The key idea here is that a hub-and-spoke network concentrates traffic through a central hub, which streamlines routing and allows for better cargo consolidation. By funneling shipments from many origins to a common hub, you can bundle more cargo into each longer flight and then distribute it onward from the hub. This reduces the number of distinct flight legs needed to connect all destinations and lets aircraft operate closer to full capacity more of the time. With higher load factors on the main legs and fewer unique routes to maintain, overall flights required for moving the same volume of cargo can be lower, which also helps reduce fuel burn per unit shipped. Direct routes between every city pair aren’t the goal of a hub-and-spoke design; that kind of direct connectivity is more characteristic of a point-to-point network, which would require many more separate flights and routes. Lower maintenance costs aren’t the primary benefit of this structure, since maintenance depends on fleet usage, costs, and other factors rather than the network design itself. Longer total transit times can occur due to transfers at the hub, which is a trade-off the hub-and-spoke model accepts in exchange for the efficiency and capacity gains.

A long-term service arrangement like this is typically tied to when the partnership is formally established and committed to a decade of collaboration. In this case, the 10-year agreement to provide the CMI service with Amazon Air began in 2020. That year aligns with a period of intensified expansion and formalization of partnerships around Amazon Air, making 2020 the point at which a decade-long commitment could realistically start. The earlier years don’t fit the timeline of initiating and locking in this specific long-term arrangement.

Time series forecasting uses data points ordered in time to predict future values, making it the category that directly relies on patterns within the historical sequence—like trend, seasonality, and serial correlation. This approach is distinct from regression, which uses external predictors to estimate the target variable, and from simulation, which generates many possible outcomes through randomized modelling. Because the question asks for the forecast model type that fits the time-ordered data approach, time series forecasting models are the appropriate category.

Time series models are built specifically to forecast future values from data that are indexed in time, using patterns like trend, seasonality, and autocorrelation to make predictions about upcoming periods. They estimate parameters from past observations and then project those patterns forward, often providing uncertainty intervals for the forecasts. Other methods shown can be used for prediction, but they don’t inherently embody the time-ordered forecasting framework. Decision trees predict outcomes by splitting on features and don’t automatically account for temporal structure unless you create time-based features. Monte Carlo methods simulate many possible outcomes to explore risk and variability rather than producing a direct time-forward forecast. Neural networks can forecast when structured for sequential data (like LSTMs), but time series modeling remains the standard category specifically designed for forecasting future observations from time-ordered data.