Trichodesmium Around Australia: A View From Space

The cyanobacterium Trichodesmium is responsible for approximately half of the ocean's nitrogen input through nitrogen fixation. Although it was first recorded near Australia in the 18th century, the knowledge of where and when large quantity of Trichodesmium around Australia could be found is still lacking. Here, using multi‐band satellite imagery acquired between 2012 and 2021, we fill this knowledge gap through the use of deep learning, designed to recognize both the spectral shapes of individual pixels and spatial morphology of surface aggregations (scums) of Trichodesmium. Trichodesmium scums were found nearly everywhere around Australia, with a cumulative footprint (i.e., where the 10‐year average density is >0.001‰) exceeding 4.6 million km2. Strong seasonality was found, with peak months between September and November. Furthermore, temperature, iron‐rich dust and black carbon aerosols, with the latter being a result of frequent bushfires, play major roles in determining the spatial distributions and seasonality of Trichodesmium.

where Trichodesmium was first reported in the 18th century (M. Furnas, 1992), a thorough knowledge of Trichodesmium distributions is still lacking.
There have been long-term monitoring programs to measure Trichodesmium around Australia. One such program is the Australia's Integrated Marine Observing System (IMOS, https://imos.org.au/; Eriksen et al., 2019). However, these programs are focused on individual sampling sites. Satellite remote sensing has also been used to map Trichodesmium surface scums (i.e., surface layer with high concentrations of cells and/or colonies to result in elevated reflectance in the near-infrared wavelengths, see Qi et al., 2018) (see summary in Table S1 and Figure S1 in Supporting Information S1), with the assumption that surface scums can serve as a proxy to represent abundance. Yet these remote sensing studies are either case studies or focused on selected regions (e.g., the Great Barrier Reef or GBR, Blondeau-Patissier et al., 2018;L. McKinna, 2015), thus not being able to provide information on all coastal waters around Australia, especially on where and why Trichodesmium blooms occur regularly there.
Here, based on the Visible Infrared Imaging Radiometer Suite (VIIRS) satellite observations and a computer deep learning (DL) model, we develop monthly maps of Trichodesmium scum density for a 10-year period between 2012 and 2021 (see Figures S2-S6 and Table S2 for methodology in Supporting Information S1). We analyze the spatial and temporal distributions changes of Trichodesmium as well as possible factors leading to such distributions, including atmospheric dust, black carbon (BC) aerosols, water temperature, winds, and ocean currents. We hypothesize that dust and BC are the two primary factors in determining the spatial distributions and seasonality of Trichodesmium around Australia.

Data and Method
The study around Australia (8°S-40°S and 110°E−160°E, Figure 1a) was based on data collected by VIIRS, the Ocean and Land Color Instruments (OLCI), and the Hyperspectral Imager for the Coastal Ocean (HICO). VIIRS was used to develop time-series of Trichodesmium density maps, while limited OLCI and HICO data were used to perform spectral discrimination analysis to verify whether the observed image features were due to Trichodesmium or something else. Briefly, for each VIIRS image, pixels containing Trichodesmium were first delineated using a DL model (specifically, Res-UNet, see application of the same approach to extract Sargassum image features from multi-band imagery over the Atlantic Ocean in Hu et al. (2023)). The DL model relies not only on the reflectance spectral shapes of individual pixels, but also on the spatial context of each pixel, and can be re-trained to improve detection accuracy through minimizing false-positive and false-negative detections, therefore showing better performance over other index-based models (e.g., the alternative floating algae index (AFAI)-based model, Hu et al., 2023). Once Trichodesmium features are detected from individual images, a spectral unmixing scheme was used to determine the sub-pixel proportion of Trichodesmium (0%-100% within a pixel), and multiple images were used to compose monthly and annual maps. Here, the detected Trichodesmium image features refer to surface scums as opposed to cells or colonies suspended in water. The scums show distinctive pigment features in their reflectance spectra (Figures S2 and S3 in Supporting Information S1), indicative of active nitrogen and carbon fixation as reported in the Arabian Sea from similar scums (Capone et al., 1998). More details on this method and its validity can be found in Supporting Information S1.
