Water treatment environmental analysis via monitoring reporting and verification sensor-geospatial fusion networks

Abstract

Monitoring, reporting, and validating (MRV) with sensor-remote sensing fusion for water treatment environmental analysis is provided by receiving data from each of a plurality of separate water sources, calculating changepoints within the data by calculating first and second standardized variables for first and second segments of the data, determining that a difference between the first and second standardized variables exceeds a threshold, and identifying a changepoint between the first and second segments, splitting the data into intervals at the changepoints, classifying the intervals using a machine learning model, and identifying predicted water metrics for a water source based on the classified intervals.

Description

TITLEWATER TREATMENT ENVIRONMENTAL ANALYSIS VIA MONITORING REPORTING AND VERIFICATION SENSOR-GEOSPATIAL FUSION NETWORKSCROSS-REFERENCES TO RELATED APPLICATIONS

[0001] The present disclosure claims priority to U.S. Provisional Patent Application 63 / 549,166 titled “DRINKING WATER TREATMENT ENVIRONMENTAL ANALYSIS VIA MONITORING REPORTING AND VERIFICATION SENSOR-GEOSPATIAL FUSION NETWORKS”, which was filed on 2024-02-02, and which is incorporated herein in its entirety.BACKGROUND

[0002] In recent years, telemetry-connected electronic sensors have been developed and applied within water service programs to perform objective and continuous temporal and spatial- level monitoring for various usages and functionalities. These sensors can be used for the monitoring, reporting, and verification (MRV) of various environmental management targets, such as the generation of carbon credits, management of land and water resources, and control of water treatment processes. A digital MRV system may facilitate project design, automated monitoring, control and data assimilation, robust verification, and data visualization; however, current systems rely on infrequent collection of data, limited spatial interpolation, and are generally poor at incorporating external data collected at inconsistent rates or from environments with sparse sensor deployment.SUMMARY

[0003] The present disclosure provides systems, methods, and apparatuses for employing sensor-geospatial fusion networks to analyze and predict environmental conditions in a drinking water treatment setting. Embodiments of the present disclosure can advantageously leverage machine learning analysis of data from various sources to perform a variety of tasks, such as but not limited to managing water treatment system behavior, attributing changes of environmental conditions to a cause, and issuing or verifying carbon credits associated with drinking water treatment.

[0004] Additional features and advantages of the disclosed method and apparatus are described in, and will be apparent from, the following Detailed Description and the Figures. The features and advantages described herein are not all-inclusive and, in particular, many additional features and advantages will be apparent to one of ordinary skill in the art in view of the figures and description. Moreover, it should be noted that the language used in the specification has been principally selected for readability and instructional purposes, and not to limit the scope of the inventive subject matter.BRIEF DESCRIPTION OF THE DRAWINGS

[0005] Figure 1 illustrates an example environment, such as a watershed, in which embodiments of the present disclosure may be practiced.

[0006] Figure 2 illustrates a functional model for tracking water quality and attributing changes in water quality to various actors, according to embodiments of the present disclosure.

[0007] Figure 3 is an example system, as may use the described models for monitoring, reporting, and validation with sensor-remote sensing fusion, according to embodiments of the present disclosure.

[0008] Figure 4A is a flowchart of an example method for analyzing sensor data, according to embodiments of the present disclosure.

[0009] Figure 4B is an example system for analyzing data and controlling a water system, according to embodiments of the present disclosure.

[0010] Figure 4C is a plot of hypothetical sensor data, according to embodiments of the present disclosure.

[0011] Figure 5 illustrates a computing device, according to embodiments of the present disclosure.DETAILED DESCRIPTION

[0012] The present disclosure provides for monitoring, reporting, and validating (MRV) with sensor-remote sensing fusion for drinking water treatment environmental analysis, in which data from a variety of sources can be combined by a machine learning-based data model to automatically estimate environmental conditions related to the drinking water treatment processand take action based upon that estimate. Embodiments of the present disclosure can process large quantities of data in near-realtime, and thus can take action in response to a change in environmental conditions much faster than a human operator could hope to achieve. Further, embodiments of the present disclosure obviate the need for lab testing and manual analysis which is common with current methods. This has the potential to drastically reduce errors from lab contamination and human error.

[0013] The climate impacts of conventional water and wastewater treatment (which, in general, are proportional to influent water quality, in-stream water quality, water demand, and carbon intensity of the local electric grid) are generally approached as manmade problems requiring manmade solutions, with nature based solutions poised as a lesser (if even considered) alternative as the effect of the manmade solutions are more easily monitored, despite having potential lower impact and requiring more costly inputs than managing the natural environment. By improving influent and in-body water quality, there can be both nearly-immediate and longterm (over 30 years) avoidances of emissions by reducing the need for further upgrades to gray infrastructure. By using the improved MRV techniques describe herein, operators are given tools to find and quantify the effects of green alternatives to manmade solutions, which may be less expensive, less energy intensive, and less carbon intensive, among various other incentives (such as carbon credits).

[0014] As discussed herein, water quality may refer to various parameters that are judged individually, in aggregate, or in direct correlation with one another. The present disclosure contemplates that various localities and professional organizations shall be understood to define various goals or levels of water quality that one of ordinary skill in the art is expected to be familiar with. The individual parameters may include chemical measurements, such as the presence, absence, or concentration of various chemicals or classes of chemicals in a given amount of water, which may directly indicate the presence of a chemical in question (e.g., higher concentration of Calcium ions to indicate a higher concentration of Calcium in the water) or provide indirect evidence of a chemical in question (e.g., (e.g., higher concentration of Calcium ions to indicate a higher concentration acids in the water that dissolve rocks carrying Calcium). The individual parameters may include physical measurements, such as the amount of water flowing through a given area, a speed of flow of the water, a number of influx and efflux directions of the water, a temperature, a turbulence, or the like. The individual parameters may include various microbial orbiological measurements, such as the presence, absences, or concentration of various microbes, fishes, crustaceans, amphibians, plants, algae, fungi, or other water-dwelling aquatic or semi- aquatic life and the markers thereof in and around the water. Additionally, the various microbial or biological measurements and chemical measurements may identify the presence, absence, or concentration of fecal matter or other makers of the effects of non-aquatic life in and around the water (e.g., due to farm or habitation runoff into a waterway) on the water. For avoidance of doubt, these various parameters may collectively be referred to as physio-bio-chemical parameters.

