Machine learning is increasingly used to simulate and predict complex urban phenomena, such as CO₂ emission patterns in cities. However, the effectiveness, biases, and uncertainty of these predictive models are not well understood. This study systematically evaluates 18 different machine learning algorithms using a rich dataset of county-level CO₂ emissions in China from 2001 to 2020, incorporating 41 diverse features spanning socioeconomic, environmental, and energy consumption factors. The analysis reveals a widespread issue: nearly all models, regardless of type, tend to systematically overestimate emissions in Chinese county-level cities across both space and time. By isolating the effects of feature selection and quantifying uncertainty through bootstrap resampling, the research demonstrates that aggressive dimensionality reduction often harms model performance, while retaining a high number of features is crucial for capturing the true complexity of urban emissions. Tree-based models, such as CatBoost and LightGBM, outperform others in accuracy and uncertainty quantification. These findings highlight the urgent need for the scientific and policy communities to prioritize diagnosing and correcting systematic biases in machine learning models, rather than focusing solely on improving accuracy, to ensure reliable guidance for urban carbon management and climate policy.
Soil phosphorus (P) availability sustains nutrient supply, plant growth, and community persistence in alpine treeline ecosystems. Most studies emphasize spatial gradients, leaving within-season reorganization of individual P fractions and functional pools poorly resolved. We sampled 0–10 and 10–20 cm soils across three mountains and four vegetation belts in June, August, and October 2021 on the southeastern Qinghai-Tibet Plateau. Hedley fractionation separated six P fractions, which we grouped into Labile P, Moderately labile P, and Stable P. Fixed-effects models, partial η2, and sequential sums-of-squares shares compared monthly and spatial effects. The fractions changed asynchronously through the growing season. Labile P increased from June to October but retained stronger mountain and vegetation signals; Moderately labile P increased late in the growing season; and Stable P declined in October. The decline in Stable P was driven mainly by Residual P, even though HCl-P increased. The first two principal components explained 83.9% of the variation and separated samples mainly by month. Month had strong effects on Moderately labile P (partial η2 = 0.322) and Stable P (0.512), whereas mountain had a stronger effect on Labile P (0.141). Month SS shares were 73.2% for NaOH-Pi, 44.4% for NaOH-Po, 45.8% for Residual P, 27.7% for Moderately labile P, and 42.5% for Stable P, whereas mountain and vegetation effects were more pronounced for NaHCO₃-Po and Labile P. Month × vegetation significantly affected several fractions. Redundancy and relative-importance analyses further showed that the dominant soil correlates changed among months. These results reveal rapid within-season redistribution among alpine treeline ecotone soil P pools and show that temporal and spatial controls differ among fractions.
Relay transport has emerged as a promising collaborative strategy in which trucks exchange trailers or drivers at designated nodes before continuing to their destinations. This operational mode helps mitigate long-standing inefficiencies in traditional direct transport, including low operational efficiency, high fuel consumption, and poor driver working conditions. This study aims to investigate the factors influencing trucking companies’ adoption of relay transport under uncertainty. Because shipment distances shape delivery requirements and operational characteristics, two stated choice experiments were designed for short-haul (300 km) and long-haul (600 km) shipments. Each experiment included three labeled alternatives: direct transport, relay with driver exchange, and relay with trailer exchange, varied by shipment characteristics, relay time, uncertain waiting time, driver overtime, and cost. Using survey data from 275 Japanese trucking managers, a cumulative prospect theory (CPT) choice model with random parameters and error components was estimated to capture companies’ heterogeneous decisions under uncertainty. The results identify factors influencing relay transport adoption in each experimental context. Uncertain waiting time, cargo type, vehicle type, relay time, shipment frequency, driver overtime, cost, and company characteristics (firm and fleet size) significantly affect adoption and choice of exchange method (drivers or trailers). These findings provide practical insights for policymakers and logistics authorities seeking to promote relay transport adoption.
The rapid uptake of battery electric vehicles (BEVs) has intensified the demand for accessible and efficient parking and charging infrastructure, particularly in dense urban environments. Leveraging household-based sharing of such facilities offers a potential solution to alleviate infrastructure constraints, yet drivers’ adoption behavior remains largely underexplored. This study develops an error component mixed logit (ECML) model, estimated with stated preference data, to jointly examine BEV drivers’ parking and charging choices across household-shared and conventional public-open facilities, explicitly accounting for unobserved heterogeneity, correlated error structures, and waiting time uncertainty. The results reveal significant preference heterogeneity that female, highly educated, and sharing-aware drivers exhibit stronger preferences for household-shared facilities. Historical driving patterns further influence BEV drivers’ parking and charging behavior, with high-mileage drivers motivated by cost savings. Moreover, activity-travel contexts, such as state-of-charge (SoC), unplanned activity duration, and time pressure significantly affect decisions, whereas cost- and time-related attributes exert negative effects, as confirmed by elasticity analysis. Nest effects captured through error components indicate correlated preferences across household-shared, public, and integrated options, emphasizing the need for coordinated facility design. These findings highlight practical interventions such as targeted pricing strategies, access improvement, and real-time information systems to enhance the adoption of household-shared facilities and improve urban BEV charging accessibility.