To interpret the observed spatial patterns and temporal changes of Trichodesmium, environmental data from different data sources were downloaded and analyzed. These include: daily sea surface temperature (SST), chlorophyll-a (Chl-a) concentration, wind speed, and sea surface current velocity; monthly mean dust and BC density in the atmosphere, mixed-layer depth (MLD); all-time averages (i.e., climatology) of sea surface nutrient (nitrogen and phosphorous) concentrations. Further details on these data types can be found in Supporting Information S1.

Trichodesmium Around Australia: Where and When?
The distribution of average surface density of Trichodesmium between 2012 and 2021 is shown in Figure 1a, and more detailed climatological monthly and annual distributions can be found in Figures S7 and S8 in Supporting Information S1, respectively. In the study region (8°S-40°S, 110°E−160°E), Trichodesmium scums can be observed almost everywhere around Australia except off the southern coast between 32°S and 40°S and 130°E−150°E. No scums were observed to the south of 40°S (not covered by the map). This is generally in agreement with the model predictions based on field observations (Davies et al., 2020). From Figure 1a, the cumulative footprint of Trichodesmium, defined as the waters with a climatological average surface density of >0.001‰, is approximately 4.6 million km 2 . This is equivalent to 60% of Australia's land area.
The distribution is uneven, with most Trichodesmium scums found north of ∼24°S, for example, in the GBR region, Gulf of Carpentaria, and Arafura Sea. In contrast, in the region south of ∼24°S, Trichodesmium density is much lower, and no Trichodesmium scums were found in the entire 10-year period for the southeast region bounded by 32°S-40°S and 130°E−150°E. Furthermore, the density generally decreased with increasing distance from shore, suggesting possibly land-based nutrient inputs. These spatial patterns are relatively stable from year to year ( Figure S8 in Supporting Information S1). The density in this cumulative map is mostly <1‰, suggesting that the appearance of Trichodesmium scums is sporadic. This is particularly true for the relatively shallow and dynamic coastal waters such as the Van Diemen Gulf, King Sound, and Shark Bay (Figure 1a). Even though Trichodesmium has been reported in these waters, Trichodesmium scums were formed much less frequently than in the adjacent, more offshore waters. This is possibly due to the more dynamic water column mixing in these shallow environments (Blondeau-Patissier et al., 2017). d) histograms of sea surface temperature, wind speed, and chlorophyll-a (Chl-a) concentration, respectively, corresponding to occurrence locations of floating Trichodesmium (red dashed lines) and their background water of Tricho-niche area (i.e., the cumulative footprint area, blue lines). Statistics for tropical/subtropical waters (north of 23.45°S) and temperate waters (south of 23.45°S) are presented in Figure S10 in Supporting Information S1. No Trichodesmium scums were found between 40°S and 44°S (including waters around Tasmania), and this area is therefore not included in the map (a). The cumulative footprint, defined as the waters with an average density of >0.001‰ in (a), is ∼4.6 million km 2 . Climatological monthly and annual mean distributions of Trichodesmium are available in Supporting Information S1. Similar to the uneven spatial distributions, Trichodesimum was not found equally in all months, but showed a clear seasonality ( Figure 2a, Figure S7, and Table S3 in Supporting Information S1) and interannual variability (Figures S8 and S9 in Supporting Information S1) regardless of whether tropical/subtropical and temperate waters were considered together or separately ( Figure S10 in Supporting Information S1). The maximum and minimum bloom sizes were found in September-November and May-July, respectively, with a maximum/minimum ratio of about 5. The peak months of Trichodesmium blooms can vary slightly from year to year ( Figure 2b and Table S3 in Supporting Information S1). For example, in 2014 and 2015, the peak month was February, but there was also a secondary peak in October. In contrast, the months with the minimum bloom size were more stable in different years, mostly between May and July.

