Why do Decision Trees struggle with extrapolation?

Decision Trees cannot extrapolate beyond training data because they make piecewise constant predictions.

Decision Trees partition the feature space into finite regions and assign a constant value (or class) to each region. This means predictions outside observed ranges are still mapped to the nearest learned region, without trend extrapolation. Unlike linear or neural models, trees do not learn continuous functional relationships. This makes them strong for interpolation but weak for forecasting or extrapolation tasks like long-term time series prediction.