Forest age estimation using UAV-LiDAR and Sentinel-2 data with machine learning algorithms- a case study of Masson pine (Pinus massoniana)

Figures & data

Figure 1. (a) Location of the study area (b) spatial distribution of Sentinel-2 images and sample plots in the study area (c) UAV laser cloud images of study area sites (d) Masson pine age survey.

Table 1. LiDAR-extracted feature variables.

Feature Type	Feature Name	Feature Description
Structural characteristics	Height	Average stand height
	CC	Canopy Cover, Ratio of the number of vegetation points in the first echo to the number of points in the first echo.
	GF	Gap Fraction, Extract the number of ground points with Z-value below the height threshold (2m) divided by the total number of points
	LAI	Leaf Area Index, $\mathit{LAI}=-\frac{cos\left(\mathit{ang}\right)\times ln\left(\mathit{GF}\right)}{k}$, where ang is the mean scan angle, GF is the gap rate, and k is the extinction coefficient (Richardson, Moskal, and Kim 2009)
Height-based metrics	H1/H5/H10/H20/H25/H30/H40/H50/ H60/H70/H75/H80/H90/H95/H99	The percentiles of the canopy height distribution (1th, 5th, 10th, 20th,25th, 30th, 40th, 50th, 60th,70th,75th, 80th, 90th, 95th, and 99th) of first returns
	Hiq	Hiq = H75−H25。
	AIH1/AIH5/AIH10/AIH20/AIH25/ AIH30/AIH40/AIH50/AIH60/AIH70/AIH75/AIH80/AIH90/AIH95/AIH99	Cumulative percentile of canopy height distribution (1th, 5th, 10th, 20th,25th, 30th, 40th, 50th, 60th,70th,75th, 80th, 90th, 95th, and 99th) of first returns
	AIHiq	AIHiq = AIH75−AIH25。
	Hmax/Hmin/Hmean/Hmedia/Hmadme	Maximum height/minimum height/average height/median height/median absolute deviation of all points within each statistical cell
	Hvar/Hstdde/Hskew/Hkurt/Hcv/Hsqrt/ Hcurt/Hcanopy/Hmad	Variance/standard deviation/skewness/kurtosis/coefficient of variation/power-of-two average/power-of-three average/canopy undulation rate/mean absolute deviation for all point heights within each statistical unit
Density-based metrics	D0/D1/D2/D3/D4/D5/D6/D7/D8/D9	The proportion of points above the quantiles (10th, 20th, 30th, 40th, 50th, 60th, 70th, 80th, and 90th) to total number of points
Intensity-based metrics	I1/I5/I10/I20/I25/I30/I40/I50/I60/I70/I75/ I80/I90/I95/I99	The percentiles of the canopy intensity distribution (1th, 5th, 10th, 20th,25th, 30th, 40th, 50th, 60th,70th,75th, 80th, 90th, 95th, and 99th) of first returns
	Iiq	Iiq = I75−I25
	AII1/AII5/AII10/AII20/AII25/AII30/AII40 /AII50/AII60/AII70/AII75/AII80 /AII90/AII95/AII99	Cumulative percentile of canopy intensity distribution (1th, 5th, 10th, 20th,25th, 30th, 40th, 50th, 60th,70th,75th, 80th, 90th, 95th, and 99th) of first returns
	Imax/Imin/Imean/Imedia/Imadme	Median of the highest intensity/lowest intensity/average intensity/median intensity/median absolute deviation for all points within each statistical cell
	Ivar/Istdde/Iskew/Ikurt/Icv/Imad	Variance/standard deviation/skewness/kurtosis/coefficient of variation/mean absolute deviation for all point intensities within each statistical cell

Table 2. Sentinel-2 extracted feature variables.