Trichodesmium Around Australia: Why?
As per literature and findings prior to this study, Trichodesmium scums were not found uniformly in different environments but predominantly in waters within certain temperature and Chl-a ranges and under relatively calm conditions (Figures 1b-1d) regardless of whether the entire study region was split into tropical/subtropical (north of 23.45°S) and temperate waters (south of 23.45°S) ( Figure S10 in Supporting Information S1). Here, the 10-year cumulative Trichodesmium footprint was used as the Tricho-niche area, serving as the background water where Trichodesmium scums could be found (i.e., average 10-year density of Trichodesmium >0.001‰ in Figure 1a). SST from Tricho-niche waters covered a relatively wide range of 10-32°C, but image pixels containing Trichodesmium scums were found in a narrower range of 20-32°C, with the most optimal temperature around 27°C. This temperature range has been reported as the optimal range for Trichodesmium growth in the laboratory (Breitbarth et al., 2007). Wind speeds over the Tricho-niche waters ranged from 0 to 15 m s −1 , but most image pixels containing Trichodesmium scums were found at <∼6 m s −1 (peak frequency ∼3.5 m s −1 ), a result consistent with the findings by Blondeau-Patissier et al. (2018). Likewise, of a relatively wide range of Chl-a found from the Tricho-niche waters (0.04-2 mg m −3 ), Trichodesmium pixels were found with a narrower Chl-a range of 0.07-2 mg m −3 and a peak value of ∼0.2 mg m −3 (Figure 1d). This suggests that, although Trichodesmium prefers oligotrophic waters (Capone et al., 1997), it may not survive if the water is extremely poor in nutrients (i.e., Chl-a < 0.07 mg m −3 ). Above this Chl-a concentration, Trichodesmium may reproduce well and fertilize other phytoplankton through nitrogen fixation. Here, Chl-a is used in a relative sense because of the likely large uncertainties of satellite data products over optically complex coastal waters (Cannizzaro et al., 2013;IOCCG, 2019).
From these results, both the spatial distribution patterns and seasonality of Trichodesmium appear to be explained well by the overall nutrient regime and environmental settings around Australia, with some factors playing more important roles than others (Figure 3).
Surrounded by the Indian Ocean and the South Pacific Ocean and being far away from other continents, Australia is the smallest continent on Earth, making it isolated from remote nutrient sources. Thus, local nutrient sources must play a major role in determining the spatial distributions and seasonality of Trichodesmium. While average surface nitrate concentration (<0.6 μmol/kg, Voss et al., 2013) suggests oligotrophic conditions, the sea surface nitrate to phosphate ratio (N:P) in surface waters around Australia is ∼2:1 (Figure 3a), much lower than the Redfield ratio of 16:1 required for phytoplankton growth (Lenton & Watson, 2000). Although phosphorus is also a critical nutrient required by Trichodesmium (Hynes et al., 2009;Sohm et al., 2008), such a low N/P ratio and oligotrophic condition, together with the optimal temperature range and light availability, provide favorable conditions for Trichodesmium to grow as long as another micronutrient, iron, is supplied by certain sources such as riverine input or atmospheric deposition.
With no large river discharging waters to the ocean as compared to many other continents (McMahon & Finlayson, 2003) and with relatively narrow river plume width as compared with shelf width (Sharples et al., 2017), riverine influence on the Trichodesmium distributions is expected to be restricted to nearshore waters. Furthermore, compared with other parts of Australia, riverine inputs of dissolved inorganic nitrogen and dissolved inorganic phosphorous to the northwest coast of Australia are lower (Sharples et al., 2017). Thus, the relatively high density of Trichodesmium along that part of coast must be driven by factors other than riverine discharge. In the south, riverine flows to the coast are greater in the austral winter (June, July, and August) than in the austral summer (December, January, and February), but there is virtually no Trichodesmium found in the south. On the east coast, there are relatively large rivers that discharge to the north and north-east. However, the river flows are extremely seasonal and have high interannual variability, dominated by large flood events in the wet season (October to May in the north) (Warfe et al., 2011). The seasonality of Trichodesmium in these waters ( Figure S7 in Supporting Information S1) is out of phase of river flow. Therefore, riverine inputs of local nutrients may be ruled out as being a major factor influencing the large-scale distributions of Trichodesmium.