[0015] Figure 1 illustrates an example environment 100, such as a watershed, in which embodiments of the present disclosure may be practiced. As illustrated, various bodies HOa-g (generally or collectively, bodies 110) are illustrated, which may include standing bodies 110 of water (e.g., lakes, reservoirs, dammed streams, aquafers), moving bodies 110 of water (e.g., rivers, streams, aqueducts, canals), and temporary bodies 110 (e.g., arroyos, seasonally dry / flooded creeks, retaining or catchment ponds, storm sewers), and include natural bodies 110, purely human-made bodies 110, and enhanced or human-engineered natural bodies 110. Surrounding lands 120a-c (generally or collectively, lands 120) associated with various owners and users may drain into theses bodies 110 due to collected precipitation (e.g., rain runoff, snowmelt, etc.). Additionally, various users 130a-c (generally or collectively, users 130) may output water and other effluents to the bodies 110, draw water and other inputs from the bodies 110, or use the water in the bodies 110 for motive force (e.g., hydroelectric power generation), navigation, or as a cooling source. The owners or users of lands 120 that drain into the bodies 110 may also be users 130 of the bodies 110, but are not required to do so. Accordingly, the entities that own or use the lands 120 and the users 130 of the bodies 110 may collectively be referred to herein as actors.

[0016] The quality and quantity of the water in the various bodies 110 may affect various nonhuman organisms living in the bodies 110 (e.g., invertebrates, fish, amphibians, water plants, algae) or using the water therein for habitat (e.g., waterfowl, beavers) or drinking purposes, affect the lands 120 bordering those bodies, and affect the ability of the users 130 to apply the water for various uses. Accordingly, various sensors can be deployed and monitored throughout the environment 100 to identify water quality, and take actions to address various issues related to water quality.

[0017] In various embodiments, the sensors may include computing devices (such as those discussed in greater detail in regard to Figure 5), that collect various data related to the quantityand characteristics of the water in the bodies 110, the land 120 as use thereof, and how the actors previously, currently, and expectedly used / use / will use the water and the surrounding lands. In addition to quantitative values for various features of the water and the land 120 (e.g., rainfall in a given time period, particulate counts (e.g., total organic carbon (TOC)) in a given time period, temperature at a given time, presence of a given biomarker or chemical, locations / thicknesses of vegetative cover, various fluorescence measures, etc.) the sensors may collect qualitative data and survey-reported data (e.g., from the actors). The sensors may, therefore, be deployed to specific portions of the environment 100 for longitudinal data collection, be intermittently present in the environment 100 (e.g., satellites collecting images of the environment 100, research teams deploying or collecting sensors or data at various intervals) to collect snapshots of data.

[0018] As will be appreciated, saturation of the environment 100 with sensors to measure every possible variable affecting water quality continuously and in real-time is not feasible. Accordingly, the environment 100 is modeled by one or more functional models that use the collected data to extrapolate various data that are not directly measured.

[0019] Figure 2 illustrates a functional model 200 for tracking water quality and attributing changes in water quality to various actors, according to embodiments of the present disclosure. A transfer model 210 receives in-site sensor data 220 from various sensors deployed throughout an environment 100 being monitored for water quality measures, and lab training data 230. The insite sensor data 220 includes various details collected directly from sensors in the environment 100, while the lab training data 230 include data generated in a laboratory setting from data or sample collected from the environment 100. As will be appreciated, in-field sensors may ordinarily lack certain capabilities that laboratory analysis tools provide, which requires extracting a sample from the environment and performing an “offline” or non-real-time analysis in a different setting to provide those values. For example, conventional in-field sensors may provide real-time data on water and air temperature at various points in the environment 100 using easy to deploy temperature probes. In contrast, performing a conventional population survey of different microbes in a body 110 may require extracting a water sample and performing a statistical analysis for the different strains of microbes visually identified therein, which is either infeasible or impossible to perform in real-time or with in-field systems alone. The present disclosure therefore augments the functionalities of the in-field systems via a transfer model 210 that is trained to correlate in-field sensor data with lab-based analyses (e.g., set as a ground truth or labeled output for a training dateset) to model what values the lab-based analyses would produce using as-of-yet uncollected infield data to thereby avoid or reduce the amount of lab-based analyses needed.

[0020] The transfer model 210 may be one of various types of machine learning (ML) models, which identifies correlations between the various values for the in-site sensor data 220 and the lab training data 230, and extrapolates modelled lab data 235 (i.e., the predictions of quantified water quality parameters modelled from the in-field data that are conventionally determined by lab instruments) from given inputs of in-site sensor data 220. Accordingly, based on training the transfer model 210 using previously collected in-site sensor data 220 and lab training data 230, the transfer model 210 may develop a function that generates a value for modelled lab data 235 based on one or more values of newly collected in-site sensor data 220. For example, the level of a given pollutant in a body 110 in parts per million (PPM) may require the use of a centrifuge and various tests that are impractical to carry out in real time in the field, and may therefore be collected and calculated in a laboratory. Several collected values of these lab-generated data are used along with data collected from the environment to train the model transfer model 210 so that in-site sensor data 220 can later be used to generate values that approximate the training data within a given confidence threshold to thereby be used to extrapolate, with confidence, values for the modelled lab data 235 as though those values were based on lab analysis.