Type	Name	Calculation Models or Descriptions	Abbreviation	Notes and References
Original Band	Aerosols	Band1	B1	/
	Blue	Band2	B2
	Green	Band3	B3
	Red	Band4	B4
	Red Edge 1	Band5	B5
	Red Edge 2	Band6	B6
	Red Edge 3	Band7	B7
	NIR	Band8	B8
	Red Edge 4	Band8A	B8A
	Water vapor	Band9	B9
	SWIR 1	Band11	B11
	SWIR 2	Band12	B12
Spectral Vegetation Indices	DVI	$\mathit{NIR}-\mathit{R}$	DVI	(Jaskierniak et al. 2021)
	EVI	$\left(2.5\left(\mathit{NIR}-\mathit{R}\right)\right)/\left(\mathit{NIR}+6\mathit{R}-7.5\mathit{B}+1\right)$	EVI	(Muhaimin et al. 2022)
	TVI	$0.5\left(120\left(\mathit{NIR}-\mathit{G}\right)-200\left(\mathit{R}-\mathit{G}\right)\right)$	TVI	(Su, Liu, and Su 2017)
	RVI	$\mathit{NIR}/\mathit{R}$	RVI	(Tan et al. 2019)
	PSRI	$\left(\mathit{R}-\mathit{B}\right)\times \mathit{RedEdge}2$	PSRI	(X. Zhang et al. 2022)
	NDII	$\left(\mathit{NIR}-\mathit{SWIR}1\right)/\left(\mathit{NIR}+\mathit{SWIR}1\right)$	NDII	(Sriwongsitanon et al. 2016)
	NDWI	$\left(\mathit{G}-\mathit{NIR}\right)/\left(\mathit{G}+\mathit{NIR}\right)$	NDWI	(Racine et al. 2014)
	NDVI	$\left(\mathit{NIR}-\mathit{R}\right)/\left(\mathit{NIR}+\mathit{R}\right)$	NDVI	(Muhaimin et al. 2022)
	MNDWI	$\left(\mathit{G}-\mathit{SWRI}1\right)/\left(\mathit{G}+\mathit{SWRI}1\right)$	MNDWI	(Qiu, Gan, and Zhao 2022)
	SAVI	$1.5\times \left(\mathit{NIR}-\mathit{R}\right)/\left(\mathit{NIR}+\mathit{R}+0.5\right)$	SAVI	(Zhen et al. 2021)
	NDBI	$\left(\mathit{SWIR}1-\mathit{NIR}\right)/\left(\mathit{SWIR}1+\mathit{NIR}\right)$	NDBI	(Muhaimin et al. 2022)
	Cire	$\left(\left(\mathit{RedEdge}3/\mathit{RedEdge}1\right)-1\right)$	Cire	(Gitelson, Gritz, and Merzlyak 2003)
Texture Features Based On The Gray-Level Co-occurrence Matrix (GLCM)	Variance	$\sum _{\mathit{i}=1}^n\sum _{\mathit{j}=1}^n{\left(\mathit{i}-\mathit{mean}\right)}^2{P}_{\mathit{ij}}$	VAR	$P_{\left(\mathit{i},\mathit{j}\right)}=\frac{{\mathit{V}}_{\left(\mathit{i},\mathit{j}\right)}}{{\sum }_{i=1}^n{\sum }_{j=1}^n{V}_{\left(\mathit{i},\mathit{j}\right)}}$ $V_{\left(\mathit{i},\mathit{j}\right)}$ is the $\mathit{i}$th row of the $\mathit{j}$th column in the $\mathit{N}$th moving window. $u_x=\sum _{\mathit{j}=1}^n\mathit{j}\sum _{\mathit{i}=1}^n{P}_{\left(\mathit{i},\mathit{j}\right)}$ $u_y=\sum _{\mathit{i}=1}^n\mathit{i}\sum _{\mathit{j}=1}^n{P}_{\left(\mathit{i},\mathit{j}\right)}$ ${\sigma }_x=\sum _{\mathit{j}=1}^n{\left(j-u_i\right)}^2\sum _{\mathit{i}=1}^n{P}_{\left(\mathit{i},\mathit{j}\right)}$ ${\sigma }_y=\sum _{\mathit{i}=1}^n{\left(i-u_j\right)}^2\sum _{\mathit{j}=1}^n{P}_{\left(\mathit{i},\mathit{j}\right)}$ (Li et al. 2023; Mohammadpour, Viegas, and Viegas 2022)
	Homogeneity	$\sum _{\mathit{i}=1}^n\sum _{\mathit{j}=1}^n\frac{1}{1+{\left(\mathit{i}-\mathit{j}\right)}^2}{P}_{\left(\mathit{i},\mathit{j}\right)}$	HOM
	Contrast	$\sum _{\left|\mathit{i}-\mathit{j}\right|=0}^{\mathit{n}-1}{\left|\mathit{i}-\mathit{j}\right|}^2\left\{\sum _{\mathit{i}=1}^n\sum _{\mathit{j}=1}^n{P}_{\left(\mathit{i},\mathit{j}\right)}\right\}$	CON
	Dissimilarity	$\sum _{\left|\mathit{i}-\mathit{j}\right|=0}^{\mathit{n}-1}\left|\mathit{i}-\mathit{j}\right|\left\{\sum _{\mathit{i}=1}^n\sum _{\mathit{j}=1}^n{P}_{\left(\mathit{i},\mathit{j}\right)}\right\}$	DIS
	Entropy	$-\sum _{\mathit{i}=1}^n\sum _{\mathit{j}=1}^n{P}_{\left(\mathit{i},\mathit{j}\right)}log\left(\mathit{P}\left(\mathit{i},\mathit{j}\right)\right)$	ENT
	Angular second moment	$\sum _{\mathit{i}=1}^n\sum _{\mathit{j}=1}^n\mathit{P}{\left(\mathit{i},\mathit{j}\right)}^2$	ASM
	Correlation	$\frac{{\sum }_{i=1}^n{\sum }_{j=1}^n\mathit{ij}{{\mathit{P}}_{\left(\mathit{i},\mathit{j}\right)}}^2-{\mu }_x{\mu }_y}{{\sigma }_x{\sigma }_y}$	COR
	Cluster Shade	$\sum _{\mathit{i}=1}^n\sum _{\mathit{j}=1}^n{\left(i+j-{\mu }_i-{\mu }_j\right)}^3{P}_{\left(\mathit{i},\mathit{j}\right)}$	SHA

