Advances in Artificial Neural Systems

Research Article

Using Ensemble of Neural Networks to Learn Stochastic Convection Parameterizations for Climate and Numerical Weather Prediction Models from Data Simulated by a Cloud Resolving Model

Table 3

NN ensemble member error statistics on training (Tr) and independent test (Ts) sets. CC is the correlation coefficient. HID = 5.


Data set	Ens. mem.	NN outputs
		Q1C (K/day)			(K/day)			Prec (mm/day)			CLD (fractions)
		Bias	RMSE	CC	Bias	RMSE	CC	Bias	RMSE	CC	Bias	RMSE	CC

Tr	2	⁻³	2.8	0.75	⁻²	4.0	0.63	⁻²	6.0	0.85	⁻⁴	0.07	0.91
	3	⁻³	2.4	0.78	⁻²	3.7	0.66	⁻²	5.7	0.86	⁻⁶	0.07	0.92
	4	⁻³	2.3	0.81	⁻³	3.7	0.68	⁻³	5.2	0.89	⁻⁴	0.07	0.92
	5	⁻³	2.3	0.80	⁻³	3.8	0.66	⁻²	5.3	0.88	⁻⁵	0.07	0.91
	6	⁻⁴	2.3	0.80	⁻³	3.8	0.64	⁻³	5.3	0.88	⁻⁵	0.08	0.89
	7	⁻⁴	2.3	0.81	⁻⁴	3.7	0.67	⁻³	5.2	0.89	⁻⁵	0.06	0.93
	9	⁻³	3.1	0.73	⁻³	4.0	0.64	⁻²	5.8	0.86	⁻⁴	0.07	0.90

Ts	2	−0.1	3.5	0.62	0.02	4.7	0.49	−1.1	8.5	0.68	0.03	0.11	0.81
	3	−0.6	3.5	0.62	−0.8	5.0	0.44	−5.1	10.6	0.66	0.01	0.11	0.81
	4	−0.5	3.0	0.70	−0.6	4.5	0.53	−4.0	8.8	0.73	0.00	0.09	0.86
	5	−0.1	2.9	0.71	−0.1	3.9	0.52	−1.8	7.8	0.74	0.01	0.08	0.87
	6	−0.3	2.9	0.70	−0.1	3.9	0.51	−2.6	8.0	0.74	0.01	0.08	0.88
	7	−0.4	2.9	0.73	−0.5	4.3	0.58	−3.3	7.9	0.77	0.00	0.07	0.92
	9	−0.7	3.8	0.65	−0.8	4.7	0.51	−4.1	8.6	0.76	0.01	0.10	0.84