Boreal forests play a critical role in the global carbon cycle as they are one of the largest terrestrial carbon sinks globally. In this study, we employ explainable machine learning (ML) techniques to investigate the influence of environmental and vegetation variables on net ecosystem exchange (NEE), focusing specifically on the effects of diffuse radiation. We utilize a sub-hourly resolution data set including satellite and in-situ observations from three boreal or hemiboreal forest research stations across the latitudes 58–68° N to capture latitudinal variability in forest carbon uptake. Using SHAP (Shapley Additive Explanations) values, we identify key drivers influencing NEE and quantify their importance across various ML model architectures. Photosynthetically active radiation (PAR), diffuse radiation, normalized difference vegetation index, and soil temperature were identified as the variables having the largest explanatory power for NEE across the ML models. ML models using only these variables result in $ {R}^2\approx 0.8 $ and RMSE$ \approx 2.3 $ $ \mu \mathrm{mol}\;{\mathrm{m}}^{-2}\hskip0.1em {\mathrm{s}}^{-1} $. Further analysis of SHAP values indicates that higher diffuse radiation (DiffPAR) is associated with more negative NEE (stronger carbon sink). SHAP analysis highlights this effect much more clearly than raw DiffPAR measurements, because it accounts for interactions with other environmental factors. This suggests that the diffuse radiation effect emerges from interactions between DiffPAR and co-varying environmental factors (such as cloudiness and total PAR). Cases identified by SHAP ($ \mathrm{DiffPAR}\ \mathrm{SHAP}<-2 $) have a median NEE $ 1.55\mu \mathrm{mol},{\mathrm{m}}^{-2},{\mathrm{s}}^{-1} $ more negative than cases with raw $ \mathrm{DiffPAR}\ge 400 $. When applied to ML models, SHAP uncovers nonlinear, context-dependent interactions between diffuse radiation and other drivers of NEE without assuming a priori relationships.

Cost planning for Product-Service Systems faces rising complexity, making life-cycle cost estimates essential. This paper investigates how machine learning (ML) can be applied for life-cycle cost estimation in product development. A literature review was conducted to identify ML-based methods, classify them across life cycle phases, and compare them against traditional methods. Results show that traditional models remain transparent but limited in early stages, while ML methods achieve higher accuracy in data-rich phases. A clear research gap exists for hybrid models and end-of-life costing.

Conformal prediction (CP) can yield statistically valid prediction intervals for any regression model, with no model modifications and small computational costs. To assess its practical value, we apply conformal methods to quantify uncertainty in machine learning emulators of six microphysical process rates (MPRs). MPRs describe small-scale processes in atmospheric clouds such as precipitation formation and aerosol–cloud interactions and help understand weather and climate. The emulators are trained on simulation output from the ICOsahedral Nonhydrostatic (ICON) model in a limited-area numerical weather prediction configuration. We compare split CP for deterministic emulators with conformalized quantile regression (CQR) for quantile regression (QR) emulators. Both CP methods yield well-calibrated and sharp prediction intervals on average, but CQR provides more consistent intervals across several orders of magnitude, making it preferable for the uncertainty quantification of climate variables.

Recent advances in machine learning (ML) offer substantial potential for product development (PD), yet adoption remains limited. A crucial step is identifying suitable ML algorithms for a given PD problem, which requires translating domain-specific formulations into appropriate ML tasks. Prior work indicates that LLMs struggle with this step due to insufficient domain knowledge. Therefore, this study investigates whether a domain-specific GraphRAG approach improves model performance by enriching prompts with structured context from a PD knowledge graph.

The operational weather radar network is highly suitable for large-scale bird monitoring and early warning systems. This study discusses, compares and contrasts current state-of-the-art technologies for large-scale bird surveillance – including manual observation, satellite tracking, dedicated avian radar and the operational weather radar network – evaluating their respective strengths, limitations and applicability for different monitoring scenarios. Key research questions focus on how to extract reliable bird movement information from weather radar data and how such data can support ecological studies and aviation safety. From the perspective of radar signal analysis, we examine the characteristics of bird echoes in terms of reflectivity, altitude distribution, flight speed and direction. We further review and compare methodologies for distinguishing bird echoes from weather and ground clutter, including clutter suppression, meteorological echo removal, feature extraction, machine learning classification and cross-validation with auxiliary data. Our synthesis demonstrates that weather radar can effectively identify and track bird movements over wide areas, providing valuable data on migration patterns and flight behaviours. These findings have practical significance for applications in avian ecology research and bird strike prevention in aviation.

