Method & system for measuring, understanding and managing oceans

Abstract

A management system for use in relation to a marine ecosystem includes data collection devices and an agent, such as an autonomous water upwelling or downwelling device, capable of causing implementation of a computer-determined change in the ecosystem, wherein a computer is programmed to operate an intelligent feedback loop and iteratively receive and process gathered data; determining from reinforcement learning whether to implement a change to the ecosystem designed for its improvement, communicate instructions to the agent for implementing the change, observing effects of the change and determining whether a further change would benefit the ecosystem; wherein the change includes causing artificial introduction or removal of water from a depth in the ecosystem by upwelling or downwelling and the rate thereof. Also included is a method of acquiring knowledge of, and managing an ocean ecosystem using the above system.

Description

METHOD & SYSTEM FOR MEASURING, UNDERSTANDING AND MANAGING OCEANS Technical Field

[0001] This disclosure relates to oceanography and benefiting marine ecosystems, through the combined use of artificial upwelling / downwelling and artificial intelligence. Background to the Disclosure

[0002] Ocean upwelling is a natural process that brings deep, cold water to the warmer surface of the ocean. This deep water, brought to the surface, is often rich in nutrients supporting the growth of phytoplankton and seaweeds and forming the basis of the marine food chain. Natural upwelling generates some of the world’s most fertile ecosystems. However, as a result of global warming, the transport of nutrient rich deep waters through natural upwelling is becoming increasingly hindered due to the temperature-induced density gradient.

[0003] Valve-regulated transfer devices that bring about the transfer of water in an open ocean environment from a first to a second depth, including to and from the surface, whether by causing artificial upwelling or downwelling will be referred to herein as vertical transfer devices, or VTDs. In some related literature sources, they may be referred to also as “pumps” because of their functionality. By way of the opening and closing of a check valve in a fluid transfer conduit, such devices cause the transfer of water from a first level to a second level. It will be appreciated that the transfer direction is not necessarily actually vertical or perpendicular to the flat water surface. By “vertical” is meant that the water intake and outlet from a VTD are located on different horizontal depth planes, which, if extended sufficiently in the horizontal direction, will have a distance of vertical separation between them. The action of a VTD is designed to replicate the natural transfer process between depths and, in the case of upwelling, enhance the upward transport of nutrient-rich deep waters by breaking the blockade of the increasing warmer layer of water. As a result, VTDs can be used to fuel primary productivity and augment the natural oceanic biological carbon pump, to sequester carbon dioxide (CO2) in deep ocean ecosystems. The cascading effects of ocean upwelling and downwelling can lead to a great biodiversity as more species find food and habitat. Artificial upwelling / downwelling can be used to restore damaged or depleted ecosystems such as coral reefs and kelp forests. The process of upwelling typically involves marine engineered hardware such as pumps, pipes or other mechanisms to lift the deep-sea water to the surface. Although, in the paragraphs that follow, reference is made mainly to upwelling, it should beappreciated by the skilled reader that the discussion may equally apply to downwelling, as the case requires.

[0004] Data buoys, used for ocean data collection, and VTD autonomous underwater vehicles (VTDVs) are increasingly being deployed for studying different oceanic phenomena such as oil spill mapping (Kinsey et al., 2011), harmful algal blooms (Das et al., 2010), phytoplankton and zooplankton communities (Kalmbach et al., 2017), and coral bleaching (Manderson et al., 2017). These VTDVs can be classified into two categories: 1) propeller- driven vehicles, such as the Dorado class, which can move fast and gather numerous sensor observations but are limited in deployment time to no more than a few hours; and 2) minimally- actuated vehicles such as drifters, profiling floats, and gliders that move slowly, but can remain on deployment for tens of days to multiple weeks. A new generation of long-range VTD autonomous underwater vehicles (LRVTDVs), i.e., Tethys, combines the advantages of both minimally-actuated and propeller-driven VTDVs (Hobson et al., 2012). These LRVTDVs can move quickly for hundreds of kilometres and can be deployed in the water for weeks at a time. However, long range data collection spanning years or even decades is still largely an unsolved problem. Furthermore, as far as we are aware, all the existing ocean measurement devices collect ecosystem data passively at a point in time and space. Although some of the sensors deployed to collect ocean data interact with the environment in order to obtain measurements such as ultrasound range finder, sonar or lidar, these interactions are directed towards isolated observations such as detecting objects. Admittedly, these active sensors interact with the physical environment by emitting energy to detect a target as opposed to passive sensors that measure radiation reflected by the target. However, it appears from the literature that such sensors are deployed only for making observations and measuring isolated oceanic conditions, which are overlayed in post-processing to deduce understanding about the respective ecosystem. Such sensors include weather stations, water quality sensors detecting pH, dissolved oxygen, turbidity, nutrient levels and the like, remote sensing devices such as found on satellites, or used for UAV hyperspectral imaging, biological sensors (camera traps, bioacoustics sensors), microclimate sensors, carbon flux sensors and oceanographic sensors, such as are used for detecting temperature, salinity, and characteristics of ocean currents. In all cases, these individual sensors are not directly measuring the ecosystem, including biological groupings and their interactions, and they do not interact with the physical ecosystem they are sensing (other than emitting a light or sound signal).

[0005] With oceans and seas being under threat from climate change, biodiversity loss, pollution and overfishing, there is a pressing need for researchers around the world tostudy marine life and habitats to help develop new strategies to preserve vital ocean ecosystems. Scientists estimate that 91% of ocean species have yet to be classified, and that more than eighty percent of our ocean is unmapped, unobserved, and unexplored.

[0006] Existing ocean data is scarce and largely fragmented. Connecting and making sense of data is the missing key to understanding our oceans and driving solutions that can halt the accelerating degradation that is being observed.

[0007] A number of technologies that harness the power of artificial intelligence (AI) are being developed to address challenges facing the world’s oceans. These include using AI for marine conservation and monitoring using sensors, drones and cameras to monitor fishing activity, coral reef health and ocean pollution. Furthermore, AI can analyse underwater acoustics to monitor marine life and identify marine species based on images and sound recordings, optimise aquaculture practices, as well as predict hurricanes and tsunamis. AI- powered VTD autonomous vessels and drones can even help clean up plastic waste more efficiently. To provide further insights, data is being aggregated, such as in mapping the ocean floor to expand knowledge of ocean topography. The above-mentioned technologies provide valuable insights, vision and awareness of independent oceanic measurements, at given points in time and space, which can then be correlated into trends and fed into predictive models. However, none of the above-mentioned technologies interact physically with the ecosystem or have a way to directly measure the collective impact on the biological groupings or the interaction between species. Understanding often involves the ability to perceive underlying patterns in data, information, or the world around us. This process is fundamental to human cognition and problem-solving. When patterns are able to be perceived and interpreted, predictions can be made and inferences drawn, in turn permitting the making of informed decisions on ocean phenomena and the beneficial management thereof. In the context of present disclosure, against which the timeframe to mitigate climate change and biodiversity loss is significantly constrained, the world urgently needs methods to fast track understanding and arrive at new solutions to manage and protect our oceans.

[0008] International patent application publication WO 2021 / 168125 discloses an inertial hydrodynamic pump that makes use of wave displacement to generate an upward or downward stream of moving water. The moving streams of water are used for generating hydro-power by means of a turbine, which is driven by the streams exiting a floating reservoir in which a body of water has been trapped as a result of external wave action.

[0009] In a paper by Hwang, et al. “AUV Adaptive Sampling Methods: A Review”, Appl. Sci. 2019, 9, 3145. https: / / doi.org / 10.3390 / app9153145, adaptive sampling using anunderwater robot is discussed. An AUV is employed to detect and monitor an upwelling event and to track the movement of the water front It is not taught that upwelling should artificially be brought about by a robot or other form of underwater vehicle or agent. Nor is any form of artificial intervention in natural upwelling or downwelling proposed, let alone studied for its effects on a predefined ecosystem.

[0010] It will be appreciated that there is a need for an ecosystem management apparatus that uses machine learning in a feedback loop for autonomously enabling intervention in ocean up- or downwelling phenomena for beneficially adjusting water flow patterns for promoting ecosystem health. Summary of Inventive Aspects

[0011] This disclosure presents a method and a system for safely interacting with, understanding and managing ocean ecosystems by combining machine learning and intelligent upwelling / downwelling in a feedback loop, whereby a computer model of a given ecosystem is established for studying the effects of altering the rate of water entering or leaving an ocean ecosystem by artificially induced upwelling / downwelling. By “ocean ecosystem” is meant an ecosystem that physically exists in an ocean environment in real time.

[0012] Incremental changes are made to the water flowrate in or out of the ocean ecosystem, data from the ecosystem is gathered and fed into the computer model on an ongoing basis for effects of the changes to be learned and analysed, and, responsive to the effects observed, adjustments are made to the upwelling / downwelling rate to drive positive or beneficial outcomes for the real ecosystem.

[0013] According to a first aspect of the invention, there is provided a method of acquiring knowledge of an ocean ecosystem, including a system of interactions that take place within it, the method including the steps of ^ Identifying an ocean zone having within it an ocean ecosystem in need of study; ^ Deploying at least one autonomous agent in the zone, the agent having data gathering capability and being associated with a computer processor programmed to receive and process the data; ^ Allowing the agent to interact with the ecosystem and collect experience data, at fixed time steps, whereby data includes at least observations of the current condition and the resulting condition of the ecosystem, Allowing the agent tocause an upwelling or downwelling event within the zone at a controllable known flow rate, ^ Allowing the computer to apply artificial intelligence to the collected data, ^ Based on artificial intelligence output, allowing the computer to generate a proposal for intervention in the ecosystem for altering a condition therein or a product thereof; ^ Allowing the computer to cause the agent to implement the proposal; and ^ Allowing the computer to monitor the ecosystem for effects of such implementation in the ecosystem.

[0014] In an embodiment, the agent comprises an autonomous vertical transfer device (AVTD).

[0015] The AVTD is operable to implement the proposal by changing the flow rate as instructed by the computer.

[0016] In an embodiment, the agent may comprise a plurality of AVTDs in interactive communication with the computer and / or each other.

[0017] The computer may be remotely located from the AVTD or AVTDs. In an embodiment, the computer is at a shore-based computer centre.

[0018] In an embodiment, the computer is located offshore. The computer may be located aboard a sea-going vessel or aboard an AVTD.

[0019] In a further embodiment, the computer includes hardware and software that is distributed between a plurality of AVTDs.

[0020] The step of allowing the computer to implement the proposal may include allowing the computer to address an issue relating at least to the ecosystem, or to further explore the ecosystem by gathering additional data.

[0021] In an embodiment, the step of allowing the computer to apply artificial intelligence includes allowing the artificial intelligence to identify an issue within the ecosystem that warrants addressing or further exploration.

