Statistical Uncertainty Estimation Using Random Forests and Its Application to Drought Forecast

Table 1

Error statistics of SPI forecast based on ARMA model and the RF model.

Station	Error statistics	SPI(3)		SPI(12)
		ARMA	RF	ARMA	RF
Beijing	BIAS	0.033	0.034	0.245	0.092
	MAE	0.637	0.415	0.606	0.3
	RMSE	0.791	0.526	0.796	0.385
	RMSE (dry days)	0.963	0.662	0.968	0.44
	% dry days detected	27.3%	54.5%	31.1%	71.1%
Shijiazhuang	BIAS	−0.019	0.002	0.034	−0.022
	MAE	0.557	0.289	0.774	0.396
	RMSE	0.736	0.382	1.015	0.507
	RMSE (dry days)	1.065	0.531	1.551	0.674
	% dry days detected	35.0%	60.0%	7.1%	42.9%
Tangshan	BIAS	0.074	0.05	0.214	0.087
	MAE	0.656	0.434	0.796	0.378
	RMSE	0.814	0.54	1.034	0.475
	RMSE (dry days)	1.036	0.651	1.247	0.553
	% dry days detected	33.3%	50.0%	17.8%	60.0%
Tianjin	BIAS	0.069	0.015	0.211	0.078
	MAE	0.597	0.288	0.523	0.264
	RMSE	0.735	0.366	0.624	0.331
	RMSE (dry days)	1.041	0.526	0.897	0.467
	% dry days detected	16.7%	50.0%	0.0%	19.0%