[0021] In various embodiments, the sensors may provide in-site sensor data 220 that include one or more of turbidity, conductivity, fluorescent dissolved organic matter (fDOM) measurements, chlorophyll a (Chl-A) measurements, temperature, flow rate / water speed, water level, and metadata related to z-score, N-day averages for various values, sensor percentile, days since deployment, days since last cleaning / maintenance, presence indicators for various chemical compounds, etc. In various embodiments, the lab data include turbidity, conductivity, Total Organic Carbon (TOC), Total Nitrogen (TN ), Kjedldahl Nitrogen content, weighted combinations of N and P lab data, presence or quantity indicators for various chemical compounds / biomarkers / species / strains, etc.

[0022] As will be appreciated, some of the developed transfer functions in the transfer model 210 may be specific to one environment 100 or portion of a given environment 100 and not applicable to other environments 100 or other portions of the given environment 100. For example, a first environment 100 located in a cold climate and a second environment 100 located in a warm climate may each produce a transfer function for microbe content based on in-site data 220 forambient temperature, the presence and type of farms, and precipitation levels, but have different outputs due to whether ambient temperatures preclude the presence of various strains of microbe in the given environment 100, the different crops grown or livestock raised on those farms, and whether the precipitation is snow or rain, among other factors. In another example, a standing body 110 may have different values calculated than a moving body 110 in the same environment 100 based on the same inputs of the in-site data 220 from that environment 100 based on the water flow properties in those different bodies 110 as different portions of the same environment 100.

[0023] A calibration model 250 receives the modelled lab data 235 and (optionally) various in-site sensor data 220 to produce interpolated data 260 matched in space in time to the environment 100. The calibration model 250 identifies the time and space where modelled lab data 235 are assigned in the environment 100, and generates modelled sensor data 225 for values measured by “virtual” sensors at locations where the physical sensors are not deployed. For example, using in-site sensor data 220 collected at time tO-tn from various physical sensors, the calibration model 250 can place the modelled lab data 235 for various extrapolated (but otherwise lab-calculated values) at specific coordinates or zones in the environment 100 at various times.

[0024] When determining values for modelled sensor data 225 for virtual sensors, the calibration model 250 may extrapolate a value based on reported values from two or more physical sensors in the environment 100. For example, a virtual temperature sensor “placed” between two physical temperature sensors to spatially interpolate a temperate at a third location in the modeled environment where no physical temperature sensor is located may be expected to report a modelled temperature value between the two physically measured temperature values, or a different value if another environmental feature that affect temperature is identified in the environment (e.g., a heat exchanger from a power plant). Similarly, the calibration model 250 can use the in-site sensor data 220 collected from times to-tn from various physical sensors to calculate values measured at times before data collection (e.g., to-x), extrapolated / forecasted after data collection (e.g., tn+x), or temporally interpolated between two or more times of data collected (e.g., at time ti when readings are taken at time to and time t2, but not time ti). When generating forecasted values, the calibration model 250 lags the data features by an equivalent number of time intervals and cross-validates the predictions against observed data (when eventually collected) for retraining and improving the calibration model.

[0025] In various embodiments, the calibration model 250 uses a multi-fold (e.g., / / -fold) stratified cross-validation structure to improve the accuracy of the predicted values over time. In cross-validation, the observed data are sequentially partitioned into independent training and testing subsets. Multiple equally-sized subsamples are generated randomly with the time series observations from one physical sensor being grouped in the same partition. The calibration model 250 is trained with at least one of the subsets and is tested on one remaining, held-out subsample as a testing dataset. This training process repeated a total of n times (“n-fold”) with each of the subsamples being used once as the testing dataset. Cross-validation allows for the calculation of performance statistics and the ability to generalize the model to new data as part of the n-fold stratified retraining process.

[0026] The calibration model 250 may also receive additional features for analysis including: rainfall data, stream flow data, location data (e.g., latitude, longitude, altitude, and combinations thereof), hydrologic unit code (HUC12) land cover classification, topographic models, temporal livestock density data, temporal nutrient / insecticide / herbicide application data.

[0027] Using these data, the calibration model 250 can identify attribution data 270, which identify from the various actors and environmental factors the causes of water quality variability due to predictors including land-management practices, water-management practices, and weather events (e.g., storms, wildfires, droughts).

[0028] Figure 3 is an example system 300, as may use the described models for MRV with sensor-remote sensing fusion, according to embodiments of the present disclosure. The system 300 includes one or both of water treatment sites 310 that extract water from a watershed, and treated water sites 320 that output water to the watershed, which the system 300 may signal or control via a machine learning model 330 (running on one or more computing devices) that receives data from a plurality of sensors 340. In various embodiments, the plurality of sensors 340 includes sensors 340 that are disposed in bodies 110 of water in the watershed that are configured to continuously collect and transmit data to the machine learning model 330. In various embodiments, the plurality of sensors 340 include data collection devices that provide qualitative data collected via survey of actors, inputs of topographical and uses of land 120 in the watershed, soil data for the watershed, precipitation data, temperature data, forecasted weather data, and other data related to the watershed that is not reported directly from the environment.

[0029] In various embodiments, the sensors 340 report various data related to water quality and water volume in the watershed. These sensors 340 may include an optical sensor to identify transmittance, reflectance and / or fluorescence of the water. In various embodiments the sensors 340 are configured to collect remote sensing data such as rainfall, biomass cover and land surface properties, and / or quantitative and qualitative survey data. The data collected by the sensors 340 may be collected via wireless transmissions (e.g., using cellular communication or satellite uplinks) so that the sensors may remain deployed in the field and not require an operator to go out to where the sensor 340 is deployed collect the data from the sensors 340.