[0022] In an embodiment, the method further includes the step of applying reinforcement learning (RL) to teach the agent to make decisions and take actions in physical control tasks.

[0023] Data may be gathered through sensors associated with the agent. In an embodiment, the sensors are embedded on the AVTD. The sensors may be adapted for gathering data including, but not limited to, temperature, salinity, pH, turbidity, and biogeochemistry.

[0024] In an embodiment, the method includes causing the agent to act within an action space that corresponds to the control inputs that are available for application to the physical system, one such input being the quantity of nutrients delivered through the process of upwelling.

[0025] In an embodiment, the reward function quantifies the success, for example increased yield, or failure – for example a negative impact on the ecosystem - of the agent’s actions. The agent aims to maximise a cumulative reward signal. The signal may, without limitation, be selected from increased yield in seaweed growth and phytoplankton bloom.

[0026] The method thus may be applied to increasing seaweed yield in an ecosystem that includes a seaweed farm. In particular, the method includes applying reinforcement learning to the regulation of controlled upwelling using an AVTD, whereby water above a coral reef is cooled to a temperature at which seaweed growth is optimized.

[0027] In a further embodiment, the processing of the data includes applying deep neural networks (DNN) or other computer-supervised learning for processing oceanographic data and identify patterns, anomalies and trends.

[0028] In an embodiment, the method is applied to cause upwelling to cool a coral reef ecosystem while implementing RL for minimising biological change to said ecosystem. Further, according to this embodiment, the method includes causing upwelled water and water proximate the water surface to mix in proportions suitable for providing a target nutrient content to support both high quality seaweed yield and reef ecosystem health.

[0029] According to a second aspect of the invention, there is provided a management system for use in relation to a marine ecosystem, the system including a computer, a suite of sensors in communication with the computer and gathering data from a marine ecosystem to be managed, and an agent capable of implementing a computer-determined change in the ecosystem;a. Wherein the computer is programmed to operate an intelligent feedback loop including iteratively receiving and processing the gathered data; determining whether to implement a change to the ecosystem for its improvement, communicating instructions to the agent for implementing the change, observing effects of the change on the ecosystem and determining whether a further change would benefit the ecosystem; b. Wherein the change includes causing artificial introduction or removal of water from a depth in the ecosystem by upwelling or downwelling and the rate thereof; and c. Wherein the agent includes an autonomously operable water transfer device capable of causing artificial upwelling or downwelling in the ecosystem.

[0030] The computer may be remotely located, such as on shore, or on one or more vessels.

[0031] The feedback loop may be based on intelligent learning of the ecosystem through exploration and / or exploitation.

[0032] In an embodiment, the change implemented via the agent is aimed at improvement or maintenance of the ecosystem.

[0033] In an embodiment, the change implemented via the agent reduces a severe weather impact in the ecosystem by altering the surface temperature of water therein.

[0034] In another embodiment, the change implemented via the agent causes a change in water condition within the ecosystem that supports or enhances defence capabilities of a country in control of the location of the ecosystem.

[0035] The change in water condition may be implemented via the agent for improving the performance of sonar in vessel detection and monitoring.

[0036] In an embodiment, the change implemented via the agent causes a change in water condition within the ecosystem that supports or enhances the collection of environmental DNA (eDNA).

[0037] The change implemented via the agent may cause the establishment of an area of optimal biodiversity for eDNA sampling.

[0038] In an embodiment, the water transfer device capable of self-navigation by virtue of including a flexible water transfer conduit of adjustable length for reaching an ocean current at a selectable depth.

[0039] In an embodiment, the conduit includes underwater data transfer cabling. In an embodiment in which the conduit comprises a fabric wall, the cabling may be sewn into the wall during manufacture of the conduit.

[0040] The preceding discussion of the background to the present disclosure is intended to facilitate an understanding of the solution herein described. However, it should be appreciated that the discussion is not an acknowledgement or admission that any of the material referred to was part of the common general knowledge in the field, whether in Australia or elsewhere in the world, at the priority date of the present application.

[0041] Further, and unless the context clearly requires otherwise, throughout the description and the claims, the words ‘comprise’, ‘comprising’, and the like are to be construed in the inclusive sense of “including, but not being limited to” - as opposed to an exclusive or exhaustive sense meaning “including this and nothing else”. Brief Description of Drawings

[0042] In order to facilitate understanding of the present invention, reference is made to the accompanying drawings, in which a preferred embodiment is illustrated. Thus: Figure 1 is a concept diagram of the system and method of this disclosure. Figure 2 is a functional component block diagram of a control system in a preferred embodiment of the system of Figure 1. Figure 3 is a schematic illustration of an autonomous artificial upwelling device (AVTD) employed in a currently preferred embodiment of the invention of the kind contemplated in Figure 2. Figure 4 shows a flow controlling valve of the AVTD of Figure 2. Figure 5 illustrates, in perspective view, embodiments of attachment cables and a winder system for the AVTD of Figure 3. In (a) a passive system is shown; in (b) an active system and in (c) the winder board is shown in more detail. Figure 6 schematically shows a manufacturing process for a tube conduit segment of an AVTD of Figure 2.Detailed Disclosure of a Preferred Embodiment

[0043] Artificial upwelling delivers nutrient rich deep-sea water to the surface, which can promote growth of seaweeds, phytoplankton and support marine life. Artificial intelligence is pivotal in analysing and optimizing the vast amount of data relating to the changes caused by such upwelling. The same applies of course in the case of artificial downwelling. Once a critical data volume is achieved, AI has the capacity, by way of the computer or computers running its programs, to generate new knowledge and support human efforts in combating challenges such as those resulting from climate change, whether naturally-occurring or human-induced, and biodiversity loss in the oceans. The critical aspect to the methodology disclosed herein is that of causing controllable artificial upwelling for interacting with a selected eco-system or multiplicity of ecosystems. Through transport of nutrients and biogeochemical species from the deep-sea to the surface, growth of micro and macroalgae are promoted, which in turn attracts other sea life.

[0044] The controlled aspect of upwelling regulates the amount and frequency of upward nutrient transport, thus providing an increased data and action space. A computer executing proprietary AI software is then fed input sensed parameters (in an embodiment, nutrients & chemical species) and measured induced output parameters (in an embodiment phytoplankton & other marine life biomass) in order to deduce patterns of ecosystem behaviour. This can be achieved using machine learning with clearly defined goals and objectives, such as finding optimal combinations of parameters to achieve and maintain a healthy productive ecosystem and / or identify combinations of oceanic conditions that result in ecological imbalances.

[0045] Alternatively, or in conjunction, supervised learning may be utilised to predict outputs based on input parameters, thereby circumventing undesirable outcomes. In this context, AI acts as a risk mitigator and knowledge generator, overcoming some of the previous concerns in scaling artificial upwelling in order to increase ocean productivity, raise fish stocks and enhance oceanic CO2 sequestration.

[0046] At any given time and location, the effects of small or larger adjustments to upwelling or downwelling flow rates are able to be monitored and interactively managed by means of artificial intelligence.

[0047] The embodiments described below focus on the use of an autonomous VTD that is self-navigating, as it seeks out ecosystems in relation to which intervention would be beneficial. It should be appreciated that in other embodiments, the self-navigating capacity ofthe VTD may be omitted or curtailed to facilitate long-term deployment within a predetermined zone, while retaining other functionalities as appropriate for the application under consideration.

[0048] Referring to Figure 1, there is shown a conceptual representation of a system of this disclosure in which external data sources (not shown) and sensors 60, which are installed on or connected with a floating autonomous VTD (or AVTD) 10, collect data relating to the condition of an ecosystem within a zone 12 of an ocean or sea of the world. The zone may extend over a portion of a sea or ocean or encompass an entire ocean or a series of oceans interactively connected. The AVTD is described in greater detail in passages below. It is controlled by a computer that monitors its geolocation and navigates it using devices and methods referred to in relation to Figure 5 and Figure 6. A preferred method of navigation involves exploitation of ocean currents, which may occur at different depths from the surface. Navigation by adjusting the length of the water conduit tube of the AVTD is described in patent application PCT / AU2023 / 050764, the relevant passages of which are incorporated herein by reference.

[0049] Data collected at an initial time S represent the condition of the ecosystem at such time. Using programmed artificial intelligence and machine learning techniques, the computer determines from an analysis of the data whether there is a need to improve the condition of the ecosystem. If a need is identified, the computer then implements an action A to cause a change in the condition to a new state, S1, later in time than S. Gathering further data pertaining to the change in condition to S1, the computer determines if the new condition at S1 represents a reward, R1. A reward represents a measured improvement in the ecosystem condition, compared to preset target parameters. If the S1 condition represents a reward, the intelligent programming of the computer “learns” that the change it implemented resulted in an improvement in the ecosystem. It then computes a further change based on its learning. The computer accumulates an archive of implemented actions and their effects and uses the “experience” so gained in optimising the ecosystem, so that it may meet pre-set performance targets. The computer may update performance targets as its experience and learning increase through the iterative steps described in this paragraph.

[0050] The technical teaming of AI with an intelligent autonomous agent, located in a zone to be investigated or studied, thus provides an unprecedented opportunity to apply techniques such as reinforcement learning (RL), to learn and effect changes in ocean environments in real-time. RL is a powerful approach for teaching autonomous agents to make decisions and take actions in physical control tasks, such as by actuating a smart component. The autonomous agent in this disclosure is the intelligent VTD, where the smartcomponent is an autonomously operable valve. By virtue of its condition, the valve controls the quantity and frequency of water flow through it. Its operation is informed by data collected by means of an embedded sensor suite aboard the VTD, as agent.

[0051] It is essential to represent the state of the ocean environment accurately, and this is achieved through the gathering of multiple streams of data, including sensor data, from which the agent is then able to make informed autonomous decisions. The sensors on the AVTD gather data about the surrounding local environment and characteristics of upwelled water. Additional parameters may be included in the relevant operating algorithm of the system, based on regional oceanographic conditions. By way of example, the characteristics of the upwelled water constitute critical input that influences the biological response to the artificially induced upwelling.

[0052] The ocean is by nature not stationary and the currents constantly may push upwelled water away from the AVTD. The typical current speed in the ocean is about 0.05m / s to 0.50m / s. Current speeds exceeding 1m / s are also quite common, especially in areas near land masses and in eddies and the like. The displacement of water in regions with typical current speeds may be from about 4km / day to 43 km / day. Many of the biological responses, for example phytoplankton bloom, occurs in about three days from the upwelling. Hence, algal bloom could drift considerable distance from the AVTD. In this context, the AVTD also gathers data from external sources to gauge the extent of algal bloom. The external data sources include publicly available high resolution numerical models, satellite observation of chlorophyll-a, the photosynthetic pigment indicating phytoplankton concentration, biogeochemical Argo profiles, moorings etc. The agent uses in-situ data and external data to predict positions of upwelled water at any given instant and algal blooms anticipated at each location. Further, it examines the satellite datasets and publicly available in-situ biogeochemical datasets to confirm the presence of algal blooms originated from artificial upwelling.

