Open Access
Add to BibSonomyAbstract
A quasi-analytical algorithm (QAA)-based algorithm which is applied to moderate imaging spectroradiometer (MODIS) land band data (469, 555, and 645 nm) is proposed and named QAA-RGR (Red-Green-bands-Ratio). The performance has been evaluated using in situ measurements data and MODIS data from the Eastern China Seas (ECS). The QAA-RGR algorithm uses the ratio of the remote sensing reflectance at 645nm (Rrs645) to the Rrs555 to estimate the absorption coefficient at 555nm. In addition, the spectral slope of the backscattering coefficient (bb) is estimated using a statistical relationship based on bb555. The other steps of the retrieval algorithm are the same as those of the extensively used QAA version 5 (QAAv5). First, the QAA-RGR algorithm was applied to an in situ measurement data set for the ECS to retrieve inherent optical properties (IOPs), and the results were compared with the QAAv5. The results demonstrate that, the two algorithms exhibit similar performance for in situ measurements. Second, the algorithm was applied to the land bands data and ocean bands data of MODIS over the ECS to obtain the distribution of IOPs at 500m and 1000m resolutions. The results of the retrieval algorithm were evaluated against the corresponding in situ measurements and compared to those from QAAv5. The results demonstrate that, the QAA-RGR algorithm is characterized by a better performance than QAAv5 for MODIS data. For QAAv5, the averaged mean absolute percentage error (MAPE) values of retrieval results of absorption coefficients and backscattering coefficients compared with in situ match-up measurements are 25.2% and 22.2%, respectively. For QAA-RGR, the averaged MAPE values are 15.9% and 18.3%, respectively. The QAAv5 retrieval results are often significantly underestimated especially for turbid coastal waters because of the easy saturation at 667nm band in addition to a large uncertainty in the estimation of Rrs of the blue bands. The QAA-RGR algorithm may be used to retrieve IOPs from MODIS measurements over the ECS for the measurement periods used in the study.
© 2015 Optical Society of America
1. Introduction
The Moderate Resolution Imaging Spectroradiometer (MODIS) has been a great success, and MODIS data have been widely applied to the study of areas of land [1–3], atmosphere [4–6], and ocean [7–9]. The spectral bands of MODIS can be divided into the two categories of land and ocean for numerous applications [10]. Ocean bands have a higher sensitivity than land bands, but the nadir spatial resolution of ocean bands is lower than that of land bands. For ocean bands, the resolution is only 1000 m. In comparison, the resolution for the land bands can reach 250 m, and the resolution of the majority of land bands is 500 m. Furthermore, the red portions of the ocean visible bands are often saturated over coastal waters. Recent studies [10–12] have demonstrated that for coastal waters, especially for turbid coastal waters, the quality of the remote sensing signals from land bands are of equally high quality as those of ocean bands, and the red portions of visible land bands are not easily saturated. Therefore, the land bands with high nadir spatial resolution are very suitable for the observation of coastal waters [10].
The inherent optical properties (IOPs) of a body of water, which mainly include the absorption (a) and backscattering coefficients (bb), play important roles in the derivation of biogeochemical parameters such as the chlorophyll concentration [13], characteristics of suspended particles [14–16], euphotic zone depth [17], primary productivity [18,19], and dissolved and particulate organic carbon concentrations [20–22].
The retrieval algorithms for IOPs are well summarized in the IOCCG report 5 [23]. The results suggest that the retrieval performance of a quasi-analytical algorithm (QAA) [24] is very high for both clear and coastal waters. In addition, the QAA is mathematically simple and physically transparent [23,24]. To date, the QAA has undergone several updates [17, 25], and the core content of these updates is the procedure for estimating the absorption coefficient at 555 nm (a555). The latest formally published version is QAAv5 [25]. These updates have made the QAA more consummate, and the performance has improved greatly [17,25]. In QAAv5, the a555 is estimated using the remote sensing reflectance (Rrs) at 443, 490, 555, and 667 nm. There are only three land bands of MODIS with a high spatial resolution for the visible channels, which are 469, 555, and 645 nm; thus the QAAv5 does not directly apply all three of these land bands. In addition, QAAv5 cannot be directly applied to coastal turbid waters because of saturation at the 667nm band. Therefore, a new estimation algorithm for a555 is needed; then, the IOPs can be derived using the Rrs values of the three high resolution bands.
This study attempts to estimate the a555 using the ratio between the Rrs values of two bands, 555 and 645 nm. Although this method has been referenced in Lee et al. (2002) [24], it is seldom used. In fact, the use of red-green-bands-ratio is sometimes helpful to the robustness of retrieving algorithm [26]. The estimation of a555 based on this method can be applied to obtain the data for coastal waters at high resolution by using MODIS land band data. In addition, the IOPs of other bands can be derived for in coastal waters using this method by analyzing a combination of ocean bands and land bands while avoiding the use of the 667 nm band.
This paper is arranged as follows. First, the data sets used to develop and evaluate the algorithm are introduced. Second, the method development of the new algorithm is introduced. Third, the performance of the newly developed algorithm and QAAv5 are evaluated and discussed using the in situ measurement data set from the Eastern China Seas (ECS) and the MODIS data for the ECS.