[0030] In various embodiments, the sensors 340 may be deployed to various bodies 110 of water that include constantly moving water (e.g., rivers), bodies of intermittent moving water (e.g., seasonally dry creeks), natural bodies of standing water (e.g., lakes), manmade bodies of standing water (e g., reservoirs), manmade bodies of moving water (e.g., aqueducts).

[0031] In some embodiments, the sensors 340 may include (or be supplemented with data from) devices used to collect farm survey details, such as the types and quantities of crops / livestock present on a parcel of land; the types and quantities of fertilizers, pesticides, and herbicides used; harvest and planting timings, and other operation details of the farm. Additionally or alternatively, the sensors 340 may include (or be supplemented with data from) devices used to collect land survey data details, such as soil type, demarcations between properties, topologies, plant cover, seasonal precipitation data, or the like.

[0032] Although illustrated with respect to a natural watershed, the present disclosure contemplates that the sensors 340 and machine learning model 330 may similarly be deployed to and used with respect to various water systems, including natural man-made lakes, canals, and seas / oceans, and various closed (or semi-closed) systems. Accordingly, the presently described systems can be used in non-facility based water treatment applications, including waters or chlorinated piped systems that are nominally self-contained (e.g., not continuously pulling from or discharging to rivers / streams / reservoirs), such as in seagoing vessels, space vessels, secure facilities, wherein the “watershed” refers to a collection area or outflow area that a defined environment may collect from or discharge to during normal operations or intermittently. For example, a vessel may include water shipments, water recyclers, showers / sinks / toilets, etc., in a first artificial watershed for potable water, and may include bilges in a second artificial watershedfor buoyancy / balance systems in the vessel that are periodically (but not continuously) opened to the natural environment to dump or intake water.

[0033] Accordingly, the treated water sites 320 may include water treatment sites 310 that output potable water, but may also include other grades of treated water. For example, a water treatment site 310 may treat water to remove a given microbe, a given living organism, a given chemical, or fecal matter may yield higher-quality, but still not potable (for human consumption) water. Water treatment sites 310 may include human-controlled treatment plants, biological filters (e.g., mangrove forests), managed wetlands, septic fields, stocked bodies of water (e.g., to introduce a given microbe, animal, plant, algae, or the like), and gated bodies of water (e.g., to remove, kill, or deter entry of various microbes, animals, plants, algae, or the like) and the like where one or more water quality parameters are intended to be altered.

[0034] Using the machine learning model 330, the collected data from the plurality of sensors 340 are used to generate a time series of estimates for water quality in the watershed, which in turn is used to activate at least one water treatment site 310 to extract or forego extraction of water from the watershed or at least one treated water site 320 to discharge or forego discharge of water into the watershed based on the time series of estimated of water quality. As will be appreciated, foregoing extraction may include a total pause in water extraction for a predefined length of time or a reduction in water extraction of at least 5% of the volume normally extracted during a similar time period of nominal extraction. In some embodiments, discharge includes diverting potable water from a water treatment site 310 into the watershed (rather than a municipal water network) after treatment or processing, opening a reservoir, or outputting water from a water treatment site 320 to the watershed. As will be appreciated, foregoing discharge to the watershed may include discharging water to a retaining pond or other body that can be separated or blocked from bodies 110 that are part of the watershed or a reduction in water output of at least 5% of the volume normally output during a similar time period of nominal output. Additionally, discharge can include water (treated or collected) and one or more treatment solution for affecting water quality downstream from the treated water site 320 within the watershed.

[0035] For example, when the water quality in the watershed is impaired by wildfire in the lands within the watershed based on a first data series from a first sensor and a second data series from a second sensor, the machine learning model 330 may reduce extraction from the bodies 110 in the watershed to improve downstream water quality (e.g., by diluting the effects of the wildfireon the water) and thereby reduce strain on downstream treatment sites or actors. Additionally or alternatively, the machine learning model 330 may increase extraction from the bodies in the watershed to reduce the effect of runoff from the land affecting the flow in the bodies 110 (e.g., due to lack of vegetation increasing water input to the bodies 110).

[0036] For example, when the water quality in the watershed is impaired by human development in the watershed based on a data series from the sensors, such as farming, building, diverting streams, or the like, the machine learning model 330 may time the extraction from or input to the bodies 110 based on human activities to reduce a strain on water treatment sites 310 and treated water sites 320 (e.g., by timing extraction to reduce intake of runoff fertilizers, pesticides, waste, or debris, by timing output to dilute the effect of runoff fertilizers, pesticides, waste or debris).

[0037] In various embodiments, the system may seek to optimize water usage according to various targets. These targets may include goals set by an operator of a water treatment sites 310 or treated water sites 320, such as reduced power usage, timed power usage to generation capacity of renewable generation systems, reduced reagent usage, improved flowrates, increases facility uptime / reduced maintenance expenses, or the like. In some embodiments, these targets may include regulatory set mandates (e.g., a maximum content in a body 110 of water for a given chemical) or green initiative goals, such as the conditions to receive (or avoid forfeiting) carbon credits.

[0038] Because not all water treatment sites 310 in a given watershed may be configured to affect all water quality parameters of interest, or that a first water treatment site 310 may be more efficient or effective at affecting a given water quality parameter than a second water treatment site 310, the machine learning model 330 is able to engage in water quality trading throughout the environment. This water quality trading may be between multiple water treatment sites 310 or treated water sites 320, but may also be between one or more water treatment sites 310 and surrounding users, or between two or more surrounding users (and no water treatment sites 310 or treated water sites 320).

