Regionalized classification of stand tree species in mountainous forests by fusing advanced classifiers and ecological niche model

Figures & data

Figure 1. (a) Yunnan, a global biodiversity hotspot region, located in southwestern China. (b) The elevation map of the Yunnan. (c) Eight distinct floristic sub-regions of Yunnan as basic unit for classification (Li et al. 2015).

Figure 2. Valid observations of Sentinel-2 images in different periods. (a) 2016.1 to 2016.12; (b) 2015.7 to 2017.12; (c) 2016.1 to 2020.12.

Table 1. Observed class and its acronym.

	Target legend	Abbreviation
Level 1	Non-forest land	/
	Forest land	/
Level 2	Pinus yunnanensis	PY
	Cunninghamia lanceolata	CL
	Quercus L.	QL
	Other broadleaves	OB
	Eucalyptus L.Herit	EL
	Pinus armandii Franch.	PAF
	Alnus cremastogyne Burk.	ACB
	Betula L.	BL
	Cupressus funebris Endl.	CFE
	PopulusL.	PL
	Pinus kesiya var. langbianensis	PKV
	Abies fabri	AF
	Pinus densata Mast.	PDM
	Sassafras tsumu	ST
	Hevea brasiliensis	HB
	Acacia dealbata Link.	ADL
	Juglans regia	JR
	Picea asperata Mast.	PAM
	Larix gmelinii	LG

Figure 3. Technique workflow of research.

Figure 4. The number of samples for each observed class.

Table 2. The vegetation indices and corresponding formulas used in this study.

Indices	Formulation	Reference
TVI	$0.5\times (120\times ({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1}-{\rho }_{\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}})-(200\times ({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}}-{\rho }_{\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}})))$	(Main et al. 2011)
MTCI	$({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}2}-{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1})/({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1}-{\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}})$	(Dash and Curran 2004)
NBR	$({\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}}-{\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}2})/({\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}}+{\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}2})$	(Long et al. 2019)
IRECI	$({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}3}-{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}1})/({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1}/{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}2})$	(Rozenstein et al. 2019)
MCARI2	$(({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}2}-{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1})-0.2\times ({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}2}-{\rho }_{\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}}))\times ({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}2}/{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1})$	(Wu et al. 2008)
LSWI	$({\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}}-{\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}1})/({\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}}+{\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}1})$	(Chandrasekar et al. 2010)
NDRE	$({\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}}-{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1})/({\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}}+{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1})$	(Ahamed et al. 2011)
Chl NDI	$({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}2}-{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1})/({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}2}+{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1})$	(Richardson, Duigan, and Berlyn 2002)
REP	$705+35\times (0.5\times ({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}}+{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}3})-{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1})/({\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}2}-{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1})$	(Schlerf, Atzberger, and Hill 2005)
NDVI	$({\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}}-{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}})/({\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}}+{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}})$	(Carlson and Ripley 1997)
NDSVI	$({\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}1}-{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}})/({\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}1}+{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}})$	(Zheng et al. 2019)
NDTI	$({\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}1}-{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}2})/({\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}1}+{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}2})$	(Zhong, Hu, and Zhou 2019)
NDSI	$({\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}1}-{\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}2})/({\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}1}+{\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}2})$	(Richardson, Duigan, and Berlyn 2002)
NDBI	$({\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}1}-{\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}})/({\rho }_{\mathrm{S}\mathrm{W}\mathrm{I}\mathrm{R}1}+{\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}})$	(Zhang, Odeh, and Han 2009)
NDWI	$({\rho }_{\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}}-{\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}})/({\rho }_{\mathrm{G}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}}+{\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}})$	(Gao 1996)
CIrededge	${\rho }_{\mathrm{N}\mathrm{I}\mathrm{R}}/{\rho }_{\mathrm{R}\mathrm{e}\mathrm{d}\mathrm{E}\mathrm{d}\mathrm{g}\mathrm{e}1}-1$	(Duan et al. 2021)

Table 3. The feature candidates for the forest mask generation.

