In this study we used the NEAT methodology to develop a GEE-based tool for the global detection of CEP (the Global Eutrophication Watch) using satellite CHL. In its screening procedure, the NEAT—a robust satellite-based preliminary assessment tool of eutrophication potential—unifies, in a single map, the temporal and spatial information of the area under consideration. It combines the level and trend of CHL to generate six patterns of water quality23. The CHL level generates two patterns based on the CHL concentration (α [mg m− 3]), the first being composed by CHL lower than the threshold α, CHL < α (L), and the other by CHL ≥ α (H). The trends have three patterns, namely: waters with decreasing trend (D), with no trend (N), or with increasing (I) trend. In this way, a composite map of six classes can be generated, viz. LD, LN, LI, and HD, HN, HI. Before moving on to the explanation of the meaning of each class, it is worth defining the terms adopted in this paper for clarity. Eutrophic potential will be used to indicate a productive system with high CHL, whereas eutrophication potential refers to the process of becoming eutrophic or a progression of an already eutrophic water body. In addition to the above definitions, we also introduce oligotrophication potential which is associated with a least productive water body. Hence, pixels flagged HD, HN and HI are eutrophic potential with HD indicative of systems under recovery, whereas in LI and HI are eutrophication potential. In HI, the conditions may worsen as the water body is already eutrophic potential. Moreover, LD is suggestive of reversed eutrophication, that is, further oligotrophication. LN and HN are indicative of L and H CHL but stable conditions over the analysis period. It is important to note that classification of waters as being L or H is subject to the consideration of the threshold α, which will vary depending on the conditions of each region. However, the same is not the case for D, N or I. Trends will most probably be impacted by the length of the analysis period and/or other environmental factors controlling the variability of CHL rather than a given α. As such, both LD, LI and HD, HI provide critical information about the eutrophication of the system under scrutiny. The global eutrophication watch, therefore, not only provides important information of areas potentially in need of preventive management efforts, but also helps in evaluating the impacts of measures taken to reduce the effects of eutrophication.
The NEAT procedure uses a threshold of 5 mg m− 3, and this threshold is computed based on the most recent 3-year mean data of the analysis period. Nevertheless, this threshold is not fixed, and users are able to adjust the level and the composite period to area specific values as different regions may have different thresholds according to the region’s background.
The estimation of trends at pixel level is based on the Sen’s slope method45—a non-parametric trend estimation method—which detects the presence of monotonic trends in a yearly data record at the 90% significance level. Nonparametric tests provide higher statistical power in the case of nonnormality, as in the case of CHL, and are robust against outliers and large data gaps. Trends estimated below a critical threshold are treated as N (no trend). Moreover, as the focus is on the detection of eutrophication potential with consideration of it being a process occurring over a long-time scale (on the order of years), the temporal trends in CHL are estimated from annual maximum obtained from monthly composites of each considered year. The choice is partly motivated by the fact that the evaluation of existence of monotonic trends can also be statistically challenged by short-term variability in CHL. So, by using annual CHL maximum from monthly composites, we effectively remove the seasonal and short-term variabilities. Doing so, we focus on the CHL peak season. Consequently, the obtained trends reflect the interannual behaviour of the phytoplankton bloom season, assuming that the bloom is manifested as high biomass.
For a global detection of CEP, the Global Eutrophication Watch uses the currently available 17-year record of CHL data from Moderate Resolution Imaging Spectroradiometer on Aqua (MODISA), reprocessing 2018, with a spatial resolution of 4 km (https://oceancolor.gsfc.nasa.gov/reprocessing/r2018/aqua/). The data set is stored in the App’s asset and its temporal extension is updated on a yearly basis. In addition, the data sets in the App will also be updated following NASA reprocessings that periodically occur when advances in algorithms or sensor calibration knowledge are shown to significantly improve product quality.
Besides the global detection of eutrophication, a case study was developed in the Bohai Sea (BS), a semi-enclosed marginal sea, one of the China seas, to demonstrate the usefulness to the introduced tool. In coastal regions like BS the retrievals of CHL based on the standard algorithm often fail. So, for this case study, we used the open access, improved CHL data from the Marine Environmental Watch of the NOWPAP (https://ocean.nowpap3.go.jp/). The improved CHL is obtained using a regionally tuned CHL algorithm developed by a project of Yellow Sea Large Marine Ecosystem on Ocean Colour (the YOC algorithm) designed to alleviate the impacts of suspended sediments on CHL retrievals46. The YOC algorithm was originally developed using the Sea-viewing Wide Field-of-view Sensor (SeaWiFS) sensor bands46. Its application to MODISA data is based on the regression between SeaWiFS and MODISA bands and band ratios. Please see ref. 22 for further. The YOC CHL has been demonstrated superior relative to the standard products47–49 and thus it is of great value for eutrophication assessment in the NOWPAP region. Users have access to the YOC data from https://ocean.nowpap3.go.jp/.
Given that the global level 3 data constitute our default asset for the global eutrophication assessment, following the case study in the BS (2.1), in 2.2 we briefly compare the trends estimated using the YOC CHL with those obtained from the global data readily obtainable from the National Aeronautics and Space Administration (NASA) of the U.S. This comparison is essential given that the NASA global standard products are more accessible than any other lower-level (such as level 2) data to non-expert users, including water quality managers and decision makers. In addition, it is the least expensive way for a rapid eutrophication assessment before a thorough investigation can follow.
The Global Eutrophication Watch (Fig. 3) on the GEE is composed of three main fields: (1) the data-set specification panel, (2) the selection of trend detection interval and (3) the specification of the CHL composite interval and the threshold selection panels. The data panel allows the selection of two default data sets, that is, MODISA and YOC CHL. In practice, only YOC CHL can be checked as MODISA is the de facto default. Moreover, this panel also includes a box for users to enter the path to an Earth Engine asset of monthly CHL for the tool to read and use for the assessment. The option is especially important given the challenges associated with CHL retrievals in the coastal waters. Unlike in the open ocean, where phytoplankton dominate the optical properties or co-vary with other optically active constituents, in coastal waters phytoplankton may vary independently of the optical constituents, and thus the global CHL product may fail to resolve phytoplankton variations50. So, this option can be understood as a plug-in that allows users around the globe to conduct the eutrophication assessment based on their own assets. This feature enables users to incorporate regionally improved CHL data while keeping the assessment procedure consistent. This has the immediate impact of allowing consistent results to be obtained from a spectrum of ecosystems with different characteristics. The next panel is used to specify the trend detection interval, the start and end years. This panel also includes a button to toggle views, that is, to split the map into two windows providing a capability for comparative assessment. The impact of inclusion of more years in the trend detection analysis, for instance, can be verified by simply using two different year intervals. Finally, the last user defined parameters are for the CHL threshold. Controls for start and end dates are available for users to indicate the time interval to be used to compute the mean CHL. This is used in conjunction with the cut-off level (threshold) to split L vs. H CHL waters.