Bowdoin College

Abstract This study explores the application of data-driven approaches to enhance the interpretation of geophysical wellbore data for carbon capture and storage (CCS) in the Illinois Basin. Extensive data from exploration and pilot-scale projects are analyzed to assess their potential for generating synthetic datasets that can effectively replicate some missing information for ongoing projects. Specifically, various machine learning (ML) models, including random forest, gradient boosting, feed-forward neural networks, and others, are employed to predict sonic velocities based on the mineralogical composition of participating rock and to identify fracture locations using petrophysical logs. ML-based approaches significantly outperform the traditional effective media interpretation, exhibiting improved accuracy in sonic velocity prediction and capturing small-scale heterogeneity of the formations. However, the prediction of fracture presence remains challenging due to data imbalance and the complex interplay between fractures and inherent heterogeneity of geologic formations. Application of t-distributed stochastic neighbor embedding (t-SNE)—an advanced technique for multidimensional data visualization—reveals that spatial heterogeneity strongly influences geophysical properties, limiting ML models’ performance in detecting fractures. Despite these challenges, ML approaches show promise in complementing traditional methodologies, enabling faster and more informed decision-making processes during early project stages. Future research is needed to address data limitations and enhance the reliability of ML models in diverse geological settings.

Seaweed farming offers a sustainable solution to malnutrition and economic challenges in southwestern Madagascar. This study assessed the seasonal performance of 12 edible seaweed species and optimized cultivation parameters for Gracilaria corticata and Gracilaria salicornia using long-line aquaculture. Environmental factors, including temperature, light intensity, salinity, and current speed, were monitored across warm and cool seasons. G. corticata, G. salicornia, and Hormophysa cuneiformis exhibited the highest retention rates (>50 %) and growth performance, with G. corticata demonstrating significant superior growth in both seasons. Smaller cuttings (0–50 g) achieved higher retention and specific growth rates, particularly in the first 30 days. Growth was significantly influenced by light intensity (4209.35 ± 3274.72 lux in the warm season; 3107.77 ± 3194.22 lux in the cool season) and temperature (30.20 ± 1.26 °C in the warm season; 24.83 ± 0.35 °C in the cool season), with optimal conditions occurring in the warm season. These findings highlight the potential of G. corticata and G. salicornia for sustainable aquaculture, contributing to species diversification and food security. Optimizing environmental conditions and cutting sizes is crucial for maximizing seaweed productivity in Madagascar.

Election campaigns increasingly pursue their strategic goals online, using digital advertising not only to persuade voters but also to raise funds, mobilize supporters, and collect data. These varied objectives reflect how campaigns operate within broader networks of candidates, parties, and outside groups. The extended party network forms the strategic backbone of modern U.S. campaigns, yet its internal division of labor in paid advertising has remained largely unmeasured. By classifying the goals of over 377,000 Facebook and Instagram ads from the 2022 US elections, we provide empirical evidence about the strategic behavior of candidates, parties, and outside groups. To this end, we build on previous work to extend and refine a taxonomy of nine goals for online election ads: acquisition, contact, donate, event, learn, persuade, poll, purchase, and vote. We analyze how advertisers allocate spending across goals and show that some — notably persuasion, donation, and learn — receive far greater spending than others. We also show that different types of sponsors often pursue different goals. For example, candidate advertisers devote a much higher share of their ads to fundraising than non-candidate advertisers, highlighting the costs of running for political office in the US. We also find that ad goals vary depending on the timing of the ad, the type of sponsor, and the characteristics of the targeted audience. Together, these findings offer the first systematic multi-sponsor account of how digital advertising reflects a functional division of labor within the extended party network that structures contemporary U.S. campaigns.

