|
Using machine learning algorithms, Landsat 8 and Sentinel 2 satellite imagery for estimating forest canopy density of Zagros Forests
|
Ahmad Abbasiwand   , Hormoz Sohrabi * , Mojdeh Miraki |
|
|
|
Abstract: (6 Views) |
| The significance of canopy cover as a crucial variable in assessing, monitoring, and managing the Zagros forests cannot be overstated. Since canopy cover reflects various ecological changes, forest managers and related institutions require accurate and up-to-date maps and data to enhance forest management efforts. The integration of satellite imagery and machine learning models has significantly improved our understanding of forest ecosystems, enabling the development of effective conservation strategies and the prevention of resource depletion. This study utilized imagery from the Sentinel-2 and Landsat 8 satellites, employing machine learning models to analyze canopy cover in the Zagros forests. Five algorithms were tested: Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Machine (SVR), Multivariate Adaptive Regression Splines (MARS), and Artificial Neural Networks (ANN). To evaluate these models, 2,631 plots (each 400 meters in size) across 27 forest areas were selected, and canopy cover percentages were calculated. The results showed that using Landsat 8 imagery, the most effective models were MARS (63.7%), RF (62.7%), and ANN (61.7%), based on adjusted coefficient of determination values. With Sentinel-2 imagery, the best-performing models were MARS (73.7%), SVR (72.7%), and RF (71.7%). Comparatively, Sentinel-2 produced superior results, with its highest accuracy (73.7%) exceeding the best result from Landsat 8 (63.7%). Overall, canopy cover modeling across the Zagros forests indicated that the Random Forest algorithm (73.7%) outperformed other methods. Additionally, Sentinel-2 imagery proved more efficient than Landsat 8, demonstrating approximately 7% higher accuracy in estimating canopy cover. The ability to analyze and map canopy cover using satellite imagery and advanced machine learning models provides forest managers with cost-effective and reliable tools for making informed decisions regarding conservation and sustainable forest management. |
Article number: 2 |
| Keywords: Forest monitoring, plant indices, random forest, canopy cover |
|
|
Full-Text [PDF 1025 kb]
(5 Downloads)
|
Type of Study: Research |
Subject:
GIS
Received: 2024/11/28
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
