The demand for environmental vulnerability assessments is increasing as the impacts of climate change become apparent (Wang et al., 2019). The vulnerability of coral reefs to repeated thermal extreme events may be influenced by spatial variability in temperature, and previous heat exposure and recovery time (Palumbi et al., 2014; Ainsworth et al., 2016; Safaie et al., 2018; Thomas et al., 2019); with recent research indicating that those coral able to spawn post a bleaching event (bleached and recovered, partially bleached, or unbleached) may have greater resistance to acute events (Hughes et al., 2019). Extreme thermal events resulting in mortality alter reef assemblages by reducing larval recruitment of key species (Hughes et al., 2019). This subsequently through time reduces overall biodiversity (e.g., species richness), which has been shown to increase the vulnerability of the ecosystem, diminishing its ability to maintain function (Beaugrand et al., 2015), resist disease (Bruno et al., 2007) and introduced species (Wernberg et al., 2011), adapt to future changes (Mora et al., 2016) and potentially hinder recovery. Reducing additional anthropogenic pressures on coral reefs may not minimize the thermal threat, however it may reduce their overall vulnerability (Wooldridge, 2009; Convention on Biological Diversity, 2015), and aid their ability to recover at the local scale.

Ultimately the scientific consensus is that the most successful way to mitigate the cause of thermal stress to coral reefs is to limit temperature increase (Ainsworth et al., 2016; Lough et al., 2018), which will require global action on reducing greenhouse gas emissions (IPCC, 2014). Whilst new earth observation technologies could assist in monitoring, maintaining or reducing greenhouse gas emissions (e.g., carbon and methane) (Doppelt, 2010) to hinder increases in SST, reviewing these strategies is beyond the scope of this framework.

Spatial remote sensing has been viewed as advantageous to environmental monitoring, providing cost-efficient observations across large (≥10 km²) and difficult to access geographical areas at extended temporal scales (decadal) (Hedley et al., 2016). However, the tropical marine environment provides difficulties for traditional aerial remote sensing data acquisition, such as the spatial resolution of satellite data (Kachelriess et al., 2014; Hedley et al., 2016), and water clarity (e.g., turbidity, atmospheric noise including clouds, and ability of spectral wavelengths to penetrate water) (Dierssen et al., 2019). As such these methods have not been able to adequately detect fine-to-moderate scale (≤10 m²) coral reef responses to environmental pressures including temperature variations. Recent advancement however has seen satellite SST being monitored and utilized (e.g., NOAA Coral Reef Watch; Strong et al., 2013) to model and predict coral reef bleaching, enabling targeted assessment in locations deemed vulnerable to bleaching or sites of interest (Marshall et al., 2017). Additionally, geomorphic zone and benthic community habitat maps have been developed from structural measurements (Heyward et al., 2018) and through object-based image analysis approaches, which identify and attribute spatially homogeneous pixel clusters within remote sensing datasets, though accuracy is limited by appropriate field data and depth, with higher accuracy in shallower areas where the aerial sensor can penetrate (e.g., <5 m). This approach combines color and textural characteristics of pixel groups in combination with physical attributes such as depth, slope, significant wave height, and spatial relationships between the mapped categories (Hedley et al., 2016; Asner et al., 2017; Roelfsema et al., 2018; Li et al., 2019). Accuracy assessment methods of remotely sensed or predicted data with in-water data should be used to understand the uncertainty around this modeled data (Phillips and Marks, 1996; Meyer et al., 2019).

Traditional high resolution and publicly available satellites have increased revisit times, such as Sentinel two which can revisit locations every 5 days, providing greater opportunity to monitor surface changes. While rapid return, nanosatellites have further increased temporal frequency of image acquisition, with up to daily scenes of the globe (Asner et al., 2017; Planet Team, 2017). These are usually constellations of hundreds of relatively small satellites that together provide daily coverage of the Earth’s surface at fine-to-moderate (≤10 m) spatial resolution (Planet Team, 2017). Whilst still hindered by the same depth issues (<5 m) radiometric issues that influence traditional satellites, with increased temporal frequency (daily) they provide greater opportunity for cloud-free, non-turbid, low, or high tidal, reduced wave conditions to be captured for greater coverage of coral reefs (Hedley et al., 2018; Joyce et al., 2018), increasing the spatial scale and resolution of ecological assessment from these satellites (Figure 3). Whilst currently limited by sensors and spectral bands (e.g., Red, Green, Blue, Infrared are standard output of rapid-return nanosatellites), the cost to produce (Kenyon and Stanton, 2017; Sarzi-Amade et al., 2017) these satellites is far less than traditional satellites.

Low altitude (<120 m) drones provide an avenue to acquire very fine (<1 m) spatial resolution data compared to satellites, which due to being flown at low heights can provide images at a low cost acquired under specific environmental conditions, such as cloud free, low tide, and clear water. However, whilst they bridge a gap between very fine scale in-field data collection and moderate-to-broadscale satellite studies, they also present limitations, including flight time, flight conditions [e.g., wind, staying within line of site (dependent on conditions), data processing, and weight limitations in mounting sensors (Joyce et al., 2018)]. As such their application needs to be specific to the ecological phenomena being captured (Figure 3). Further research is required for best practice experimental design for drones in ecological studies, though Joyce et al. (2018) provides some recommendations for use in marine environments, including optimizing flight times and direction to minimize sun glint and reflection, in conjunction with flying at low tide to further minimize water interference.

Machine learning is a subset of artificial intelligence and statistics that builds models based on supervised input or labeled data. The modeling method itself determines statistical relationships and patterns in datasets with limited predefined model structure or statistical inference. Utilizing earth observation data with machine learning could enable virtual monitoring stations to be set up to alert changes in baseline conditions in near-real-time (Bakker and Ritts, 2018), initiating a move to the response phase quickly.

