Harmful algal blooms (HABs) occur in many coastal regions around the world and appear to be increasing in severity and extent (Hallegraeff, 1993, 2003; Grattan et al., 2016; Gobler, 2020). HABs have caused severe economic losses to aquaculture, fisheries, and tourism while creating major environmental and human health impacts (Anderson et al., 2000; Landsberg, 2002; Heisler et al., 2008). Toxic bloom-forming algae can cause wildlife mortality or human seafood poisoning, and even HAB species that do not produce toxins are able to cause harm through development of high biomass, leading to foams or scums, depletion of oxygen as blooms decay, or destruction of habitat for fish or shellfish by shading of submerged vegetation (Sellner et al., 2003). Such impacts from HABs pose a serious threat to aquatic ecosystems and can disrupt their associated food web (Fogg, 1969; Paerl, 1988). Therefore, considerable attention has been focused on methods to reduce the risks of HAB impacts (Sengco and Anderson, 2004; Anderson, 2009; Anderson et al., 2012).

Satellite ocean color sensors offer a means of detecting and monitoring HABs in the ocean and coastal zone. The potential value of remote sensing for HABs was first described by Mueller (1981), after an experimental ocean color sensor attached to an aircraft detected a bloom of Karenia brevis. As the instrument was developed to simulate the Coastal Zone Color Scanner (CZCS), launched in late 1978, this indicated the capability for satellite detection of blooms. Various approaches were further developed for detection and monitoring of HABs from satellite remote sensing (reviewed by Klemas, 2012; Blondeau-Patissier et al., 2014).

There are many advantages of satellite ocean color products compared to in situ monitoring: wide temporal and spatial coverage, and inexpensive; however, there are significant limitations. First, the coarse spectral resolution of current ocean color satellites is only sufficient to distinguish certain clear anomalies in bloom coloring, as the visible spectrum is mostly determined by optical properties of varying concentration of chlorophyll pigments. Second, data from satellite ocean color only reflect the state of the waters at the moment the measurements are taken, whereas the blooms are subject to physical forcing from tidal and wind-driven currents which will advect them away from this state. It is thus very challenging from satellite data alone to provide information regarding the future development and movement of HABs. Another limitation of ocean color remote sensing is that satellite observations are often hampered by weather conditions, such as clouds and sea fog, which can substantially reduce the number of valid ocean color pixels. As a result, satellite imagery may only partially capture features of HABs.

In recent years, the use of Lagrangian particle tracing (or other numerical models) for HAB monitoring has gained more interest (Olascoaga et al., 2008; Wynne et al., 2011; Son et al., 2015; Kwon et al., 2019; Li et al., 2020; Fernandes-Salvador et al., 2021). Lagrangian particle tracing models are useful for determining sources, trajectories, and destinations of drifting water parcels, with high temporal and spatial resolution. Thus, this approach could compensate for some of the limitations of satellite ocean color. However, this approach has so far been used to track particle locations along limited trajectories, preventing a synoptic view of the variability of water properties associated with HABs, e.g., algal concentration. Recently, a new methodology was proposed for synthesizing ocean color data (or in situ observations) with ocean circulation velocity fields from an operational model (Jönsson et al., 2009; Jönsson and Salisbury, 2016). This method could significantly improve our capability for monitoring of HABs.

Therefore, this paper aims to expand the Jönsson et al. (2009) scheme for synthesizing satellite observations and Lagrangian data to include extrapolation for better interpretation of the development of HABs. This improved scheme can not only fill gaps in HAB patches in the satellite images captured on cloudy days but also provides an early warning of harmful algal risk. An application example is presented showing the development of a Karenia mikimotoi bloom along the southern coast of England.

The applicability of this prediction scheme is demonstrated by analyzing the predictions of water properties in two applications in the coastal and offshore regions of southeastern England.

It is quite clear, as seen both in our findings and in earlier studies (e.g., Jönsson et al., 2009; Jönsson and Salisbury, 2016), that biological processes in coastal are strongly affected by physical advection of water parcels, and that combining satellite derived products and simulated velocity fields using particle tracking provides a powerful approach for such analyses. While earlier works by Jönsson et al. (2011) and Jönsson and Salisbury (2016) are primarily focused on estimating the rate of change in satellite properties, we present results showing that the method can be expanded to predict locations and concentrations of the properties. This novel application can be used to better predict HAB events and to identify the extent and location of blooms more precisely, especially in regions with strong tidal cycles where surface currents are predictable.