[0039] For example, if a managed septic field is used as a treatment site 310 for multiple users, the model 330 can allocate usage (e.g., in total amount, flow within a given time period, etc.) between the multiple users to avoid or reduce runoff from the treatment site 310 into a local waterway. Similarly, if the land of two different users drain into a shared waterway with nointervening treatment sites 310, the model 330 can advise the users on how and when to apply fertilizer to avoid excessive runoff into the shared waterway that would negatively affect other users who are downstream from the advised users, but upstream from any treatment sites 310.

[0040] In another example, if two operators of water treatment facilities at different locations in a given watershed are collectively tasked with reducing a microbe count in a waterway to or below a given point downstream to both facilities via individual control of the two facilities, the machine learning model 330 can identify how the two facilities can most effectively reach the goal, which may include identifying users within the watershed to communicate with to curtail certain activities at various times (e.g., to manage or reduce run off waters from agricultural lands to a manageable amount by the facilities). Additionally or alternatively, the machine learning model 330 may be used to trade quality metrics throughout the watershed so that various actors may more efficiently reach the water quality goals.

[0041] In various embodiments, the machine learning model 330 generates the time series of estimates for control of the water system by identifying or calculating changepoints within the data. A changepoint may be identified by calculating a first standardized variable for a first segment of the data and a second standardized variable for a second segment of the data and determining that a difference between the first and second standardized variables exceeds an optimal threshold. Once this difference has been identified as exceeding the optimal threshold, the machine learning model 330 identifies a changepoint between the first segment and the second segment split the data into intervals at the changepoints and may then classify the intervals. These time series of estimates may identify one or more of predicted water quality, water volume, estimated carbon credits, and environmental benefits a water source based on the classified intervals using the data from a subset of the plurality of separate water sources.

[0042] For example, the machine learning model 330 may use these estimates to show compliance with or attainment of various carbon credit targets (e.g., to receive credit for these carbon credits) based on water quality or bioaccumulation in the watershed affected by water management policies. In another example, the machine learning model 330 may use the estimates to identify potential sources to receive some or all of a carbon credit, or be penalized (or identified as a target to work with) when land use policies by those entities affect water management policies in reaching (or nor reaching) a carbon credit target. Accordingly, the machine learning model 330 may attribute various effects in the bodies 110 of water to various actors, and help direct actionsto improve land usage in the watershed with specific actors in need of positive or negative reinforcement.

[0043] In another example, the machine learning model 330 may identify the effects of a wildfire or other disaster (e.g., flood, hurricane, tornado) affecting the land of the watershed, and identify, using the time series of estimates, ways to reduce the effect on the water and downstream lands and actors of that disaster.

[0044] In various embodiments, the machine learning model 330 is configured to control various systems linked within the watershed to water quality based on the time series of estimates. The machine learning model 330 can determine which of the quality-linked systems or combinations thereof will have the largest, fastest, most cost-effective (or some combination thereof) positive effect on water quality for the users or the watershed as a whole and direct the operation of those quality-linked systems at various times to meet various water quality goals. These quality-linked systems may include potable water treatment facilities, wastewater treatment facilities, irrigation equipment, farm equipment (such as fertilizer applicators or harvesters), dams (for water retainment or redirection), fences (e.g., to control the movement or location of livestock, wildlife, or humans), in-stream or in-lake algae treatment systems, water sources (e.g., pumps at wellheads), broadcast systems (e.g., to transmit advisories to persons or entities in the watershed or in neighboring watersheds), and the like.

[0045] For example, the machine learning model 330 may control an amount of water output by irrigation equipment located in the watershed, including at least one of a timing, a duration, and a location of irrigation. For example, the machine learning model 330 may control an amount of fertilizer or pesticide output by farm equipment located in the watershed, including at least one of a timing, a duration, an intensity, a chemical composition, and a location of application. For example, the machine learning model 330 may control movement of livestock within the watershed, including activating virtual or real electric fences. For example, the machine learning model 330 may transmit a boil-water advisory to persons and entities located in the watershed. For example, the machine learning model 330 may change an activation level (e.g., turn on, turn off, increase or decrease level of usage) of an algae treatment technology in the river or stream or a reservoir in the watershed. For example, the machine learning model 330 may release water from a dam fed by or feeding into the river or stream or control the dam to retain additional water from a current level. For example, the machine learning model 330 may direct a water utility to changea drinking water source used to supply users with. For example, the machine learning model 330 may direct a wastewater utility to change an activation level (e.g., turn on, turn off, increase or decrease level of usage) of treatment equipment or change a discharge level (e.g., turn on, turn off, increase or decrease level of usage) at one or more locations in the watershed.

[0046] Figure 4A is a flowchart of an example method 400A for analyzing sensor data, according to embodiments of the present disclosure. These sensor data include data collected from source sensors (e.g., sensors associated with various water sources) and remote sensors that may be deployed at various locations in a watershed. In various embodiments, the remote sensors may include survey data collection devices, satellite imaging devices, and other sensors that collect data related to the watershed that are not directly deployed in a body of water.

[0047] Method 400A begins at block 402, where a source sensor collects source data comprising at least one of a water volume, a water quality, or a water system function for a water source. For example, a turbidity sensor 340 attached to an intake pipe at a water treatment plant may collect data 220 on suspended particle volumes in the intake pipe. These data 220 may be combined with source data 220 from other source sensors 340. For example, a temperature sensor 340 might simultaneously take temperature readings of influent water, a flow meter sense 340 might simultaneously measure a flowrate through the intake pipe, a sensor 340 might measure and record an acidity of the influent water, and a pump status sensor 340 might sense which pumps are operating at what time.

[0048] At block 404, a machine learning-based data model receives the source data from one or more of the source sensors. For example, a neural network-style machine learning model 330 trained on historical data from the water treatment site 310 may receive the source data 220. The neural network may be trained with historical data from a different water treatment sites 310 or treated water sites 320, and may have adjustments such as low-rank adaptations applied to compensate for particular differences between a geographical location of the source sensor 340 and a geographical origin of the training data.

