covering parts of Europe (including the Alps) and Africa were used since they fit the study needs (Fig. 1). Investigations were based on six regions of interest (ROI) as shown in Figs. 1 and 2. The ROIs from 1 to 6 enable us to investigate the geolocation accuracy at different SatZs, topography, as well as latitudes and longitudes. Their locations and extents are consistent for the scenes from NOAA-17 and MetOp-A (Fig. 1), which enables the comparison of geolocation accuracy between these two sensors. The size of ROI was set as large as possible in order to get more significant and comprehensive results. On the other hand, areas covered by cloud and water have to be avoided, resulting in the different sizes of these ROIs. Half of the ROIs (ROIs 2, 4, 6) serve as a good example for a typical mountainous area on Earth. The other half of ROIs (ROIs 1, 3, 5), on the other hand, mainly cover relatively flat areas. Since the NOAA-17 scene was almost unaffected by cloud, another ROI (ROI 7) was selected to check the geolocation accuracy at nadir. The MetOp-B scene was influenced by cloud but served as a good example to illustrate the combined effect of topography and large SatZs (Fig. 2). Although there are also six ROIs (ROIs (a–f)) selected, their sizes and extents are totally different from the above two scenes. In order to include the terrain area, two subsets were used (Fig. 2a and c). Each grid in the ROI represents the minimum unit (namely the patch) based on which we conduct the geometric quality analysis.
Top-of-atmosphere reflectance true color composite (AVHRR
GAC bands 2-1-2) surrounding the study area
Top-of-atmosphere reflectance true color composite (AVHRR
GAC bands 2-1-2) from the MetOp-B satellite on 12 March 2017
The assessment was performed by comparing the AVHRR GAC scenes with geo-located reference data, i.e., MOD13A1 (V006). An approach named the correlation-based patch matching method (CPMM) is proposed to find the best match between small image patches taken from the reference images and the AVHRR GAC images. This method is expected to be more suitable for the geometric accuracy assessment of coarse-resolution images than the current methods, i.e., the CCM, LFM and co-registration using shorelines. The framework of CPMM is shown in Fig. 3, and the detailed description of this method is provided below.
Flowchart of the correlation-based patch matching method (CPMM).
The AVHRR GAC dataset is stored in a Network Common Data Form (NetCDF), with latitude and longitude assigned to each pixel. In order to achieve a higher accuracy of image matching, the data need to be reprojected. The AVHRR GAC scene was reprojected into the Lambert conformal conic (LCC) projection by building the geographic lookup table (GLT) using the latitude and longitude data in ENVI. The spatial resolution of the AVHRR GAC map in the LCC projection is 4 km. Based on the reprojected data, the NDVI was calculated using the band combinations as indicated by Eq. (1). Similarly, the NDVI band of MOD13A1 in the hierarchical data format (HDF) format was extracted and converted to LCC projection from its raw sinusoidal projection using the MODIS Reprojection Tool (MRT). The nearest-neighbor (NN) resampling scheme was employed in this procedure. The spatial resolution of the MODIS NDVI in the LCC projection is 500 m. Thus, the geometric assessment is performed at the 4 km resolution of AVHRR NDVI based on the 500 m MODIS NDVI data.
In the process of matching the AVHRR GAC data with reference MODIS data, a
patch size of
For each patch in the ROI, the AVHRR GAC data within the patch were
extracted. Then the patch was shifted in the
It is expected that the results from each patch are different. Therefore, the general accuracy of each ROI was determined by summarizing the measured shifts of each respective patch statistically. Here, the histogram was employed to show the distribution of geometric errors in the across-track and along-track directions. And the quantitative indexes, such as the number of patches, their mean and standard errors, were calculated. The averaging is expected to reduce the uncertainties caused by random factors and produce accurate shift measurement estimates (Bicheron et al., 2011). The final shifts of the scene were calculated by averaging the measured shifts of all patches on the scene.
The influence of potential variables on the geometric accuracy was studied, including SatZs, topography, latitudes and longitude. To achieve this, the information of these factors was also extracted for each patch on the scene. The geometric errors induced by SatZ were highlighted by checking the relationship between errors and SatZ. The effect of topography was investigated by checking the relationship of geometric errors in the across-track direction over terrain areas compared to relatively flat areas. The effect of latitude and longitude was determined by analyzing their relationship with measured shifts in the along-track and across-track directions, respectively.
