Open Access

Regional-Scale Analysis of Vegetation Dynamics Using Satellite Data and Machine Learning Algorithms: A Multi-Factorial Approach

Abolfazl Abdollahi
Abolfazl Abdollahi
, 
Biswajeet Pradhan
Biswajeet Pradhan
 and 
Abdullah Alamri
Abdullah Alamri
   | Oct 21, 2023
International Journal on Smart Sensing and Intelligent Systems's Cover Image
About this article
Previous Article
Next Article
Cite
Article Category: Article
Published Online: Oct 21, 2023
Page range: -
Received: Jul 19, 2023
DOI: https://doi.org/10.2478/ijssis-2023-0013
Keywords
© 2023 Abolfazl Abdollahi et al., published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1:

Overall flowchart of vegetation classification process in the GEE platform.
Overall flowchart of vegetation classification process in the GEE platform.

Figure 2:

Greater Sydney region, Australia.
Greater Sydney region, Australia.

Figure 3:

The schematic diagram of RF for season-based vegetation mapping.
The schematic diagram of RF for season-based vegetation mapping.

Figure 4:

Confusion matrix used in the proposed RF model's training process: (a) with only spectral indices; and (b) all input variables. (i), (ii), and (iii) present the normalized confusion matrix for summer, autumn, and winter seasons, respectively.
Confusion matrix used in the proposed RF model's training process: (a) with only spectral indices; and (b) all input variables. (i), (ii), and (iii) present the normalized confusion matrix for summer, autumn, and winter seasons, respectively.

Figure 5:

Visualization results of season-based vegetation mapping achieved by the proposed model for summer season: (a) original multi-temporal Sentinel-2 image, (b) results of RF+spectral indices, (c) results of RF+spectral indices+topographic factors, and (d) results of RF+spectral indices+topographic factors+texture information.
Visualization results of season-based vegetation mapping achieved by the proposed model for summer season: (a) original multi-temporal Sentinel-2 image, (b) results of RF+spectral indices, (c) results of RF+spectral indices+topographic factors, and (d) results of RF+spectral indices+topographic factors+texture information.

Figure 6:

Visualization results of season-based vegetation mapping achieved by the proposed model for autumn season: (a) original multi-temporal Sentinel-2 image, (b) results of RF+spectral indices, (c) results of RF+spectral indices+topographic factors, and (d) results of RF+spectral indices+topographic factors+texture information.
Visualization results of season-based vegetation mapping achieved by the proposed model for autumn season: (a) original multi-temporal Sentinel-2 image, (b) results of RF+spectral indices, (c) results of RF+spectral indices+topographic factors, and (d) results of RF+spectral indices+topographic factors+texture information.

Figure 7:

Visualization results of season-based vegetation mapping achieved by the proposed model for winter season: (a) original multi-temporal Sentinel-2 image, (b) results of RF+spectral indices, (c) results of RF+spectral indices+topographic factors, and (d) results of RF+spectral indices+topographic factors+texture information.
Visualization results of season-based vegetation mapping achieved by the proposed model for winter season: (a) original multi-temporal Sentinel-2 image, (b) results of RF+spectral indices, (c) results of RF+spectral indices+topographic factors, and (d) results of RF+spectral indices+topographic factors+texture information.

Figure 8:

Input variable importance in season-based vegetation mapping achieved by the RF model for the Greater Sydney region: (a) for summer, (b) for autumn, and (c) for winter.
Input variable importance in season-based vegetation mapping achieved by the RF model for the Greater Sydney region: (a) for summer, (b) for autumn, and (c) for winter.

Area of pixels in each class for each season in square km.

Class Area (km2)
Summer Non-vegetation 1,364.3155
Grass 1,691.4873
Trees 9,225.3798
Crops 86.8179
Autumn Non-vegetation 1,335.9587
Grass 1,537.0343
Trees 9,440.5423
Crops 54.4652
Winter Non-vegetation 1,370.4691
Grass 1,485.4252
Trees 9,440.9971
Crops 71.1091

Spatial and spectral resolutions of Sentinel-2 satellite data.

Band Central wavelength (nm) Spatial resolution (m)
Coastal aerosol 443 60
Blue 490 10
Green 560 10
Red 665 10
Vegetation red edge 705 20
Vegetation red edge 740 20
Vegetation red edge 783 20
NIR 842 10
Vegetation red edge 865 20
Water vapor 945 60
SWIR-Cirrus 1,380 60
SWIR 1,610 20
SWIR 2,190 20

Quantitative results were achieved by the suggested RF method for season-based vegetation mapping.

Precision (%) Recall (%) F1 score (%) OA (%) Kappa (%)
Summer RF+spectral indices Non-vegetation 82.45 91.22 86.61 90.65 86.11
Grass 91.54 83.96 87.59
Trees 98.93 97.64 87.59
Crops 67.01 75.01 70.78
RF+ spectral indices+topographic factors Non-vegetation 82.99 91.39 86.99 91.29 87.06
Grass 92.80 84.46 88.43
Trees 99.24 97.94 98.58
Crops 67.87 78.15 72.65
RF+ spectral indices+topographic factors+texture information Non-vegetation 84.27 92.08 88.00 92.56 88.96
Grass 93.50 87.31 90.30
Trees 99.37 98.28 98.82
Crops 74.98 80.89 77.82
Autumn RF+spectral indices Non-vegetation 81.45 88.72 84.93 90.08 85.27
Grass 90.10 83.50 86.68
Trees 98.93 97.83 98.38
Crops 66.61 73.64 69.95
RF+ spectral indices+topographic factors Non-vegetation 80.70 92.30 86.11 90.60 86.06
Grass 90.33 83.98 87.04
Trees 98.96 97.85 98.40
Crops 71.69 73.46 72.57
RF+ spectral indices+topographic factors+texture information Non-vegetation 81.87 92.49 86.86 91.64 87.60
Grass 91.97 85.70 88.72
Trees 99.44 98.21 98.83
Crops 73.50 76.56 75.00
Winter RF+spectral indices Non-vegetation 81.88 91.61 86.47 91.35 87.17
Grass 90.99 84.73 87.75
Trees 99.25 98.29 98.77
Crops 73.84 77.10 75.44
RF+ spectral indices+topographic factors Non-vegetation 82.62 91.93 87.03 92.08 88.27
Grass 91.05 88.04 89.52
Trees 99.37 98.37 98.87
Crops 78.46 75.87 77.14
RF+ spectral indices+topographic factors+texture information Non-vegetation 83.36 92.57 87.72 92.89 89.46
Grass 92.78 86.96 89.78
Trees 99.50 98.63 99.06
Crops 80.16 82.97 81.54

The number of input variables for the RF method used to create season-based vegetation maps.

Category Description Input variables number
Topographic Elevation, slope, aspect 3
Spectral bands Blue, green, red, vegetation red edge, near-infrared, and SWIR 10
Spectral indices NDVI, NDBI, MNDWI, NDTI 4
Textural information Variance, contrast, dissimilarity, homogeneity, correlation 5×10
Total variable 67
eISSN:
1178-5608
Language:
English
Publication timeframe:
Volume Open
Journal Subjects:
Engineering, Introductions and Overviews, other
Journal RSS Feed