Mapping stem volume in fast-growing eucalypt plantations: integrating spectral, textural, and temporal remote sensing information with forest inventories and spatial models

Table 3 Statistics of the multilinear models for stem volume prediction (m³ ha⁻¹) obtained from different variable selection methods (see Section “2.4”)

		Calibration			Validation
Method	Selected auxiliary variables	RMSE	R²_adj	AICc	RMSE	R²
mod_fw	INT_2_DS, INT_2_MAX, NDVI_2_MIN	20.5	0.88	1519.39	19.1	0.88
mod_bw	RAT_B3_B1, INT_0_MIN, NDVI_2_MIN	31.6	0.72	1666.57	27.3	0.74
mod_sw	MAX_MIN, INT_2_DS, NDVI_2_MIN	21.2	0.87	1531.46	18.5	0.88
mod_spls	MUL_B1_HOM_B1, INT_0_DS, INT_1_MAX, NDVI_2_MAX	18.5	0.90	1487.06	18.1	0.89
mod_sl	RAT_B2_B1, RAT_B3_B2, RAT_B3_B4, MAX_MIN_NDVI, INT_2_MIN	30.2	0.74	1656.22	26.3	0.76
mod_rd	RAT_B3_B2, RAT_B4_B3, INT_2_MIN, MAX_MIN_NDVI	30.5	0.74	1657.00	26.2	0.76
mod_ls	RAT_B1_B2, RAT_B2_B4, RAT_B3_B2, INT_0_DM, INT_0_2, INT_2_MAX	14.0	0.94	1399.54	12.44	0.95
mod_en	RAT_B1_B2, RAT_B2_B4, RAT_B3_B2, RAT_B2_CON_B2, INT_0_2, INT_1_DM, INT_2_MAX	14.2	0.94	1401.06	14.6	0.93

Statistics are given in the table for the calibration dataset and the validation dataset (20% of the data). In bold the model which gave the best statistics. mod_fw: forward stepwise selection, mod_bw: backward stepwise selection, mod_sw: sequential replacement, mod_spls: sparse partial least squares, mod_sl: Spike and Slab Lasso; mod_rd: Ridge; mod_ls: Lasso; mod_en: Elastic Net

ISSN: 1297-966X