We find our results to be generally physically coherent and that any patches of high phytoplankton biomass are predicted and advected in a reasonable way. The nature of the method is such that we expect significant methodological errors to be seen as noise or spurious changes in our predicted fields (see Jönsson et al., 2009 for further discussion). We see for example, as expected, generally sharper Chl gradients in coastal waters compared to the open ocean in the Celtic Sea and English Channel. The method is also able to successfully predict the location of a coccolithophore bloom. The skill of the method discussed here is in line with what earlier studies have reported from the California Current (Jönsson and Salisbury, 2016) and Gulf of Maine (Jönsson et al., 2009) when estimating rates of change, which suggests that the extrapolation approach has a similar ability to provide useful information.

While the prediction tool presented in this study has the potential to improve regional HAB monitoring by more precisely assessing how the current state will develop in the near future, we readily agree that there are limitations. It is inevitable that there will be inaccuracies due to imperfections in the modeling and methodology. Errors in the prediction method could arise from many aspects, e.g., imperfect velocity of fields, extrapolation of model velocities in time and space, numerical integration of the trajectory path, and the omission of vertical velocity.

In this study, Lagrangian particle trajectories were computed according to the effect of both advection and diffusion. However, because each trajectory only contains a limited number of particles, the final locations of simulated particles could not be representative of all possible particle locations due to diffusion. The problem could be especially acute for some areas with large horizontal current shear. Thus, it would be important to keep in mind that uncertainty from diffusion always exists in these areas. The problems could be solved by seeding many more particles at each initial location, though the additional burden of computing their trajectories is currently unfeasible for this real-time application. A practical solution could be to introduce a probability density function indicating possible locations for each traced particle. However, it is beyond the scope of this study and substantial studies of this question are anticipated in the future. Regarding errors from advection, this was assessed in the previous study of Jönsson et al. (2009) by comparing the changes in Chl and sea surface temperature (SST) between the start and end positions of particle trajectories advected between two satellite images. The results showed that the advective errors are small relative to total changes due to other factors.

The extrapolation procedure will also introduce errors to the predictions. The predicted values were estimated via linear extrapolation with a weighting function. To reduce error from the extrapolation, when the extrapolated value for a trajectory exceeded 50% of the mean value of all extrapolations, this trajectory was disregarded and was not included for the calculation of final prediction. Considering uncertainty from the IOPs retrieval algorithm, errors in predicted IOPs are inevitable, but errors in R_rs could possibly be compensated by reconstruction from these IOPs. Furthermore, it is also worth noting that although individual trajectories have errors, the statistics obtained from thousands of particles are still very informative.

Still, while not perfect, the resulting predictions provide an enhanced set of information for managers and stakeholders to include when assessing the risk of HABs that has not been available until now. The prediction tool is also agnostic to any of its component modules. We are able to easily leverage new improved ocean circulation models and satellite products into the framework to provide as accurate predictions as possible.

This study set out to develop a prediction scheme for monitoring HABs by merging satellite observations and Lagrangian particle tracking. Two case studies in regions along the coast of England have shown that the Lagrangian methodology is effective for the interpretation of satellite data for early warning of HAB risk. The accuracy of the predictions relies on many factors. Particularly, the uncertainties from modeling and methodology, e.g., velocity of fields, extrapolation of model velocities in time and space, numerical integration of the trajectory path, and the omission of vertical velocity, may have big impacts on determining the accuracy of final predictions. Notwithstanding these imperfections, this work offers a powerful tool for monitoring and observing the state of the marine ecosystem. The synoptic, time-resolved quantification is invaluable to our understanding of HABs developments. Animated sequences generated by this method promote greater understanding and usage of satellite ocean color data for communicating with aquaculture farmers. Future research will extend the approach to predict quantitative risk maps for key high-biomass HAB species. Hence our future priority is to seek further applications of this new technique to support the aquaculture industry with improved early warning of potential HABs.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

JL developed the prediction scheme, implemented the code, and wrote the first draft of the manuscript. PM proposed initial ideas and helped shape the research. BJ provided suggestions on code and algorithm development. MB helped carry out numerical simulations. All authors discussed the results and commented on the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.