Figure 4 shows the correlation distribution over the
Variations in the correlation with respect to each shift
combination. Only the results of one patch from the NOAA-17
The geolocation shifts of each patch are slightly different as shown in
Figs. 5–7. The
As shown in Fig. 5, it can be seen that the scene of NOAA-17 generally shows
westward shifts in the across-track direction, since the majority of patches in
all ROIs show negative shifts. Nevertheless, the magnitudes of shifts for
different ROIs vary from one to another. ROI 2 shows the smallest shift with
a mean value of
The distribution of shifts in the across-track (
When combining the results of all ROIs together (Fig. 5h), the shifts in the
across-track direction generally follow an approximately normal distribution
with a mean value of
The shifts in the along-track direction are mainly negative throughout these
ROIs, indicating that the NOAA-17 scene is dominated by south shifts in the
along-track direction. Nevertheless, a considerable number of patches also
show slight north shifts over ROIs 1, 3 and 4 (Fig. 5a, c and d), where the
shifts are distributed around 0 with mean values of
Furthermore, it can be stated that the distribution of shifts in the along-track direction is less widely spread than that in the across-track direction, demonstrating the smaller uncertainty of geocoding in the along-track direction, as indicated by the smaller SD values throughout these ROIs (Table 2). Moreover, the geolocation errors in the across-track direction are greater than the along-track direction (Fig. 5), which is expected due to the applied clock drift correction.
Summary of the results for the scene of NOAA-17. The unit of the shift is kilometers.
Similar to the results of NOAA-17, the MetOp-A scene mainly presents westward shifts
in the across-track direction, indicated by the widely distributed negative
values throughout these ROIs (Fig. 6a–f). These shifts are basically
concentrated around
The distribution of shifts in the across-track (
Since ROIs 1–6 on the MetOp-A scene are identical to those on the NOAA-17 scene in terms of spatial extents, their shifts in the across-track direction are generally comparable. When excluding the results of ROIs 4 and 5, the ROIs on the MetOp-A scene generally show larger average shifts but smaller SDs than the NOAA-17 scene in the across-track direction (see Tables 2 and 3). However, it does not necessarily mean that the MetOp-A scene has a smaller uncertainty than the NOAA-17 scene in the across-track direction, because the ROIs on the MetOp-A scene are slightly closer to the nadir area than those on the NOAA-17 scene (Fig. 1b and d). Given the larger SatZ and the smaller average shifts of the NOAA-17 scene, it is reasonable to conclude that the NOAA-17 scene shows a slightly better geolocation accuracy than the MetOp-A scene in the across-track direction.
Looking at the shifts in the along-track direction, the MetOp-A scene does
not show strong systematic north or south shifts, but rather a general
distribution of the shifts around 0 (Fig. 6a–f). The shifts are generally
small within a range of
Summary of the results for the scene of MetOp-A. The unit of the shift is kilometers.
By comparing Fig. 6a–f with Fig. 5a–f, it becomes obvious that large
differences exist between the shifts in the along-track direction of the MetOp-A
and NOAA-17 scenes. In the first place, systematic south shifts occur on the
NOAA-17 scene but not on the MetOp-A scene. Secondly, the magnitudes of
shifts on the MetOp-A scene are generally smaller than those on the NOAA-17
scene, as the former are concentrated around 0 while the latter are
concentrated around
Similar to the scenes of NOAA-17 and MetOp-A, the MetOp-B scene generally
shows westward shifts in the across-track direction, indicated by the
predominant occurrence of negative values (Fig. 7a–f). Nevertheless, unlike
the results for the terrain areas on the NOAA-17 and MetOp-A scenes, the ROI c
located in the terrain area on the MetOp-B scene (Fig. 2a) shows the
largest shifts throughout these ROIs with an average of
The distribution of shifts in the across-track (
Since the extent of the ROIs in the MetOp-B scene is not consistent with those on NOAA-17 and MetOp-A scenes, only their overall performances in the across-track direction are compared here. By comparing Fig. 7g with Fig. 6g and Fig. 5h, it is obvious that the MetOp-B scene shows larger shifts and greater uncertainties than NOAA-17 and MetOp-A scenes in the across-track direction. This is partly due to the larger range of SatZs of these ROIs and partly due to the worse geolocation accuracy of the MetOp-B scene in the across-track direction.
The MetOp-B scene is dominated by north shifts in the along-track direction,
indicated by the predominantly positive shift values (Fig. 7a–f). It is
interesting to find that ROI c, which is located at terrain area and with
large SatZs, shows the largest shifts with an average of 1.85 km in the
along-track direction. Given that terrain does not affect the geolocation
accuracy in the along-track direction, the main cause of the largest shift
may be the largest SatZ of ROI c among these ROIs. Furthermore, by comparing
the results of ROIs d and e with those of ROIs b, c and f, it can be found that the
shifts of ROIs with smaller SatZs are more concentrated around 0 (Fig. 7d
and e), while the shifts of ROIs with larger SatZs are more widely spread
(Fig. 7b, c and f). This shows that the effect of large SatZs on shifts
in the along-track direction cannot be neglected. When combining the results
of all ROIs, the MetOp-B scene shows shifts with an average of 0.96 and a
standard deviation of 1.7. Only 52 % of the shifts are distributed within
the range of
It can be seen that the shifts in the along-track direction are still significantly smaller than those in the across-track direction. Furthermore, the uncertainties of the shifts in the along-track direction are generally smaller than those in the across-track direction, when excluding the results of ROI a due to its limited number of patches (Table 4). This further verifies that after removing clock drift errors, the geolocation errors in the along-track direction are generally more accurate and have fewer uncertainties than the across-track direction.
