In times of an increasingly uncertain climate and growing demand for water resources, there is a need for assessing how aquatic systems worldwide are responding to co-occurring human and climatic disturbances. Over the past half century, remote sensing technology, high performance and cloud computing infrastructure, machine learning techniques, open data practices, and widened training in computer programming have created extraordinary opportunities to expand aquatic science research across spatial and temporal scales. Remote sensing of chlorophyll concentrations has enabled near-global monitoring of algal blooms. Data science and machine learning methods have empowered the prediction of water quality dynamics, while also informing management actions. Process-based models have improved our understanding of environmental fluid dynamics, which can be consequential for ecosystem regime shifts. Open data practices have increased the amount of publicly available data that many calibration and validation techniques ultimately demand. The nexus of data science, remote sensing, modeling, and open science in the aquatic sciences is a dynamic, rapidly progressing space, where novel aquatic science questions can be answered at unprecedented scales.
Given these developments, there can be a lingering question of “What’s next?” To date, many aquatic remote sensing, modeling, and data science efforts have focused on measuring or predicting individual parameters or variables. While these individual efforts are herculean initiatives, a ripe frontier for this nexus is linking limnological, hydrological, oceanographic, and ecological processes and principles with remote sensing, data science, and modeling techniques to understand fundamental emergent properties of aquatic systems and inform monitoring at local-to-global and subdaily-to-decadal scales. Additionally, the increase of open science and data democratization brings the potential for a more diverse, inclusive, and accessible scientific community and more holistic research.
To build a conversation around multifaceted developments in remote sensing, data science, modeling, and the aquatic sciences, this session invites contributors to share how they use one or a combination of remote sensing, data science, modeling, or open science practices to expand the fields of limnology, oceanography, hydrology, or ecology. We envision this session will host a range of presentation topics, including but not limited to novel methods for atmospheric corrections, scaling of cloud and other high-volume computing environments, assessing water quantity and quality across spatial and temporal scales, quantifying long-term changes in stratification dynamics, merging process-based modeling and deep learning, and applying data-intensive techniques for basic and applied research questions.
While submissions may stem from methodological and technical hurdles encountered in remote sensing and data science fields, we challenge submissions to focus on how their efforts de-silo the aquatic sciences from the remote sensing and data science arenas.
We enthusiastically encourage submissions by early career researchers as well as by researchers from BIPOC, LGBTQIA+, and other marginalized identities. An intentional focus on research that breaks down barriers to entry for underrepresented scientists in the fields of remote sensing and aquatic data science, through open science, will yield insight into the power and potential of the next frontier in emergent properties of aquatic systems.
Lead Organizer: Michael Meyer, U.S. Geological Survey (mfmeyer@usgs.gov)
Co-organizers:
Elisa Calamita, Eawag (elisa.calamita@eawag.ch)
Kate Fickas, Esri (kfickas@esri.com)
Robert Ladwig, University of Wisconsin - Madison (rladwig2@wisc.edu)
Rachel Pilla, Oak Ridge National Laboratory (pillarm1@ornl.gov)
Presentations
02:00 PM
MODULAR COMPOSITIONAL LEARNING FOR WATER TEMPERATURE MODELING V2.0: MORE LAKES, RESTRAINTS, AND MEMORY (7768)
Primary Presenter: Robert Ladwig, Aarhus University (rladwig@ecos.au.dk)
One-dimensional water temperature models provide a valuable tool to quantify future climate change impacts on lakes and reservoirs at the global scale. Recently, the novel paradigm of Knowledge-Guided Machine Learning (KGML) has highlighted the increased performance of 1D water temperature projections by combining process-based modeling with deep learning. Modular compositional learning (MCL), a design based on the KGML paradigm, highlighted that a deep learning model embedded in a process-based model framework can improve predictions of physical limnological variables with a high signal-to-noise ratio while limiting physically unrealistic density violations. In the hybrid MCL model, turbulent diffusive transport is simulated through the deep learning model to decrease the uncertainty of empirical process parameterizations. Here, we improve upon the current hybrid MCL model by changing its training algorithm from sequential to recurrent. This update allows the hybrid MCL model to utilize past information for future predictions. Further, we incorporate a loss function to penalize energy conservation violations. This loss function helps to prevent physically unrealistic results by the deep learning model. Finally, we train the deep learning model for turbulent diffusive transport on a set of lakes from Denmark, Sweden, Switzerland, Germany, and the US to increase its generalizability. The modified hybrid MCL model produces more robust results than its previous version with improved water temperature projections across a variety of lake types, highlighting the utility of this approach for hydrodynamic modeling.