The local nutrient sources appear to be from dust ( Figure 3b) and BC (Figure 3c). With many large deserts on the Australian continent (∼18% of being permanent deserts and another ∼35% being episodic deserts, Geoscience Australia, 2023), Australia is a major source of dust, with mean dust pathways from the continent to northwest and southeast (De Deckker, 2019) ( Figure 3b). Likewise, bushfires or biomass burning are widespread in the northern and southeastern Australia every year because of the dry climate, resulting in high concentrations of BC aerosols with similar southeast-northwest orientation (Figure 3c). Aerosol deposition is an important source of nutrient and iron input to the ocean (Jickells & Moore, 2015;Polyviou et al., 2018;Schulz et al., 2012), and both dust and BC can supply nutrients in the form of C, N, P, and iron to the ocean (Mallet et al., 2017), stimulating growth of Trichodesmium and other phytoplankton (Sohm et al., 2011). For example, on the West Florida Shelf, following a Saharan dust event, total dissolved iron concentrations were found to increase by 30-100 folds from the background levels, with Trichodesmium concentrations increased by 100 folds (Lenes et al., 2001). In the GBR region, pigment concentration of phytoplankton (likely Trichodesmium or other cyanobacteria) was found to increase by 1.5-2 folds after a dust storm event in October 2002 (Shaw et al., 2008). Furthermore, biomass burning may substantially enhance the solubility of mineral dust around Australia, where the soluble iron can be up to ∼12% in the mixed aerosols (Winton et al., 2016). The soluble iron supplies from the atmosphere may help stimulate Trichodesmium growth in the oligotrophic waters around Australia.
The distributions of dust and BC do not always correlate with the Trichodesmium patterns found in Figure 1a. For example, off southeast Australia, both dust and BC concentrations appear relatively high, yet there is little or no Trichodesmium found. Such a disparity may be explained by the strong ocean currents off southeast Australia (>0.5 m s −1 , Figure 3d), which can prevent Trichodesmium accumulation on the surface and may also impede Trichodesmium growth in the water column. In contrast, in similar latitude but off southwest Australia, although wind and SST are similar to those off southeast Australia (Figures 3e and 3f), small amount of Trichodesmium scums can still be found due possibly to weaker currents. In addition to influencing the spatial distributions, dust and BC appear to play an important role in driving the Trichodesmium seasonality as well, as all three variables are in phase (Figure 2a) with peak months of September-November and minimum months of May-July. A multi-variate correlation analysis showed strong and statistically significant correction between Trichodesmium and the combined variables of BC and dust ( Figure S11 in Supporting Information S1), and the regression coefficients for BC and dust are both positive, suggesting their positive influence on Trichodesmium growth. Such a high correlation was also found in most high-density locations ( Figure S11c in Supporting Information S1). The question is whether this is simply a coincidence as opposed to causality. Because other phytoplankton (through Chl-a) shows the opposite phase from dust and BC (i.e., peak Chl-a is found between May and July, Figure 2a) and because iron-rich dust and BC are known to stimulate Trichodesmium growth, the latter appears to be a logical inference. In particular, the smaller size of the Tricho-niche water area between January and March, when dust concentration is nearly as high as that in September-November, may also be explained by the iron mechanism. This is because although dust may contain a greater amount of Fe, most of the Fe in dust aerosols is insoluble and cannot be utilized by phytoplankton (Winton et al., 2016). In contrast, BC contains a higher proportion of soluble Fe, and what's more important is that BC from biomass burning can enhance dusty mineral solubility (Hamilton et al., 2020;Hand et al., 2004;C. Luo et al., 2005;Winton et al., 2016). Thus, the relatively lower BC between January and March may be a reason to explain the lower Trichodesmium amount.