[0049] At block 406, the machine learning model fuses the source data with remote sensing data. For example, the machine learning model 330 may combine the source data 220 from the source sensor 340 with weather data from a nearby weather station and survey data from a surrounding geographical area. Additional remote data may further be combined, including but not limited to information about current construction projects, biomass cover and logging data,and data about other factors which may affect the water treatment site. The fusion may be configured to incorporate methodologies associated with carbon credit generation and issuance to ensure that data are correlated according to the relevant methodologies for accreditation for carbon credits.

[0050] At block 408, the machine learning-based data model 330 outputs a near-realtime estimation of environmental conditions. For example, the neural network may deduce from a high turbidity, high water flow, recent rain, and data indicative of lots of construction nearby that an unusually large amount of sediment is being washed from construction sites into a river which serves as influent for the water treatment site 310. This information can then be used in a number of ways. For example, the machine learning-based data model 330 may predict when high levels of soil runoff from construction sites will occur and cause the water treatment site 310 to take in less water during these periods, as the high turbidity increases the amount of carbon output required to process the water. Accordingly, the treatment site 310 may temporarily forego treating water until sediment levels are predicted to fall below an acceptability threshold. This information may also be output to an interface such as a monitor or saved for future manual analysis, allowing for the collected data and neural network analysis to potentially inform policy decisions surrounding and use in the watershed.

[0051] The machine learning-based data model 330 may be configured to detect when a metric for water quality corresponding to data from one or more sensors 340 is likely to depart from a predefined desired range or value, and may send commands to a water treatment site 310 or treated water site 320 to proactively prevent such a departure. Responsive to a departure actually occurring, the machine learning-based data model 330 may be configured to send commands to the water treatment site 310 or treated water site 320 to attempt to bring the appropriate metric back into compliance with the predefined desired range or value.

[0052] Figure 4B is an example system 400B for analyzing data and controlling a water system, according to embodiments of the present disclosure. A first source sensor 410 and a second source sensor 412 monitor a water system 450. Source data 414 from the first source sensor 410 and the second source sensor 412 are fed into a machine learning-based data model 430 along with remote data 420. The machine learning-based data model 430 uses the source data 414 and the remote data 420 to generate output data 440 about current and / or future environmental conditions.The machine learning-based data model may also generate input commands for the water system 450.

[0053] The first source sensor 410 and the second source sensor 412 may be any variety of sensor, including but not limited to a flow meter, a density sensor, a viscosity sensor, a turbidity sensor, an acidity sensor, a chronometer, a thermometer, a barometer, a rain gauge, an anemometer, a weather vane, a light sensor, and a color sensor. The source data 414 may be fed to the machine learning-based data model continuously in near-realtime, or the source data 414 may be divided into blocks which may be sent to the machine learning-based data model periodically. The source data 414 may be recorded for future reference and / or model training. While Figure 4B is illustrated with a first source sensor 410 and a second source sensor 412, actual embodiments may include additional or fewer source sensors as desired. These various source sensors may be associated with the same water source (e g., a well, a borehole, a water treatment facility, a water tower) or different water sources.

[0054] The remote data 420 may include but is not limited to water quality metrics gathered from bodies of water away from the source, weather data, survey data (both quantitative and qualitative), air quality data, land use data, and biomass coverage data. The output data 440 may include but is not limited to an estimation of current environmental conditions, a prediction of future environmental conditions, a number of carbon credits generated, an attribution of any irregularities in the source data 414 to a suspected cause, and commands for controlling the water system 450. The output data 440 may be recorded for future reference and / or model training.

[0055] The machine learning-based data model 430 may be trained with any form of machine learning algorithm, including but not limited to supervised, semi-supervised, unsupervised, reinforcement algorithms, and combinations thereof. Similarly, the machine learning-based data model 430 may be any machine learning model, including a support vector machine, a neural network, a random forest, a k-nearest neighbor algorithm, and a symbolic regression model. The machine learning-based data model 430 may be continuously trained on the source data 414, and may be periodically updated as better-trained models become available. A “template” machine learning-based data model 430 may be combined with a low-rank adaptation tailored to the water system 450 to achieve quick functionality upon installation, with the potential for continuous in- place training once installation is completed.

[0056] In some embodiments, the sensor 412 is associated with a second water system 450, separate from that which is illustrated. In such an embodiment, the second water system 450 may be linked to the water system 450, or may be isolated. The machine learning-based data model 430 may be able to send commands to one, both, or neither of the water systems 450.

[0057] Figure 4C is a plot 400C of hypothetical sensor data, according to embodiments of the present disclosure. The data is presented to illustrate how the machine learning-based data model might identify changepoints in a data set. A line 460 tracks the data over a period of approximately 16 hours. A first changepoint 480, a second changepoint 482, a third changepoint 484, and a fourth changepoint 486 separate a first interval 470, a second interval 472, a third interval 474, a fourth interval 476, and a fifth interval 478.

[0058] It will be appreciated that the changepoints 480-486 are located at points where the data appear to have been pushed out of an equilibrium. The machine learning-based data model may be configured to detect these changepoints 480-486 and attribute the changepoints 480-486 to respective causes. The machine learning-based data model may also be configured to predict and / or cause changepoints 480-486 by interpolating the data and / or sending commands to an associated water system (see Figure 4B).

[0059] Figure 5 illustrates a computing device 500, as may be used for MRV with sensorremote sensing fusion, according to embodiments of the present disclosure. The computing device 500 may include at least one processor 510, a memory 520, and a communication interface 530.

[0060] The processor 510 may be any processing unit capable of performing the operations and procedures described in the present disclosure. In various embodiments, the processor 510 can represent a single processor, multiple processors, a processor with multiple cores, and combinations thereof.

