Fig. 4 | Scientific Reports

Fig. 4

From: Understanding machine learning weather prediction by designing a cost-efficient model with knowledge-oriented modules

Fig. 4

The ACC skill as a function of forecast lead days (x-axis) and latitude (y-axis) for the 7 meteorological variables (T2M, T850, U850, V850, MSLP, Z500, and Q700), and the effects of GeoCyclic Padding. (a) Longitudinally averaged ACC of KARINA. The solid red lines indicate two-tailed 95% confidence intervals, derived by Student’s T-test. (b) ACC difference between KARINA and KARINA w/o pad (i.e., Effect of GeoCyclic Padding). (c) same as b, but for KARINA w/o SENet. In the second and third rows, positive (negative) values indicate a gain (loss) in forecasting performance due to the application of each method. The figures were generated using the Python Matplotlib (version 3.7.2; http://matplotlib.org/).

Back to article page