First measurements of ocean and atmosphere in the Tropical North Atlantic using 
Caravela, a novel uncrewed surface vessel

Figure S1. 500hPa specific humidity anomaly (10kgkg) in coloured contours and 700hPa vertical velocity anomaly (10Pas) in line contours for all phases of the Boreal Summer Intraseasonal Oscillation (BSISO) during June–September 1998–2018. The anomalies are computed against the summer mean climatology (June–September 1979–2018). The vertical velocity anomalies are smoothed using a 2-D Gaussian filter with a smoothing radius of ~200km. Solid (dashed) contour lines indicate anomalous decent (ascent). Coloured and line contours are not shown where the mean surface pressure is less than 500hPa and 700hPa respectively. Phase 0 represents BSISO events featuring an amplitude less than one. Figure S2. 850hPa vector wind anomalies (ms) for all phases of the Boreal Summer Intraseasonal Oscillation (BSISO) during June–September 1998–2018. The anomalies are computed against the summer mean climatology (June–September 1979–2018). Contours are greyed out where the mean surface pressure is less than 850hPa. Figure S3. As in Figure S1, but for all phases of the Monsoon Intraseasonal Oscillation. Note the different contour interval for 500hPa specific humidity anomaly than shown in Figure S1. Figure S4. As in Figure S2, but for all phases of the Monsoon Intraseasonal Oscillation. Figure S5. As in Figure S1, but for all phases of the Madden-Julian Oscillation. Note the different contour interval for 500hPa specific humidity anomaly than shown in Figure S1. Figure S6. As in Figure S2, but for all phases of the Madden-Julian Oscillation.


Supporting Information
. 500hPa specific humidity anomaly (10 -3 kgkg −1 ) in coloured contours and 700hPa vertical velocity anomaly (10 -2 Pas −1 ) in line contours for all phases of the Boreal Summer Intraseasonal Oscillation (BSISO) during June-September 1998-2018. The anomalies are computed against the summer mean climatology (June-September 1979-2018. The vertical velocity anomalies are smoothed using a 2-D Gaussian filter with a smoothing radius of ~200km. Solid (dashed) contour lines indicate anomalous decent (ascent). Coloured and line contours are not shown where the mean surface pressure is less than 500hPa and 700hPa respectively. Phase 0 represents BSISO events featuring an amplitude less than one. Figure S2. 850hPa vector wind anomalies (ms −1 ) for all phases of the Boreal Summer Intraseasonal Oscillation (BSISO) during June-September 1998-2018. The anomalies are computed against the summer mean climatology (June-September 1979-2018. Contours are greyed out where the mean surface pressure is less than 850hPa. Figure S3. As in Figure S1, but for all phases of the Monsoon Intraseasonal Oscillation. Note the different contour interval for 500hPa specific humidity anomaly than shown in Figure S1. Figure S4. As in Figure S2, but for all phases of the Monsoon Intraseasonal Oscillation. Figure S5. As in Figure S1, but for all phases of the Madden-Julian Oscillation. Note the different contour interval for 500hPa specific humidity anomaly than shown in Figure S1. Figure S6. As in Figure S2, but for all phases of the Madden-Julian Oscillation.

Introduction
In the tropics, air-sea interactions are an important driver of weather and climate variability and can seed extreme weather events. Robust, accurate and widespread observations at the air-sea interface can improve our understanding of air-sea interaction, help to validate coupled climate models and improve the initial conditions for weather forecasts. A crucial component of the air-sea interaction is the exchange of heat and moisture at the surface. When observing these fluxes, satellites and vessels can only take us so far. To make the next step in understanding air-sea interactions, a comprehensive network of flux measurement platforms, able to sample for extended periods of time, is needed (Cronin et al., 2019).
Nowadays, there are a range of instruments spread across global oceans to capture in situ measurements as part of systems like the Global Ocean Observing System (GOOS) and the EUMETNET Surface Marine Programme. Some examples relevant to the collection of observations at the air-sea interface include the Argo network, moored and drifting buoys, and ships. However, these systems have their limitations: vessels are costly and thus only provide sparse coverage, while moorings rely on deployment and maintenance from a ship, another costly procedure. Argo floats have provided a step change in global coverage of ocean observations, but they typically only surface at 10-day intervals and so are unsuited to studying air-sea interactions on short time scales. Drifting buoys also require deployment by vessel and cannot be targeted to a region of interest, and most drifting platforms only measure near-surface ocean temperature and atmospheric pressure, in addition to recording their position. There are efforts to incorporate a range of meteorological and ocean sensors onto drifters (Centurioni et al., 2019) but their Lagrangian nature still limits their use when a set location is to be studied.
The development of autonomous surface vessels allows targeted measurements of a wide suite of surface ocean and atmospheric data in particular regions of interest, over long time periods. These vessels will be a key component of future global in situ arrays of observation platforms for air-sea fluxes with high spatial resolution and minimal reliance on ship time. Ideally, these surface vessels would be non-polluting and powered by renewable resources, such as waves, wind and sun.