[0061] The memory 520 is an apparatus that may be either volatile or non-volatile memory and may include RAM, flash, cache, disk drives, and other computer readable memory storage devices. Although shown as a single entity, the memory 520 may be divided into different memory storage elements such as RAM and one or more hard disk drives. As used herein, the memory 520 is an example of a device that includes computer-readable storage media, and is not to be interpreted as transmission media or signals per se.

[0062] As shown, the memory 520 includes various instructions that are executable by the processor 510 to provide an operating system 522 to manage various features of the computingdevice 500 and one or more programs 524 to provide various functionalities to users of the computing device 500, which include one or more of the features and functionalities described in the present disclosure. One of ordinary skill in the relevant art will recognize that different approaches can be taken in selecting or designing a program 524 to perform the operations described herein, including choice of programming language, the operating system 522 used by the computing device 500, and the architecture of the processor 510 and memory 520. Accordingly, the person of ordinary skill in the relevant art will be able to select or design an appropriate program 524 based on the details provided in the present disclosure, n various embodiments, the program 524 may include or make use of a machine learning model that is trained to make determinations as set forth in the present disclosure, and may be retrained or updated based on data collected as set forth in the present disclosure.

[0063] The communication interface 530 facilitates communications between the computing device 500 and other devices, which may also be computing devices as described in relation to Figure 5. In various embodiments, the communication interface 530 includes antennas for wireless communications and various wired communication ports. The computing device 500 may also include or be in communication, via the communication interface 530, one or more input devices (e.g., a keyboard, mouse, pen, touch input device, etc.) and one or more output devices (e.g., a display, speakers, a printer, etc.).

[0064] Although not explicitly shown in Figure 5, it should be recognized that the computing device 500 may be connected to one or more public and / or private networks via appropriate network connections via the communication interface 530. It will also be recognized that software instructions may also be loaded into a non-transitory computer readable medium, such as the memory 520, from an appropriate storage medium or via wired or wireless means.

[0065] Accordingly, the computing device 500 is an example of a system that includes a processor 510 and a memory 520 that includes instructions that (when executed by the processor 510) perform various embodiments of the present disclosure. Similarly, the memory 520 is an apparatus that includes instructions that, when executed by a processor 510, perform various embodiments of the present disclosure.

[0066] Certain terms are used throughout the description and claims to refer to particular features or components. As one skilled in the art will appreciate, different persons may refer to thesame feature or component by different names. This document does not intend to distinguish between components or features that differ in name but not function.

[0067] As used herein, the term “optimize” and variations thereof, is used in a sense understood by data scientists to refer to actions taken for continual improvement of a system relative to a goal. An optimized value will be understood to represent “near-best” value for a given reward framework, which may oscillate around a local maximum or a global maximum for a “best” value or set of values, which may change as the goal changes or as input conditions change. Accordingly, an optimal solution for a first goal at a given time may be suboptimal for a second goal at that time or suboptimal for the first goal at a later time.

[0068] As used herein, various chemical compounds are referred to by associated element abbreviations set by the International Union of Pure and Applied Chemistry (IUPAC), which one of ordinary skill in the relevant art will be familiar with. Similarly, various units of measure may be used herein, which are referred to by associated short forms as set by the International System of Units (SI), which one of ordinary skill in the relevant art will be familiar with.

[0069] As used herein, “about,” “approximately” and “substantially” are understood to refer to numbers in a range of the referenced number, for example the range of -10% to +10% of the referenced number, preferably -5% to +5% of the referenced number, more preferably -1% to +1% of the referenced number, most preferably -0.1% to +0.1% of the referenced number.

[0070] Furthermore, all numerical ranges herein should be understood to include all integers, whole numbers, or fractions, within the range. Moreover, these numerical ranges should be construed as providing support for a claim directed to any number or subset of numbers in that range. For example, a disclosure of a range from 1 to 10 should be construed as supporting ranges of any two numbers X and Y that fall into the initial range of from 1 to 10 where X > 1 and Y < 10.

[0071] As used in the present disclosure, a phrase referring to “at least one of’ a list of items refers to any set of those items, including sets with a single member, and every potential combination thereof. For example, when referencing “at least one of A, B, or C” or “at least one of A, B, and C”, the phrase is intended to cover the sets of: A, B, C, A-B, B-C, and A-B-C, where the sets may include one or multiple instances of a given member (e.g., A-A, A-A-A, A-A-B, A- A-B-B-C-C-C, etc.) and any ordering thereof. For avoidance of doubt, the phrase “at least one ofA, B, and C” shall not be interpreted to mean “at least one of A, at least one of B, and at least one ofC”.

[0072] As used in the present disclosure, the term “determining” encompasses a variety of actions that may include calculating, computing, processing, deriving, investigating, looking up (e.g., via a table, database, or other data structure), ascertaining, receiving (e.g., receiving information), accessing (e.g., accessing data in a memory), retrieving, resolving, selecting, choosing, establishing, and the like.

[0073] Without further elaboration, it is believed that one skilled in the art can use the preceding description to use the claimed inventions to their fullest extent. The examples and aspects disclosed herein are to be construed as merely illustrative and not a limitation of the scope of the present disclosure in any way. It will be apparent to those having skill in the art that changes may be made to the details of the above-described examples without departing from the underlying principles discussed. In other words, various modifications and improvements of the examples specifically disclosed in the description above are within the scope of the appended claims. For instance, any suitable combination of features of the various examples described is contemplated.