Type	Proxies	Number of features
Spectral Bands	Blue, Green, Red, RedEdge1, RedEdge2, RedEdge3, NIR, Narrow NIR, SWIR1, SWIR2	10
Vegetation Index	TVI, LSWI, NDWI, NDBI, NBR, CIrededge, NDSI, NDTI, Chl NDI, REP, NDRE, MCARI2, MTCI, IRECI, NDVI, NDSVI	16
Texture	Sum Average, Correlation, Variance, Dissimilarity, Contrast, Cluster shade (Calculated for NDVI)	6=1×6
Tasseled Cap Transformation	Wetness, Greenness	2
Topographic	Elevation, Aspect, Slope	3

Table 4. The feature candidates for forest stand species classification.

Type	Proxies	Number of features
Spectral Bands	Blue, Green, Red, RedEdge1, RedEdge2, RedEdge3, NIR, Narrow NIR, SWIR1, SWIR2	10
Vegetation Index	TVI, LSWI, NDWI, NDBI, NBR, CRE, NDSI, NDTI, Chl NDI, REP, NDRE, MCARI2, MTCI, IRECI, NDVI, NDSVI	16
Texture	Sum Average, Correlation, Variance, Dissimilarity, Contrast, Cluster shade (Calculated for Blue, Green, Red, RedEdge1, RedEdge2, RedEdge3, NIR, Narrow NIR, SWIR1, SWIR2)	60=10×6
Tasseled Cap Transformation	Wetness, Greenness	2
Topographic	Elevation, Aspect, Slope	3
Time Series	NDVI, REP	146=73×2
WorldClim	Bio01, Bio02, Bio03, Bio04, Bio05, Bio06, Bio07, Bio08, Bio09, Bio10, Bio11, Bio12, Bio13, Bio14, Bio15, Bio16, Bio17, Bio18, Bio19	19

Figure 5. Schematic diagram of two decision fusion strategies.

Table 5. Feature-classifier combinations explored in this study.

Type	Feature setting	Classifier/fusion rule	Abbreviation
Component	Feature Set 1	MaxEnt	M
	All Feature Set	XGBoost	XALL
	All Feature Set	SVM	SALL
	All Feature Set	RF	RALL
	Feature Set 2	XGBoost	XOPT
	Feature Set 2	SVM	SOPT
	Feature Set 2	RF	ROPT
Ensemble	The probability outputs of M, XOPT, SOPT, ROPT	Weighted voting	MXSR
	The probability output of M and Feature Set 2	XGBoost	MX
	The probability output of M, SOPT and Feature Set 2	SVM	MS
	The probability output of M, ROPT and Feature Set 2	RF	MR

Figure 6. The numbers of optimal features and overall accuracies in the 8 sub-regions. Darker red tones mean higher accuracy for species classification.

Figure 7. Forest/non-forest classification result of Yunnan province.

Table 6. Accuracy assessment for different classification scenarios in each sub-region. The highest accuracies over all sub-regions are in bold.

Classifier	I	II	III	Ⅳ	Ⅴ	Ⅵ	Ⅶ	Ⅷ	Mean
	Overall accuracy (%)
M	53.44	49.11	49.58	42.59	42.59	41.33	51.68	48.68	47.38
XALL	67.76	65.70	67.92	67.35	68.39	65.17	67.20	64.28	66.72
SALL	68.95	66.61	69.42	65.39	68.31	65.37	68.74	66.39	67.40
RALL	68.57	66.57	68.25	65.39	66.79	65.21	66.56	64.80	66.52
XOPT	67.30	65.07	66.35	64.14	66.08	63.20	65.39	59.63	64.65
SOPT	67.58	66.63	65.30	66.23	66.56	63.60	67.74	60.35	65.50
ROPT	67.83	65.53	66.92	64.87	66.41	64.30	66.14	60.35	65.29
MXSR	71.89	72.22	71.55	66.46	70.99	67.70	71.07	64.19	69.51
MX	73.76	71.15	72.88	69.49	71.37	68.79	74.02	68.43	71.24
MS	72.93	71.65	72.31	70.16	70.79	68.13	75.16	70.47	71.45
MR	74.44	72.21	74.12	70.34	72.52	70.12	75.24	71.98	72.62