2. Data
The data sets analyzed for this paper consist of four separate data sets, the synthetic data set [23], the NASA bio-Optical Marine Algorithm Data set (NOMAD) data set [27], the data set of measured ECS data (ECS data set), and the MODIS data. The synthetic data set which is constructed using optical and bio-optical models includes numerous water bodies and is often used to develop, test and evaluate the ocean color retrieval algorithm. The NOMAD data set is the in situ measured data set for the global oceans. Detailed descriptions of the synthetic data set and NOMAD data set can be found at http://www.ioccg.org/groups/OCAG_data.html and in Werdell and Bailey [27], respectively; hence, they are not described here. The emphasis of this section is the description of the ECS data set.
In contrast to the synthetic and NOMAD data sets, the ECS data set is a specific data set that focuses on the water optical properties of the ECS.
The ECS data set was collected during 2006 and 2014 with a total of 9 cruises performed in the ECS. The measured parameters mainly include the remote sensing reflectance, total absorption coefficient, and backscattering coefficient, among others. In total, there were 204 in situ measurement locations that included remote-sensing reflectance (Rrs) and absorption coefficient measurements, and among these sites, there were 112 sites that also included backscattering coefficients. The distribution of these sites is shown in Fig. 1.
Fig. 1 Distribution of the in situ measurement locations. The red circles represent the locations where Rrs and were measured, and the green dots represents the locations where Rrs, a, and bb were measured.
The in situ measurement locations covered nearly all of the water types in the ECS. In addition, the in situ measured optical data covered all of the seasons in the ECS; thus the variation in the ECS is represented well in the data set, supporting the evaluation of the applicability of retrieval algorithms. The measurement methods and instruments employed to measure the optical data are described as follows.
The measurement of Rrs prior to the year of 2013 has been well described by Chen et al. [11], and in the year of 2013, the Rrs were measured using the Compact-Optical Profiling System (C-OPS) (Biospherical Instruments Inc.) and hyperspectral instruments from the Trios Radiation Measurement Sensor with Enhanced Spectral Resolution (RAMSES) (Trios Inc., Rastede, Germany) using the profiling method [28].
All of the absorption coefficients data were measured using a Wetlabs AC-S meter. The bb data collected prior to 2013 were measured using a Wetlabs BB-9 sensor, and the data in 2013 were measured using a HydroScat-6 sensor (HS-6, Hydro-Optics, Biology, & Instrumentation Laboratories, Inc.). Notably, the post processing schemes of the data measured using AC-S, BB-9 and HS-6 sensors were realized strictly according to the corresponding AC-S, BB-9 and HS-6 manuals [29–31]. Specially, for AC-S, temperature and salinity correction was performed using the synchronous measured temperature and salinity data, subsequently, scattering correction of temperature and salinity corrected absorption coefficients was made based on the proportional relationship between scattering errors and scattering coefficients at the corresponding bands [29]; for BB-9 and HS-6, the path compensation correction were made using the corrected AC-S data [30,31]. It's also noted that, in order to make the in situ measurements comparable to those derived from ocean color remote sensing retrieval algorithms [24], all of the IOPs data used in this paper are the optically average values of the vertical profile data using the approach of Gordon [32].
The MODIS data are primarily composed of several scenes and several pixels. A detailed description of the MODIS data and its processing is provided in section 3.2.
3. Methods
3.1 Development of the new QAA algorithm
Following the framework of the previous QAA, the high resolution, visible land bands of MODIS at 555 and 645 nm were used to develop the new QAA named QAA-RGR based on Red-Green-bands-Ratio.
As reported by Lee et al. [24], a555 can be estimated using the ratio between Rrs640 and Rrs555. For this study, the coefficients of Eq. (19) in Lee et al. [24] were tuned using the IOCCG synthetic data set according to the Rrs at 555 and 645 nm. As shown in Fig. 2, there is a strong correlation between a555 and the ratio of Rrs645 to Rrs555 (coefficient of determination of 0.97) and the data points for a555 and Rrs645/Rrs555 are convergent even for turbid waters. The formula for the estimation of a555 is expressed in Eq. (1). The coefficients in Eq. (1) were fit by the least-squares method.
The ECS data set is also plotted in Fig. 2 and exhibits a good consistency with the IOCCG data set. The NOMAD data set is not shown in Fig. 2 because there are no Rrs data for 645nm.
Once the a555 value is derived, the bb555 can be derived from Eq. (2).
Here, u is a function of Rrs, which can be derived from Eq. (3) as given by Lee et al. [24].
The key step is the estimation of the spectral slope of bb. In the QAA, the value of Y characterizes the spectral distribution of the particle backscattering coefficients (bbp) and is derived from an empirical relationship [24]; bb is expressed well by Eq. (3), using the same pattern as bbp.
The value of Y is strongly correlated with the specific bb values at a given wavelength.
As shown for the IOCCG synthetic data set in Fig. 3, the NOMAD data set and ECS data set exhibit high consistency in the relationship between the spectral slope of spectral bb and bb555. The coefficient of determination for the relationship between Y and bb555 values can reach 0.94.
When bb555 is less than 0.03, Y increases with a decrease in bb555, and when bb555 approaches the backscattering coefficient of pure water (approximately equal to 0.000923 m−1) [33], Y approaches the spectral slope of backscattering coefficient of pure water (approximately equal to 4.32). The Y and bb555 of pure water are denoted by the red circle in Fig. 3. When bb555 is greater than 0.03, Y remains approximately constant and suggests a more regular variation than is actually observed. Notably, the Y values were calculated by fitting the wavelengths and their bb values according to the function type defined in Eq. (3); The wavelengths used in the fitting process for the ECS, NOMAD, synthetic data sets and pure water are described as follows: ECS data set, 412, 440, 488, 532, 595, and 676nm for BB-9, and 410, 442, 488, 532, 550, and 640nm for HS-6; NOMAD data set: 412, 443, 490, 510, 555, and 665nm; synthetic data set: 400-800nm (interval = 10nm); pure water: 400-710nm (interval = 5nm).