[0053] Observations may be recorded at equally spaced intervals, where each observation is influenced by the prior actions and states. For example, a first observation in a time series may be an initial data set measured by the VTD sensor system prior to the causing of an artificial upwelling or downwelling event. A subsequent or second data set is gathered one time interval later, possibly coinciding with post upwelling / downwelling event, so that a comparison of the data sets enables the effect of the event to be measured. The measurement may therefore enable real time reaction or interaction from the agent, such as by altering the flow of water through the device or changing the depth of intake or outlet.

[0054] The data may include, without limitation, temperature, salinity, pH, biogeochemistry, nutrient stoichiometry and other measurements in the water and the air above. For time series data, the state typically consists of historical observations and actions. The action space in time series represents the available choices the agent can make at each time step. In one embodiment of this disclosure the action space corresponds to the control inputs that can be applied to the physical system, such as amount of nutrients delivered through the virtue of upwelling (or downwelling, as the case may require). A software-based reward function quantifies the success or failure of the agent’s actions. In the ocean environment context, a success would be increased yield, whereas a failure would be negative impact to an ecosystem.

[0055] Therefore, the agent aims to maximise a cumulative reward signal, such as increased yield in seaweed growth or phytoplankton bloom. The agent then needs to balance exploration with exploitation. Exploration may involve trying new actions to learn more about the environment, while exploitation may involve implementing actions postulated or believed to be high in yield. The agent is programmed with policies that dictate which actions to take in different states. These are learned and updated through interactions with the environment. As a safety mechanism, Model-Based RL can be utilised to build an internal model of an environment for simulating an action plan. For example, simulators may be used in physical control tasks to allow extensive training and testing in a safe controlled environment, before deploying the learned policies to live VTDs.

[0056] The advantages of RL in understanding and managing the oceans include without limitation: ^ Its adaptability in situations where the parameters of what is considered optimal may evolve over time, as agents adapt to changing environments and continually update a strategy being implemented, to take into account new experiences; ^ Its suitability for complex, unstructured environments where traditional machine learning may struggle, such as in playing complex video games, where the space is vast and continuous; and ^ Its ability to make optimal decisions and find and implement the best strategy or policy in situations where human intuition or known algorithms may not provide an obvious solution.

[0057] In an embodiment, RL is combined with another AI strategy, such as Deep Neural Networks (DNN). DNNs are able to process vast amounts of oceanographic data and identify patterns, anomalies and trends, resulting in new insights into ocean behaviour. Theyhave been found to excel at predicting ocean-related phenomena such as forecasting surface temperature, storms and waves. DNNs as applied in this disclosure automatically extract features such as key variables that affect natural upwelling, and they are used for analysing satellite imagery and remote sensing data and detecting phytoplankton blooms. DNNs also assist in identifying and tracking marine species and thus measure additionality, and assess health of coral reefs, kelp forests and fish populations.

[0058] Referring to Figure 2 and Figure 3, there is shown a preferred embodiment of a system for performing the method of this disclosure in relation to an environment defined by a zone 12 of depth D. The environment may include an ecosystem to be improved by controlled modification. The modification may be accomplished in discrete steps that are determined using feedback loop techniques. Control centre 100 in Figure 2 is a remote computing service that includes a computer that operates independently of an agent, being in this instance an AVTD 10, operable to transfer water from a first depth to a second depth, whether by up- or down-welling. The control centre is preferably shore-based for power supply considerations, but may be located on a floating platform such as a ship or other vessel, or even on an AVTD, depending on the adequacy of power supply. The control centre establishes a human-machine interface that enables an operator to manage operation of AVTD 10 from afar, whether singly or in collaboration with other VTDs forming a cluster and operating within an ocean zone 12, depicted in Figure 3. The operations that may be managed include computer resources analysing telemetric messages received from sensors in the environment, evaluating optimal control actions and balancing energy requirements for the on-board electronic subsystems with energy resources. Control centre 100 may send and receive messages to and from the device, including commands that set the operating priority or configuration of the device and its subsystems 110. Subsystems 110 include but are not limited to communications, data integrity & pre-processing, executive control, power, flow management, sensor, environmental monitoring, safety and recovery subsystems.

[0059] A communications subsystem 114 aboard the AVTD is an electronics and computational subsystem that transmits data from the AVTD to a remote computing facility, in particular (although not always necessarily) to control centre 100. Subsystem 114 in this embodiment: a. Sends messages to remote command centre 100 when the network service permits communication; b. Receives messages from remote command centre 100 and routes them to the appropriate subsystem aboard AVTD 10; c. Cryptographically signs messages before transmission to prevent tampering;d. Cryptographically verifies messages received to ensure that they originated from control centre 100; e. Communicates the state of the communication subsystem to executive control system 112 on board the AVTD; and f. Sends messages with processed or raw data to control centre 100.

[0060] A data integrity and pre-processing subsystem 106, associated with the control centre, is programmed to sign messages cryptographically before transmission to and from an AVTD, to prevent tampering, and to verify incoming messages to ensure they originated from the AVTD and have not been modified. Once data integrity has been assured, subsystem 106 preprocesses the data for analysis phase. This may involve cleaning, normalising, and transforming the data to make it suitable for unsupervised machine learning algorithms. Furthermore, this subsystem stores, processes, and interprets data about the physical state of the device and its surrounding environment; and ensures the security and integrity of data within the device. The data integrity and pre-processing subsystem 106 therefore may perform some or all of the following, applying AI with RL or DNN techniques, whichever may be most advantageous in particular circumstances: a. It accepts commands that determine operating priorities or operating parameters. For example, it can turn data logging for specific sensors associated with a specific AVTD on or off; b. It accepts commands that set operating parameters. For example, it can increase or decrease the frequency at which data is collected from different sensors; c. It accepts commands that determine how data is pre-processed on the device before transmission. It can change the content or level of detail of information returned in data messages; d. It communicates the state of the data subsystem to the remote command centre; e. It collects data from each sensor at an interval specified in configuration and writes that data to local storage; and f. It processes the data as it may be directed by configuration and operating priorities to create new data.

[0061] In one embodiment, the executive control subsystem 112 is an electronic and computational subsystem that is configured and programmed to coordinate the operation of other subsystems within the device to achieve operating goals. The executive control subsystem may carry out one or more of the following tasks in collaboration with the subsystems already discussed: a. It accepts commands from control centre 100 that set operating priorities. It then sends commands to the appropriate control subsystems 110 to enact those priorities. These goals may evolve and be modified and adapted as the number of operating devices increases and data accumulation renders new knowledge; b. Using communications subsystem 114, it communicates the operating state of each autonomous device 10 to control centre 100; and c. It autonomously assumes control of operating priority for AVTD 10 when safety and recovery conditions require immediate response. It returns priority when safety and recovery conditions no longer require immediate action.

[0062] AVTD 10 includes a power subsystem 116 that collects, stores, and manages the use of energy within the device to achieve operating goals. Power subsystem 116 is programmed to perform the following functions, without necessarily being limited to these: a. It collects solar and wind energy and stores it in a local energy storage system within subsystems cluster 110. In another embodiment, alternate sources of energy may be utilised and converted to power the system; b. It monitors and forecasts the availability of energy in the local storage system to create an energy budget; c. It analyses total and distributed energy consumption for the electronics system of an individual AVTD, processing collected historical data from which it assists in predicting future energy consumption; d. It accepts commands that determine operating priorities; e. It implements operating priorities by turning specific electrical subsystems on and off to ensure that those priorities can be achieved within budgeted energy parameters. For example, when the AVTD is in recovery mode, flow sensorsare not required and can be disabled to conserve power. However, in recovery mode, light beacons and other signalling systems are required to execute and assist the recovery procedure and these are required to remain operational; f. It sends messages and data to communicate its own condition and that of its associated energy store to executive control system 112 and remote-control centre 100.

[0063] In one embodiment, AVTD 10 may respond to signals emanating from power subsystem 116 by reserving and optimising power for a requested operation, transmitting its location at a high rate of frequency and activating a flashing beacon. To conserve energy, some of the sensors may be temporarily switched off or may have their frequency of sensing reduced.

[0064] A further subsystem included on board AVTD 10 is a flow management subsystem 120. This is an electronic and computational subsystem that monitors and controls the rate of water flow through the upwelling tube 20 of AVTD 10 via commands that may retain the programmable flow controlling valve 34, mounted within the tube or at the tube inlet or outlet piece 32, in permanently open, shut or partially open state to control the upwelling rate. Flow management subsystem 120 thus: a. Accepts commands that determine operating priorities and parameters for AVTD 10; b. Activates the valve control system for valve flaps 70, shown in Figure 4, to increase, decrease, or prevent water flow through the upwelling tube 20 of AVTD 10; and c. Monitors the flow state through signals put out by data subsystem 118 and adjusts valve controls to achieve the operating goals as these evolve. This can occur on a scheduled basis at the require frequency or autonomously adjust to the operating requirements in response to sensed parameters.

[0065] The sensor subsystem 124 is an electronic and computational subsystem that collects data about the physical state of the AVTD and its surrounding environment. With reference to Figure 5, sensor subsystem 124 includes sensors 60 associated with the tube segments 36 via joiners 38, discussed below. A plurality of sensors can be mounted across the AVTD, providing telemetry across the water column passing via tube 20. Sensors within sensor subsystem 124 include - but are not necessarily limited to - temperature (for examplea Measurement Specialties TSYS01 sensor), pressure (for example a Keller Series 4LD pressure sensor), angle of inclination, acceleration, and geo-location positioning. These sensors are well understood in the art and are readily available commercially. Other sensors measure biogeochemical properties such as dissolved oxygen, nitrate, phosphorus, iron, chlorophyll and other chemical species.

[0066] An environmental and carbon capture monitoring system 104 is shown in the preferred embodiment as being located at control centre 100, remote from the environment 12 being monitored and managed. Monitoring system 104 is pre-loaded with at least initial marine-relevant algorithms, GPS enabled tracking, and to-date sensor readings. These marine algorithms evolve and improve in accuracy through supervised learning and AVTD interaction with the environment. They include two distinct classes of algorithms: optimal growing conditions for individual species and potential environmental threats to the same.

[0067] Algorithms for optimal growing conditions may pertain to marine plants that sequester carbon such as microalgae called phytoplankton and various types of seaweeds or macroalgae. For example, such algorithms may be employed for evaluating whether the immediate environment in the zone surrounding the deployed AVTD is favourable for marine life growth, based on the sensed conditions and previous state of knowledge. This may be calculated via simple known ratios such as phytoplankton C:N:P uptake ratio (146:19:1), or carbon to phosphorus ratios C:P 117:1 that can be trapped in seaweed and consequently buried into deeper layers of the ocean environment. These and other known nutrient ratios provide valuable information on whether the immediate environment surrounding a deployed AVTD 10 may be optimal for inducing phytoplankton blooms or growing kelp and other marine aquaculture. Through supervised learning, these models can evolve to capture other hidden factors and refine existing ones.