Figure 8. The spatial details of the forest stand species map in Yunnan Province in 2016. Site a (103.279°E, 26.851°N); Site b (103.626°E, 25.910°N); Site c (103.031°E, 25.434°N); Site d (104.568°E, 23.026°N); Site e (101.347°E, 21.946°N); Site f (99.173°E, 25.318°N); Site g (99.688°E, 27.743°N).

Figure 9. Confusion matrix heatmap of tree species classification.

Figure 10. The mean spectral reflectance of 19 tree species, calculated from the Sentinel-2 bands.

Figure 11. JM distances between species with (a) spectral bands and vegetation indices shown in Table 4, (b) spectral features and texture features, and (c) spectral features and time series features.

Figure 12. Distribution map of forest stand species in the altitude range.

Table 7. Comparison of classification accuracies of different methods in each sub-region.

Region	M	ROPT	MR	MR-M	MR-ROPT
I	53.44%	67.83%	74.44%	21.00%	6.61%
II	49.11%	65.53%	72.20%	23.09%	6.67%
III	49.58%	66.92%	74.12%	24.54%	7.20%
Ⅳ	42.59%	64.87%	70.34%	27.75%	5.47%
Ⅴ	41.46%	66.41%	72.52%	31.06%	6.11%
Ⅵ	41.33%	64.30%	70.12%	28.79%	5.47%
Ⅶ	51.68%	66.14%	75.24%	23.56%	9.10%
Ⅷ	48.68%	60.35%	71.98%	23.30%	11.63%
Mean	47.23%	65.29%	72.62%	25.39%	7.33%

Figure 13. The spatial details of classification results in forest stand species. (a) Original image, (b) the Forest Management Inventory data, (c) M classification, (d) ROPT classification, and (e) MR classification.

Figure 14. Brief review on research cases in tree species classification.

Li, R., N. J. Kraft, J. Yang, and Y. Wang. 2015. “A Phylogenetically Informed Delineation of Floristic Regions Within a Biodiversity Hotspot in Yunnan, China.” Scientific Reports 5 (1): 1–7. doi:10.1016/S0034-4257(96)00178-2.

 Google Scholar

Main, R., M. A. Cho, R. Mathieu, M. M. O’Kennedy, A. Ramoelo, and S. Koch. 2011. “An Investigation into Robust Spectral Indices for Leaf Chlorophyll Estimation.” Isprs Journal of Photogrammetry & Remote Sensing 66 (6): 751–761. doi:10.1016/j.isprsjprs.2011.08.001.

 Web of Science ®Google Scholar

Dash, J., and P. Curran. 2004. “The MERIS Terrestrial Chlorophyll Index.” International Journal of Remote Sensing 25 (23): 5403–5413. doi:10.1080/0143116042000274015.

 Web of Science ®Google Scholar

Long, T., Z. Zhang, G. He, W. Jiao, C. Tang, B. Wu, X. Zhang, G. Wang, and R. Yin. 2019. “30 M Resolution Global Annual Burned Area Mapping Based on Landsat Images and Google Earth Engine.” Remote Sensing 11 (5): 489. doi:10.3390/rs11050489.

 Web of Science ®Google Scholar

Rozenstein, O., N. Haymann, G. Kaplan, and J. Tanny. 2019. “Validation of the Cotton Crop Coefficient Estimation Model Based on Sentinel-2 Imagery and Eddy Covariance Measurements.” Agricultural Water Management 223: 105715. doi:10.1016/j.agwat.2019.105715.