02:15 PM
ADVANCING UNDERSTANDING OF LAKE WATER QUALITY THROUGH TIME WITH MODULAR COMPOSITIONAL LEARNING (7905)
Primary Presenter: Bennett McAfee, University of Wisconsin–Madison (bennettjmcafee@gmail.com)
Water quality is integral to the ecosystem services of lakes and is an emergent property of complex spatial and temporal dynamics, which makes predicting changes in lake water quality challenging. Modular Compositional Learning (MCL) is a framework within the Ecology Knowledge-Guided Machine Learning (Eco-KGML) paradigm that allows ecosystem phenomena, such as water quality, to be simulated by coupled modules that are either process-based models or machine learning models. Lake metabolism, an important metric of water quality, is a prime candidate for modeling with MCL. Metabolism can be broken down into component modules representing atmospheric gas exchange, net primary production, ecosystem respiration, and transport of organic matter within the water column. We have created modules within a MCL framework to represent these four processes and linked them with an existing MCL model of lake thermal structure to model dynamics of dissolved oxygen, dissolved organic carbon, and particulate organic carbon along the water column at hourly intervals over the course of 5 years within Lake Mendota (Wisconsin, USA). By swapping out process-based versions of each module with machine learning versions, we can determine where and when there are ecosystem processes integral to observed water quality that the processed-based models are not currently capturing. Consequently, MCL offers novel opportunities for us to revise our understanding of the drivers of water quality to finer spatial and temporal scales.
02:30 PM
Understanding and Predicting Sea Surface Temperature Responses to Tropical Cyclone Wind Pump Using Explainable Machine Learning (7805)
Primary Presenter: Hongxing Cui, Guangdong Remote Sensing Center for Marine Ecology and Environment, Southern Marine Science and Engineering Guangdong Laboratory (Guangzhou); Hong Kong University of Science and Technology (hcuiaf@connect.ust.hk)
This study develops a predictive model for the sea surface temperature (SST) cooling generated by tropical cyclones (TCs) in the northwest Pacific (NWP), leveraging machine learning techniques such as random forest and extreme gradient boosting (XGBoost). By incorporating 12 predictors related to the characteristics of TCs and the pre-storm ocean states, the model adeptly predicts the spatial structure and temporal evolutions of SST cooling and surpasses a conventional numerical model in accuracy. Important predictors include TC intensity, speed, size, and pre-storm ocean conditions like mixed layer depth and SST. These predictors were identified through feature importance assessments and SHapely Additive exPlanations (SHAP), which contribute to attribute-oriented explainability in the proposed method. The model has the ability to accurately simulate the spatial structure and temporal evolution of SST cooling for different TC intensities, along with its explanatory power regarding the ocean-atmosphere interactions during TCs. This makes the model an insightful tool for analyzing the complex responses of oceanic conditions to TC wind pump effects.
