气候变化领域集成服务门户(CCPortal): Corrigendum to “A global approach for chlorophyll-a retrieval across optically complex inland waters based on optical water types” [Remote Sens. Environ., 229: 159–178](S0034425719301786)(10.1016/j.rse.2019.04.027)

CCPortal

DOI	10.1016/j.rse.2020.111837
	Corrigendum to “A global approach for chlorophyll-a retrieval across optically complex inland waters based on optical water types” [Remote Sens. Environ., 229: 159–178](S0034425719301786)(10.1016/j.rse.2019.04.027)
	Neil C.; Spyrakos E.; Hunter P.D.; Tyler A.N.
发表日期	2020
ISSN	00344257
卷号	246
英文摘要	The authors regret that recognised errors in the formulation of the three-band algorithms and in processing of one of two-band algorithms led to a misrepresentation of the algorithms performance results. The authors would like to apologise for any inconvenience caused. Abstract The paper “A global approach for chlorophyll-a retrieval across optically complex inland waters based on optical water types” presents a framework for optimising the parameterisation of chlorophyll-retrieval algorithms and selecting the most appropriate algorithm(s) for the optical water type (OWT) of interest. Importantly, this approach is shown to improve the retrieval of chlorophyll-a (Chla, mg/m3) when optimised algorithms are dynamically selected from an ensemble based on OWT membership using a global dataset of in-situ spectra gathered from the inland water remote sensing community. However, we regret that recognised errors in the formulation of the three-band algorithms and in processing of one of two-band algorithms led to a misrepresentation of the algorithms performance results. These algorithms now show improved performance when compared to those presented in the original article. The errors have affected 4 of 47 algorithms assessed for Chla retrieval performance and 3 of 13 algorithm(s) we recommended for application to the various OWTs considered in the paper. Importantly, these results do not affect the key findings in terms of approach and the main conclusions of the paper that optimising algorithms for specific OWTs improves our ability to retrieve chlorophyll concentrations on a global scale. 1. Introduction The identified error in three-band algorithms was due to incorrect formulation of a reflectance ratio and another error in the processing in the algorithms previously published by Anatoly Gitelson, Alexander Gilerson and their colleagues in references below. Here we address the correction of the errors in relation to the sections of the original paper. 2.5 Model B Model B refers to the three-band algorithm developed by Dall'Olmo et al. (2003) and used (Moses et al., 2009; Gitelson et al., 2011) for retrieval chlorophyll concentration from MERIS satellite data using Rrs measured at 665, 708 and 753 nm: (equation numbers remain as in the original paper): [Formula presented]where a = 232.329 and b = 23.174 are determined empirically. As suggested by Dall'Olmo et al. (2003), the combination of three bands alters the model sensitivity to the presence of optically active constituents by removing the effects of SPM and CDOM (Rrs(665) and Rrs(708) are comparably influenced by SPM and CDOM and Rrs(753) is mainly driven by backscattering). 2.7 Model D Model D refers to the three-band algorithm (Dall'Olmo et al., 2003) calibrated using close range field measurements of Rrs and Chla taken from Fremont lakes, Nebraska USA (Gurlin et al., 2011): [Formula presented]where a = 315.50, b = 215.95 and c = 25.66. 2.8 Model E Model E refers to the advanced two-band semi-analytical algorithm proposed by Gilerson et al. (2010). This model was reprocessed excluding imaginary numbers from the analysis with the same equations as in the original article. [Formula presented] 2.9 Model F Model F has been removed from this analysis. 2.10 Model G Model G refers to an advanced three-band semi-analytical algorithm proposed by Gilerson et al. (2010). The algorithm is based on a semi-analytical expression for the ratio of NIR to red reflectance in combination with water absorption and a⁎ ph665 (=0.022 × Chla−0.1675); [Formula presented]where aw753 = 2.494 m−1 (Pope and Fry, 1997). Substituting the expression for a⁎ ph665 gives; [Formula presented] Or for regional calibration of a⁎ ph665; [Formula presented]where a and b are determined empirically. 