[0068] The threat conditions may be described in relation to specific thresholds, parameter combinations and ratios, based on latest peer-reviewed research and current learning models. The algorithms generate alerts concerning current or predicted water quality, nutrient concentrations, presence of certain microorganisms or other oceanic conditions, which may negatively impact on the surrounding marine life or ecosystems. In the event of predicted or observed negative ecological impact, safety control mechanisms would be initiated to temporarily suspend the AVTD pumping operation. This could be achieved by fixing valve 34 in open or closed position for a duration of time until the threat has dissipated, and baseline conditions have been restored.

[0069] One example of an ecological threat that may be addressed using the system and methods of this disclosure is that of harmful algal blooms (HABs), some of which can produce toxins harmful to marine life including fish and shellfish and posing serious health risks to humans if these are consumed. Known conditions that may cause algal blooms include (1) warmer temperatures, (2) excessive nutrient inputs, particularly nitrogen and phosphorus from agricultural runoff, seaweed discharge and industrial pollution, (3) stagnant or slow moving water leading to stratification of the water column, with wind surface layers being warmer and nutrient rich, (4) factors that reduce the predation of algal grazers such as zooplankton and shellfish, (5) algal species composition, (6) turbidity and light conditions which suppress growth of some algae and allow harmful algae to adapt to low-light conditions, (7) weather and climate patterns such as El Nino, La Nina, and others. In one embodiment, an algorithm for detecting a harmful algal bloom involves steps selected from the following: (a) Data collection – A subsystem of the AVTD sensors subsystem 124 gathers real- time temperature, chlorophyll and water quality data, combined with concurrent satellite imagery and other remotely sensed data, and data concerning the baseline presence of harmful algae prior to operative deployment of the AVTD; (b) Data pre-processing – involving data cleaning, noise removal and interpolation of missing values; (c) Feature extraction – involving identifying relatively warmer temperatures, nutrient levels, chlorophyll concentrations and spectral characteristics that are present in that or may impact environment 12; (d) Algorithm selection – involving training supervised and unsupervised machine learning models to classify water conditions and analysing spectral signatures to enable detection of changes in water colour; (e) Training – involving training machine learning models on historical data with labelled and non-labelled harmful algae bloom samples; (f) Monitoring – involving using the AVTD for continually monitoring environmental conditions and transmitting live data that will update and improve the accuracy of the detection algorithm; and (g) Environmental alert – involving running the algorithm to analyse the features extracted from the real-time data to detect harmful algal blooms and generate alertsthat can trigger environmental awareness and actions such as to cause the AVTD to cease operation.

[0070] Other invasive or disruptive species that can be harmful to ecosystems include coral predators such as crown-of-thorns starfish that feed on coral polyps leading to coral degradation, coral disease such as stony coral tissue loss, lionfish in Atlantic and Caribbean oceans with voracious appetite for small reef fish, invasive green crabs, zebra and quagga mussels, coral predators, eutrophication or excess of nutrient runoff from urban areas resulting in algal blooms and oxygen depletion, marine debris and pollution, and the like. Each threat would give rise to a new algorithm or class or algorithms for addressing it.

[0071] One or more AVTDs combined with AI provide a powerful tool for measuring carbon capture in the oceans. The AVTD sensor subsystem 124, when teamed up with satellite data and imaging, can monitor ocean surface properties including sea surface temperature, chlorophyll-a concentration and ocean colour. These properties indicate presence of phytoplankton in the ocean. Machine learning algorithms may then be trained to detect and quantify phytoplankton concentrations, based on patterns and spectral characteristics of sensed data, and even to recognise the distinct optical properties that define them. Furthermore, AI can estimate phytoplankton biomass by correlating sensed data with in-situ measurements and consequently make predictions about productivity and sequestration capacity. Temporal AI may then be harnessed to analyse the phytoplankton bloom seasonality. Finally, AI models can be used to model the conversion of CO2 into organic matter via phytoplankton photosynthesis, assess the environmental impact of ecosystem services provided by phytoplankton to fisheries, as well as carbon sequestration and oxygen production.

[0072] The safety subsystem 126 is an electronic and computational subsystem that monitors the state of the AVTD 10 to determine whether the device is in or near approaching a critical state, which may partially or fatally compromise the operations and functionality of AVTD 10. Based on the determination made, safety subsystem 126 takes countermeasures to prevent the partial or fatal loss of the operating capacity of AVTD 10. If a critical state likely to lead to an emergency situation is being predicted, safety subsystem 126 may respond by applying controls on the opening of valve 34 or the length of tube 20, thereby changing the direction of travel of AVTD 10, or its operational capacity to mitigate the conditions at hand. If deemed appropriate by the processor or by human intervention, safety subsystem 126 may open all valves 70 fully to reduce load. For example subsystem 126 may:a. Identify risks to the operation of the AVTD, such as from a change in structural loading, which may suggest possible upwelling tube rupture, loss of a tube segment, a valve failure, or from a change in measured upwelling rate that may suggest a breach of the tube, or a valve failure and the like; or b. Detect collisions involving the AVTD by employing an accelerometer to determine whether its float or support superstructure may have collided with another floating object in the water or one that is airborne.

[0073] The recovery subsystem 128 is an electronic and computational subsystem that facilitates the location and extraction of the AVTD from the ocean. It accepts commands from executive control processor 112 that determines operating priorities and parameters. It may cause signals of the presence or proximity to another object of the AVTD to be emitted, for example using a sequence of short sound blasts.

[0074] The collation of field and other broader environmental and climatic data, the generation of statistics about conditions and performance, and the computation of key indicators define a computer-managed system that is agile to expand and to update a suite of algorithms for analysing telemetry data.

[0075] The AVTDs, when free-drifting, navigate the oceans, flowing generally with the currents they encounter. However, should control be required, for example to disable the activity of an AVTD due to an excess of nutrients or an unfavourable ratio of nutrients in the garden zone of deployment, or due to the device inadvertently floating towards a marine park or other restricted area, the controls may be activated from control centre 100 to send instructions that override prevailing commands being transmitted from the onboard computer 112. For example, control actions issued by control centre 100 may include instructions for: a. Engaging energy-saving algorithms or implementing manual override to prioritise and control sensor operations. This may include temporarily switching off a sensor or reducing the frequency of its measurements or communications; b. Having the AVTD automatically apply controls if unfavourable environmental conditions occur, such as toxic algal blooms. Controls may include temporarily disabling all or a subset of VTDs in a concert of such devices, to restore nutrient balance prior to resuming upwelling; c. Controlling buoyancy of the water column in tube 20 of an AVTD 10 by injecting salt or air bubbles at selected depth locations;d. Applying control on to a targeted subset of AVTDs within a cluster or concert of such devices, based on GPS locations and individual sensor readings. This may be done to achieve uniform distribution of temperature and nutrients in a marine farming area, or to deliberately create different distribution areas for comparative testing; e. Controlling water upwelling or downwelling speed by disabling some of the valve flaps 70, utilising the venetian blind-type valve design. For example, to halve the speed of upwelling, valve shutters covering about half the cross-sectional flow area in the valve housing would be closed. Alternatively, the opening angle of all valve flaps may be adjusted simultaneously to achieve a correspondingly equivalent effect. Flow control may be implemented by installing a flow management valve such as a magnetically actuated venetian blind-type valve as illustrated in Figure 4 at the top exit ring 62 in Figure 5. There is no difference between top- or bottom-of-tube control as far as function is concerned, but benefits of the latter will be shorter control cable 66 (see Figure 5), and equipment that does not have to be deep depth rated; and f. Setting the flow valve 34 to remain permanently open for AVTD retrieval procedures (for example, in case of maintenance being required) in order to empty the water column in tube 20 of tons of captured water. Similar control measures may be applied in cases where external conditions are such that they pose safety concerns for the operations and life-cycle of an AVTD 10.

[0076] An AVTD, when employed as agent in accordance with this disclosure, includes the following connected elements, which are identified in Figure 3. The numerical label 10 is used to generally denote a pair of AVTDs drifting within a selected study zone 12 of water having a surface 14 and depth D, and being part of a greater body of open ocean water 16. ^ A float 18, which is an apparatus that maintains the buoyancy of AVTD 10 under varying weather, current and wave conditions; ^ An upwelling tube 20, being a conduit for containing and transporting water from a target depth 22 to the ocean surface 14 of zone 12. The tube is connected to float 18. The tube would in most embodiments be multi-segmental, depending on the depth to which it must extend from the float in or below study zone 12. The segments are connected by joiner pieces discussed below; ^ An inlet 24 associated with an inlet end piece 32 at the lower end of the tube; and^ A one-way valve 34, mounted within inlet end piece 32, that operates to prevent reverse flow of water through the upwelling tube, by opening when the AVTD descends on a wave and closing when the AVTD ascends on a wave.

[0077] The valve and inlet structure are described in co-pending application PCT / AU2023 / 050764, incorporated herein by reference.

[0078] In this disclosure, AVTDs may be either anchored, or free drifting, as already mentioned. The anchored devices are fixed in a location and may be utilised in applications such as bringing cold water to cool a reef environment or to supply nutrients for marine permaculture farm. Free-drifting devices traverse large zones of deep ocean, held in by the ocean currents in which they are deployed, in a similar manner to the way the gravity field of a planetary body draws in a satellite. The free-drifting AVTD is constructed to pump water from target depths within the mesopelagic layer of a particular ocean region, where the nutrients required to facilitate phytoplankton growth are found. The upwelling tube and its support structure including the float are fabricated to reach a total target depth from which water is to be forced upwards in an artificial upwelling current. The efficiency of the device is measured in terms of the upwelling rate achieved, relative to the cost of production. For examples of artificial upwelling devices and their applications, reference may be made to patent application publications US2005155922, US2021301800, US2010300560, WO2021 / 168125A1 and unpublished application PCT / AU2023 / 050764.

[0079] The AVTD contemplated in the preferred embodiment of this disclosure comprises fully-integrated electronics with a suite of sensors 60 (shown in Figure 5), with software and processing capacities that connect and exchange data with a remotely located control centre, such as control centre 100 of Figure 2, over the internet or other digital communications networks. This aspect to referred to herein as the artificial upwelling internet of things, or “AVTD IoT”.