Table 3. Optimized model parameters.

Model name	Parameter name	Parameter value
AdaBoost	n_estimators	250
	learning_rate	1
RF	n_estimators	50
	max_depth	10
	min_samples_split	10
	min_samples_leaf	1
ERT	n_estimators	5
	max_depth	10
	min_samples_split	2
	min_samples_leaf	4

Figure 2. Flowchart of the steps used in our study.

Figure 3. Results of feature variable screening based on the Boruta algorithm.

Red box plots represent the Z scores of rejected features, and green box plots represent the Z scores of confirmed features. Yellow box plots indicate the Z scores of features that are yet to be determined. Blue box line plots correspond to the minimum, average, and maximum Z scores of shadow features. WiBjXX represents the texture information for the jth band window size of the image for i. XX refers to VAR, HOM, CON, DIS, ENT, ASE, COR, and SHA.

Table 4. Feature variables filtered based on the Boruta algorithm.

Feature Types		Number	Feature Name
LiDAR parameter	Average stand height	1	Height
	Height-based metrics	9	AIH90/AIH95/AIH99/H5/H60/H80/H90/H95/H99
	Density-based metrics	2	D2/D3
Sentinel-2 parameter	Original Band	5	B1/B5/B8A/B11/B12
	GLCM	3	W9B3VAR/W9B4ENT/W5B8SHA

Figure 4. Comparison of the accuracy of the forest age estimation models.

Figure 5. Forest age estimation results (a) accuracy evaluation of forest age model based on single LiDAR data (b) accuracy evaluation of forest age model based on single Sentinel-2 data (c) Accuracy evaluation of forest age model with LiDAR combined with Sentinel-2.

Table 5. Distribution of the absolute values of prediction errors.

