Original scientific paper
Assessment of Stand Structural Elements on the Basis of Spectral Reflectance Values of an IKONOS Satellite Image
Ante Seletković
; Šumarski fakultet Sveučilišta u Zagrebu, Zavod za izmjeru i uređivanje šuma, Svetošimunska 25, HR-10 000 Zagreb, HRVATSKA
Renata Pernar
; Šumarski fakultet Sveučilišta u Zagrebu, Zavod za izmjeru i uređivanje šuma, Svetošimunska 25, HR-10 000 Zagreb, HRVATSKA
Mario Ančić
; Šumarski fakultet Sveučilišta u Zagrebu, Zavod za izmjeru i uređivanje šuma, Svetošimunska 25, HR-10 000 Zagreb, HRVATSKA
Jelena Sučić
; Šumarski fakultet Sveučilišta u Zagrebu, Zavod za izmjeru i uređivanje šuma, Svetošimunska 25, HR-10 000 Zagreb, HRVATSKA
Abstract
Rapid technological advances in the second part of the 20th century have brought about immense changes in data collection methods, especially in developed countries. In addition to standard terrestric data collection methods, forest data are increasingly being collected with remote sensing methods. The application of remote sensing reduces the amount of field work and offers the possibility of saving time and money.
Satellite remote sensing is a very efficient method, while satellite images are a useful means for the surveillance and study of forest vegetation. Remote sensing is based on the prediction of the relationship between spectral reflectance and the observed variable. To interpret this kind of relationship, it is important to know the reflective characteristics of the target object (spectral features of vegetation, leaf, etc.), the size of area presentation (spatial resolution), and other factors such as topography, sun height, slope and orientation between the surveyed area and the survey sensor. A detailed and high quality analysis of satellite imagery, recognition and selection of certain data is highly dependent on image resolution, and especially on spatial resolution.
Acquiring information on forests from satellite photographs of high spatial resolution has been the subject of a large number of research activities. One of the approaches involves classification based on pixel value, or regression analyses in which spectral records are used to predict classes or continuous variables of stand structure.
The basic objective of this work is to explore the relationship between spectral reflectance recorded in the IKONOS satellite image of high spatial resolution and any individual stand parameter, as well as regression models for the evaluation of stand parameters.
The main material for this work was provided by the IKONOS satellite image of high spatial resolution for the Spačva basin area. The survey was conducted on 18 October 2006. An IKONOS satellite image of the Spačva basin with an area of 132 km² was delivered in 5 spectral channels: PAN (1x1m), and 4 MS Bundle. In the area of Vinkovci Forest Administration, within the management units encompassed by satellite surveying (MU Vrbanjske Šume, Kragujna, Slavir and Otočke Šume), field measurements were conducted in management classes of pedunculate oak and narrow-leaved ash through all age classes (2nd to 7th age class), for the purpose of subsequent comparison of interpretations and assessments of structural elements. Based on field measurements and Management Plan data, a database with compartments and subcompartments was established within two management units: Vrbanjske Šume and Slavir, where the largest number, 504 compartments in all, belong to the management class of pedunculate oak. For each compartment, the mean value of spectral reflectance within four channels was read from the satellite image. These read values were associated to the database mentioned earlier.
Descriptive statistics was performed for all the variables. In all the analyses, the significance level of 5% was considered statistically significant. Canonical correlation analysis was used to examine the relationship between a linear combination of channel value and a linear combination of field data value for each feature separately (number of trees, diameter at breast height, height, basal area and volume). Since there is no uniform solution, a linear combination should be selected which provides the best correlation among the monitored variables.
The Pearson’s correlation was used to study the relationship between arithmetic means of spectral reflectance values by channels and variables. The results of the assessment in the satellite image were compared with field measurement data, and with the data taken from the Management Plan. These data were used to generate the relationships between spectral reflectance recorded in the satellite image and a particular stand parameter, and consequently, the regression models for the assessment of stand parameters. Two regression models were assessed for each stand parameter within the management class of pedunculate oak, separately for each age class. The first model was assessed by means of the stepwise procedure, wherein the independent variables were the values of all four channels. A combination of the channels that best explains the dependent variable is selected. The second regression model contained a combination of all the four channels together (Table 2-6).
The best correlation between the observed variables and spectral values of individual channels was determined by means of canonical correlation.
Table 1 shows that canonical correlation for all the observed features in the third age class is higher than 0.9. In the third age class, the first canonical solution for the combination with four channels with the combination of tree number and correlation of 0.95 explains (overlaps) 90% of the variability. The poorest correlation of 0.46 proved to be for breast diameter and for all four channels in the sixth age class, with variability overlap of 21.19%. In terms of canonical correlation for tree number by all age classes, good correlations are observed in the 2nd (0.91), 3rd (0.95), and 5th (0.91) age class, where 80-90% of variability is explained. In contrast, correlation in the 6th and 7th age class was from 0.75 and 0.65, with determination coefficients of R² = 56.4% and 42%. Very similar results were obtained for volume, where the poorest correlation was found in the 6th (0.69) and 7th (0.76) age class.
The results of canonical correlation analysis allow us to conclude that the observed variables are in good correlation with all the four channels for all age classes, except for age class 4, which cannot be assessed due to low spatial participation of this age class in the sample.
As for the possibility of stand parameter assessment in satellite images, the results of regression models for the third age class proved to be the best. These results relate to the number of pedunculate oak trees and the total number of trees, the total basal area and the total volume. In general, the results of regression model assessment for the third age class show that all the observed parameters (number of trees, breast diameter, height, basal area and volume) are very well assessed in models with all four channels and determination coefficients of 60% and 80%. Stand parameters of the second and fifth age class are also well assessed. It should be pointed out that, according to research results, there is no justification to assess stand parameters for all the observed variables in the sixth and seventh age class due to low values of determination coefficients.
Since data based on spectral information from pixels were used to assess stand structural elements, a possible cause of poorer assessment results of stand parameters for the 6th and 7th age class may be attributed to the spatial resolution of the IKONOS satellite image. Namely, images of high spatial resolution frequently show isolated pixels (classified as a class) inserted into the area that represents the second class, which makes further analysis and application of the satellite image more difficult.
Keywords
IKONOS; spectral reflectance values; regression models; assessment of stand parameters
Hrčak ID:
68188
URI
Publication date:
8.4.2011.
Visits: 1.952 *