[0080] In one embodiment, the AVTD includes a valve 34 illustrated in Figure 3,with two or more flaps 70 and associated underwater electromagnets 78. This embodiment has two operative conditions: OPEN and CLOSED, achieved through the activation and deactivation of said magnets. An alternative embodiment may deliver a plurality of intermediate valve conditions, delivering precision flow control. Utilising the VTD IoT and the remote operable valves, the AVTD becomes enabled to perform and act autonomously. In the autonomous mode of operation, valve 34 is opened or closed by a signal initiated from control centre 100 or in response to the data gathered by the sensors 60 attached or connected to the VTD in the sensed environment.

[0081] The binary-state valve operation is depicted in Figure 3, with valve 34 in CLOSED condition being shown in Figure 4(a), and the valve mounted within a joiner ring 38 being shown in Figure 4(b). Joiner ring 38 is used for connecting consecutive segments of an upwelling tube 20 in a multisegmented AVTD. An OPEN state, as suggested by broken lines in Figure 4(b) depicting the path of travel of the flap lower extremities, is achieved when magnets 78 mounted on the flaps adjacent to hinge 74 are energized to be activated. A CLOSED state is achieved when magnets 80 are activated, as shown in Figure 4(a).

[0082] In another embodiment, a submersible linear actuator is used to vary the available stroke for the valve, such that the device flow can be more precisely controlled than with electromagnets and simple two-position, fully open or fully closed flaps. Examples of suitable such actuators are the 2000 Series Subsea Linear Actuator from 2G Engineering of Sun Prairie, Wisconsin USA, and the NEMA Submersible Linear Actuator from Ultra Motion of Cutchogue, NY, USA). In this embodiment, the control centre or on-board computer may command the position of the linear actuator (or rod length), thus limiting the stroke of the valve.

[0083] The check-type valve 34 discussed in the example above operates automatically with wave action. It is also within the scope of this invention to provide a remotely controllable valve, actuated by a control system (still to be described below) for the apparatus, according to whether the apparatus is required to function as an upwelling or downwelling pump or vertical transfer device. An example of such a valve is a solenoid valve, energised by electrical energy stored in a battery and replenished by solar, wind or wave power, or a combination thereof, collected by the devices installed on the superstructure.

[0084] The up- or down-welling tube 20 in the preferred embodiment of this disclosure is of adjustable length. With reference to Figure 5, adjustment in one embodiment is enabled by virtue of a segmented, modular construction, wherein the tube comprises segments 36 sequentially interconnected by connectors 38, not all of which are shown, for ease of illustration. The segmented tubing design in preference to a single uninterrupted length of tubing material facilitates the system to respond in appropriate manner to ocean surface, epipelagic, and mesopelagic depth layers. For example, the upper segments of tubing may be located in the top 50m of the ocean, where they may encounter the most significant wear and abuse, such as from wave action. Below them, the mid-segments sit at between the 50m and 200m depths, and the lower segments are below the 200m level, each experiencing different kinds of exposure to currents and pressures. The segmented tubing modules permit construction of source material units to suit the conditions as required, rather than utilising more expensive materials in areas where these may not be required. As a result, thesegmentation makes the construction more economical and handling easier. The modular construction serves also to facilitate the replacement and repair of segments in the field.

[0085] The tubes are made of a resiliently flexible and highly ductile but substantially water-impervious membrane that withstands prolonged, continuous exposure to seawater. The membrane is preferably of low thermal transmissibility for thermally insulating the fluid contents of the tube from the external environment and minimising heat loss or gain. Examples of suitable materials of manufacture include thermoplastic polyurethanes (TPUs) and plasticized canvas fabric. Alternative examples will be apparent to the person of skill in the art. Fabric tubes are readily collapsible in an axial direction and are preferred over tubes of rigid materials such as metals, plastics and composites. Rigid tubes tend to be made in shorter lengths than flexible, collapsible versions. In certain embodiments, rigid tubes may be interspersed or alternated with flexible, collapsible tubes.

[0086] The joiner may be made of any suitable material for ocean conditions. Examples include (without limitation) stainless steels of suitable grade (for example SS316 or SS304), and aluminium. The weight of the joiner needs to be effective to maintain the attached tube segment above it at its selected depth.

[0087] In the embodiment of Figure 5, the segmented tube is suspended from the float superstructure by means of a set of high-strength, low-stretch cables. The cables carry the weight of the upwelling tube segments, joints, valves and inlet end piece 32. They transmit the structural load of pumped water, and associated twisting and bending forces to the float superstructure. The cables connect to the underside of the float superstructure. Metaphorically, the cables serve as a skeleton of the AVTD and may be mounted on the interior or exterior of tubing material. In the embodiment illustrated, the exterior skeleton version has an advantage over the interior location, being more accessible for maintenance and its causing no additional internal upwelling friction within the tube segments. Conversely, interior structural cabling can minimise twisting and tangling. The sensors are placed on the joining rings across the water column, where, in the simplest embodiment, with a single segment, data is captured at the inlet and outlet of the device. Data is transported via specialty underwater cabling. In one embodiment, data cables are incorporated during the tube manufacturing process into the tubing material by sewing, providing protection to the cabling, as well as preventing twisting, tangling and resultant physical damage.

[0088] Referring again to Figure 5, segmentation may be used for reducing structural load by splitting the total volume of water within the water column inside the tube 20 into smaller volumes. This is achieved by providing a check valve 34 that is associated with eachsegment, or with a selected number of segments. The valve is installed within the joiner 38 between segments so that each segment establishes a module having the tube segment, a joiner and a check valve. The valve is installed in the joiner and, when closed, isolates the segment above it from the segment below it.

[0089] A joiner 38 between adjacent segments, or as weighting piece located at the lower end in tube structures having no terminal bell intake, is illustrated. The joiner is an open- ended structural ring having a central conduit 62 to provide passage for water through it and structural attachment points in the form of eyes 64 for the tube segments to be attached to cables 66, as shown in Figure 5. The end portion of the fabric segment 36 is drawn over the end of the joiner in the manner of a sock. The tube material has our looped straps 68 fastened to lugs (not shown) on the tube surface. The loops from the straps extend toward the anchor points 64 on joiner ring 38 and are secured to the eyes 64 by shackles of conventional design (not shown) to prevent the tubes from sliding off the joiner. Of course, alternative fasteners to shackles may be used, subject to suitability for underwater conditions.

[0090] The segmented tube is suspended from the float superstructure by means of a set of high-strength, low-stretch cables 66, as illustrated in Figure 5. The cables carry the weight of the upwelling tube segments, joints, valves and inlet end piece 32. They transmit the structural load of pumped water, and associated twisting and bending forces to the float superstructure. The cables connect to the underside of the float superstructure. Metaphorically, the cables serve as a skeleton of the AVTD and may be mounted on the interior or exterior of tubing material. In the embodiment illustrated, the exterior skeleton version has an advantage over the interior location, being more accessible for maintenance and its causing no additional internal upwelling friction within the tube segments. Conversely, interior structural cabling delivers the advantage of being able to minimise twisting and tangling. Sensors 60 are placed on the joining rings across the water column, where, in the simplest embodiment, with a single segment, data is captured at the inlet and outlet of the device. Data is transported via specialty underwater cabling. In one embodiment, data cables are incorporated during the tube manufacturing process into the tubing material by sewing, providing protection to the cabling, as well as preventing twisting, tangling and any physical damage.

[0091] Referring to Figure 6, a tube segment 36 is shown laid out in (a) as a flat sheet in folded, inoperative condition and in (b) in operative condition, in which it defines a flexible, round cylinder. In manufacturing, a sheet of the flexible fabric from which the segment is to be formed is cut to size and rectangular shape. Opposite longitudinal edges 43 and 45 are folded back toward the longitudinal midline 47 so that a short overlap results. Data cables arelocated between the overlapping edge portions 43, 45. The overlap is sufficiently wide for double sewing of the tubing material to create an elongate custom data pocket extending the vertical length of the folded sheet and enclosing the data cables. An enlarged image is shown of the protruding end in the callout box of Figure 6. On completion of the manufacturing process, only the cable ends 49 protrude from the opposite longitudinal ends of the tubing material sheet and are then connected during final assembly and attached to the electronics subsystem.

[0092] The measurements captured by the sensors suits of an individual AVTD and / or collaborating clusters of such AVTDs, when communicated wirelessly using a known protocol to a remote-control centre, are processed, analysed and optimised in the control centre 100 (Figure 2). In one embodiment, the control centre is configured to: a) Transmit once-off manual commands, such as to switch a particular identified sensor on or off; b) Instigate automatic regulation on an individual AVTD, such as energy conservation and battery monitoring; and c) Transmit complex commands as a result of utilising algorithms embedded within an AI system, ranging from simple heuristics to complex optimization routines.

[0093] As discussed in relation to Figure 2, contained within the on-board computer system of the AVTD are subsystems 110 configured for regulating functions including those of executive control 112, power 116, data 118, flow management 120, navigation 122, sensor management 124, safety 126, recovery 128 and communications 114.

[0094] Unlike previous experiments where introduction of non-native species to new environments have had unintended consequences, such as the cane toad and rabbits in Australia, the brown tree snake in Guam or the Zebra Mussel in North America, VTDs, and AVTDs in particular, provide a safe and controllable means for interacting with the ecosystem. They upwell native species to the surface layer of the ocean and can control the quantities in which this is done. AI can optimise the levels of upward transfer of nutrients and minimise any potential harm. The devices can be disabled in a worst-case scenario, without consequences.

[0095] By combining an AVTD agent with AI-driven data collection, a deeper understanding of ocean ecosystems may be achieved, and strategies to address ocean acidification, overfishing and climate change can be explored. Through controlled upwelling and an ability to vary flow rate, the AI algorithms can predict optimal levels and ratios ofnutrients to stimulate marine plant growth, and hidden / unknown factors can be examined and optimised. Utilising deep neural networks and AVTD flow control, it is possible now to predict outcomes for set actions. The RL models are able to self-evolve, continually optimising and consequently improving solutions. Reinforcement learning can support autonomous control with action being related to the amount of upwelled water, and environment feedback sensed through the AVTD. Resulting action may be recommended, based on the predictions generated.

[0096] An example of the application of the present disclosures is in assisting resistance to climate change, whether or not human-induced, and involves optimising the sun- blocking effect of fog. Artificial upwelling in large water bodies such as in ocean environments contributes to local cooling by lowering surface water temperatures and, in favourable conditions, allowing the formation of fog that blocks sunlight and reflects incident heat. This combination of direct cooling and sunlight reflection helps manage the impact of marine heat waves, offering localized climate control that supports marine ecosystems sensitive to temperature spikes.

[0097] An algorithm is generated that manages a team or cluster of AVTDs working collaboratively, and combines the operation of the AVTDs with reinforcement learning (RL) in a feedback loop to optimize fog or sun-blocking effects. The outcome is an altered environment that can help cool local areas during marine heat waves. The goal of this algorithm is to control AVTD parameters to achieve and then to enhance cooling through fog formation or increased reflectivity from sun-blocking conditions.

