Noise: Residuals after a model

Figure 1 shows deviations of individual observations for CD 96 from the linear regression line fitted to the data using CD76 as predictor. This error term (difference between individual observations and the modeled values) might be considered noise. If the linear regression could be considered a close representation of underlying signal, the fitted model should result in an error term that would be distributed normally with mean 0 and a variance that can be expressed as an uncertainty in the forecasts.

Scatter plot showing regression line and its errors
Figure 1. Errors of the linear regression model for modeling temperature at CD 92 using CD 76.