We apply machine learning methods to predict Thoroughbred yearling auction prices at the Keeneland September Sale (2020–2024). Our sample includes 5,788 yearling prices with pedigree data. We use both linear and tree-based models to predict log prices. We use cross-validation to tune model hyperparameters and select Ridge regression (α = 1.451) as the primary model for interpretation given its stability and interpretability. The Ridge regression explains approximately 54% of out-of-sample variation (R2≈ 0.5403). Sire and Dam Reputation emerge as the dominant predictors. Results provide pricing benchmarks and show how reputation and session structure shape Thoroughbred yearling auction prices.

In this article I am concerned with interrogating the intersection between the Newtonian and Cartesian intellectual inheritances of AI and machine learning, and ideas about the ethics of war. As militaries turn to new and emerging technologies to maintain or achieve a technological edge over their perceived adversaries, they create new imaginaries of future war—alongside the technologists, academics, and defense scientists crafting new terms, ideologies, and frameworks for making sense of these technologies. In this article I will argue that the intellectual inheritances of machine learning strengthen certain pre-existing tendencies of thinking about ethics and war that function to push the experience of war, particularly for those subjected to it, to one side. The first of these is ethics as code, which in its most extreme form seeks to quantify ethics. The second is ethics as identity, in which we see the reduction of complex ethical debates to a simple belief that “we” are the ethical actors and the “other” is not. To combat the expansion of militarism that these narratives enable we must foreground the experience of war, both of those subject to it and of those creating the conditions for war.

In this study, experimental deep reinforcement learning (DRL) control of aerofoil flow separation is conducted at a chord-based Reynolds number of ${\textit{Re}}_c=1.5\times 10^5$. A dielectric barrier discharge plasma actuator mounted at the leading edge and a hotwire placed in the separated shear layer act as the flow disturber and the state monitor, respectively. The closed-loop control law is parameterised by a radial basis function network and executed in real time on a field-programmable gate array at 1 kHz. With the aid of a deep deterministic policy gradient algorithm, a satisfying closed-loop control strategy can be derived in less than one minute, and the resulting lift coefficient increment (21 %) is similar to that achieved by the best open-loop control. For stable and effective DRL control, the sensor should be placed at the position with strongest velocity fluctuation, and both time delay and cashing period should be accounted in the reward design. Incorporating historical sensor measurements are beneficial to DRL control, and the optimal history length is approximately equal to the ratio of the local Taylor microscale to the control period. The final control law sought by DRL dictates that strong plasma actuation should only be applied when the flow separation region contracts to a minimum. Statistically, control benefits can be ascribed to both the increase of average reward at each cluster and the elevation of occurrence probabilities of high-reward clusters. Physically, suppression of the aerofoil flow separation is attributed to the excitation of large-scale shear layer vortices, which promotes the momentum exchange between the free stream and the reverse flow.

This work aims to investigate experimentally the control of a turbulent boundary layer over a flat plate based on a genetic-algorithm (GA)-based control system. A novel actuator is developed in house, consisting of one array of 11 spanwise-aligned synthetic minijets through longitudinal miniature slits, each independently controlled in terms of its exit momentum coefficient Cμ, frequency fe and phase ϕ, while nine wall wires are placed downstream of the actuator to measure a variation in wall friction. A GA is employed for the unsupervised learning of near-optimal control strategies, i.e. the optimal ϕ of the synthetic jets, with fe and Cμ optimized first. The learning process unveils three typical control strategies that may attain substantial drag reduction (DR) in both actuation and downstream regions, i.e. conventional uniform forcing (CUF), GA-I and GA-II. The latter two are characterized by the triangle- and trapezoid-like distributions of ϕ, respectively, thus each producing a spanwise motion. The three strategies achieve spanwise-averaged local DR by 45 %, 52 % and 52 %, respectively. The downstream drag-reducing areas exhibit an appreciable difference between the three cases, with CUF recovering drag more rapidly than the other two. It has been found that the three cases are associated with different DR mechanisms and flow physics, which account for the distinct control performances.

In a non-generative approach to artificial intelligence in an artistic practice, this work looks at mapping processing data from artificial neural networks (NNs) onto sound and visuals. One aim of this practice-based piece of research is to paradoxically offer insights into how these ubiquitous, yet notoriously opaque algorithms operate, by exposing the audience to the intrinsic unintelligibility of their processes. The other is to use these vast amounts of abstract data as a creative starting point for audiovisual artworks, referring to aesthetic traditions that have emerged from the need to make use and potentially make sense of such extensive masses of information, and from ones that have developed sounds that have gradually become associated with digital and post-digital worlds and other exterior and abstract concepts. At the heart of the whole work is a link and cross-fertilisation between the use of sounds and visuals aesthetically associated with errors and digital malfunction and the use of actual ‘waste’ data (from NN training), which acts as a trace of their operation.