02:45 PM
MAPPING PLASTIC POLLUTION HOTSPOTS IN LAGOS: INTEGRATING DRONE TECHNOLOGY AND MACHINE LEARNING FOR COASTAL ECOSYSTEM MONITORING (7835)
Primary Presenter: Ngozi Oguguah, Nigerian Institute for Oceanography and Marine Research (ngozimoguguah@yahoo.com)
Plastic pollution in coastal ecosystems is of utmost concern, especially in population-dense and metropolitan cities such as Lagos, Nigeria. The study aimed to map plastic debris hotspots analyse water quality indices, and target plastic-related pollutants in drainage channels that empty into the Lagos Lagoon, Nigeria. Drainage channels were purposively sampled across 4 districts/zones in Lagos characterised by proximity to anthropogenic inputs and population density. Utilizing Sentinel-2 satellite images, a plastic debris detection technique was developed using the Normalized Difference Water Index (NDWI) and Normalized Difference Built-up Index (NDBI). Water quality indices and target pollutants (temperature, dissolved oxygen, pH, VOCs, Pb, Cd) were analysed using in situ water quality test instruments and standards methods respectively. Fieldwork, drone images and a 2018 NASA plastic debris dataset were used to evaluate the performance of the developed technique. The results were promising as the technique successfully identified plastic debris. After that, plastic debris hotspot maps were created using GIS techniques. An automatic plastic debris detection method was also developed using drone images and machine learning techniques. The YOLOv8 model was utilized for plastic debris detection and classification. This study contributes to pollutant monitoring in coastal ecosystems for evidence-informed interventions through autonomous detection methods, machine learning and Geospatial techniques.
03:00 PM
Deep Reinforcement Learning for Macro-Scale Dissolved Organic Carbon Predictions from Multi-Modal Data (8341)
Primary Presenter: Daniel Dominguez, Colorado State University (Daniel.Dominguez@colostate.edu)
Dissolved organic carbon (DOC) is a key water quality constituent providing a key energy source for freshwater ecosystems, altering light regimes in freshwaters, and controlling nutrient cycling dynamics. However, it is underrepresented in freshwater data collections and databases when compared to other datasets on nutrients, algae concentration, sediment and other key water quality parameters. To fill this data-gap we propose estimating DOC concentrations with remote sensing and deep learning. We aim to construct a generalized model capable of predicting DOC across freshwater estuaries, lakes, and rivers leveraging the Landsat archive and our dataset that pairs Landsat reflectance data with in-situ DOC collections. To accomplish our goal, we introduce an innovative approach employing deep reinforcement learning (DRL) for accurate predictions of DOC concentrations from remote sensing data across the contiguous US. DOC varies spatial-temporally, and are highly responsive to terrestrial inputs, biological activity, land-use, human activity, and water body metabolism. Because of these highly varied controls on DOC dynamics, we complement our direct remote sensing data with additional inputs to produce reliable national data. Although model performance approximately matches current models with an MAE of 1.93 mg/L overall, our approach provides specific error rates for types of water bodies in ecoregions. This work forms the foundation of future work exploring controls on DOC variation in waterbodies of the USA, with an unbiased national data source.
03:15 PM
Machine learning for water quality forecasting and intervention in managed reservoirs (8268)
Primary Presenter: Bethel Steele, Colorado State University (b.steele@colostate.edu)
Temperature and clarity are critical indicators of reservoir ecosystem function, and they are often regulated at the local, state, and federal levels. Water resource managers face the challenge of balancing reservoir operations with these regulations, often working with sparse datasets and few, if any, tools for forward-looking operation management. As the occurrence of extreme climatic events, natural disasters and interannual variability increases, some water resource managers are turning to forecast models to assist with decision making. Understanding stakeholder needs and the available mitigation actions allows us to create applications guiding day-to-day water management within regulatory and delivery requirements. In this study, we utilize a dataset comprised of water temperature, local meteorology, stream discharge, and reservoir level in an autoregressive neural network to make estimations of water temperature at a managed reservoir in Northern Colorado. This model consistently outperforms the baseline of "yesterday-is-today" at a one-day time horizon, with a mean absolute error of 0.37°C for the top 1 meter and 0.29°C for 0-5 meters during preliminary hindcasting application. We are currently testing the model's sensitivity to changes in pumping operations, which is the water quality mitigation action used within this system. By leveraging machine learning techniques, we aim to provide stakeholders with valuable insights and tools to navigate the complexities of reservoir management while adhering to regulatory requirements.
SS01B - The Next Frontier in Aquatic Sciences: Linking Remote Sensing, Data Science, Modeling, and Open Science to Understand Ecosystems’ Emergent Properties
Description
Time: 2:00 PM
Date: 5/6/2024
Room: Hall of Ideas G