 Web of Science ®Google Scholar

Wu, C., Z. Niu, Q. Tang, and W. Huang. 2008. “Estimating Chlorophyll Content from Hyperspectral Vegetation Indices: Modeling and Validation.” Agricultural and Forest Meteorology 148 (8–9): 1230–1241. doi:10.1016/j.agrformet.2008.03.005.

 Web of Science ®Google Scholar

Chandrasekar, K., M. Sesha Sai, P. Roy, and R. S. Dwevedi. 2010. “Land Surface Water Index (LSWI) Response to Rainfall and NDVI Using the MODIS Vegetation Index Product.” International Journal of Remote Sensing 31 (15): 3987–4005. doi:10.1080/01431160802575653.

 Web of Science ®Google Scholar

Ahamed, T., L. Tian, Y. Zhang, and K. C. Ting. 2011. “A Review of Remote Sensing Methods for Biomass Feedstock Production.” Biomass & bioenergy 35 (7): 2455–2469. doi:10.1016/j.biombioe.2011.02.028.

 Web of Science ®Google Scholar

Richardson, A. D., S. P. Duigan, and G. P. Berlyn. 2002. “An Evaluation of Noninvasive Methods to Estimate Foliar Chlorophyll Content.” The New Phytologist 153 (1): 185–194. doi:10.3390/rs12203324.

 Web of Science ®Google Scholar

Schlerf, M., C. Atzberger, and J. Hill. 2005. “Remote Sensing of Forest Biophysical Variables Using HyMap Imaging Spectrometer Data.” Remote Sensing of Environment 95 (2): 177–194. doi:10.1016/j.rse.2004.12.016.

 Web of Science ®Google Scholar

Carlson, T. N., and D. A. Ripley. 1997. “On the Relation Between NDVI, Fractional Vegetation Cover, and Leaf Area Index.” Remote Sensing of Environment 62 (3): 241–252. doi:10.1016/S0034-4257(97)00104-1.

 Web of Science ®Google Scholar

Zheng, G., A. Bao, X. Li, L. Jiang, C. Chang, T. Chen, and Z. Gao. 2019. “The Potential of Multispectral Vegetation Indices Feature Space for Quantitatively Estimating the Photosynthetic, Non-Photosynthetic Vegetation and Bare Soil Fractions in Northern China.” Photogrammetric Engineering & Remote Sensing 85 (1): 65–76. doi:10.14358/PERS.85.1.65.

 Web of Science ®Google Scholar

Zhong, L., L. Hu, and H. Zhou. 2019. “Deep Learning Based Multi-Temporal Crop Classification.” Remote Sensing of Environment 221: 430–443. doi:10.1016/j.rse.2018.11.032.

 Web of Science ®Google Scholar

Zhang, Y., I. O. Odeh, and C. Han. 2009. “Bi-Temporal Characterization of Land Surface Temperature in Relation to Impervious Surface Area, NDVI and NDBI, Using a Sub-Pixel Image Analysis.” International Journal of Applied Earth Observation and Geoinformation 11 (4): 256–264. doi:10.1016/j.jag.2009.03.001.

 Web of Science ®Google Scholar

Gao, B. -C. 1996. “NDWI—A Normalized Difference Water Index for Remote Sensing of Vegetation Liquid Water from Space.” Remote Sensing of Environment 58 (3): 257–266. doi:10.1016/S0034-4257(96)00067-3.

 Web of Science ®Google Scholar

Duan, B., S. Fang, Y. Gong, Y. Peng, X. Wu, and R. Zhu. 2021. “Remote Estimation of Grain Yield Based on UAV Data in Different Rice Cultivars Under Contrasting Climatic Zone.” Field Crops Research 267: 108148. doi:10.1016/j.fcr.2021.108148.

 Web of Science ®Google Scholar