The opposite phase between Chl-a and dust/BC is perhaps the strongest reason to support the hypothesis that dust and BC can collectively control the seasonality of Trichodesmium. From long-term in situ measurements, phytoplankton communities around Australia are dominated by diatoms and dinoflagellates (Eriksen et al., 2019), which are driven by different environmental factors. For example, as in other subtropical oceans such as the Gulf of Mexico (GOM, Muller-Karger et al., 2015) and the South China Sea (SCS, Ji et al., 2018), non-Trichodesmium phytoplankton (and their associated Chl-a) around Australia may grow fast from nutrient enrichment due to deeper mixing in May-July induced by high winds ( Figure S12 in Supporting Information S1). However, because waters around Australia are limited in Fe (Zehr & Capone, 2020), without Fe inputs from the atmosphere, the enhanced nutrients from deep waters would benefit diatoms and dinoflagellates as opposed to Trichodesmium. In contrast, Trichodesmium may benefit more than diatoms and dinoflagellates with Fe inputs from the atmosphere.
Similar to other cyanobacteria (Paerl & Huisman, 2008) and macroalgae (Qi et al., 2022), Trichodesmium also prefers a certain temperature range for optimal growth (Figure 1b), which may explain the overall low abundance in the south than in the north (Figures 1a and 3f). However, temperature may play a less important role in controlling the seasonality of Trichodesmium (Figure 2a) because the annual variation of mean SST is 24-28°C, optimal for Trichodesmium growth.
The collective role of dust and BC in controlling both spatial distributions ( Figure 1a) and seasonality of Trichodesmium ( Figure 2a) is also reflected in the 10-year monthly time series (Figure 2b). With no significant trend (Mann-Kendall test, p = 0.72, Hussain & Mahmud, 2019), inter-annual variations are observed in all three variables, yet a multi-variant regression analysis indicates statistically significant correction (R = 0.70, p < 0.01) between Trichodesmium and combined dust and BC (Figure S11b in Supporting Information S1). In contrast, large-scale climate events such as El Niño and La Niña do not appear to affect the inter-annual changes in Trichodesmium of either tropical/subtropical waters or temperate waters, as indicated by the lack of correlation between them ( Figure S13 in Supporting Information S1).
Overall, while temperature is an important factor to determine where and when Trichodesmium can grow, dust and BC appear to be the other two major factors to drive both spatial and temporal variations of Trichodesmium around Australia.

Impacts of Episodic Events
The dominant roles of dust and BC, as argued above, are supported by episodic events, such as the 2019-2020 bushfire event. This event, from June 2019 to March 2020, is one of the most extensive and long-lasting fires in recent years (Ward et al., 2020). It burned ∼46 million acres of land covering forests, farms, and buildings, leading to mortality of 173 people and >1 billion animals (Borchers Arriagada et al., 2020;Ward et al., 2020). About ∼715 million tons of CO 2 (195 Tg C) were released into the atmosphere during the fire period (van der Velde et al., 2021), together with nitrogen, phosphorus, and iron depositions into the ocean. The event resulted in significantly higher BC around most of Australia, and particularly to the southeast of Australia during the austral summer of 2019-2020 ( Figure 4a) as emissions of BC were not uniform (Tang et al., 2021). If the dust/BC hypothesis was true, there would be increased Trichodesmium in this region. Indeed, increased Trichodesmium was found in this otherwise Tricho-poor region (Figure 4b). In other regions where Trichodesmium is usually high, increased Trichodesmium was also found. For waters off southeast Australia (blue rectangular box in Figure 4b), time-series analysis shows that the timing of the 2019-2020 Trichodesmium bloom event and increased Chl-a appeared synchronized with the positive BC anomaly (Figure 4c), suggesting a possible causality. In contrast, water temperature and MLD in this region did not show such a correspondence ( Figure S14 in Supporting Information S1). In this region, higher-than-usual Trichodesmium in this Tricho-poor region in three consecutive years after the 2019-2020 bushfire event was abnormal. Using a Mann-Kendal test, no trend in Trichodesmium areal coverage was found for the period of 2012-2018 before the bushfire event, yet the increasing trend was statistically significant (p < 0.05) when the post-fire years of 2019-2021 were included. To our knowledge, although there are no field data available to explain this pattern, these preliminary results suggest that iron recycling from the ash deposition could be a possible reason. However, this does not necessarily suggest that iron enrichment from bushfire events is the only possible cause of Trichodesmium anomalies. For example, in March 2015, an all-time high Trichodesmium areal coverage was found, yet there was no positive BC anomaly. The exact reason causing the March 2015 anomaly remains to be investigated.

