The true bottleneck of artificial intelligence (AI) is not access to the data, but rather labeling this data. We have tons of raw agriculture image data coming from various sources and manual labelling remains to be a crucial step to keep the data well organized which requires considerable amount of time, money, and labor. This process can be made more efficient if we can automatically label the raw data. We propose contrastive learning representations for agriculture images (AgCLR) model that uses self-supervised representation learning approach on unlabeled real-world agriculture field data, to learn the useful image feature representations from the images. Contrastive learning is a self-supervised approach that enables model to learn attributes by contrasting samples against each other without the use of labels. AgCLR leverages the state-of-the-art SimCLRv2 framework to learn representations by maximizing the agreement between differently augmented views of same sample. We have incorporated critical enablers like mixed precision, multi-GPU distributed parallel computing, and use of Google Cloud's Tensor Processing Units (TPU) for optimizing the training process. We achieved 80.2% accuracy while classifying the test data. We further applied AgCLR to unrelated task to determine the alleys and rows in corn field videos for corn phenotyping and we observed two cluster formations for alleys and rows when plotted embeddings in a 3-dimensional space. We also developed a content-based image retrieval tool (pixel affinity) to identify similar images in our database and results were visually very promising.

The concepts of resistance, recovery, and resilience are in diverse fields from behavioral psychology to planetary ecology. These “three Rs” describe some of the most important properties allowing complex systems to survive in dynamic environments. However, in many fields—including ecology—our ability to predict resistance, recovery and resilience remains limited. Here, we propose new disturbance terminology and describe a unifying definition of resistance, recovery, and resilience. We distinguish functional disturbances that affect short-term ecosystem processes from structural disturbances that alter the state factors of ecosystem development. We define resilience as the combination of resistance and recovery—i.e., the ability of a system to maintain its state by withstanding disturbance or rapidly recovering from it. In the Anthropocene, humans have become dominant drivers of many ecosystem processes and nearly all the state factors influencing ecosystem development. Consequently, the resilience of an individual ecological parameter is not an inherent attribute but a function of linkages with other biological, chemical, physical, and especially social parameters. Because every ecosystem experiences multiple, overlapping disturbances, a multidimensional resilience approach is needed that considers both ecosystem structure (configuration of linkages) and disturbance regime. We explore these concepts with a few case studies and recommend analytical tools and community-based approaches to strengthen ecosystem resilience. Disregarding cultural and social dimensions of disturbance regimes and ecosystem structures leads to undesirable outcomes, particularly in our current context of intensifying socioecological crises. Consequently, cultivating reciprocal relationships with natural disturbance regimes and ecosystem structures is crucial to Earth stewardship in the Anthropocene.

A document by Jessica Gephart. Click on the document to view its contents.

Natural methane gas release from the seafloor is a widespread phenomenon that occurs at cold seeps along most continental margins. Since their discovery in the early 1980s, seeps have been the focus of intensive research, partly aimed to refine the global carbon budget. However, deep-sea research is challenging and expensive and, to date, few works have successfully monitored the variability of methane gas release over long time periods (> 1 yr). Long-term monitoring is necessary to study the mechanisms that control seabed gas release. The M³ project, funded by the German Ministry of Education and Research, aims to study the temporal and spatial variability of gas emissions at the Southern Hydrate Ridge (SHR) by acoustically monitoring and quantifying gas effluxes over several years. Located 850 m deep on the Cascadia accretionary prism offshore Oregon, the SHR is one of the most studied seep sites and persistent but variable gas release has been observed for more than 20 years. Since 2015, the Ocean Observatories Initiative’s (OOI) Cabled Array observatory, provides power supply and two-way communication to the SHR, making it an ideal site for continuous long-term monitoring work. In this work, we present how we will take advantage of the OOI infrastructure and deploy several instruments on the seabed for at least 1.5 year. A multi-beam “overview” sonar mounted on a rotor will identify every gas bubble stream located within 200 m from the sonar location. A scanning “quantification” sonar will be used to estimate the amount of gas that is released from discrete gas streams. A camera system and a CTD probe will help process and analyze the hydro-acoustic data. All instruments will be powered and controlled from land through the OOI infrastructure. We present the instrument design, the operation protocol, as well as the data processing steps and expected results.

