Exploring possibilities for solar irradiance prediction from solar photosphere images using recurrent neural networks | Journal of Space Weather and Space Climate

Open Access

Table 2

RNNs performance for TSI prediction (mean values), and the same metrics applied to physical models’ (SATIRE and EMPIRE) daily reconstructed values. This performance was obtained for the test period of 2012/11/20 to 2012/12/18.

	LSTM				GRU
	MSE	MAPE (%)	R² (%)	R (%)	MSE	MAPE (%)	R² (%)	R (%)
Six hours	0.005	0.004	89.9	97.5	0.005	0.004	89.5	97.0
Twelve hours	0.006	0.005	87.8	95.9	0.006	0.005	87.4	95.8
Eighteen hours	0.009	0.006	82.3	93.3	0.009	0.006	82.2	93.2
One day	0.010	0.006	78.9	91.2	0.012	0.007	76.7	90.2
Two days	0.022	0.009	57.4	80.9	0.020	0.009	61.1	81.2
Three days	0.026	0.010	53.0	76.8	0.024	0.010	54.9	76.4
SATIRE	0.06	0.018	−25.4	99.2	0.06	0.018	−25.4	99.2
EMPIRE	0.03	0.011	36.5	90.3	0.03	0.011	36.5	90.3

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.