Below-canopy downwelling shortwave radiation (SWR) influences snow ablation, soil microclimate, and carbon uptake, yet remains difficult to map in heterogeneous forests. To address this, we measured below-canopy SWR using a 122-m cable-borne mobile tram equipped with an upward-facing pyranometer along a transect spanning a subalpine forest-to-fen ecotone. Subsequently, airborne lidar-derived forest structure was used in a Lidar-to-Radiation canopy ray tracing model to simulate spatial patterns of SWR at 30-min timesteps across the full growing season (21 Mar. - 21 Sep. 2023). Simulated SWR demonstrated strong agreement with field measurements (Spearman ρ = 0.95). At 1-m horizontal resolution, the model captured tree- and gap-scale variability and growing-season contrasts. When applied to two neighboring 1 ha forest plots, simulated SWR had a monotonic, log-linear relation with sky-view factor (SVF) (R2 = 0.96 and 0.78). SVF adequately explained the mean tendency but not the fine-scale or temporal variation of SWR. Within a single SVF class (±5% SVF interval), growing-season SWR spanned wide ranges (e.g., 46-166 W m-2 at 10% SVF and 311-398 W m-2 at 50% SVF), suggesting additional geometric controls such as relative crown position, height, and shape that were not captured by SVF. Thus, ecohydrological extremes—early snow disappearance, soil moisture deficits, or persistent shade—can coexist within a narrow SVF range. Results demonstrate the ability of a three-dimensional ray tracing model to capture sunflecks, canopy edge shadows, and terrain masking in a subalpine forest. Model errors likely reflect sensitivity to woody element representation, meter-scale uncertainty in lidar geolocation and sensor positioning, and model parameters. Coupled observation and model systems provide a transferable method to scale results from meteorological stations to transects and landscapes. Integrating such detailed measurements with process-based modeling is essential for understanding how canopy structure shapes energy availability, soil moisture and plant water availability, and land-atmosphere interactions.
Cross-domain named entity recognition (NER) aims to adapt knowledge learned from the source domain to recognize named entities in the target domain with limited labeled data. One of its core challenges is the significant semantic gap existing between domains, which hinders the transfer of learned entity-type relations to new domains. Although syntactic knowledge can partly reduce domain discrepancies, existing works are mostly limited to token-level dependency relations and fail to directly capture the boundaries and associations of entities and their types as complete chunks, thereby limiting the ability of syntax to help models perceiving entity types in different domains. To address this issue, we proposed a cross-domain NER framework based on chunk-level dependency syntax (CDS) guided pre-training. Initially, we utilize large language models to generate task-oriented corpus that describes the relationships between entities and their types. Then, we design a CDS construction strategy that integrates a part-of-speech rules filter with dependency constraint to automatically identify entity and type chunks from the generated text. These chunks serve as self-supervised signals to guide the model in chunk-level masked pre-training, enabling it to learn chunk-level association patterns between entities and their types. Subsequently, we fine-tune the model on labeled data from both the source and target domains to enhance its adaptability across domains, and then apply it to perform NER in the target test set. Experimental results on six public datasets across different domains demonstrate that our proposed method consistently outperforms other baseline models, achieving average Micro-F1 improvements of 2.46% and 2.18% over the best comparative baselines in single-source and multi-source cross-domain settings, respectively.
Extracorporeal membrane oxygenation (ECMO) is increasingly being utilized for life-threatening cardiac and/or respiratory failure refractory to conventional treatment. Acute kidney injury (AKI) and fluid balance disorders commonly occur both before and during ECMO, with approximately half of cases receiving treatment with continuous renal replacement therapy (CRRT). Acute kidney injury, fluid balance disorders, and CRRT influence both short- and long-term outcomes in this population. The 36th Acute Disease Quality Initiative (ADQI) meeting was held in June 2025 to develop multidisciplinary international expert recommendations for AKI, fluid balance, and CRRT during ECMO across the age spectrum. This work encompassed five working groups: 1) epidemiology, risk factors, and outcomes of AKI and CRRT, 2) fluid management and outcomes, 3) indications for CRRT and fluid removal during ECMO, 4) best practices for performing CRRT during ECMO, and 5) biomarkers, extracorporeal blood purification, and drug pharmacokinetics and pharmacodynamics. As part of this work, knowledge gaps and research priorities were identified.
Using tick-by-tick data, this study examines intraday liquidity of WTI, Brent, and INE crude oil futures. US EIA inventory announcements significantly affect intraday returns and, to a lesser extent, liquidity across all three markets. We document strong commonality in intraday liquidity, largely driven by supply-side factors, and find that WTI is the dominant source of liquidity spillovers, a pattern not mirrored in return spillovers. These findings highlight different mechanisms behind liquidity commonality and return comovements. The results are robust to additional control variables and alternative liquidity measures.
Endocrine-disrupting chemicals (EDCs) pose urgent threats to public and environmental health, disproportionately affecting marginalized communities. Yet most EDC research training remains siloed within disciplines like epidemiology, toxicology, or molecular biology, limiting cross-disciplinary collaboration. Despite rising awareness of EDCs’ widespread impact, traditional academic training fails to equip scientists for the interdisciplinary, community-engaged, and policy-oriented work these challenges demand. We propose a training framework based on four pillars: (1) Interdisciplinary Science to integrate diverse approaches; (2) Community Engagement Reflexivity to ensure research aligns with community needs; (3) Science Communication to share findings clearly and combat misinformation; and (4) Public Health Policy to turn evidence into action. The aim is not to master every method but to foster collaboration across fields. We highlight this model through a case study of the Marine Biological Laboratory’s “Endocrine Disrupting Chemicals: Hazards and Opportunities” (ECHO), a two-week course uniting scientists from various disciplines for immersive training. This approach empowers researchers to bridge science and policy, promote environmental justice, and address the complex realities of chemical exposures and environmental inequities.
