Correlation Plots
We can also see this in the correlation plots. Here, the correlation trend with lead time for the fast-response basin group (right side) shows a trend away from the perfect correlation coefficient of 1 as lead time increases. Also note that when compared to the slow-response basin group, the fast-response basins show less favorable correlation. The change to less favorable correlation also occurs more quickly with the fast- response group.
So there is a degradation of correlation between observations and forecasts with increasing lead time. As we saw with mean error and mean absolute error, this degradation in forecast quality occurs more rapidly with the fast-response basins than it does with the slow-response basins.
This is why it is important to produce different scatter plots (and different verification metrics) for various forecast time horizons. The performance of the forecasts will vary greatly with lead time. When pooling together all forecasts and observations from short to long forecast horizons, one might miss important information about forecast performance.