A new strategy for improving the accuracy of forest aboveground biomass estimates in an alpine region based on multi-source remote sensing

Figures & data

Figure 1. Study area; (a-c) show the geographical location of the study area and the phenological observation sites; (d) is a Landsat 8 OLI false-color composite (R: NIR band, G: red band, B: green band) acquired on 15 December 2019.

Figure 2. Field photos of common dominant tree species groups in the study area.

Figure 3. Flowchart of this study.

Figure 4. Spectral characteristics of different dominant tree species over time characterized by the MCCDC coefficients.

Figure 5. Comparison of the phenological parameters of different dominant tree species groups.

Table 1. Predictor variables for AGB modeling.

Type	Variables
SR data	Blue, red, green, NIR, SWIR1, SWIR2
Vegetation indices	NDVI
	EVI
	SAVI
Tasseled cap transformations	Brightness, wetness, greenness
Texture measures	Mean, variance, homogeneity, contrast, dissimilarity, entropy, second moment, correlation
PCA	PCA variables
Topographic variables	Elevation, slope, aspect
Seasonal variation	MCCDC coefficients
Phenological parameters	SOS, EOS, LOS
PALSAR data	HH, HV, and the difference and ratio of HH and HV

Table 2. Accuracy and confidence interval of the dominant tree species groups obtained from the model with only spectral variables in 2007 and 2016 (confidence level = 95%).

	2007		2016
	User’s accuracy	Producer’s accuracy	User’s accuracy	Producer’s accuracy
pine	75%±1.2%	85%±1.0%	74%±1.2%	80%±1.1%
oak	79%±1.4%	80%±1.4%	74%±1.5%	74%±1.5%
Walnut	71%±2.3%	43%±2.5%	64%±2.3%	45%±2.0%
other dominant tree species	84%±1.7%	28%±2.1%	78%+±2.5%	34%±2.9%
non-forest	89%±1.0%	91%±0.9%	88%±1.0%	91%±0.9%
overall accuracy	80%±0.7%		78%±0.7%

Table 3. Accuracy and confidence interval of the dominant tree species groups obtained from the model with the MCCDC coefficients and phenological parameters in 2007 and 2016 (confidence level = 95%).

	2007		2016
	User’s accuracy	Producer’s accuracy	User’s accuracy	Producer’s accuracy
pine	87%±1.4%	92%±0.8%	87%±1.0%	88%±1.1%
oak	88%±1.1%	91%±1.0%	82%±1.3%	93%±0.9%
walnut	91%±1.4%	86%±1.7%	91%±1.4%	87%±1.6%
other dominant tree species	87%±1.6%	74%±2.0%	88%±2.0%	71%±2.8%
non-forest	94%±0.7%	91%±0.9%	94%±0.7%	90%±0.9%
overall accuracy	89%±0.5%*		88%±0.5%*

*indicates significant differences in the accuracy of the two methods at the 95% confidence level.

Figure 6. Spatiotemporal distribution of the dominant tree species groups obtained from the model with the MCCDC coefficients and phenological parameters.

Figure 7. Percentage of different dominant tree species groups from 2000 to 2019.

Figure 8. Variable importance ranking of the predictor variables.

Table 4. Evaluation indices of models with different variable combinations.

	R^2	RMSE (t/ha)	MAE (t/ha)	RMSE%
Spectral variables+PALSAR	0.52	18.62	13.66	52.49
Spectral variables + phenology variables+PALSAR	0.56	14.05	13.12	32.91

Table 5. AGB accuracy of the proposed AGB model for different dominant tree species.

	R^2	RMSE（t/ha）	MAE（t/ha）	RMSE%
pine	0.61	9.75	8.35	32.47
oak	0.62	12.92	16.43	25.99
walnut	0.78	4.80	3.69	35.13
other dominant tree species	0.65	24.45	18.62	56.67

Figure 9. Scatter plot of predicted and observed AGB; (a) without considering dominant tree species and (b) considering dominant tree species. * indicates significant differences in the predicted AGB between (a) and (b).

Figure 10. Spatial distribution of forest AGB from 2000 to 2019.

Figure 11. AGB statistics from 2000 to 2019.

Figure 12. Total AGB of different dominant tree species from 2000 to 2019.

Table

Dominant tree species	biomass conversion equation
Pinus yunnanensis	w = 0.8569 v^0.8564
Other pines and conifer forests	w = 0.5272 v^1.0793
Quercus and Hardwood	w = 0.6573 v^1.0502
Eucalyptus	w = 2.5878 v^0.6261
Populus	w = 0.7629 v^0.9505
Softwood	w = 2.1529 v^0.6085
Abies and Picea	w = 2.137 v^0.7532
Cunninghamia lanceolata	w = 0.1807 v^1.2771
Cupressus funebris	w = 1.2237 v^0.8592
Cryptomeria fortunei	w = 0.8707 v^1.0239
Phoebe and Cinnamomum	w = 1.0726 v^1.0128

Data availability statement

The data that support the findings of this study (i.e. the land cover and probability maps) are available from the corresponding author and USGS (https://glovis.usgs.gov/).