[0098] An overview of the algorithm is set out below, using headings to distinguish functions and steps that may be carried out. Not all the steps listed are necessarily essential, their inclusion being determined by required outcomes and changing conditions as detected by local and remove sensors and incoming environmentally specific and relevant data.

[0099] A first function is to define initial conditions and known environmental variables. These may include without limitation surface water temperature, humidity levels, air temperature, dew point, wind speed, and solar radiation intensity. The target outcomes in this step include achieving a cooling effect through water surface temperature reduction, fog formation from increasing local humidity and adjusting temperature to the dew point. Another target outcome may be sun blocking, achieved by maximizing the albedo effect through promoting fog formation, which creates a reflective layer of the water that will reduce solar heat absorption. A second step is to set a reward function by defining a reward structure and positive and negative rewards. The reward structure encourages actions that achieve thetarget outcomes, in this example cooling and sun-blocking effects. A positive reward would recognize actions that lower surface temperature, form fog or create a reflective sun-blocking layer, while a negative reward would pertain to action that increase temperature, reduce cooling effects or lead to unstable environmental conditions in the ecosystem under consideration.

[0100] A third step is to define actions to be taken by each AVTD being utilised and controlled or operated by the control centre. These action may include: a. Adjusting the upwelling depth, being the depth from which colder water is upwelled; b. Controlling water flowrate by regulating the volume of cold water pumped to the surface. This can be achieved by valve positioning adjustment. c. Coordinating upwelling timing by making use of pulsed or continuous upwelling, depending on the target fog or cooling effect.

[0101] This third step may include group coordination when multiple AVTDs are being deployed. It may, for example, involve allowing each device to work in tandem, with individual actions contributing to a collective cooling and sun-blocking outcome.

[0102] A fourth step involves reinforcement learning in conjunction with a state and a feedback loop. In terms of representing the state of the ecosystem under consideration, each AVTD in use monitors local environmental conditions, forming a state from real-time data on temperature, humidity, solar radiation, and upwelling efficiency. The RL agent then will select upwelling parameters for each device based on the environmental state, optimizing for cooling and fog generation. A feedback loop is established according to which the agent receives continuous data, updating the state with the coming of each loop cycle, based on variables such as temperature, humidity, and fog conditions.

[0103] In a fifth step, the operations that the algorithm may execute may include the following: a. Observe the environmental state: Collect initial data from designated or selected sensors (temperature, humidity, solar radiation) to define the state.b. Select team actions: The RL agent selects and designates actions for each AVTD to optimize cooling and fog formation, based on current incoming environmental data being received. c. Apply AVTD actions: Each AVTD executes designated upwelling actions, including adjusting flow rate, flow pulse timing, or intake depth, based on the RL agent’s policy. d. Evaluate action results and receive a determined reward: Monitor the effects on surface temperature, humidity, and fog density. The agent receives a reward based on how well cooling and sun-blocking goals are achieved. e. Update operating policy and coordinate the next actions: The RL agent refines its policy to improve group actions and coordination, learning to balance upwelling intensity, timing, and positioning for optimal cooling and sun-blocking.

[0104] A sixth step involves continuous learning and optimization. This may include group coordination or implementing seasonal adjustments to upcoming designated actions. In regard to group coordination, the RL agent learns to coordinate actions between AVTDs for maximum impact, balancing cooling needs and fog density. Seasonal adjustments may include making adjustments to parameters on a seasonal basis, or introducing longer-term environmental changes to maintain effectiveness over time.

[0105] Alternatively, to optimize the number of agents required in a team or cluster of AVTDs, an adaptive RL approach may be applied. This approach adjusts the number of active agents based on real-time environmental feedback, so as to balance resource usage, energy efficiency, and effectiveness in achieving the desired cooling and nutrient distribution. This approach enables a flexible, data-driven strategy to determine the optimal number of AVTDs that would be required in a team, minimizing resources while achieving environmental goals effectively.

[0106] The concepts of this disclosure may also be implemented, without limitation, in models for: 1. Early detection and mitigation of weather events 2. Optimizing seaweed production3. Defence enhancement 4. Environmental DNA monitoring Example 1: A model for early detection of weather events

[0107] In early weather detection, artificial upwelling (AU) is combined with reinforcement learning (RL). This approach leverages the capability of AU to affect local atmospheric conditions and then applies RL to interpret environmental data patterns resulting. This approach aims to create an adaptive system that not only helps stabilize conditions in specific areas but also provides insights into developing weather changes by identifying early indicators of severe weather patterns.

[0108] According to the above approach, a system framework is created, making use of the mechanism of artificial upwelling, whereby AVTDs bring cooler, nutrient-rich water to the sea surface, affecting local temperature, humidity, and, potentially, cloud formation. These environmental changes influence atmospheric conditions, providing a valuable data source for early weather monitoring.

[0109] Reinforcement learning for data interpretation using an RL model processes real-time data collected by sensors (for example temperature, humidity, wind speed and atmospheric pressure) positioned on AVTDs and in surrounding areas. The model is trained to detect early signs of developing weather changes by continuously analyzing patterns in this data.

[0110] Core components of the RL and operations management system include sensors, a data feedback loop and a data aggregation and pattern recognition model.

[0111] The AVTDs in the system are equipped with sensors to monitor water temperature, humidity, pressure, wind speed, and salinity. These devices gather data that could indicate precursors to severe weather. By way of non-limiting examples, these may include rapid temperature drops and changes in wind patterns.

[0112] The data feedback loop provides for environmental data to be fed into an RL model in real time, creating a feedback loop where data is continuously analysed, and the AVTDs’ operational parameters are adjusted based on detected trends.

[0113] The data aggregation and pattern recognition module allows for the RL system to learn from historical and real-time data and detect patterns linked to specific weather events,such as tropical storms, heavy rainfall, or tornadoes, enabling proactive adjustments to the AVTDs.

[0114] In implementing the early weather detection system, the algorithm workflow executes the following procedures: a. Initialization of environmental and baseline conditions i. Baseline conditions are determined by causing the computer module to collect initial data on water temperature, humidity, air pressure, wind speed, and any other relevant factors. ii. Historical patterns are determined from gathering date from previous severe weather events, thereby to give the RL model context for detecting anomalies. b. Collection of real-time data and state observation i. Environmental conditions are continuously monitored with the AVTD sensors, thereby tracking changes in local parameters. ii. The RL agent’s understanding of the current state is updated in real time, using data feedback to detect any deviations from baseline conditions. c. RL decision-making and adaptive control i. State interpretation is performed, by way of the RL model analysing sensor data to assess if any atmospheric conditions are deviating from what is considered normal for the ecosystem or environment that is under observation. ii. Actions are selected for implementation based on detected trends, with the RL agent adjusting AU parameters, such as upwelling depth, timing, or flow rate, to influence conditions if needed. Adjustment may be considered appropriate for reasons such as to cool surface temperatures or create cloud cover. iii. Predictive adjustment is performed in situations such as where the RL model detects early signs of severe weather. These signs may include phenomena such as rapid temperature drops and atmosphericpressure changes. These may trigger more intensive AU which would then further cool surface temperatures or increase humidity, potentially mitigating the conditions conducive to storms. d. RL and continuous feedback i. A reward system assigns positive rewards for accurate early detections of developing weather events, or for promoting conditions that maintain stability. ii. Policy updates are the result of the RL model refining its understanding of which actions, namely AU adjustments, contribute to early detection and potential stabilization of conditions. These allow the model to learn how best to predict and respond to early weather indicators. e. Reporting and prediction i. The RL system sends alerts or reports when early indicators of severe weather are detected, potentially aiding local weather stations or meteorological agencies in issuing early warnings. ii. Over time, as the data store increases, the RL model learns to distinguish between minor atmospheric variations and true weather precursors, improving detection accuracy.

[0115] Benefits of artificial upwelling and RL for early weather detection include: a. Localized early weather event detection: The system can monitor environmental changes in real time, providing highly localized data that is valuable for early detection of developing weather patterns. b. Preventative actions: When detecting early signs of severe weather, the system has the capacity to intervene. In an embodiment, the intervention may involve causing upwelling of cooler water in sufficient and effective quantity and flowrate to cause cooling in the surface temperatures of both water and the air layer above it, thereby to reduce the severity of atmospheric instability. c. Continuous learning: The RL model continuously learns from new data, becoming more accurate over time in identifying patterns and potential early indicators of severe weather.An Example 1 scenario for detecting and mitigating potential severe weather events early:

[0116] In an exemplary scenario, an AVTD detects a sudden drop in water temperature and corresponding shifts in humidity and air pressure, these being conditions that are known to often precede storm cell development. The RL model recognizes these changes as similar to patterns observed before previous storm events and adjusts the AU settings to increase cooling, aiming to stabilize the local atmosphere. Simultaneously, the system alerts weather monitoring agencies of potential severe weather in the area.

[0117] In summary regarding this exemplary weather adjustment model, the integration of artificial upwelling and reinforcement learning for early weather detection provides a proactive approach to climate management through the local adjustment of conditions in a defined ecosystem. Through adaptive data collection, continuous learning, and localized adjustments, the system and methods of this disclosure enable detection of severe weather indicators early and enable the support for preventative measures, creating a valuable tool for both monitoring and potentially mitigating local climate risks.

[0118] Artificial upwelling integrated with reinforcement learning and applied to weather detection introduces both localized environmental manipulation and adaptive learning to the field. It adds value to current weather systems by enhancing prediction accuracy in marine and coastal areas, actively managing local atmospheric conditions, and serving as a learning system that improves over time. This approach could provide a significant advantage for early detection, especially in hard-to-predict regions, and offer a unique tool for both monitoring and mitigating extreme weather risks.

[0119] The application of the principles set out in this disclosure in supporting the management of marine ecosystems may be summarized further in terms of the following functionalities: a. Localized Monitoring: The AVTDs act as “micro-weather stations,” capturing detailed surface data that complements broad, macro-level models. b. Proactive Intervention: Unlike passive systems, AU can actively stabilize local conditions by cooling surfaces, especially at the ocean / air interface, potentially reducing severe weather risks. c. Adaptive Learning: RL dynamically refines predictions based on real-time data, enabling precise early detection with each event.d. Improved Coastal Accuracy: AU fills data gaps in marine / coastal areas, enhancing early warnings for high-risk zones. e. Climate Risk Tool: AU and RL together offer proactive climate management, potentially reducing severe weather impacts in vulnerable areas. Example 2: A model for managing seaweed production

[0120] Using artificial upwelling (AU) with reinforcement learning (RL), conditions in a seaweed farm can be adjusted towards an optimized state and to maintain an ideal balance between upwelled and surface layer water, ensuring nutrient-rich waters remain at the surface long enough for absorption by seaweed. The focus here is on mixing upwelled water with surface layer water effectively to stabilize salinity and prevent nutrient-rich water from sinking too quickly before achieving a desired beneficial effect on seaweed health.