Diatoms are prominent eukaryotic photoautotrophs in today’s oceans. While dominant in nitrate-rich conditions, they face competitive exclusion by other phytoplankton when ammonium forms the bulk of bioavailable nitrogen. The extent to which this competitive exclusion defines diatom abundance worldwide and the consequences of potential future ammonium enrichment remain unexplored and unquantified. Here, using phytoplankton abundance proxies from the Tara Oceans dataset and an ocean-biogeochemical model, we demonstrate that ammonium enrichment reduces diatom prevalence in marine ecosystems at the global-scale. Under a high emission scenario, we anticipate 98% of the euphotic zones to experience ammonium enrichment by 2081-2100 and attribute a majority (70%) of future diatom displacement to competitive exclusion by other phytoplankton as bioavailable nitrogen supply shifts from nitrate to ammonium. Overall, the form of nitrogen emerges as a significant but previously underestimated stressor affecting diatoms and ocean ecosystems globally.

Automated monitoring and evaluation systems for plant phenotyping are one of the keys to advance and strengthen crop breeding programs. In this study, the improvements of the camera-based sensor system and a weather station from a previous study-assembled mainly from Raspberry Pi products-board with dual cameras (RGB and NoIR) providing high spatial and temporal resolution data-is outlined. Hardware for the internet connection and the power supply system of the sensor were upgraded. Previously, the sensor could automatically capture plant images following user-defined time points; thus, an image processing algorithm (edge computing) was developed and installed to extract digital phenotypic traits from the images after capturing process. With the development, the new sensor system could be integrated with the internet, and a cloud server was configured to store data online (digital traits and raw images). A real-time monitoring system was created to visualize the time series data of a trait development and plant images throughout the season. With such a system, plant breeders will be able to monitor multiple trials for timely crop management and decision-making process, which is also resources efficiency.

Seismology is a data-driven science with a huge amount of data gathered for over a century. Though seismic data recording started in 1900, the growth of seismic data has obviously been exponentially in the last three decades. This data growth can be easily noticed if one takes a close look at just one of the largest seismological data centers in the US, the Data Management Center (DMC) of the Integrated Research Institutions in Seismology (IRIS). Data at the DMC grew from less than 10 Tebibytes in 1992 to about 800 Tebibytes in 2022. With the availability of such a large amount of seismic data, it is paramount to develop new seismic data processing and management tools to help analyze and find new and better seismic models. Developing new big seismic data processing and management tools will be helpful to make the best use of such growing big seismological data sets. The main goal of this investigation is the development of efficient data manipulation and processing tools for retrieval, processing, merging, aggregation, and management of big seismic data from disparate data sources. In this study, such big seismic data processing tools are being developed using python programming language and open source python libraries, and the tools we are developing will be helpful to extract, split, and convert, merge and process big seismic data. In addition, python is very suitable for data science and has powerful libraries to process and manage data and applications. Significant contributions have been made in recent python based libraries for seismic data processing, though there are still some rooms for improvement when it comes to seismic applications to merge, convert, manage  and process big seismic data from disparate data sources and converting different file formats. Seismic data from different networks surrounding the Rio Grande Valley have been collected from different data sources. Our attempt is to test the developed tools and evaluate their performance. This study has made important progress in this regard and the results are promising. 

A document by Pradeebane VAITTINADA''AYAR. Click on the document to view its contents.

The Amazon rainforest has a great influence on the global energy balance and carbon fluxes, responsible for the net removal of approximately 4 million tons of carbon per year, via photosynthetic activity. Climate change and deforestation have impacts on the carbon budget in Amazonia, transforming CO2 sink areas into sources. Given the complexity of the factors that govern the carbon exchange in the Amazon and its influence on biological processes, the use of Data science strategies can promote a better understanding about the main environmental factors for different scenarios, and also, assist in public policies to mitigate the global warming effects. This study aims to identify the environmental factors that determine the temporal variability of carbon exchanges between the biosphere and the atmosphere in the Tapajós National Forest, in the Amazon, applying Data Science strategies in an integrated set of environmental data from energy and carbon fluxes and remote sensing data. The specific objective is to assess the influence of a selected set of environmental variables on the variability of carbon exchanges, with the use of an artificial neural networks classification model to identify the variables with great impact on source, sink and neutrality scenarios in Tapajós National Forest. Data Science strategies were applied to an integrated dataset of ground-based carbon flux measurements and remote sensing data, considering the period between 2002 and 2006. An artificial neural network (ANN) classification model was developed to identify the environmental variables with great impact on carbon source, sink and neutrality conditions. The average global score of ANN model was 65%. It was possible to identify the predictor variables with greatest impact to the carbon sink condition: radiation at the top of the atmosphere, sensible and latent energy fluxes and leaf area index. Thus, the ANN model with an ensemble of Data Science strategies can improve a better understanding of variability CO2 fluxes and be a powerful tool to promote new knowledge.

