A general deep learning model for bird detection in high-resolution airborne imagery.

Advances in artificial intelligence for computer vision hold great promise for increasing the scales at which ecological systems can be studied. The distribution and behavior of individuals is central to ecology, and computer vision using deep neural networks can learn to detect individual objects in imagery. However, developing supervised models for ecological monitoring is challenging because it requires large amounts of human-labeled training data, requires advanced technical expertise and computational infrastructure, and is prone to overfitting. This limits application across space and time. One solution is developing generalized models that can be applied across species and ecosystems. Using over 250,000 annotations from 13 projects from around the world, we develop a general bird detection model that achieves over 65% recall and 50% precision on novel aerial data without any local training despite differences in species, habitat, and imaging methodology. Fine-tuning this model with only 1000 local annotations increases these values to an average of 84% recall and 69% precision by building on the general features learned from other data sources. Retraining from the general model improves local predictions even when moderately large annotation sets are available and makes model training faster and more stable. Our results demonstrate that general models for detecting broad classes of organisms using airborne imagery are achievable. These models can reduce the effort, expertise, and computational resources necessary for automating the detection of individual organisms across large scales, helping to transform the scale of data collection in ecology and the questions that can be addressed.

Kelpwatch: A new visualization and analysis tool to explore kelp canopy dynamics reveals variable response to and recovery from marine heatwaves.

Giant kelp and bull kelp forests are increasingly at risk from marine heatwave events, herbivore outbreaks, and the loss or alterations in the behavior of key herbivore predators. The dynamic floating canopy of these kelps is well-suited to study via satellite imagery, which provides high temporal and spatial resolution data of floating kelp canopy across the western United States and Mexico. However, the size and complexity of the satellite image dataset has made ecological analysis difficult for scientists and managers. To increase accessibility of this rich dataset, we created Kelpwatch, a web-based visualization and analysis tool. This tool allows researchers and managers to quantify kelp forest change in response to disturbances, assess historical trends, and allow for effective and actionable kelp forest management. Here, we demonstrate how Kelpwatch can be used to analyze long-term trends in kelp canopy across regions, quantify spatial variability in the response to and recovery from the 2014 to 2016 marine heatwave events, and provide a local analysis of kelp canopy status around the Monterey Peninsula, California. We found that 18.6% of regional sites displayed a significant trend in kelp canopy area over the past 38 years and that there was a latitudinal response to heatwave events for each kelp species. The recovery from heatwave events was more variable across space, with some local areas like Bahía Tortugas in Baja California Sur showing high recovery while kelp canopies around the Monterey Peninsula continued a slow decline and patchy recovery compared to the rest of the Central California region. Kelpwatch provides near real time spatial data and analysis support and makes complex earth observation data actionable for scientists and managers, which can help identify areas for research, monitoring, and management efforts.

Quantity and fate of synthetic microfiber emissions from apparel washing in California and strategies for their reduction.

Synthetic microfibers have been identified as the most prevalent type of microplastic in samples from aquatic, atmospheric, and terrestrial environments across the globe. Apparel washing has shown to be a major source of microfiber pollution. We used California as a case study to estimate the magnitude and fate of microfiber emissions, and to evaluate potential mitigation approaches. First, we quantified synthetic microfiber emissions and fate from apparel washing in California by developing a material flow model which connects California-specific data on synthetic fiber consumption, apparel washing, microfiber generation, and wastewater and biosolid management practices. Next, we used the model to assess the effectiveness of different interventions to reduce microfiber emissions to natural environments. We estimate that in 2019 as much as 2.2 kilotons (kt) of synthetic microfibers were generated by apparel washing in California, a 26% increase since 2008. The majority entered terrestrial environments (1.6 kt), followed by landfills (0.4 kt), waterbodies (0.1 kt), and incineration (0.1 kt). California's wastewater treatment network was estimated to divert 95% of microfibers from waterbodies, mainly to terrestrial environments and primarily via land application of biosolids. Our analysis also reveals that application of biosolids on agricultural lands facilitates a directional flow of microfibers from higher-income urban counties to lower-income rural communities. Without interventions, annual synthetic microfiber emissions to California's natural environments are expected to increase by 17% to 2.1 kt by 2026. Further increasing the microfiber retention efficiency at the wastewater treatment plant would increase emissions to terrestrial environments, which suggests that microfibers should be removed before entering the wastewater system. In our model, full adoption of in-line filters in washing machines decreased annual synthetic microfiber emissions to natural environments by 79% to 0.5 kt and offered the largest reduction of all modeled scenarios.

SubsidyExplorer: A decision-support tool to improve our understanding of the ecological and economic effects of reforming fisheries subsidies.

The magnitude of subsidies provided to the fishing sector by governments worldwide is immense-an estimated $35.4 billion USD per year. The majority of these subsidies may be impeding efforts to sustainably manage fisheries by incentivizing overfishing and overcapacity. Recognizing the threat these subsidies pose, the World Trade Organization has set a goal of reaching an agreement that would end fisheries subsidies that contribute to overcapacity, overfishing, and illegal fishing. However, negotiations have been hampered by uncertainty around the likely effects of reforming these subsidies. Here we present a novel method for translating a bioeconomic model into an interactive online decision support tool that draws upon real-world data on fisheries subsidies and industrial fishing activity so users can directly compare the relative ambition levels of different subsidy reform options.