[0121] Below is an example of an enhanced algorithm for AU in combination with RL in a seaweed farm. It has the following functionalities. a. Defined environmental variables and target conditions o Key variables include, without limitation, nutrient levels (e.g., nitrogen, phosphorus), water temperature, salinity, tidal movements, and current speed. o The target outcome is enhanced and optimized nutrient concentration at the surface by achieving balanced salinity levels, suitable for maximizing nutrient retention and seaweed growth. b. Multi-agent system and feedback loop with integrated RL o Distributed AVTDs: Multiple AVTDs are deployed across the seaweed farm, each equipped with sensors to monitor nutrient levels, temperature, salinity, and tidal movement in real time. o RL state representation: Each AVTD contributes data from which the computer executes instructions so as to form a comprehensive view of the farm’s conditions, including the mixing state of upwelled and surface water. o Adaptive coordination: The RL model coordinates the operations of the individual AVTDs to adjust their individual respective upwelling actions,and achieving an optimal mix that keeps nutrients at the surface without allowing salinity to disrupt seaweed growth. c. Actions for the individual AVTD o To control upwelling depth and flow rate: Each device is programmed or instructed to adjust its upwelling water intake depth and flowrate to bring nutrient-rich water to the surface at just the right amount as determined from the control centre. o To optimize mixing intensity: AVTDs are controlled to blend upwelled water with surface water, achieving a balanced salinity that prevents nutrients from sinking and keeps them accessible for seaweed. o Timing with tides: By timing upwelling around tidal flows determined by using public or privately generated data, the RL model helps to maximize nutrient retention and surface mixing without nutrient drift. d. The reward function o A positive reward is outputted when the system achieves defined ideal nutrient concentration and salinity balance, and maintains these conditions at the surface for optimal seaweed absorption. o A negative reward is outputted when the system has executed actions that lead to nutrient sinking or imbalanced salinity, being actions that are recognized as having a potential negative impact on seaweed growth. o An energy efficiency reward is outputted for rewarding system actions that achieve defined desired surface conditions with minimal energy consumption across the set of AVTDs in operation. e. Learning and workflow optimization function o Step 1: Monitor Initial Conditions – AVTDs collect real-time data on nutrient levels, temperature, and salinity, while accounting for tidal and current conditions. o Step 2: RL Decision on Upwelling and Mixing – The RL agent decides the depth, flow rate, and mixing intensity needed to bring nutrients tothe surface while balancing salinity, coordinating across multiple devices. o Step 3: Apply Optimized Upwelling and Mixing – Each AVTD executes the chosen actions, maintaining nutrient availability at the surface by mixing upwelled water appropriately with surface water. o Step 4: Feedback and Reward – The RL agent evaluates the effectiveness based on nutrient retention and salinity balance, adjusting its approach based on results. o Step 5: Continuous Adjustment – The RL agent refines actions over cycles, learning to optimize nutrient retention through a stable mix of upwelled and surface waters.

[0122] Benefits obtainable from implementing the system on a seaweed farm include: a. Maximized nutrient retention: Through effective mixing by introducing upwell, AVTDs keep nutrient-rich water accessible at the surface, where it can be efficiently absorbed by seaweed. b. Stable growth conditions: By using upwell for balancing and then maintaining balanced salinity, the system prevents nutrient-rich water from sinking, ensuring consistent conditions conducive to seaweed growth. c. Energy efficiency: RL-driven adjustments cause energy use to be reduced and minimised, while maximizing nutrient retention and mixing efficiency. d. Enhanced adaptability: The RL system continuously adapts its outputs in response to changes in tides and currents, ensuring optimal nutrient availability under varying conditions in the seaweed cultivation ecosystem.

[0123] This integrated AU and RL approach provides an adaptive, energy-efficient solution to sustain surface-level nutrients in seaweed farms that rely on a flourishing seaweed ecosystem. The optimized mixing strategy ensures that upwelled nutrients remain accessible, supporting a thriving seaweed farm ideal for biofuel, bioplastics, and other needs that support a sustainable industry. Example 3: Defence support

[0124] Artificial upwelling offer several strategic advantages for defence applications, especially in supporting underwater operations, enhancing surveillance capabilities, and sustaining marine ecosystems near critical infrastructure. The advantages are achieved as set out below: a. Enhanced surveillance and detection i. Improved performance of sonar: By using artificial upwelling to lower water temperatures and increase nutrient levels, the density and salinity of the water in the ecosystem are influenced. This may be used for creating conditions conducive to improved sonar performance, aiding in detecting and monitoring surface vessels and underwater vessels, and suspected hostile submarines in particular. ii. Reduced biofouling: By cooling local waters, artificial upwelling can limit the growth of biofouling organisms on defence infrastructure within a defined target ecosystem. The infrastructure may include, without limitation, sonar arrays, underwater drones, and other equipment. With biofouling reduced, or controlled, equipment performance is enhanced and maintenance requirements are reduced. b. Strategic support for marine operations i. Oxygenation for underwater vehicles: Artificial upwelling in a defined ecosystem in a targeted zone is used for increasing oxygen levels, thereby creating more favourable conditions for unmanned underwater vehicles (UUVs) and other equipment that rely on oxygen-rich waters for efficient operation. ii. Extended habitats for marine surveillance drones: Sustained artificial upwelling can support marine life habitats, attracting fish and other sea life to an ecosystem. The ecosystem is then monitored for changes. RL learns, identifies and then detects change patterns that might indicate the presence of potentially hostile vessels or other threats. c. Environmental security and sustainability i. Ecosystem stabilization near defence bases: Implementing artificial upwelling managed by RL in marine ecosystems located proximate to coastal defence installations supports and helps protect marinebiodiversity, which in turn contributes to overall environmental stability near defence sites. ii. Support for defence-adjacent communities: By promoting sustainable fisheries or reducing the impact of harmful algal blooms by mitigating their effects with cooler, nutrient-rich waters, artificial upwelling can support coastal communities that may be found to have strategic importance in a conflict. d. Climate adaptation and strategic resilience i. Mitigation of thermal pollution: For military installations with high-energy output, such as naval stations or power-intensive operations, implementing artificial upwelling managed by RL helps mitigate thermal pollution, maintaining local marine health and reducing environmental impact. ii. Coastal defence against climate impacts: Causing upwelling to help cool and protect coastal ecosystems will assist in protecting natural barriers against erosion and storm surges, providing a buffer to protect critical defence infrastructure from climate impacts.

[0125] Artificial upwelling, therefore, not only enhances operational readiness and environmental security but also contributes to a defence force’s overall resilience in the face of changing environmental and strategic conditions.

[0126] An outline of an algorithm that uses artificial upwelling combined with data feedback and reinforcement learning (RL) to optimize sonar operations is presented below, with its objective being to control upwelling in real-time to create favourable conditions for operation of detection systems such as sonar, using a reinforcement learning agent that receives feedback from sonar performance data. a. Initialize the ecosystem condition: This step involves defining the environment variables within the ecosystem. The variables include, without limitation, water temperature, salinity, pressure, sound speed profile, and ambient noise level. Initialize the artificial upwelling system, sonar device, and RL agent. b. Set the reward function: This step involves defining a reward function that encourages the RL agent to achieve optimal sonar conditions. This may involve:i. Maximizing sonar range: A higher reward is allocated for achieving a greater range of effective sonar performance. ii. Improving signal-to-noise ratio (SNR): A higher reward is allocated for achieving increased SNR within the target ecosystem. iii. Stability: A reward is allocated for achieving stable sound speed and reducing variability within the target ecosystem. c. Define actions for the upwelling system: This step involves specifying actions that the RL agent can take to control the artificial upwelling system. Without limitation, these may include: i. Adjusting the depth of intake and the flowrate of water being upwelled. ii. Controlling temperature and salinity gradients within the ecosystem by varying the input flowrate or selecting different depths for water intake. iii. Modulating upwelling intensity based on sonar performance feedback. d. Initiate and continue data collection and the ML feedback loop: i. Sonar data: Continuously collect sonar performance data, including range, SNR, and acoustic propagation characteristics. ii. Environmental feedback: Gather real-time data on water temperature, salinity, and sound speed to monitor the impact of upwelling in the ecosystem under consideration. iii. State representation: Use sonar and environmental data to define the current state of the ecosystem, which information is fed to the RL agent. e. The reinforcement learning process: i. Ecosystem state observation: The RL agent observes the current state of the ecosystem, differentiating it from the broader environment. ii. Action selection: Guided by the RL observations of emergent patterns, the agent selects an action to take which will adjust the upwelling parameters based on the observed state. Actions are chosen using a policy optimized to increase sonar performance.iii. Implementation of an action: The upwelling system undergoes adjustment as a result of implementation of the selected action, such as increasing flowrate or changing intake depth. iv. Receive the reward: After the action is applied, the agent receives a reward based on the measured impact on sonar performance, for example improved SNR or extended effective range. v. Update policy: The RL agent updates its policy using the reward feedback, learning to select actions that optimize sonar conditions over time. f. Learning Iterations: i. This involves repeating the above steps, with the agent exploring different upwelling configurations to find optimal conditions for sonar operations. ii. Over multiple episodes, the agent learns a policy that maximizes sonar performance under varying environmental conditions. g. Deployment and real-time adjustment: i. Once trained, the next step it so deploy the RL model in the field. Once deployed, the model then continuously adjusts upwelling in response to changing conditions, using learned policies to maintain optimal sonar performance. ii. Adaptation: The agent continues learning from new data in real-time, adapting to seasonal changes, marine life, and unexpected environmental conditions.

[0127] The defence model workflow may be summarized as follows: a. Initialize: Set the ecosystem environment and reward function. b. Observe: Gather data on sonar performance within the ecosystem and environment. c. Act: The RL agent adjusts upwelling system.d. Reward: The agent receives feedback on action effectiveness. e. Update: Policy is refined to improve sonar conditions. f. Deploy: Use learned policies in real-time, with continued adaptation.