The oceanic storage of anthropogenic CO2 (Cant) that humans have emitted into the atmosphere has been pivotal for counteracting climate change. Yet multi-decadal trends in the ocean interior storage of Cant have not been assessed at global scale. Here, we determine storage changes of Cant by applying the eMLR(C*) regression method to ocean interior observations collected between 1989 and 2020. We find that the global ocean storage of Cant grew by 29 ± 3 Pg C dec-1 and 27 ± 3 Pg C dec-1 (±1σ) from 1994 to 2004 and 2004 to 2014, respectively. Although the two growth rates are not significantly different, they imply a reduction of the oceanic uptake fraction of the anthropogenic emissions from 36 ± 4 % to 27 ± 3 % from the first to the second decade. We attribute this reduction to a decrease of the ocean buffer capacity and changes in ocean circulation. In the Atlantic Ocean, the maximum storage rate shifted from the Northern to the Southern Hemisphere, plausibly caused by a weaker formation rate of North Atlantic Deep Waters and an intensified ventilation of mode and intermediate waters in the Southern Hemisphere. Between 1994 and 2004, the oceanic Cant accumulation exceeded the net air-sea flux by 8 ± 4 Pg C dec-1, suggesting a loss of natural carbon from the ocean during this decade. Our results reveal a substantial sensitivity of the ocean carbon sink to climate variability and change.

The development of distributaries in large river deltas plays an important role in the geology, hydrology, and ecology of the coastal ocean, as large rivers are a dominant mechanism by which particulate, suspended, and dissolved material is delivered from the continents to the global ocean. And yet, there is relatively little, near-real time observational data on the development of distributaries in large river deltas-in part because the development of modern observation, coincides with an era when rivers have been controlled by large engineering projects (i.e. the 20th and 21 st centuries). This article reports on Neptune Pass, the largest new distributary to form in the Mississippi River in nearly a century. It developed between 2019 and 2021 when a small canal rapidly expanded by at least an order of magnitude. The system now carries about 15-17% of the flow of the Mississippi River, > 3,000 m 3 s-1 when the Mississippi River is at moderately high flows-comparable to the 10th largest river in North American and the 100th largest river on Earth. Neptune Pass is building a delta, and this study sought to examine whether it is building is comprised largely of material eroded from the Neptune Pass (redistributed sediment hypothesis), or includes material recently derived from the Mississippi River (new sediment hypothesis). These hypotheses were tested using a combination of marine-geophysical surveys, remote sensing techniques, and sediment core collections. Results indicate that the delta in Quarantine Bay was 56 to 79% larger than the material excavated from Neptune Pass, corroborating the new sediment hypothesis, and indicating that it is a net land building system. These findings provide key insights that are critical to the restoration and safe management of the Mississippi River and its delta, the largest system of its kind in North America.

Anticipating climate impacts and risks in present or future climates requires predicting the statistics of high-impact weather events at fine-scales. Direct numerical simulations of fine-scale weather are computationally too expensive for many applications. While deterministic-based (deep-learning or statistical) downscaling of low-resolution climate simulations are several orders of magnitude faster than direct numerical simulations, it suffers from several limitations. These limitations include the tendency to regress to the mean, which produces excessively smooth predictions and underestimates the magnitude of extreme events. They also fail to preserve statistical measures that are key for climate research. We use a conditional GAN (cGAN) architecture to downscale daily precipitation as a Regional Climate Model (RCM) emulator. The cGAN generates plausible residuals on top of the predictable expectation state produced by a deterministic deep learning algorithm. The skill of cGANs is highly sensitive to a hyperparameter known as the weight of the adversarial loss (\(\lambda_{adv}\)), where the value of  \(\lambda_{adv}\) required for accurate results varies with season and performance metric, casting doubt on the reliability of cGANs as usually implemented. However, by applying a simple intensity constraint to the loss function, it is possible to obtain reliable performance results across \(\lambda_{adv}\) spanning two orders of magnitude. CGANs are considerably more skillful in capturing climatological statistics, including the distribution and spatial characteristics of extreme events. With this modification, we expect cGANs to be readily transferable to other applications and time periods, making them a useful weather generator for representing extreme event statistics in present and future climates.