The relationship between Y and bb555 as shown in Eq. (4) was fit by the least-squares method when bb555 was less than or equal to 0.03, and for the bb555 values greater than 0.03, Y was empirically thought to be 0.4 according to its variation tendency.
Here, spectral bb is used to construct the backscattering spectral model instead of spectral bbp, mainly because of the large uncertainty of the fit when the particle content is low. Notably, the ECS data set was mainly used to develop the relationship between Y and bb555 in conjunction with the synthetic and NOMAD data sets. The development of the relationship between a555 and the ratio of Rrs645 to Rrs555 was completely based on the synthetic data set. The purpose of showing the ECS data set in Fig. 2 is to demonstrate the applicability of Eq. (1) (developed based on the synthetic data set).
Furthermore, if the Rrs at another wavelength is known, the value of absorption coefficient at this wavelength can be derived using Eq. (5).
The newly developed retrieval algorithm for a(λ) and bb(λ) QAA-RGR is summarized in Table 1.
Table 1. Steps of QAA-RGR for deriving a(λ) and bb(λ). Note that steps 1, 2, 4, 6, and 7 are adopted from Lee el al [24].
3.2 Evaluation scheme
The performance of the newly developed QAA-RGR algorithm was first evaluated using the ECS data set. Then, the performance of the QAA-RGR when applied to MODIS data for the ECS was evaluated. In addition, the performance of QAAv5 was also evaluated for comparison. The following statistical parameters for the evaluation of results are used in this paper: correlation coefficient (R), the mean absolute percentage error (MAPE), and the slope and intercept of the linear relationship between the true values and the derived values fit by the least-squares method. The MAPE is defined as follows:
Here, qtrue and qderived are the in situ data and derived values, respectively, and N is the number of values that are compared.
To illustrate the performance of QAA-RGR when applied to MODIS data, the following three steps were implemented:
- (1) Apply QAA-RGR to the 500 m resolution MODIS land bands data to obtain the IOPs and thus the distribution of IOPs at 500 m resolution.
- (2) Apply QAA-RGR to the combined MODIS ocean and land band data. First, use the MODIS land band data (555 and 645 nm) to derive the a555, on the basis of obtaining the bb555 from steps 1, 2, and 4 listed in Table1, and further obtain the distribution of IOPs at 1000 m resolution. Here, the resolution of land bands is reduced to 1000 m.
- (3) Apply QAAv5 to the MODIS ocean band data to obtain the IOPs for the ocean bands to obtain the distribution of IOPs with 1000 m resolution.
Here, two scenes over the ECS were selected from MODIS-Aqua. One scene was located near the Yangtze River Estuary which is extremely turbid, on 20 April, 2006 as shown in Fig. 6(a), and the other was located near the Shandong Peninsula on 21 November, 2011. These two in situ measurements were collected when the MODIS-Aqua passed over this region as shown in Fig. 8(a). Note that, for the MODIS data, the L1A data from MODIS-Aqua were downloaded from the following website: http://oceancolor.gsfc.nasa.gov. The processing from L1A to L1B, and subsequently from L1B to L2, were accomplished using the Sea-viewing Wide Field-of-view Sensor (SeaWiFS) Data Analysis System version 6.4 with the latest updates. For the processing from L1B to L2, the short-wave infrared (SWIR) atmospheric correction algorithm [34,35] was employed to obtain the Rrs.
4. Results and discussion
4.1 Performance of QAA-RGR for the in situ ECS data set
Figures 4 and 5 present the comparison results for QAA-RGR and QAAv5 when applied to the ECS data set. The statistical parameters are given in Tables 2 and 3. For the derivation of a(λ), the QAA-RGR and QAAv5 exhibit approximately the same performance for the ECS data set. The mean R values of the comparison results (estimated vs. measured values) for QAA-RGR and QAAv5 are 0.95 and 0.95, respectively, and the mean MAPE values are 15.8% and 15.8%, respectively. For the derivation of bb, although QAA-RGR performs slightly worse than QAAv5, it still exhibits good retrieval results. The mean R values of bb comparison results for QAA-RGR and QAAv5 are 0.94 and 0.96, respectively, and the mean MAPE values are 19.5% and 21.7%, respectively.
Fig. 4 Comparisons between the a(λ) values estimated using QAA-RGR (a) and QAAv5 (b) and measured values for the ECS data set at 443, 469, and 490 nm.
Fig. 5 Comparisons between the bb(λ) values estimated using QAA-RGR (a) and QAAv5 (b) and the measured values for the ECS data set at 412, 443, 490, 532, and 555 nm.
Table 2. Statistical parameters for the comparisons between the a(λ) estimated using QAA-RGR and QAAv5 and the measured values for the ECS data set at 443, 469, and 490 nm.
Table 3. Statistical parameters for the comparisons between the bb(λ) values estimated using QAA-RGR and QAAv5 and the measured values for the ECS data set at 412, 443, 490, 532, and 555 nm.