[0128] This reinforcement learning-based approach allows the upwelling system to create optimal sonar conditions dynamically, taking into account environmental variability and real-time data to adaptively support sonar operations. Example 4: Recovery of eDNA

[0129] Artificial upwelling, combined with data feedback loops and reinforcement learning, can significantly enhance the detection and monitoring of environmental DNA (eDNA) in marine environments. By creating optimal conditions for collecting eDNA and using reinforcement learning to adaptively control upwelling parameters, it is possible to maximize the effectiveness and reliability of eDNA sampling. This may be achieved in the following way: a. Enhancing eDNA concentration and distribution i. The purpose of upwelling for eDNA collection: Artificial upwelling brings nutrient-rich water from deeper layers to the surface zone, stimulating the local marine ecosystem. This leads to higher biological activity, potentially increasing the eDNA concentration from organisms attracted to the nutrient boost. ii. Controlling targeted sampling areas: By controlling where and how upwelling occurs, eDNA collection can be focused on areas with optimal biological diversity, improving the chance of detecting target species and obtaining representative samples. b. Adaptive Control for Optimal eDNA Sampling Conditions i. Real-time environmental feedback is obtained via the sensors monitoring environmental parameters critical to eDNA stability, such as temperature, pH, and salinity. These factors affect eDNA degradation rates and persistence in the water column. ii. RL optimization: The RL agent receives feedback on eDNA concentrations and degradation rates. The agent adjusts upwelling parameters and learns the effects of the adjustments so as to enablethe creation of conditions that minimize eDNA degradation in the ecosystem and maximize detection reliability. iii. Examples of actions that the RL agent may take include varying upwelling intensity, adjusting water intake depth, or modulating timing of pulsed upwelling, based on processing feedback from environmental sensors and eDNA monitoring tools. c. Improving sampling frequency and quality i. Data feedback loops for sampling timing help identify peak times when eDNA concentration in the ecosystem is highest. The RL agent learns the patterns of concentration change and adjust upwelling actions to enhance sampling schedules. ii. Dynamic response to environmental changes: If conditions for eDNA stability suddenly worsen, for example if temperature suddenly spikes, the system is able to adjust upwelling parameters in real time to preserve eDNA quality or switch to harvesting from areas with better conditions, ensuring consistently high-quality samples. d. Enhancing Species Detection and Diversity Analysis i. Species-specific optimization is achievable by correlating eDNA data with specific species patterns, allowing the RL model then to optimize upwelling to attract certain types of marine life to the sampling zone, increasing the likelihood of detecting the eDNA of specific species that have been pre-selected for collection. ii. Biodiversity hotspot mapping: Artificial upwelling can create localized biodiversity hotspots, enhancing the diversity of eDNA available for collection. Over time, this data allows for adaptive biodiversity monitoring and the identification of ecologically significant areas. e. 5. Data-Driven Learning for Long-Term eDNA Trends i. Pattern recognition and seasonal adaptation: The system learns seasonal or diel cycles that impact eDNA concentration and quality. RL helps identify these patterns, enabling adaptive management of upwelling parameters for consistent, long-term monitoring.ii. Feedback-driven evolution of sampling protocols: As new data is collected, the RL agent can refine sampling protocols to improve eDNA yield and reliability, continually adapting to environmental changes and evolving eDNA dynamics.

[0130] An example of algorithm workflow for eDNA optimization with artificial upwelling may include the following steps: a. Initialize the environment and reward structure by defining environmental variables (non-limiting examples being water temperature and pH) and setting reward parameters based on eDNA concentration, stability, and species diversity in samples. b. Continuous monitoring and state representation is carried out by collecting real- time data on environmental conditions and eDNA metrics. This information is fed into the RL model to define the current state of the target ecosystem. c. Action selection requires that the RL agent selects upwelling parameters (e.g., intensity, depth, timing) based on current conditions, aiming to maximize eDNA concentration and sample diversity. d. Applying the action and monitoring resulting changes: The selected upwelling settings are implemented and the changes in eDNA concentration and quality are measured over stipulated time periods. The agent is allocated a reward based on the effectiveness of its actions in relation to the relevant target values. e. The policy is updated using feedback from the measured eDNA values and other environmental data, refining the agent’s ability to create ideal sampling conditions over time.

[0131] Some benefits of using artificial upwelling with data feedback and RL for optimising eDNA collection include: a. Increased eDNA collection efficiency: By dynamically adapting upwelling parameters, the system optimizes conditions for capturing high-quality, representative eDNA samples. b. Improved species detection and biodiversity monitoring: Enhanced conditions attract a wide range of species, improving the breadth and accuracy of biodiversity assessments.c. Consistent sampling quality: Adaptive control helps maintain favourable eDNA stability, even in fluctuating environmental conditions, allowing for more reliable long-term monitoring.

[0132] In this way, artificial upwelling controlled by reinforcement learning can create an adaptive, data-driven system to optimize eDNA sampling for ecological monitoring, biodiversity research, and environmental assessments

[0133] Upon reading this disclosure, those of skill in the art will appreciate still additional alternative structural and functional designs for a system and method for an automated and autonomous intelligent feedback system for beneficially, favourably and advantageously conditioning a zone containing a marine ecosystem through the disclosed principles herein. Thus, while particular embodiments and applications have been illustrated and described, it is to be understood that the disclosed embodiments are not limited to the precise construction and components disclosed herein. Various modifications, changes and variations, which will be apparent to those skilled in the art, may be made in the arrangement, operation and details of the method and apparatus disclosed herein without departing from the spirit and scope defined in the appended claims. Parts List Below is a list of various components numbered in the accompanying drawings: 10 – Autonomous vertical transfer device (AVTD) for causing upwelling or downwelling. 12 – Zone for study 14 – Water surface in study zone 16 – Water body in which the study zone is located 18 – A float included in the superstructure of the AVTD 20 – Water transfer conduit, or upwelling / downwelling tube 22 – A target depth for water intake of delivery 24 – Water inlet to conduit 20 32 – Conduit tube inlet or outlet piece 34 – Flow controlling valve for flow through the conduit 20 36 – Conduit tube segments 38 – Segment connectors 43, 45 – Longitudinal edges of conduit fabric sheet 47 – Centre line of fabric sheet49 – Cable end 60 – Sensor 62 -Tube exit ring 64 – Cable anchor point on segment joiner ring 38 66 – Control cable for drawing up or letting out conduit tubing for adjusting its length. 68 – Looped straps 70 – Valve flaps, hinged 72 – Motor driven cable take-up spool 74 – Hinge to valve flap 76 – Valve mounting bracket in conduit or conduit end piece 78 – Electromagnet 80 – Magnet 82 – On-board computer processor 100 – Control centre

Claims

CLAIMS 1. A method of acquiring knowledge of, and managing an ocean ecosystem, the method including the steps of: ^ Identifying an ocean zone having within it an ocean ecosystem in need of study; ^ Deploying at least one autonomous agent in the zone, the agent having data gathering capability and being associated with a computer processor programmed to receive and process the data; ^ Allowing the agent to interact with the ecosystem and collect experience data, at fixed time steps, whereby data includes at least observations of the current condition and the resulting condition of the ecosystem, ^ Allowing the agent to cause an upwelling or downwelling event within the zone at a controllable known flow rate, ^ Allowing the computer to apply artificial intelligence to the collected data, ^ Based on artificial intelligence output, allowing the computer to generate a proposal for intervention in the ecosystem for altering a condition therein or a product thereof; ^ Allowing the computer to cause the agent to implement the proposal; and ^ Allowing the computer to monitor the ecosystem for effects of such implementation in the ecosystem.

2. The method of claim 1, wherein the agent comprises an autonomous vertical transfer device (AVTD).

3. The method of claim 2, wherein the AVTD is operable to implement the proposal by changing the flow rate as instructed by the computer.

4. The method of claim 3, wherein the agent may comprise a plurality of AVTDs in interactive communication with the computer and / or each other.

5. A method according to any one of the preceding claims, wherein the step of allowing the computer to cause the agent to implement the proposal includes allowing thecomputer to identity and address an issue relating at least to the ecosystem, or to further explore the ecosystem by gathering additional data.

6. A method according to any one of claims 1 to 5, wherein the step of allowing the computer to apply artificial intelligence includes allowing the artificial intelligence to identify an issue within the ecosystem that warrants addressing or further exploration or exploitation.

7. A method according to any one of the preceding claims, including the step of applying reinforcement learning (RL) to teach the agent to make decisions and take actions in physical control tasks.

8. A method according to any one of the preceding claims, wherein processing of the data includes applying deep neural networks (DNN) for identifying patterns, anomalies and trends in said data.

9. A method according to any one of claims 6 to 8 applied to increasing seaweed yield in an ecosystem that includes a seaweed farm.

10. A method according to any one of claims 6 to 9 applied to cause upwelling to cool a coral reef ecosystem while implementing RL for minimising biological change to said ecosystem.

11. A management system for use in relation to a marine ecosystem, the system including a computer, a suite of sensors in communication with the computer and gathering data from a marine ecosystem to be managed, and an agent capable of implementing a computer-determined change in the ecosystem, a. Wherein the computer is programmed to operate an intelligent feedback loop including iteratively receiving and processing the gathered data; determining whether to implement a change to the ecosystem for its improvement, communicating instructions to the agent for implementing the change, observing effects of the change on the ecosystem and determining whether a further change would benefit the ecosystem; b. Wherein the change includes causing artificial introduction or removal of water from a depth in the ecosystem by upwelling or downwelling and the rate thereof; and c. Wherein the agent includes an autonomously operable water transfer device capable of causing artificial upwelling or downwelling in the ecosystem.

12. The system of claim 11, wherein the feedback loop is based on intelligent learning of the ecosystem through exploration and / or exploitation.

13. The system of claim 11 or claim 12, wherein the change implemented via the agent is aimed at improvement or maintenance of the ecosystem.

14. The system of any one of claims 11 to 13, wherein the change implemented via the agent reduces a severe weather impact in the ecosystem by altering the surface temperature of water therein.

15. The system of any one of claims 11 to 13, wherein the change implemented via the agent causes a change in water condition within the ecosystem that supports or enhances defence capabilities of a country in control of the location of the ecosystem.

16. The system of claim 15, wherein the change in water implemented via the agent improves the performance of sonar.

17. The system of any one of claims 11 to 13, wherein the change implemented via the agent causes a change in water condition within the ecosystem that causes the mixing of upwelling water with surface layer water in proportions that optimise nutrient content in the mixed waters that enhances seaweed yield, where the ecosystem includes a seaweed farm.

18. The system of any one of claims 11 to 13, wherein the change implemented via the agent causes a change in water condition within the ecosystem that supports or enhances the collection of environmental DNA (eDNA).

19. The system of claim 18, wherein the change implemented via the agent causes the establishment of an area of optimal biodiversity for eDNA sampling.

20. A system according to any one of claims 11 to 19, wherein the sensors are embedded on the water transfer device.

21. A system according to any one of claims 11 to 20, wherein the water transfer device is capable of self-navigation by virtue of including a flexible water transfer conduit of adjustable length for reaching an ocean current at a selectable depth.

22. The system of claim 21, wherein the conduit includes underwater data transfer cabling.

23. The system of claim 22, wherein the data transfer cabling is incorporated into a wall of the conduit during manufacturing thereof.

24. The system of claim 23, wherein the wall comprises a fabric into which the cabling is sewn.

25. Use of the system of any one of claims 11 to 14 for enhancing defence capabilities.

26. Use of the system of any one of claims 11 to 14 for improving seaweed yield form a seaweed farming ecosystem.

Images