Data source – Models	The absolute value of prediction error (N = 31)
	0a	1a	2a	3a	4a	5a	>5a	$\mathit{\epsilon }\le 5$
S – AdaBoost	2	6	3	4	2	1	12	0.58
S – RF	4	5	1	5	2	2	11	0.61
S – ERT	3	4	5	1	3	3	12	0.61
L – AdaBoost	4	6	7	6	2	2	4	0.87
L – RF	2	7	5	6	2	3	6	0.81
L – ERT	4	4	5	5	4	4	9	0.84
S+L – AdaBoost	4	6	9	4	1	2	5	0.84
S+L – RF	7	3	6	3	4	3	5	0.84
S+L – ERT	1	6	4	7	2	3	8	0.74

Figure 6. Age distribution of Masson pine based on LiDAR combined with Sentinel-2 data.

Figure 7. Sentinel-2 spectral curves and normalized LiDAR point cloud heights for trees of different ages.

Figure 8. (a) Importance ranking of modeling parameters and (b) correlation between the top 6 variables in terms of importance ranking and stand age.

Table 6. Visible vegetation index.

Name	Calculation Models	References
NGRDI	$\left(\mathit{G}-\mathit{R}\right)/\left(\mathit{G}+\mathit{R}\right)$	(Tucker 1979)
GLI	$\left(2\mathit{G}-\mathit{R}-\mathit{B}\right)/\left(2\mathit{G}+\mathit{R}+\mathit{B}\right)$	(Louhaichi, Borman, and Johnson 2001)
RGRI	$\mathit{R}/\mathit{G}$	(Deng et al. 2016)
ExG	$2\mathit{G}-\mathit{R}-\mathit{B}$	(W. Mao, Wang, and Wang 2003)

Figure 9. RF forest age model estimation results (a) accuracy evaluation of forest age model based on single UAV imagery data (b) accuracy evaluation of forest age model with LiDAR combined with UAV imagery.

Figure 10. Uncertainty analysis of forest age prediction.

Richardson, J. J., L. M. Moskal, and S. H. Kim. 2009. “Modeling Approaches to Estimate Effective Leaf Area Index from Aerial Discrete-Return LIDAR.” Agricultural and Forest Meteorology 149 (6–7): 1152–1160. https://doi.org/10.1016/j.agrformet.2009.02.007.

 Web of Science ®Google Scholar

Jaskierniak, D., A. Lucieer, G. Kuczera, D. Turner, P. N. J. Lane, R. G. Benyon, and S. Haydon. 2021. “Individual Tree Detection and Crown Delineation from Unmanned Aircraft System (UAS) LiDAR in Structurally Complex Mixed Species Eucalypt Forests.” ISPRS Journal of Photogrammetry and Remote Sensing 171:171–187. https://doi.org/10.1016/j.isprsjprs.2020.10.016.

 Web of Science ®Google Scholar

Muhaimin, M., D. Fitriani, S. Adyatma, and D. Arisanty. 2022. “Mapping Build-Up Area Density Using Normalized Difference Built-Up Index (Ndbi) and Urban Index (Ui) Wetland in the City Banjarmasin.” IOP Conference Series: Earth and Environmental Science 1089 (1): 012036.

 Google Scholar

Su, T., Q. Liu, and X. Su. 2017. “Remote Sensing Classification of Crops Based on Multiple Vegetation Indices Time Series with Machine Learning.” Jiangsu Agricultural Sciences 45 (16): 219–224. https://doi.org/10.15889/j.issn.1002-1302.2017.16.054.

 Google Scholar

Tan, Y., J. Y. Sun, B. Zhang, M. Chen, Y. Liu, and X. D. Liu. 2019. “Sensitivity of a Ratio Vegetation Index Derived from Hyperspectral Remote Sensing to the Brown Planthopper Stress on Rice Plants.” Sensors 19 (2): 375. https://doi.org/10.3390/s19020375.

 PubMed Web of Science ®Google Scholar

Zhang, X., X. Hou, M. Wang, L. Wang, and F. Liu. 2022. “Study on Relationship Between Photosynthetic Rate and Hyperspectral Indexes of Wheat Under Stripe Rust Stress.” Spectroscopy and Spectral Analysis 42 (3): 940–946. https://doi.org/10.3964/j.issn.1000-0593(2022)03-0940-07.