Summary of the results for the scene of MetOp-B. The unit of the shift is kilometers.
The comparison of Fig. 7g with Figs. 6g and 5h reveals that the MetOp-B
scene is significantly inferior to the MetOp-A scene in terms of the
geolocation accuracy in the along-track direction, with the former being
concentrated around 1 and the latter around 0. Furthermore, the uncertainty
of the shifts of the MetOp-B scene (SD
From the results above, it can be concluded that NOAA-17 and MetOp-A scenes
show distinct advantages over the MetOp-B scene in both directions. However,
the NOAA-17 scene is slightly better than the MetOp-A scene in the
across-track direction, with average shifts of
From the above results, it is known that SatZ plays an important role in determining the geolocation accuracy of the satellite scene. To investigate how and to what extent it influences the geolocation accuracy, Fig. 8 displays the shifts in both directions as a function of SatZ for all three satellites. Furthermore, the influences of latitude and longitude on geolocation accuracy are also explored.
Influence of SatZ on the geolocation accuracy in the
across-track
As shown in Fig. 8a–c, it can be seen that the shifts in the across-track
direction vary considerably for all SatZs, and this is particularly evident
in the results of MetOp-B (Fig. 8c). This demonstrates that besides the SatZ
effects, the geolocation accuracy is also influenced by other factors.
Furthermore, the spread at each fixed SatZ tends to become larger at larger
SatZs (larger than 20
Compared to the shifts in the across-track direction (Fig. 8a–c), the
shifts in the along-track direction show smaller variability at each fixed
SatZ (Fig. 8d–f). From Fig. 8d–e, it can be seen that the shifts in the
along-track direction are relatively stable at each level of SatZ for SatZs
smaller than 15
The mean shift for each range of SatZ in the across-track direction. The unit of the shift is kilometers.
For NOAA-17, the shifts tend to be smaller with the longitudinal range of
10–15
The variation in the shifts (in the along-track direction) with latitude
also depends on the situation (Fig. 8j–l). The magnitudes of shifts with
larger latitude (larger than 45
The AVHRR GAC test data in this paper draw on datasets from the ESA CCI cloud
project (
The geometric accuracy of satellite data is crucial for most applications as geometric inaccuracy can bias the obtained results. Therefore, the assessment of the geolocation accuracy is important to provide satellite data of high quality enabling successful applications. In this study, a correlation-based patch matching method was proposed to characterize and quantify the AVHRR GAC geo-location accuracy. This method presented here yields significant advantages over existing approaches and enables the achievement of a sub-pixel geo-positioning accuracy of coarse-resolution scenes. It is free from the impact of false detection due to the influence of mixed pixels and not limited to a certain landmark (e.g., shoreline) and therefore enables a more comprehensive geometric assessment. This method was utilized to characterize the geolocation accuracy of AVHRR GAC scenes from the NOAA-17, MetOp-A and MetOp-B satellites.
The study is based on several ROIs comprising numerous patches over
different land cover types, latitudes and topographies. The scenes from
these satellites all present westward shifts in the across-track direction, with
an average shift of
From the results above, it can be found that the geolocation accuracy in the
along-track direction is always higher and with fewer uncertainties than the
across-track direction, which is consistent with previous related studies.
This is understandable since the GAC dataset from the ESA cloud CCI project
has been corrected for clock drift errors but has no ortho-correction,
which is not feasible due to the onboard sampling characteristics. SatZ
plays a decisive role in determining the magnitude as well as the
uncertainty of the shifts in the across-track direction. Larger SatZ
generally induce greater shifts and uncertainties in this direction. The
combined effect of SatZ and topography on geolocation accuracy in the
across-track direction has also been shown. And significant terrain effects
appear only in the case of large SatZs (
Although this assessment was only conducted for a single scene of each satellite, the highly variable ROIs take the influential factors of geometric accuracy into account. Therefore, the presented conclusions are transferable to other regions or seasons. However, it is noteworthy that this method is not applicable to homogeneous surfaces (e.g., water, desert), where the correlations are almost the same in any simulated displacement cases. In general, this study provides an important preliminary geolocation assessment for AVHRR GAC data. It is a first step towards a more precise geolocation and thus improves application of coarse-resolution satellite data. For instance, it identifies the threshold of SatZ under which the GAC data should be preferred in applications. Furthermore, the CPMM geolocation assessment method proposed by this study is also applicable to other coarse-resolution satellite data.
XW was responsible for the main research ideas and writing the manuscript. KN contributed to the data collection. SW contributed to the manuscript organization. All the authors thoroughly reviewed and edited this paper.
The authors declare that they have no conflict of interest.
The authors are grateful to the ESA CCI (Climate Change Initiative) cloud project team (Martin Stengel, Rainer Hollmann) for making the datasets available for this study.
This work was jointly supported by the National Key R&D Program of China (grant nos. SQ2018YFB0504804 and 2018YFA0605503) and the National Natural Science Foundation of China (grant no. 41801226).
This paper was edited by Prasad Gogineni and reviewed by two anonymous referees.