The National Water Model (NWM) is a key tool for flood forecasting and planning and water management. Key challenges facing NWM include calibration and parameter regionalization when confronted with big data. We present two novel versions of high-resolution (~37 km2) differentiable models (a type of physics-informed machine learning): one with implicit, unit-hydrograph-style routing and another with explicit Muskingum-Cunge routing in the river network. The former predicts streamflow at basin outlets whereas the latter presents a discretized product that seamlessly covers rivers in the conterminous United States (CONUS). Both versions used neural networks to provide multiscale parameterization and process-based equations to provide structural backbone, trained them together (“end-to-end”) on 2,807 basins across CONUS, and evaluated them on 4,997 basins. Both versions show the great potential to elevate future NWMs for extensively calibrated as well as ungauged sites: the median daily Nash-Sutcliffe efficiency (NSE) of all 4,997 basins is improved to around 0.68 from 0.49 of NWM3.0. As they resolve heterogeneity, both greatly improved simulations in the western CONUS and also in the Prairie Pothole Region, a long-standing modeling challenge. The Muskingum-Cunge version further improved performance for basins >10000 km2. Overall, our results show how neural-network-based parameterizations can improve NWM performance for providing operational flood predictions while maintaining interpretability and multivariate outputs. We provide a CONUS-scale hydrologic dataset for further evaluation and use. The modeling system supports the Basic Model Interface (BMI), which allows seamless integration with the next-generation NWM.

A document by Yosuke Yamazaki. Click on the document to view its contents.

Neural Weather Models (NWM) are novel data-driven weather forecasting tools based on neural networks that have recently achieved comparable deterministic forecast skill to current operational approaches using significantly less real-time computational resources. The short inference times required by NWMs allow the generation of a large ensemble potentially providing benefits in quantifying the forecast uncertainty, particularly for extreme events, which is of critical importance for various socio-economic sectors. Here we propose a novel ensemble design for NWMs spanning two main sources of uncertainty: epistemic —or model uncertainty,— and aleatoric —or initial condition uncertainty. For the epistemic uncertainty, we propose an effective strategy for creating a diverse ensemble of NWMs that captures uncertainty in key model parameters. For the aleatoric, we explore the “breeding of growing modes” for the first time on NWMs, a technique traditionally used for operational numerical weather predictions as an estimate of the initial condition uncertainty. The combination of these two types of uncertainty produces an ensemble of NWM-based forecasts that is shown to improve upon benchmark probabilistic NWM and is competitive with the 51-member ensemble of the European Centre for Medium-Range Weather Forecasts based on the Integrated Forecasting System (IFS) —a gold standard in weather forecasting,— in terms of both error and calibration. In addition, we report better probabilistic skill than the IFS over land for two key variables: surface wind and air surface temperature.

A document by Srishti Singh. Click on the document to view its contents.

A document by Michel Wortmann. Click on the document to view its contents.

Accurately representing vertical turbulent fluxes in the planetary boundary layer is vital for moisture and energy transport. Nonetheless, the parameterization of the boundary layer remains a major source of inaccuracy in climate models. Recently, machine learning techniques have gained popularity for representing oceanic and atmospheric processes, yet their high dimensionality limits interpretability. This study introduces a new neural network architecture employing non-linear dimensionality reduction to predict vertical turbulent fluxes in a dry convective boundary layer. Our method utilizes turbulent kinetic energy and scalar profiles as input to extract a physically constrained two-dimensional latent space, providing the necessary yet minimal information for accurate flux prediction. We obtained data by coarse-graining Large Eddy Simulations covering a broad spectrum of boundary layer conditions, from weakly to strongly unstable. These regimes are employed to constrain the latent space disentanglement, enhancing interpretability. By applying this constraint, we decompose the vertical turbulent flux of various scalars into two main modes of variability: wind shear and convective transport. Our data-driven parameterization accurately predicts vertical turbulent fluxes (heat and passive scalars) across turbulent regimes, surpassing state-of-the-art schemes like the eddy-diffusivity mass flux scheme. By projecting each variability mode onto its associated scalar gradient, we estimate the diffusive flux and learn the eddy diffusivity. The diffusive flux is found to be significant only in the surface layer for both modes and becomes negligible in the mixed layer. The retrieved eddy diffusivity is considerably smaller than previous estimates used in conventional parameterizations, highlighting the predominant non-diffusive nature of transport.