Autonomous vehicles
The use of autonomous vessels in air-sea interaction studies allows for measurements very close to the water surface, with minimal disturbance to the surrounding air and water parcels. Other advantages include: the ability to launch and recover the vessels from the shore, cutting down costs and reliance on ship time for study; the lack of emissions and low carbon footprint; and the ability to reach previously inaccessible areas. Examples of autonomous vessel deployments to date include Saildrones as part of the SPURS-2 campaign (Zhang et al., 2019), to demonstrate their feasibility as air-sea interaction observational platforms, a wave glider studying air-sea interaction in Drake passage (Thomson and Girton, 2017) and the OCARINA platform developed by Bourras et al. (2014), deployed off the west coast of France as part of FROMVAR. It is apparent that the use of surface vehicles in flux deter-First measurements of ocean and atmosphere in the Tropical North Atlantic using Caravela, a novel uncrewed surface vessel mination is still in its very early stages. The focus is currently on data acquisition, quality testing and determining the combinations of conditions under which autonomous surface vehicles struggle to operate.

AutoNaut
Following these studies, the University of East Anglia (UEA) worked with AutoNaut Ltd to develop an uncrewed surface vessel, named Caravela. Previous uses of AutoNaut vessels include scientific deployments (Johnston and Pierpoint, 2017), surveillance (Johnston and Poole, 2017) and environmental campaigns through AutoNaut's involvement in The Ocean Cleanup. 1 The UEA's Caravela is a 5m-long surface vessel (shown during deployment in Figure 1 and described in Figure 2), with 0.8m draft and 1.5m high mast. AutoNaut's Wave Foil Technology generates the vessel's forward motion. This uses sprung foils at the front and aft of the vessel, which articulate to draw energy from the vessel's pitch and roll (Johnston and Pierpoint, 2017). The larger the waves, the more energy generated and the faster the forward motion of the vessel. An auxiliary thruster is also fitted on the aft foil to aid propulsion in difficult conditions. Caravela is a robust vessel, designed to withstand rough ocean conditions and in the event of capsize, self-right. Caravela can be operated in three different ways depending on proximity to the pilot. Up to 200m from the pilot, a joystick can be used to drive Caravela and engage the thruster, to allow for controlled movement at launch and retrieval sites. Up to around a kilometre from the pilot, Caravela can operate under 'local controls' in which the pilot sends commands from the piloting interface (called RCW) to Caravela over UHF radio. Finally, when beyond line of sight, the pilot can send commands from RCW over the Iridium satellite network. Within these piloting regimes, Caravela can operate under three modes: station mode, where Caravela circles a location at a specified radius, typically 25m; heading mode, in which a heading is set and kept regardless of course over ground; or track mode in which a series of waypoints are set and Caravela automatically adjusts heading to reach these points. This large range of operational modes opens many possibilities for meteorological and oceanographic measurement with an AutoNaut vessel.
Caravela has a modular monohull, allowing for integration of different sensor types and minimising risk of damage in case of water ingress. Four lithium-ion batteries recharged by three solar panels spanning Caravela's surface are responsible for powering the onboard computer and sensors. The sensor package fitted on Caravela was selected to support the determination of air-sea fluxes and is described in Table 1, with locations on the platform shown in Figure 2.
Caravela's novel aspect is its ability to transport and release a profiling ocean glider, specifically a Seaglider. The Seaglider is a separate underwater autonomous vehicle piloted over Iridium, capable of profiling the ocean to 1000m through changes in its buoyancy. The Seaglider is visible inside Caravela's release mechanism in Figure 2. The benefits of developing a Seaglider transport system are that the Seaglider can be deployed in a remote or challenging area, without the cost of sending a ship or endangering personnel. Additionally, the Seaglider can rest in Caravela's glider release mechanism for a long period of time without significant battery wastage. This opens the opportunity to time the deployment of a Seaglider to study an event, like a monsoon or phytoplankton bloom, again without reliance on ship availability for deployment. The release mechanism does not allow for Seaglider recovery to Caravela, so it is still necessary to consider ship availability to collect the Seaglider. However, this provides many more opportunities for deployment than if a ship was needed for both deployment and recovery.