4.2 Performance when applied to MODIS-Aqua
Figures 6(b) and 6(c) show the distributions of the values of a469 and bb532 derived from MODIS land bands (555 and 645 nm) Rrs data with 500m resolution using QAA-RGR. Notably, the resolutions of the 469, 555, and 645 nm bands are 500 m, 500 m, and 250 m, respectively; hence, the resolution of the 645nm band is reduced to 500m resolution. Figures 6(d) and 6(e) present the a490 and bb532 values retrieved from the combined ocean and land bands data using QAA-RGR after the resolution of the land bands (555 and 645 nm) was reduced to 1000 m. It’s noted that the resolution reduction of land bands is performed using the spatial averaging method [36]. The a490 and bb532 derived by QAAv5 using the MODIS traditional ocean band Rrs data with 1000 m resolution are also shown in Figs. 6(f) and 6(g). Clearly, the distribution patterns of the a469 and bb532 derived by QAA-RGR using land band data with 500 m resolution coincide well with the water color (Yellow: extremely turbid; Light green: turbid; Green: clear; Blue: more clear) as evident from the RGB image in Fig. 6(a), especially for bb532. This finding is not surprising because bb is highly correlated with the concentration of suspended particulate matters [16, 37]. For the waters that are adjacent to the coast, the values of a469 and bb532 are high because of the high content of suspended sediments, the strong absorption in the blue bands and the strong scattering characteristics of suspended sediment. In contrast, the waters that are farther away from the coast are characterized by low values of a490 and bb532 because of the low content of suspended sediments. Second, the distributions of a490 and bb532 derived by QAA-RGR from the combined ocean and land band data with 1000 m resolution are highly consist with the a469 and bb532 derived by QAA-RGR using the land band data with 500 m resolution in terms of the distribution pattern and amplitude, except for a certain level of stripe noise in the images obtained from QAA-RGR results using land bands with 500 m resolution. This stripe noise can be removed using the method proposed by Rakwatin et al. [38]. Third, considering the a490 and bb532 values derived by QAAv5 using the ocean band data with 1000 m resolution, the distribution patterns clearly differ from those derived by QAA-RGR. For example, for the waters adjacent to the coast, which are very turbid and appear yellowish, the a490 and bb532 values of these waters are lower than the values of the waters located at the top right corner of Fig. 6(a) which appear relatively clearer. In addition, the a490 and bb532 values derived by QAAv5 are obviously lower than the values derived by QAA-RGR, especially for the highly turbid waters adjacent to the coast.
Fig. 6 Distribution of IOPs across the Yangtze River Estuary for the MODIS-Aqua image acquired on 20 April, 2006. (a) is the RGB image, and location T in the RGB image is used to illustrate the derivation performance of Rrs667 in the following section; (b) and (c) are a469 and bb532 values derived by QAA-RGR for the land bands data with 500 m resolution, respectively; (d) and (e) are a490 and bb532 values derived by QAA-RGR for the combined ocean and land band data with 1000 m resolution, respectively; (f) and (g) are a490 and bb532 derived by QAAv5 for the ocean band data with 1000 m resolution, respectively.
To quantitatively illustrate the retrieval performance of QAA-RGR and QAAv5, the data points that are located on the red line marked 1 in Fig. 6(a) were selected. First, the retrieval results from QAA-RGR using the land band data with 500 m resolution are approximately the same as those retrieved by QAA-RGR using the combined ocean and land band data with 1000 m resolution (Fig. 7). Second, the a490 and bb532 values derived by QAAv5 are obviously lower than those derived by QAA-RGR, especially for the waters at longitudes west of 122.6°E on red line 1. For these waters, the a490 and bb532 values derived by QAA-RGR are approximately 4-fold greater than the values derived by QAAv5, but for the waters east of 122.6°E on red line 1, the a490 and bb532 values derived by QAA-RGR are only slightly greater than the values derived by the QAAv5.
Fig. 7 Comparisons of a490 (a) and bb532 (b) derived by QAA-RGR with 500 m resolution and a 1000 m resolution and by QAAv5 with 1000 m resolution for the data points located on the red line 1 shown in Fig. 6(a).
Which is the most reliable of the values derived by QAA-RGR and QAAv5? Figure 8(a) is an image of the area surrounding the Shandong Peninsula acquired by MODIS-Aqua on 21 November, 2011, and Fig. 8 follows a similar layout to Fig. 6. Similar results are observed. For example, for the QAA-RGR, the IOPs derived using land band data with 500 m resolution are highly consistent with those derived using the combined ocean and land band data with 1000 m resolution. For QAAv5, the values of the derived IOPs are all slightly lower than those derived by QAA-RGR.
Fig. 8 Distribution of IOPs around Shandong Peninsula for a MODIS-Aqua image acquired on November 21, 2011. (a) is the RGB image; (b) and (c) are a469 and bb532 values derived by QAA-RGR from the land band data with 500 m resolution, respectively; (d) and (e) are a490 and bb532 values derived by QAA-RGR from the combined ocean and land band data with 1000 m resolution, respectively; (f) and (g) are a490 and bb532 values derived by QAAv5 from the ocean band data with 1000 m resolution, respectively.