 Web of Science ®Google Scholar

Sriwongsitanon, N., H. Gao, H. H. Savenije, E. Maekan, S. Saengsawang, and S. Thianpopirug. 2016. “Comparing the Normalized Difference Infrared Index (NDII) with Root Zone Storage in a Lumped Conceptual Model.” Hydrology and Earth System Sciences 20 (8): 3361–3377. https://doi.org/10.5194/hess-20-3361-2016.

 Web of Science ®Google Scholar

Racine, E. B., N. C. Coops, B. St-Onge, and J. Bégin. 2014. “Estimating Forest Stand Age from LiDAR-Derived Predictors and Nearest Neighbor Imputation.” Forest Science 60 (1): 128–136. https://doi.org/10.5849/forsci.12-088.

 Web of Science ®Google Scholar

Qiu, M., S. Gan, and L. Zhao. 2022. “Analysis of Index Method for Extracting Water Area of Erhai Lake Using Sentinel-2 Lmage.” Urban Geotechnical Investigation & Surveying 6:117–122. https://doi.org/10.1672/8262(2022)06-117-06.

 Google Scholar

Zhen, Z., S. Chen, T. Yin, E. Chavanon, N. Lauret, J. Guilleux, M. Henke, W. Qin, L. Cao, and J. Li. 2021. “Using the Negative Soil Adjustment Factor of Soil Adjusted Vegetation Index (Savi) to Resist Saturation Effects and Estimate Leaf Area Index (Lai) in Dense Vegetation Areas.” Sensors 21 (6): 2115. https://doi.org/10.3390/s21062115.

 PubMed Web of Science ®Google Scholar

Gitelson, A. A., Y. Gritz, and M. N. Merzlyak. 2003. “Relationships Between Leaf Chlorophyll Content and Spectral Reflectance and Algorithms for Non-Destructive Chlorophyll Assessment in Higher Plant Leaves.” Journal of Plant Physiology 160 (3): 271–282. https://doi.org/10.1078/0176-1617-00887.

 PubMed Web of Science ®Google Scholar

Li, H., G. Zhang, Q. Zhong, L. Xing, and H. Du. 2023. “Prediction of Urban Forest Aboveground Carbon Using Machine Learning Based on Landsat 8 and Sentinel-2: A Case Study of Shanghai, China.” Remote Sensing 15 (1): 284. https://doi.org/10.3390/rs15010284.

 Web of Science ®Google Scholar

Mohammadpour, P., D. X. Viegas, and C. Viegas. 2022. “Vegetation Mapping with Random Forest Using Sentinel 2 and GLCM Texture Feature—A Case Study for Lousã Region, Portugal.” Remote Sensing 14 (18): 4585. https://doi.org/10.3390/rs14184585.

 Web of Science ®Google Scholar

Tucker, C. J. 1979. “Red and Photographic Infrared Linear Combinations for Monitoring Vegetation.” Remote Sensing of Environment 8 (2): 127–150. https://doi.org/10.1016/0034-4257(79)90013-0.

 Web of Science ®Google Scholar

Louhaichi, M., M. M. Borman, and D. E. Johnson. 2001. “Spatially Located Platform and Aerial Photography for Documentation of Grazing Impacts on Wheat.” Geocarto International 16 (1): 65–70. https://doi.org/10.1080/10106040108542184.

 Google Scholar

Deng, J., G. Ren, Y. Lan, H. Huang, and Y. Zhang. 2016. “Low Altitude Unmanned Aerial Vehicle Remote Sensing Image processing Based on Visible Band.” Journal of South China Agricultural University 37 (6): 16–22. https://doi.org/10.7671/j.issn.1001-11X.2016.06.003.

 Google Scholar

Mao, W., Y. Wang, and Y. Wang. 2003. “Real-Time Detection of Between-Row Weeds Using Machine Vision.” 2003 ASAE Annual Meeting, 1. St. Joseph: American Society.

 Google Scholar

Data availability statement

The Sentinel-2 dataset used in this study is publicly available for download from The Copernicus Open Access Hub (https://scihub.copernicus.eu/).The other data that support the findings of this study are available from the corresponding author, [Huaqiang Du, E-mail: [email protected]], upon reasonable request.