nonparametric algorithms

Cover-management factor (C) is one of the most important influential factor on soil erosion using the Revised Universal Soil Loss Equation (RUSLE) model. C-factor is challenging to determine based on the proposed procedures due to lack of accurate information. Vegetation cover map can be used to estimate C-factor, but preparing a suitable mapping of vegetation cover is challenging in many situations. Therefore, in this study vegetation cover map was prepared and compared using the k Nearest Neighbor (k-NN) algorithm, linear regression (LR) and linear stepwise regression (LSR) in the study area. In regression methods, 17 vegetation and environmental indices were prepared and their relationships were investigated. The results of comparing the three methods showed that the k-NN method has better results than other regression methods due to its highest overall accuracy (83.3%) and kappa coefficient (75.9%) therefore, it was used to produce C-factor map. Results showed that the k-NN was very promising for mapping vegetation canopy cover in the arid and semi-arid areas. The results showed that among vegetation indices NDVI had the highest correlation (0.82) with percentage vegetation cover. Also, in the k-NN method, the Euclidean distance metrics in k = 9 has better results than the other two Fuzzy and Mahalanobis distances and can be used to estimation of vegetation cover map.

Estimating forest attributes is essential for understanding the condition and function of the forest to be applied in forest planning and management. The purpose of this study was to estimate the structural attributes of conifer-dominated plantations using radio detection and ranging (RADAR) polarimetric data and nonparametric algorithms in the Arabdagh region of Golestan province. Field-based structural attributes were collected from 319 circular plots with 400 m2 areas designed within a random cluster method. Within each plot, diameter at breast height (for all trees) and height (for some trees) were measured. The precise position of plots ere also recorded. Then stand volume, basal area, and the number of stem per ha were calculated. The required preprocessing and processing were conducted on raw RADAR data, followed by the extraction of plot-based values from the derived indices. Model training was done on 75% of plots using random forest, support vector machine, and K nearest neighbor algorithms. Results were validated with the remaining 25% of the plots. The results showed the lowest Root Mean Square Error and Bias for Random Forest algorithm for basal area 50.62% and -1.7%, respectively. Moreover, the support vector machine model achieved 58.82% and -7.94% for volume as well as 52.07% and -5.1% for no. of trees per hectare. As a whole, this study showed that the full polarization PALSAR-2 data has a moderate ability to estimate the quantitative structural attributes due to the high amplitude of changes in the quantitative forest characteristics.

In this study we evaluated the potential of ALS data in estimation of height, volume, basal area and DBH and canopy cover area of individualtrees for the part of Shast kalate of Gorgan.

In this study 125 tree that located in dominant story and without overlay with adjacent trees, were selected. Tree species, tree diameter at breast height (DBH) and tree crown diameter were measured. The height of trees was measured using a Vertex VL 402. Center coordinates of sample trees were determined using Digital Golobal Position System. After separating the crown border of a single tree and providing a polygon of the boundaries of them using aerial digital images, all of height and density metrics were created.Then, we explored the possibility of defining relationships between combination of airborne laser scanning data and height, volume, basal area, DBH and canopy cover area of individual trees using machine learning algorithms (random forest (RF), support vector machine (SVM), k-nearest neighbor (k-NN) and artificial neural network (ANN)).

The best RMSE% on an independent validation data for height, volume, canopy basal area, diameter, and basal area were 13.39, 56.88, 33.17, 22.34 and 25.88%. Also, the results demonstrate that the ANN algorithm can be useful for modeling biophysical properties of individual tree in the North of Iran.

Overall, the results showed the ALS data has the ability to estimate of tree height, basal area, diameter at breasts, and canopy cover but this data hasn’t the ability to estimate the volume very accurate. Also, the results showed that between all algorithms, the ANN algorithm have a better performance than other algorithms.