Two in situ measurements marked as green circles on red line 2 in Fig. 8(a) were collected near the overpass time of MODIS-Aqua. The location of the green circle on the left was measured at 1:17 p.m. local time on 21 November, 2011, and the location on the right was measured at 2:50 p.m. The measurement times of the two points are very close to the overpass time of MODIS-Aqua (at approximately 1:30 p.m. local time). It's noted that, for the left measurement point, the distribution of a490 and bb532 profiles are homogeneous vertically; and for the right point, the distributions of a490 and bb532 profiles are homogeneous vertically at the depths from 0m to 40m. Figures 9(a) and 9(b) present the a469 and bb532 values along red line 2 derived by QAA-RGR and by QAAv5. The in situ measurements are also presented in Fig. 9. The a490 and bb532 values derived by QAA-RGR are greater than those derived by QAAv5. In addition, for the waters where in situ measurements were collected, the a490 and bb532 values derived by QAA-RGR are similar to the in situ measurement values, but the values derived by QAAv5 are all lower than the in situ measurement values. For QAA-RGR, when applied to the land bands data with 500 m resolution for the area surrounding the Shandong peninsula, which is considerably clearer than the area located near the Yangtze River Estuary, the noise level of the retrieval results is slightly greater than that of the QAAv5 results, which uses ocean band data with high accuracy. However, when the resolution of the land band data are reduced to 1000 m, the noise level of the retrieval results derived by QAA-RGR using the combined ocean and land band data is similar to that from the QAAv5 results. However, the distribution pattern of the retrieval results derived by QAA-RGR using the land band data with 500 m resolution are highly consistent with that derived by QAA-RGR using the combined ocean and land band data with 1000 m resolution. This finding can also be further verified from the comparison of the results for the sample locations denoted on red line 2 (marked in Fig. 8(a)), as shown in Fig. 9.
Fig. 9 Comparisons of a490 (a) and bb532 (b) derived by QAA-RGR with a 500 m resolution and 1000 m resolution and by QAAv5 with a 1000 m resolution for the sample locations which are located on red line 2 shown in Fig. 10(a).
Therefore it can be concluded that good retrieval results from MODIS land bands with high resolution can be obtained using the QAA-RGR, and for the retrieval of IOPs using MODIS data with 1000 m resolution, the application of the QAA-RGR in conjunction with the use of ocean band data after the resolution of the land band data is reduced to 1000 m can produce more reliable results than the use of QAAv5, which only uses ocean band data.
To further verify these conclusions, a greater number of in situ measurements with corresponding MODIS-Aqua data were selected. The temporal and spatial match up method proposed by Bailey and Werdell [39] was employed. It’s noted that, the value of the match-up point from MODIS is not the mean of the valid pixels defined in Bailey and Werdell [39]; rather, is the value of the pixel closest to the in situ location in order to avoid the influence of ocean fronts.
Finally, 16 in situ measurements for a and bb were successfully matched with MODIS-Aqua data. Figures 10 and 11 present the evaluation results for the comparisons of the IOPs derived by QAA-RGR and by QAAv5 from the in situ measurements. Clearly, the IOPs derived by QAA-RGR using the 500 m and 1000 m resolution data are very similar to those of the in situ measurements; all the data points in the scatter plot are located close to the 1:1 line. In contrast, for the QAAv5, the derived values are obviously lower than the in situ measurements in general. The statistical results (Tables 4 and 5) further demonstrate that, for the ECS, QAA-RGR performs better than QAAv5 when applied to MODIS data.
Fig. 10 Comparisons of the values of a(λ) estimated using QAA-RGR (a) and using QAAv5 (b) with the measured values for the match-up points at 443, 469, and 490 nm.
Fig. 11 Comparisons of the values of bb(λ) estimated using the QAA-RGR (a) and using the QAAv5 (b) with the measured values for the match-up points at 490, 532, and 555 nm.
Table 4. Statistical parameters for the comparisons of the values of a(λ) estimated using QAA-RGR and using QAAv5 with the measured values for the match-up points at 443, 469, and 490 nm.
Table 5. Statistical parameters for the comparisons of the values of bb(λ) estimated using QAA-RGR and using QAAv5 with the measured values for the match-up points at 490, 532, and 555 nm.
From the above-presented evaluation results, it can be observed that the QAAv5 exhibits a good retrieval performance for the in situ data set for the ECS, but when applied to the MODIS-Aqua data over the ECS, the derived values are often underestimated, especially for extremely turbid coastal waters.
For the Rrs spectra derived from the MODIS data for the ECS, the Rrs values of the blue bands are frequently overestimated compared to those of green and red bands. In addition, as evident in the Fig. 2 in Wang et al. [35], the Rrs values of the blue bands are obviously overestimated compared with the Rrs values of the green and red bands. In addition, Lamquin et al. [40] also demonstrated this phenomenon by performing in situ match up comparisons. In the QAAv5, because the numerator of X defined in Eq. (5) in Lee et al. [25] that is used to estimate a555 is the sum of the Rrs values of the two blue bands, the overestimation of Rrs values of the blue bands for the Rrs spectra will directly lead to an overestimation of the X value. Furthermore, the a555 will be underestimated, and finally, the values of a and bb derived by QAAv5 will be underestimated. For the above referenced match up points, the in situ measured Rrs645/Rrs555 values are compared with the MODIS derived values in Fig. 12, and the X values defined in Eq. (5) in Lee et al. [25] derived from the in situ and MODIS data are also compared. Considering the Rrs645/Rrs555 ratio, the Rrs of the in situ and MODIS-Aqua data are approximately the same values. For X, a number of the values derived from the MODIS-Aqua data are similar to the values from the in situ data set, but other values from the MODIS data are obviously overestimated, which will directly lead to an underestimation of the IOPs derived by QAAv5. For the overestimation of the Rrs of the blue bands by MODIS, Lamquin et al. [40] concluded that this phenomenon was mainly attributed to calibration issues with the blue bands [41]. In this paper, the latest calibration coefficients were used when processing MODIS data, but this phenomenon still exists. In fact, because the blue bands are far away from the SWIR bands, extrapolating using SWIR bands in the atmospheric correction inevitably induce large errors for the blue bands. So, atmospheric correction may play an important role in the estimation of Rrs of blue bands. Hence, underestimation of QAAv5 when applied to MODIS may be mainly induced by the effect of atmospheric correction and is worth of further research in future.
