Fig. 4: The quantified contribution (bivariate-mean-square difference; BMSD) of each input variable for 5, 10, 15, 20, 25, and 30-day forecast lead time.
From: Deep learning reveals moisture as the primary predictability source of MJO
The BMSDs are calculated between the original prediction and the prediction after removing the effect of a single variable using the CNNOBS for each forecast lead time. The gray bars indicate the sum of the BMSDs for all variables. The red, yellow, green, blue, and purple bars indicate the proportion of BMSD for PW, TS, U200, U850, and OLR, respectively, to the total sum. The x-axis is the lead days.