Concluding Remarks
Although recurrent Trichodesmium blooms around Australia are well known, especially in the GBR region, the knowledge of where and when blooms occurred has been incomplete until now. The improved knowledge is attributed to the synoptic and frequent satellite measurements to extract Tricho-specific signals using a DL technique. These observations fill the knowledge gap for previously undocumented areas along the east, west, and 9 of 12 southern coastlines, suggesting a broader input into the nitrogen cycle of these waters than previously appreciated. Such a knowledge paves the pathway toward the understanding of why Trichodesmium around most of Australia is abundant and why strong seasonality is found. For example, while the lack of Trichodesmium off southern Australia could be explained by low water temperature, dust and BC aerosols appear to be the other major factors controlling both the spatial distributions and seasonality of Trichodesmium, as they collectively serve as a major iron source to stimulate Trichodesmium growth. However, much remains to be done to have a better understanding of Trichodesmium dynamics around Australia. For example, other than dust and BC aerosols, are there any other iron inputs from ocean circulation or upwelling? With the detailed Trichodesmium maps made available through this study, analysis of ocean circulation through hydrodynamic modeling and incorporation of more observational data could help address this question in the near future.
Climate change is a globally pervasive process and Australian coastal waters are not immune to its influences. Upper ocean warming and stratification and acidification are ongoing (Hutchins & Fu, 2017), as well as processes in the terrestrial realm that affect coastal waters such as desertification, changes in coastal runoff and wildfires expansion, in both time and space (Hoegh-Guldberg et al., 2014). How these factors will affect Trichodesmium populations (and other diazotrophs) is an area of ongoing research (e.g., Fu et al., 2014), where the findings presented here may serve as the baseline to understand future changes. For example, declining wind stress and increasing SST have been shown in the south and southeast regions of Australia, with the Tasman Sea being the fastest warming region (Duran et al., 2020). Although minimal Trichodesmium was found in these regions from the current study, should such wind and SST trends continue in the future, increased Trichodesmium would be a consequence, and, likewise, expansion of Trichodesmium to higher latitudes could also occur (Breitbarth et al., 2007). With the continuity missions of VIIRS and other satellite missions such as the hyperspectral Plankton, Aerosol, Cloud, ocean Ecosystem (PACE) (https://pace.gsfc.nasa.gov), we expect to extend the data record to future years to understand how Trichodsmium around Australia responds to environmental conditions under a changing climate.

Data Availability Statement
The processed data used in this study can be accessed through Mendeley Data: Qi (2023). This work was supported by the Joint Polar Satellite System (JPSS)/NOAA program for ocean color calibration and validation (ST13301CQ0050/1332K-P22FNEED004), by the U.S. NASA (80NSSC20M0264, 80NSSC21K0422, 80NSSC21K0439), and by the LDEO Climate and Life fellowship. We thank Dr. Barbara Robson of the Australian Institute of Marine Science for her assistance in interpreting the factors supporting Trichodesmium blooms around Australia. We are deeply grateful to the two anonymous reviewers who provided extremely detailed comments and suggestions to improve the presentation of this work. The scientific results and conclusions, as well as any views or opinions expressed herein, are those of the author(s) and do not necessarily reflect those of NOAA or the Department of Commerce.