Fig. 12 Comparisons in situ measured and MODIS-Aqua-derived values of Rrs645/Rrs555 and X for the match-up points.
In addition, for the extremely turbid waters that are located adjacent to the coast and appear yellowish (see Fig. 6(a)), the ocean band at 667 nm of MODIS with the radiance value greater than 4.2 mW cm−2 μm−1 sr−1 is saturated [34]. In practical terms, the Rrs667 used in QAAv5 is derived using an empirical relationship (Eq. (9) in Lee et al. (2009)) [25]. The larger uncertainty of the estimation of Rrs667 using this empirical relationship is apparent in Fig. 1(b) in Lee et al. [25]; in which, it is clearly evident that Rrs667 is underestimated at high values. As shown in Fig. 13, for the Rrs spectrum of the location marked T in Fig. 6(a), the Rrs at 667 nm of the location marked by the red circles is obviously underestimated, and the true value of Rrs667 should be closer to that of Rrs645. For X, the Rrs667 is located in the denominator; thus the underestimation of Rrs667 will also directly lead to the overestimation of the value of X in addition to the overestimation of Rrs for the blue bands. Consequently, the IOPs derived by QAAv5 will be further underestimated. For extremely turbid coastal waters, the IOPs results derived by QAAv5 are more severely underestimated compared to the waters where the 667 nm band is not saturated.
Fig. 13 Rrs spectra at 443, 469, 488, 555, 645, and 667 nm at location T shown in Fig. 6(a), derived from MODIS-Aqua data collected over the ECS on April 20, 2006.
A QAAv6, referred to in Mitchell et al. [42], has also been developed. The main body of QAAv6 is still QAAv5, and the main change is that the reference band is shifted to 667 nm for turbid waters. When applied to MODIS data, band saturation will also severely limit the use of QAAv6. As noted, although the value of Rrs667 can be estimated, the accuracy of this estimation is a severe concern, and the overestimation the values of Rrs for the blue bands will also impact the performance of QAAv6.
For QAA-RGR, when it is applied to MODIS land band data to derive the distribution of IOPs with 500 m resolution, the regions that satisfy the criteria proposed by Wang et al. [43] are recommended for good retrieval results based on a consideration of the low signal-noise-ratio (SNR) of the SWIR bands.
5. Conclusions
The proposed algorithm QAA-RGR was applied to in situ measurements data and MODIS land band and ocean band data for the ECS and compared with QAAv5. The conclusions are summarized as follows:
- (1) For the in situ measurements data, QAA-RGR and QAAv5 are characterized by a similar retrieval performance for IOPs.
- (2) When QAA-RGR is applied to MODIS land band data and to combined MODIS ocean band and land band data, the distribution of IOPs can be obtained at 500 m and 1000 m, respectively. Both results exhibit high consistency. Specifically, the distribution of IOPs derived by QAA-RGR from the MODIS land band data at 500m resolution exhibit a high quality.
- (3) When QAAv5 is applied to MODIS ocean band data, the retrieval results are underestimated, especially for coastal turbid waters. Two reasons of this underestimation are as follows: First, QAAv5 is very sensitive to the Rrs of blue bands. For the MODIS data, there is a large uncertainty in the estimation of Rrs of the blue bands associated with atmospheric correction. Second, the 667 nm ocean band, which is used in QAAv5 is easily saturated in turbid coastal waters, and there is a large uncertainty in the estimation of Rrs667 using an empirical relationship.
The advantage of QAA-RGR is the derivation of a555 using the land band data at 645 and 555 nm. On one hand, the use of the Rrs of the blue bands, which have larger uncertainties, can be avoided. On the other hand, the use of band 667 nm which is easily saturated also can be avoided.
The QAA-RGR was developed using a synthetic data set and may be applied extensively to other water bodies. This paper focuses on the ECS. Hence, the application of the QAA-RGR to other bodies of waters is not discussed. To ensure the robustness of the QAA-RGR for other water bodies, further testing should be conducted prior to use.
Acknowledgments
We are grateful to all of the people who worked hard collecting the in situ data. The provision of the HS-6 data by Keping Du (Beijing Normal University) and the C-OPS data by Haili Wang (Xiamen University) are highly appreciated. Additional thanks to the NASA Ocean Biology Processing Group for the provision of the MODIS-Aqua data and SeaDAS software packages. Financial support of this study was provided by NSFC- Shandong Joint Fund for Marine Science Research Centers under the grant number U1406404, and the National Natural Science Foundation of China under the grant number 41276041, 40876005, and 60638020. Valuable Comments from two anonymous reviewers are much appreciated.
References and links
1. M. A. Friedl, D. K. McIver, J. C. Hodges, X. Y. Zhang, D. Muchoney, A. H. Strahler, C. E. Woodcock, S. Gopal, A. Schneider, A. Cooper, A. Baccini, F. Gao, and C. Schaaf, “Global land cover mapping from MODIS: algorithms and early results,” Remote Sens. Environ. 83(1), 287–302 (2002). [CrossRef]
2. X. Zhang, M. A. Friedl, C. B. Schaaf, A. H. Strahler, J. C. Hodges, F. Gao, B. C. Reed, and A. Huete, “Monitoring vegetation phenology using MODIS,” Remote Sens. Environ. 84(3), 471–475 (2003). [CrossRef]
3. F. A. Heinsch, M. Zhao, S. W. Running, J. S. Kimball, R. R. Nemani, K. J. Davis, et al.., “Evaluation of remote sensing based terrestrial productivity from MODIS using regional tower eddy flux network observations,” IEEE Trans. Geosci. Rem. Sens. 44(7), 1908–1925 (2006). [CrossRef]
4. J. Bendix, B. Thies, and J. Cermak, “Fog Detection With Terra-MODIS and MSG-SEVIRI,” In Proceedings of the 2003 Meteorological Satellite Conference (pp. 427–435).