Abstract Background Timely colonoscopy completion in primary care can prevent diagnostic delays in colorectal cancer. Factors that influence why patients experience timely or delayed colonoscopy completion are unclear. We sought to identify potentially intervenable factors associated with earlier (or later) colonoscopy test completion in primary care. Methods All colonoscopy orders placed by primary care clinicians in two clinics within a single hospital system between January 1 2018 and December 31 2021 were examined for time to completion using Cox Proportional Hazards Model, where the hazards of completion were adjusted for variables potentially associated with the outcome, including sociodemographic, individual health-related, and care delivery factors. Results Among 10,576 colonoscopy tests ordered, 56% were completed within one year. After multivariable adjustment, earlier colonoscopy completion was associated with receiving care at a community health center, preferred spoken language other than English, male sex, Black race, any college education, a diagnosis of rectal bleeding, and documented use of an electronic patient portal. Completion occurred later among patients with Medicaid insurance, subsidized commercial insurance, or depression, and among patients whose tests were ordered by a nurse practitioner or a resident, or during a telehealth appointment. Results were similar for a sensitivity analysis restricted to those patients within a Medicaid ACO, except for the finding that Black race was no longer associated with test completion. Conclusions Based on our findings, certain factors associated with –early or delayed- colonoscopy test completion lend themselves to possible interventions. Targeted care navigation (e.g. reminder calls, assisted scheduling, transportation assistance) may be needed for patients whose colonoscopy tests are ordered at telehealth appointments or by NPs or residents, or who have certain insurance types (e.g. Medicaid and high-deductible insurance). Harnessing strategies used at community health centers to improve culturally competent test scheduling may also be a promising area of future work to help patients book and attend colon cancer test appointments. More study is needed to understand the observed relationship between electronic portal use and test completion and why some patient characteristics such as non-English preference are associated with earlier completion of ordered tests in our study.

Abstract Observations on the North Slope of Alaska have revealed patches of Sphagnum peat within the widespread matrix of tussock tundra on mineral soils. Little is known about the developmental history of these Sphagnum patches and whether they represent incipient peatlands established in response to warming‐related environmental changes. Nine peat cores were collected from nine Sphagnum ‐dominated peat patches spanning an approximately 300‐km longitudinal gradient on the North Slope to determine their development and establishment history. Stratigraphically constrained cluster analysis was applied to plant macrofossil data, carbon‐to‐nitrogen ratios, and total organic matter measured from bulk peat to delineate developmental phases, and radiocarbon dating was used to constrain the timing of Sphagnum peat patch establishment. We compared these data to changes in testate amoeba community composition and amoeba‐inferred water‐table depth and pH in six of the peat cores. We also compared Sphagnum peat‐patch development and establishment history to paleoclimate and local instrumental temperature records. Results indicated a predictable pattern that describes the transition from moist tussock tundra to Sphagnum peat. Furthermore, although Sphagnum has been present on the North Slope for millennia, our data suggest that Sphagnum ‐dominated peat patches constitute recent landscape features, mainly established in the 1800s and 1900s, and with rapidly increasing Sphagnum abundance in the past 50 years. Sphagnum expansion was associated with pronounced changes in testate amoeba communities, including an increase in mixotrophic taxa and species associated with densely growing Sphagnum, and community changes consistent with drying and increased acidity. The recent development of Sphagnum ‐dominated peat patches has been associated with warming air and soil temperatures, active layer deepening, and earlier snowmelt. Sphagnum expansion has also been observed in other arctic regions, and understanding the extent and growth potential of Sphagnum peat patches has implications for understanding and anticipating changes in carbon cycling, edaphic conditions, permafrost thermal regimes, and floristic diversity.

ABSTRACT Studies in proxy wars, interventions, counterinsurgency, and security force assistance often suggest that foreign patrons employ coercive methods like aid conditionality or direct rule to pressure defiant local partners. Yet large-scale U.S. military interventions show persistent aversion to coercing critical partners. Why? Examining primary source U.S. documents from Vietnam and Iraq, I argue increased U.S. commitment to partner survival raises perceived risks of coercive approaches. Promoting local institutional competency to expedite U.S. withdrawal, ensuring future access to local partners, and endorsing local sovereignty to sustain a non-colonial intervention narrative are deemed more important than promoting specific policy reforms through coercion.

When a large language model (LLM) has been trained on text featuring social biases, those biases implicitly impact the outputs of the model. Training an LLM on sanitized content, i.e., those pieces of content which remain after being subjected to state censorship (including alterations, deletions, and self-imposed censorship), results in what we term censorship bias. A model impacted by censorship bias may be less likely to reflect views that are routinely prohibited and more likely to reflect views that are not. This may particularly be an issue when interfacing with a model in a language that is predominantly used in a region with strong censorship laws. In this work, we outline what censorship bias is, introduce a novel methodology for identifying and measuring it, and apply that methodology to evaluate the most popular current LLMs. As part of the contributions of this work we designed and evaluated CensorshipDetector, a Chinese language text classification model which we use as part of our experimental design. Our evaluation of CensorshipDetector found it to be 91% accurate at differentiating between sanitized content and non-sanitized content. Our testing revealed evidence of censorship bias across all of the models we evaluated. Finally, we outline the potential harms of censorship bias, namely the exportation of information manipulation that would have primarily harmed a domestic audience to diaspora, as well as recommendations to various stakeholders to limit the harms of censorship bias and prevent it in the future.