Estimating forest structural attributes by means of ASTER imagery and CART algorithm (Case study: Shastkolateh forest, Gorgan)

• Main Authors: Nuroddin Noorian, Shaban Shataee, Jahangir Mohammadi, Salam Yazdani
• Format: Article
• Language: fas
• Published: Research Institute of Forests and Rangelands of Iran 2014-11-01
• Series: تحقیقات جنگل و صنوبر ایران
• Subjects: Forest structural attributes; ASTER imagery; CART algorithm; Shastkolateh forest
• Online Access: http://ijfpr.areeo.ac.ir/article_12424_8de17f370b8298b0cd24c400ddf5ff43.pdf

Large-area estimation of forest structural attributes by remotely-sensed data is crucial for cost effective inventory of the stands, and in turn for sustainable forest management. The objective of this research was to investigate the capability of Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) imagery for predicting forest structural attributes over Shastkolateh experimental forest in Gorgan. By means of random cluster sampling method, 112 DGPS-established square plots with an area of 0.09 ha were inventoried which were also homogenous by type and aspect. In those plots, the stand volume, basal area and tree stem density were measured. The image data was geometrically and atmospherically corrected. Moreover, information within the data was used to create additional band ratios, principal components, texture indices, and tasseled cap components, which were then added to the original datasets. Classification and Regression Trees (CART) algorithm was applied for modeling the ground inventory data. The models were assessed for their performance by means of root mean square error (RMSE) and Bias using hold-out samples. The results showed the best values of adjusted R-squared to be 76, 73 and 80% for stand volume, basal area and tree stem density, respectively. Whereas the models of standing volume, basal area and stem density retuned  RMSE vauues of 40.22, 38.67, and 58.68, the models were associated with bias values of 17.5 %, 8% and 2.72%, respectively. Results therefore indicate the moderate potential of ASTER imagery for sample plot-based estimation of forest structural attributes.