Access to free and low cost satellite programs (e.g., Landsat missions, Sentinel, Planet Labs, MODIS) has exponentially increased the amount of data available to researchers, enabling access to long-term global datasets that can be utilized for baseline change studies (Zhu et al., 2019). However, their use by environmental researchers has been limited by their capacity to process and analyze such large datasets, and methods to downscale information to scales relevant to ecology and management (Zhu et al., 2019).

The recent development in cloud processing [e.g., Microsoft Azure, Amazon Cloud, and Google Earth Engine (GEE)] and machine learning has been significant, increasing the amount of data available and processing capability, providing an avenue to link fine-scale ecological site data with broader-scale remote sensing data, locally and at a global scale. The development of cloud processing services has been fundamental in determining environmental baselines (Hansen et al., 2013), highlighting locations of possible ecosystem vulnerability, and aiding in the development of The Sustainable Development Goals through near-real time monitoring (Zhu et al., 2019). However, machine learning models are highly data driven and most require vast amounts of training data before they can be applied (Sun et al., 2017). Currently marine earth observation data lacks large libraries of standardized in situ datasets, essential for prototyping, training, tuning and benchmarking machine learning models. While cloud computing and access to specialized computing hardware [i.e., Graphics Processing Units (GPUs)] may be essential to train models and process the large volume of required data, the models are fast to run once they are appropriately trained and can be used for near-real time analysis. It is therefore essential to pre-train the models in advance and in anticipation of thermal impact events, for this reason we have included this in the readiness phase of the framework.

Coral monitoring of small reef areas is currently achieved by in situ data collection following well-established ecological survey methods (English et al., 1994; Hill and Wilkinson, 2004). Here we focus on methods applicable to the shallow reef flat and intertidal environments, as these areas are the most suitable to aerial remote sensing. Such methods tend to use sensors to measure point data such as temperature at a fine spatial (e.g., cm) and very fine temporal resolution (e.g., minutes), or transects undertaken by a diver to collect information on benthic condition and community composition (Jonker et al., 2008). In deeper areas not addressed here, manta-tows can be used to cover larger areas of the reef (Hill and Wilkinson, 2004). The advent of (Remotely Operated Vehicles) ROVs has allowed assessments of reefs that are less accessible to scientific divers, increasing the capacity to monitor remote and mesophotic reefs – though not the focus of this framework (Kahng and Kelley, 2007; Armstrong et al., 2019). Corals are either assessed in real time in situ by a person or photos are assessed after the survey (Jonker et al., 2008). Observations in the field are thus able to provide detail on the taxonomic composition and condition of coral communities within a defined area. The ability to collect geo-located photo quadrats representing a footprint along transects collected by divers in combination with automated photo analysis has increased our ability to get large amounts of data consistently whilst maintaining a record that could be revisited to assess the benthos condition and composition (Roelfsema and Phinn, 2010; Beijbom et al., 2015; Roelfsema et al., 2015; González-Rivero et al., 2016). How this information relates to corals outside of the survey area can be difficult and requires extensive sampling designs and uncertainty models (Phillips and Marks, 1996; Meyer et al., 2019). Sampling design in these shallow environments also needs to consider tide in relation to the application of monitoring techniques. An optimal location would occur in a reef flat zone where low tide water level locally is less than 50 cm, and away from the wave zone to allow for optimal drone photos, and high tide water level locally is appropriate to allow for in-water snorkel or SCUBA diver transects (dependent on water level and conditions) to collect in-water diver data and images.

Emerging technologies by first responders can also be utilized as also detailed in section “Spatial Remote Sensing”. For instance, imagery can be captured with a small lightweight inexpensive drone across a range of habitat types and shallow water depths to quantitatively capture the extent of bleaching (Levy et al., 2018), or enable targeted in-water field assessments. Whilst drones can cover a larger area than a diver along a transect (approximately 500 m Euclidean line), they can only cover a relatively small part of a whole reef or reef systems. Drones can also be used in zones such as the reef flat, and emergent shallow reefs, which are difficult to access and often not prioritized in the field, though are most exposed to high thermal conditions (Le Nohaïc et al., 2017), and where other remote monitoring through virtual monitoring stations may be possible (e.g., from satellite data). More traditional aerial remote sensing has included photography from helicopters and planes, this method is capable of documenting reef systems, and has been used to document bleaching events, including the recent Great Barrier Reef bleaching events (Hughes et al., 2018b,c). Through combining current response strategies with the readiness phase of ongoing data analysis, the most vulnerable and resistant sites can be detected, providing necessary but often overlooked spatial information for studying coral resilience (van Oppen and Lough, 2018), microenvironment dynamics (e.g., fine-scale hydrodynamics), and coral reef restoration (Foo and Asner, 2019).

Fine-scale aerial drone imagery (and in-water imagery) provide an opportunity to map the structural complexity of shallow and intertidal coral reef areas. The geolocated and overlapping images can use machine learning software with location and scale reference data to create a structural surface layer, termed “structure-from-motion” (SFM). Common in terrestrial studies to monitor canopy height of forests (Cunliffe et al., 2016), SFM is novel in application to coral reefs though it provides an opportunity to derive rugosity measurements at known locations, when collected with accurate Global Positioning System (GPS) data. This then has the potential to be monitored through time to provide a measurement of coral cover structural complexity. Previous studies have used LiDAR to collect similar elevation and structural measures, however this is expensive, not easy to obtain in remote areas, and is still limited to non-turbid, relatively shallow (<25 m) depths.