[0074] Within the claims, reference to an element in the singular is not intended to mean “one and only one” unless specifically stated as such, but rather as “one or more” or “at least one”. Unless specifically stated otherwise, the term “some” refers to one or more. No claim element is to be construed under the provision of 35 U.S.C. § 112(f) unless the element is expressly recited using the phrase “means for” or “step for”. All structural and functional equivalents to the elements of the various embodiments described in the present disclosure that are known or come later to be known to those of ordinary skill in the relevant art are expressly incorporated herein by reference and are intended to be encompassed by the claims. Moreover, nothing disclosed in the present disclosure is intended to be dedicated to the public regardless of whether such disclosure is explicitly recited in the claims.

Claims

CLAIMSThe invention is claimed as follows:

1. A system comprising: a plurality of separate water sources, wherein each water source of the plurality of separate water sources includes: a sensor indicating a measure of at least one of water quality and water volume; a data transmission system configured to transmit data from each of the plurality of separate water sources; and a treated water detection system including a microcontroller having a processor and a memory storing instructions, wherein the instructions, when executed by the processor, cause the processor to: receive the data from each of the plurality of separate water sources; calculate changepoints within the data by: calculating a first standardized variable for a first segment of the data and a second standardized variable for a second segment of the data; determining that a difference between the first and second standardized variables exceeds an optimal threshold; and identifying a changepoint between the first segment and the second segment, split the data into intervals at the changepoints; classify the intervals using a machine learning model; and identify predicted water quality, water volume, estimated carbon credits, and environmental benefits a water source based on the classified intervals using the data from a subset of the plurality of separate water sources.

2. The system of claim 1, wherein the data from the plurality of separate water sensors are transmitted via satellite or cellular telemetry.

3. The system of claim 1, wherein the data comprise remote sensing data that include at least one of a rainfall, a biomass cover, a land surface property, or survey data.

4. The system of claim 1, wherein the source sensor is associated with a first water system, and wherein the machine learning model is further configured to fuse data from a second source sensor associated with a second water system.

5. The system of claim 4, wherein fusing the data from the second source sensor includes incorporation of generally-accepted carbon credit issuance standards, statistical methods, and methodologies.

6. The system of claim 1, wherein the machine learning model is further configured to yield a near real-time estimation of water volumes.

7. The system of claim 6, wherein the machine learning model is further configured to yield a prediction of water volumes.

8. The system of claim 1, wherein the machine learning model is further configured to yield a near real-time estimation of avoided biomass consumption.

9. The system of claim 1, wherein the machine learning model is further configured to yield a prediction of avoided biomass consumption.

10. The system of claim 1, wherein the machine learning model is further configured to yield a near real-time estimation of water quality.

11. The system of claim 1, wherein the machine learning model is further configured to yield a prediction of water quality.

12. The system of claim 1, wherein the machine learning model is further configured to yield a near real-time estimation of carbon credit generation.

13. The system of claim 1, wherein the machine learning model is further configured to: determine that an environmental metric can be brought closer to a predefined desired range or value by modifying a configuration of a target water system; and modify the configuration of the target water system, responsive to the determination.

14. The system of claim 13, wherein the environmental metric is one of a water quality, a water volume, or a rate of carbon credit generation.

15. The system of claim 1, wherein the machine learning model is further configured to:predict that an environmental metric will depart from a predefined desired range or value unless a configuration of a target water system is modified; and modify the configuration of the target water system proactively, responsive to the prediction.

16. The system of claim 15, wherein the environmental metric is one of a water quality, a water volume, or a rate of carbon credit generation.

17. A method, comprising: measuring, via a source sensor, source data comprising at least one of a water volume, a water quality, or a water system function associated with a water source; receiving the source data from the source sensor at a machine learning model; fusing the source data with remote sensing data via the machine learning model; and outputting, via the machine learning model, a near-realtime estimation of environmental conditions.

18. The method of claim 17, wherein the source sensor is associated with a first water system, and wherein the machine learning model is further configured to fuse data from a second source sensor associated with a second water system.

19. The method of claim 17, wherein the remote sensing data includes at least one of a rainfall, a biomass cover, a land surface property, or survey data.

20. The method of claim 17, wherein the environmental conditions include at least one of a treated water volume, an avoided biomass consumption, a water quality, or a carbon credit generation rate.

21. The method of claim 17, further comprising outputting, via the machine learning model, a prediction of future environmental conditions.

22. The method of claim 17, further comprising: determining that an environmental metric can be brought closer to a predefined desired range or value by modifying a configuration of a target water system; and modifying the configuration of the target water system, responsive to the determination, wherein the environmental metric is one of a water quality, a water volume, or a rate of carbon credit generation.

23. The method of claim 17, further comprising: predicting that an environmental metric will depart from a predefined desired range or value unless a configuration of a target water system is modified; and modifying the configuration of the target water system proactively, responsive to the prediction, wherein the environmental metric is one of a water quality, a water volume, or a rate of carbon credit generation.

24. A non-transitory computer-readable medium storing instructions which, when executed by a processor, cause the processor to:measure, via a source sensor associated with a water source, source data comprising at least one of a water volume, a water quality, or a water system function; receive the source data from the source sensor at a machine learning model; fuse the source data with remote sensing data via the machine learning model; and output, via the machine learning model, a near-realtime estimation of environmental conditions.

25. The non-transitory computer-readable medium of claim 24 storing further instructions which, when executed by the processor, cause the processor to: output, via the machine learning model, a prediction of future environmental conditions.

26. The non-transitory computer-readable medium of claim 24 storing further instructions which, when executed by the processor, cause the processor to: determine that an environmental metric can be brought closer to a predefined desired range or value by modifying a configuration of a target water system; and modify the configuration of the target water system, responsive to the determination.

27. The non-transitory computer-readable medium of claim 24 storing further instructions which, when executed by the processor, cause the processor to: predict that an environmental metric will depart from a predefined desired range or value unless a configuration of a target water system is modified; and modify the configuration of the target water system proactively, responsive to the prediction.

Images