5. P. Minnis, Q. Z. Trepte, S. Sun-Mack, Y. Chen, D. R. Doelling, D. F. Young, D. A. Spangenberg, W. F. Miller, B. A. Wielicki, R. R. Brown, S. C. Gibson, and E. B. Geier, “Cloud detection in nonpolar regions for CERES using TRMM VIRS and Terra and Aqua MODIS data,” IEEE Trans. Geosci. Rem. Sens. 46(11), 3857–3884 (2008). [CrossRef]
6. C. D. Papadimas, N. Hatzianastassiou, N. Mihalopoulos, X. Querol, and I. Vardavas, “Spatial and temporal variability in aerosol properties over the Mediterranean basin based on 6‐year (2000–2006) MODIS data,” J. Geophys. Res. 113(D11), D11205 (2008). [CrossRef]
7. R. L. Miller and B. A. McKee, “Using MODIS Terra 250 m imagery to map concentrations of total suspended matter in coastal waters,” Remote Sens. Environ. 93(1-2), 259–266 (2004). [CrossRef]
8. C. Hu, F. E. Muller-Karger, C. J. Taylor, K. L. Carder, C. Kelble, E. Johns, and C. A. Heil, “Red tide detection and tracing using MODIS fluorescence data: A regional example in SW Florida coastal waters,” Remote Sens. Environ. 97(3), 311–321 (2005). [CrossRef]
9. C. Hu, “An empirical approach to derive MODIS ocean color patterns under severe sun glint,” Geophys. Res. Lett. 38(1), 1944 (2011). [CrossRef]
10. B. A. Franz, P. J. Werdell, G. Meister, E. J. Kwiatkowska, S. W. Bailey, Z. Ahmad, and C. R. McClain, “MODIS Land Bands for Ocean Remote Sensing Applications,” Proc. Ocean Optics XVIII, Montreal, Canada. 9–13 October (2006).
11. S. Chen, T. Zhang, and L. Hu, “Evaluation of the NIR-SWIR Atmospheric Correction Algorithm for MODISs-Aqua over the Eastern China Seas,” Int. J. Remote Sens. 35(11–12), 4239–4251 (2014). [CrossRef]
12. C. Hu, L. Feng, Z. Lee, C. O. Davis, A. Mannino, C. R. McClain, and B. A. Franz, “Dynamic range and sensitivity requirements of satellite ocean color sensors: learning from the past,” Appl. Opt. 51(25), 6045–6062 (2012). [PubMed]
13. K. L. Carder, F. R. Chen, Z. P. Lee, S. K. Hawes, and D. Kamykowski, “Semianalytic Moderate-Resolution Imaging Spectrometer algorithms for chlorophyll-a and absorption with bio-optical domains based on nitrate-depletion temperatures,” Adv. Space Res. 33, 1152–1159 (1999). [CrossRef]
14. M. Twardowski, E. Boss, J. B. Macdonald, W. S. Pegau, A. H. Barnard, and J. R. V. Zaneveld, “A model for estimating bulk refractive index from the optical backscattering ratio and the implications for understanding particle composition in case I and case II waters,” J. Geophys. Res. 106(C7), 14129–14142 (2001). [CrossRef]
15. E. Boss, W. S. Pegau, M. Lee, M. Twardowski, E. Shybanov, G. Korotaev, and F. Baratange, “Particulate backscattering ratio at LEO 15 and its use to study particle composition and distribution,” J. Geophys. Res. 109(C1), C01014 (2004). [CrossRef]
16. G. Neukermans, H. Loisel, X. Mériaux, R. Astoreca, and D. McKee, “In situ variability of mass-specific beam attenuation and backscattering of marine particles with respect to particle size, density, and composition,” Limnol. Oceanogr. 57(1), 124–144 (2012). [CrossRef]
17. Z. P. Lee, A. Weidemann, J. Kindle, R. Arnone, K. L. Carder, and C. Davis, “Euphotic zone depth: Its derivation and implication to ocean-color remote sensing,” J. Geophys. Res. 112(C3), C03009 (2007). [CrossRef]
18. J. R. C. Zaneveld, J. C. Kitchen, and J. L. Mueller, “Vertical structure of productivity and its vertical integration as derived from remotely sensed observations,” Limnol. Oceanogr. 38(7), 1384–1393 (1993). [CrossRef]
19. Z. P. Lee, K. L. Carder, J. Marra, R. G. Steward, and M. J. Perry, “Estimating primary production at depth from remote sensing,” Appl. Opt. 35(3), 463–474 (1996). [CrossRef] [PubMed]
20. D. A. Siegel, S. Maritorena, N. B. Nelson, D. A. Hansell, and M. Lorenzi-Kayser, “Global distribution and dynamics of colored dissolved and detrital organic materials,” J. Geophys. Res. 107(C12), 3228 (2002). [CrossRef]
21. W. M. Balch, H. R. Gordon, B. C. Bowler, D. T. Drapeau, and E. S. Booth, “Calcium carbonate measurements in the surface global ocean based on Moderate-Resolution Imaging Spectroradiometer data,” J. Geophys. Res. 110(C7), C07001 (2005). [CrossRef]
22. M. J. Behrenfeld, E. Boss, D. Siegel, and D. M. Shea, “Carbon-based ocean productivity and phytoplankton physiology from space,” Global Biogeochem. Cycles 19(1), GB1006 (2005). [CrossRef]
23. “Remote Sensing of Inherent Optical Properties: Fundamentals, Tests of Algorithms, and Applications, in Reports of the International Ocean-Colour,” IOCCG report5 (2006).