Deployment of an AutoNautdata quality
The first full scientific deployment of Caravela took place from January to March 2020, as part of the Eurec4a campaign (Bony et al., 2017;Stevens et al., 2020). Eurec4a was developed to investigate the coupling between clouds, circulation and climate but expanded to cover many themes within meteorology and oceanography that feed into climate research. UEA's contribution involved the preparation of Caravela in Barbados, where the vessel was deployed with the aid of the Barbados Coastguard on 22 January 2020.
Caravela travelled from Barbados to the study site and back over 33 days. This included 11 days occupying a 10km wide hourglass-shaped sampling pattern at the study site (upper right of Figure 3). On the outward journey, Caravela covered approximately 150km before the Seaglider was released to travel independently to the study site. Average speed over ground was approximately 0.34ms −1 whilst carrying the Seaglider, compared with 0.49ms −1 across the whole deployment. Unfortunately, we suspect entanglement of the Seaglider in Sargassum slowed Caravela, hence releasing the Seaglider earlier than planned on the outward journey. Fortuitously, Caravela and the Seaglider arrived separately at the study site within a day of one another, giving us an almost co-located dataset between the two platforms during outward transit.
The Airmar 120WX (Airmar) and Rotronic Hygroclip HC2A (Hygroclip) both measured air temperature on Caravela. When analysed, we discovered these instruments showed poor agreement. Both instruments were shaded and well ventilated. The Airmar is situated at the top of the mast (1.5m), approximately 0.5m above the Hygroclip. Comparisons with data from the R/V Meteor (Figure 4) shows consistency with the Hygroclip sensor. The Hygroclip time series stopped on 18 February due to sensor failure but we are satisfied with the quality of data obtained from the Hygroclip before failure. However, the Airmar does not provide the accuracy required to detect small temperature variations important in heat fluxes. It is often used as a sailing or fishing instrument and whilst useful for these applications, is not appropriate for our needs in terms of air temperature measurement. The Airmar instrument is also responsible for apparent wind data. This analysis is in progress so comprehensive assessment of the quality of Airmar wind data will be addressed in subsequent publications.
A time series of Caravela's sea surface temperature (SST) data whilst at the study site was compared with Seaglider SST data ( Figure 5). This uses three different Seagliders deployed in the study site throughout Eurec4a, the one released from below Caravela and two deployed from the R/V Meteor. We would expect to see a diurnal cycle in SST, which is clearly visible between 10-14 February. The measurements from the two platforms are consistent, albeit with substantial spatial and temporal variability evident.
Downwelling longwave (5-30μm) and shortwave (360-1120nm) radiation were measured by Caravela throughout the Eurec4a campaign because accurate measurements of these parameters are vital for heat flux estimation. The total air-sea heat flux is the sum of four fluxes: net longwave and shortwave radiative fluxes; surface latent heat flux; and sensible heat flux. Variability in incoming solar radiation throughout the day impacts surface heat flux and causes the diurnal cycle in SST ( Figure 5). Figure 6 shows the diurnal cycle in shortwave (i.e. solar) radiation. We see significant variation of around 100Wm −2 in the longwave radiation ( Figure 6).

Future work
Caravela offers continuous measurements of surface fluxes and surface conditions that are co-located with the HALO aircraft's flight circle during the Eurec4a campaign, as well as complements the measurements taken from the R/V Meteor on a meridional transect at 57°14.7′W. Caravela's data will be valuable outside of our heat and momentum flux research, providing a stationary time series in Eurec4a where many other platforms had large spatial coverage. Having Caravela provide measurements in the marine boundary layer, co-located with flights by the HALO aircraft, will enhance atmospheric analysis within the wider scope   UEA's future work with Caravela will first build on the observations from Eurec4a, calculating local heat and momentum fluxes between the ocean and atmosphere. Using time series of upper ocean heat content from the profiling gliders, we will estimate an upper ocean heat budget at the study site with the purpose of determining the dominant SST variability driver in the region. This would allow us to differentiate between SST variability based on surface heat fluxes and subsurface processes like mixing, entrainment or advection. We intend to do this work with observations alone, utilising data from Caravela, the Seaglider transported to the study site by Caravela and the two other Seagliders that were deployed from the R/V Meteor. Deriving the ocean mixed layer heat budget based solely on ocean and atmosphere observations is rare. If the analysis is successful, we intend to undertake a similar deployment in Antarctica, where in situ observations are even more scarce. We hope this work will provide a foundation for future air-sea interaction research based only on the use of autonomous observations.