Combining airborne laser scanning and Landsat data for statistical modeling of soil carbon and tree biomass in Tanzanian Miombo woodlands

Table 2 Results from plot level regression analysis of soil carbon (SOC), above ground tree biomass (AGB) and below ground tree biomass (BGB) using Landsat 8 OLI, ALS and the combination of these data sources

Data source	Model^a,b	R²(adj)	RMSE	RMSE
Data source	Model^a,b	%	Mg ha⁻¹	%
OLI 140731	SOC = −113.8 + 0.0637 B7 + 22.93 B5/B4	34.6	16.2	27.9
ALS	SOC = 74.97 − 0.000500 XL1 + 0.425 XL2 + 0.02500 XL3	42.4	15.2	26.2
ALS + OLI 140731	SOC = −4.2 + 36.93 B7/B4 − 0.000429 XL1 + 0.00433 XL4	56.0	13.3	22.9
OLI 140512	AGB = 35.3 − 0.0661 B6 − 18.41 B5/B4 + 55.20 B6/B4	38.1	30.6	66.2
ALS	AGB = 5.92 − 5.05 P60 + 1.248 PFR50 + 0.576 PFR75	64.4	23.3	50.3
ALS + OLI 140512	AGB = 80.8 − 0.02178 B5 − 3.38 P60 + 1.499 PFR75	66.0	22.7	49.1
OLI 140512	BGB = 39.4 − 0.02009 B5 + 7.12 B6/B4	40.1	11.6	62.2
ALS	BGB = 1.99 − 1.960 MAD − 0.943 P70 + 0.6644 PFR500	71.5	8.1	43.4
ALS + OLI 140512	BGB = 19.43 − 0.00549 B5 − 0.03186 XLS1 + 0.6417 PFR500	71.8	8.1	43.3

The R² _adj statistic is for the model and RMSE values from ‘leave one out cross validation’ (LOOCV)
XL and XLS = combination of different variables. For full variable explanation see [37]
XL1 = “return 1 count above −1.00” + “total return count above −1.00”
XL2 = “percentage first returns above 1.50”/“P80”
XL3 = “P90” * “percentage first returns above 1.50” * “return 1 count above −1.00”/“total return count above −1.00”
XL4 = “P90” * “percentage first returns above 1.50” * “return 1 count above −1.00”/“return 2 count above −1.00”
XLS1 = “percentage first returns above 1.50” * “P80”/(“NDVI + 1”); MAD = elev MAD median
^aAll regression coefficients were statistically significant at 5% level
^bB = Landsat 8 OLI band (1, 2,…,8); P = Height percentiles of lidar vegetation echoes (0, 10,…,90); PFR = Percentage first lidar returns above heightbreak in dm (50, 75, 100)

ISSN: 1750-0680