24. Z. Lee, K. L. Carder, and R. A. Arnone, “Deriving inherent optical properties from water color: a multiband quasi-analytical algorithm for optically deep waters,” Appl. Opt. 41(27), 5755–5772 (2002). [CrossRef] [PubMed]
25. Z. P. Lee, B. Lubac, J. Werdell, and R. Arnone, “An Update of the Quasi-analytical Algorithm (QAA_v5),” http://www.ioccg.org/groups/Software_OCA/QAA_v5.pdf, 2009.
26. C. Le, C. Hu, D. English, J. Cannizzaro, Z. Chen, L. Feng, R. Boler, and C. Kovach, “Towards a long-term chlorophyll-a data record in a turbid estuary using MODIS observations,” Prog. Oceanogr. 109, 90–103 (2013). [CrossRef]
27. P. J. Werdell and S. W. Bailey, “An improved in-situ bio-optical data set for ocean color algorithm development and satellite data product validation,” Remote Sens. Environ. 98(1), 122–140 (2005). [CrossRef]
28. R. C. Smith, C. R. Booth, and J. L. Star, “Oceanographic biooptical profiling system,” Appl. Opt. 23(16), 2791–2797 (1984). [CrossRef] [PubMed]
29. W. E. T. Labs, Inc. (2009), “Ac Meter Protocol Document,” Revision P. www.wetlabs.com
30. W. E. T. Labs, Inc. (2010), “Scattering Meter ECO BB-9 User’s Guide,” Revision K. www.wetlabs.com
31. H. O. B. I. Labs, Inc. (2008), “HydroScat-6 Spectral Backscattering Sensor USER’S MANUAL,” Revision H. www.hobilabs.com.
32. H. R. Gordon, “Diffuse reflectance of the ocean: influence of nonuniform phytoplankton pigment profile,” Appl. Opt. 31(12), 2116–2129 (1992). [CrossRef] [PubMed]
33. A. Morel, “Optical properties of pure water and pure sea water,” in Optical Aspects of Oceanography, N. G. Jerlov and E. S. Nielsen, eds. (Academic, New York, 1974), pp.1–24.
34. M. Wang, “Remote sensing of the ocean contributions from ultraviolet to near-infrared using the shortwave infrared bands: simulations,” Appl. Opt. 46(9), 1535–1547 (2007). [CrossRef] [PubMed]
35. M. H. Wang, J. W. Tang, and W. Shi, “MODIS-derived ocean color products along the China east coastal region,” Geophys. Res. Lett. 34(6), L06611 (2007). [CrossRef]
36. B. A. Franz, P. J. Werdell, G. Meister, E. J. Kwiatkowska, S. W. Bailey, Z. Ahmad, and C. R. McClain, “MODIS Land Bands for Ocean Remote Sensing Applications,” Proc. Ocean Optics XVIII, Montreal, Canada.9–13 (2006).
37. E. Boss, L. Taylor, S. Gilbert, K. Gundersen, N. Hawley, C. Janzen, T. Johengen, H. Purcell, C. Robertson, D. W. H. Schar, G. J. Smith, and M. N. Tamburri, “Comparison of inherent optical properties as a surrogate for particulate matter concentration in coastal waters,” Limnol. Oceanogr. Methods 7(11), 803–810 (2009). [CrossRef]
38. P. Rakwatin, W. Takeuchi, and Y. Yasuoka, “Stripe noise reduction in MODIS data by combining histogram matching with facet filter,” IEEE Trans. Geosci. Rem. Sens. 45(6), 1844–1856 (2007). [CrossRef]
39. S. W. Bailey and P. J. Werdell, “A multi-sensor approach for the on-orbit validation of ocean color satellite data products,” Remote Sens. Environ. 102(1-2), 12–23 (2006). [CrossRef]
40. N. Lamquin, C. Mazeran, D. Doxaran, J. H. Ryu, and Y. J. Park, “Assessment of GOCI radiometric products using MERIS, MODIS and field measurements,” Ocean Science Journal 47(3), 287–311 (2012). [CrossRef]
41. G. Meister, B. A. Franz, E. J. Kwiatkowska, and C. R. McClain, “Corrections to the calibration of MODIS Aqua ocean color bands derived from SeaWiFS data,” IEEE Trans. Geosci. Rem. Sens. 50(1), 310–319 (2012). [CrossRef]
42. C. Mitchell, A. Cunningham, and D. McKee, “Remote sensing of shelf sea optical properties: Evaluation of a quasi-analytical approach for the Irish Sea,” Remote Sens. Environ. 143, 142–153 (2014). [CrossRef]
43. M. H. Wang, S. Son, and W. Shi, “Evaluation of MODIS SWIR and NIR-SWIR atmospheric correction algorithms using SeaBASS data,” Remote Sens. Environ. 113(3), 635–644 (2009). [CrossRef]