3.1 Full dataset comparison Fig. 4 (replaces original Fig. 4) shows a quantitative comparison of Chla generated from each of the examined models against the in-situ measurements. Corresponding error metrics are presented in Fig. 5 (replaces original Fig. 5). Contrary to results presented in the original article, the three-band algorithms perform well under ORG and CLUS formats with correlation coefficients r in excess of 0.7 and slopes close to 1 for models B, D and G. It should also be noted that after reprocessing, Model E also shows good performance. The corresponding performance scores as determined by the objective scoring system are shown in Fig. 6 (replaces original Fig. 6). Models B, D and G now produce maximum quantitative error metric scores of 12, 13 and 10 respectively, each corresponding to CLUS format and model E has a score 11 in the original form. With a total score of 14, still the highest scoring algorithms when comparing modelled and measured Chla for the full validation dataset were models A, C, J and R. [Figure presented] [Figure presented] [Figure presented] 3.2 Performance per OWT The second stage of the algorithm validation focussed on performance within a specific OWT group. Chla concentrations generated using the OWT training subsets (CLUS parameterisations) were compared to outputs from CAL and ORG models for only the corresponding OWT assigned spectra. Results from objective scoring per OWT group are shown in Fig. 7. We now see that models B, D, E and G perform consistently across all OWTs and are identified as top performers in four of the 13 OWTs (6, 8, 11 and 12). Interestingly, top performance was demonstrated by the ORG version of model G (OWT 8), highlighting the strength of the semi-analytical approach. As per results presented in the original article, recalibrating the models using the full dataset did not provide any advantages in terms of performance. Corresponding error and regression statistics for the maximum OWT model scores are shown in Table 4, which replaces original Table 4. [Figure presented] [Table presented] 3.3 Recommendation for a dynamic OWT switching algorithm Based on objective scoring and individual error statistics, the recommended algorithm selection for inland waters exhibiting water-leaving reflectance characteristics similar to those described by Spyrakos et al. (2018) are shown in Table 5. The main changes from the original paper are the inclusion of model G_ORG, B_CLUS and model E_ORG as the recommended algorithms for OWTs 8, 11 and 12 respectively. Of the 13 groups, six models identified as high performers in terms of Chla retrieval appear in their CLUS form, where the recalibration of the model was based on the OWT group subset data. Four OWTs are more accurately characterised in terms of Chla retrieval by their original published algorithms (ORG) and three OWTs by parameterising the proposed models using the complete training dataset (CAL). OWT recommended algorithms (as shown in Table 5) were combined to form a dynamic switching algorithm, which selects the optimum Chla model for a given OWT. Estimates generated by the dynamic switching algorithm are compared to in-situ measurements of Chla concentration in Fig. 8c (replaces original Fig. 8c). The dynamic switching algorithm estimates log-transformed Chla from Rrs with a mean absolute error of 0.18. [Table presented] [Figure presented] 4.1 Discussion The empirical and semi-analytical three-band algorithms were shown to perform comparably to their two-band counterparts. 5 Conclusion An error in the original paper in formulating the three-band algorithms and another one in processing lead to misrepresentation of the results for models B, D, E and G. These have been reassessed for quantitative scoring and have subsequently been recommended for the final dynamic switching algorithm. Updated results have demonstrated that overall performance was improved by focussing algorithm development within distinct OWT clusters and that optimising algorithms for specific OWTs improves our ability to retrieval chlorophyll concentrations on a global scale. The optimisation of algorithms for application to specific OWTs provides a more effective framework to improving the accuracy of chlorophyll-a retrieval than simply re-parameterising algorithms using large datasets encompassing multiple OWTs. © 2020 Elsevier Inc.
语种	英语
来源期刊	Remote Sensing of Environment
文献类型	期刊论文
条目标识符	http://gcip.llas.ac.cn/handle/2XKMVOVA/179299
作者单位	University of StirlingFK9 4LA, United Kingdom
推荐引用方式 GB/T 7714	Neil C.,Spyrakos E.,Hunter P.D.,等. Corrigendum to “A global approach for chlorophyll-a retrieval across optically complex inland waters based on optical water types” [Remote Sens. Environ., 229: 159–178](S0034425719301786)(10.1016/j.rse.2019.04.027)[J],2020,246.
APA	Neil C.,Spyrakos E.,Hunter P.D.,&Tyler A.N..(2020).Corrigendum to “A global approach for chlorophyll-a retrieval across optically complex inland waters based on optical water types” [Remote Sens. Environ., 229: 159–178](S0034425719301786)(10.1016/j.rse.2019.04.027).Remote Sensing of Environment,246.
MLA	Neil C.,et al."Corrigendum to “A global approach for chlorophyll-a retrieval across optically complex inland waters based on optical water types” [Remote Sens. Environ., 229: 159–178](S0034425719301786)(10.1016/j.rse.2019.04.027)".Remote Sensing of Environment 246(2020).