In an exponentially weighted average, when the time step $t$ is small, the estimation only considers a few data points. This can cause an initial bias towards smaller values, especially if the initial value is set to $v_0 = 0$. Bias correction adjusts these early estimates to provide a more accurate trend. The bias-corrected value $v_t'$ is calculated as:

$v_t' = \frac{v_t}{1 - \beta^t}$

where $v_t$ is the uncorrected exponentially weighted average and $\beta$ is the weighting parameter. For example, as shown in the accompanying diagram, without correction, the early temperature estimates (the purple curve) indicate artificially low values due to the influence of the zero initialization, whereas the bias-corrected estimates (the green curve) better approximate the true underlying trend.