Інвентаризація інвазій комах на підставі космозображень як засіб моніторингу лісових ландшафтів



1. Introduction. Rapid changes in the environment bring about the need to landscape monitoring in a quick and repetitive way. Remote sensing, aerial as well as satellite, fulfils the requirements necessary for it to become a monitoring tool i.e. it provides objective and repeatable data.

Especially landscape engineering can be expected to take interest in remote sensing.

Owing to the data provided by remote sensing the Earth's surface can be observed in a synthetic and objective way. Various types of sensors installed on aircraft and satellite platforms acquire the data. Their variety guarantees fast data acquisition. This is crucial when disasters are monitored. And disasters are likely to occur more frequently due to climate changes and according to specialists we are exposed to them more and more often.

It is characteristic that in forests affected by stressors insect infestation appears subsequent to other factors impairing trees and forest stands, such as: fires, pollution, biocenosis simplification due to forest management etc.

The paper deals with the application of satellite images acquired by Landsat Thematic Mapper sensor in monitoring feeding areas of conifer sawfly (Diprionidae) in Kozienice Forest in central Poland. The analyses were carried out with the help of a geographic information system, which served to integrate inventory and image data, and enabled processing, satellite image interpretation, classification and spatial analysis.

2. Search of spectral bands useful for distinguishing defoliation classes. As the first stage, univariate analysis of variance was carried out in order to establish whether there exist any differences of statistical significance between the groups (categories). For this analysis the groups consisted of 20% defoliation classes: 0-19%, 20-39%, 40-59%, 60-79%, above 80%. Univariate analysis of variance, in which the equality of means in the chosen defoliation classes was investigated, was carried out separately for seven raw Landsat TM bands and several band ratios. The band ratios were the following: TM4/TM3 called the biomass index,

TM5/TM4 - the damage index, BR =, called the brightness index and the vegetation index NDVI =.

Furthermore, before the univariate analysis of variance was attempted, control plots were assessed for their homogeneity and therefore for their representativeness. Because of the pixel structure of raster images, spectral reflectance mean calculated for a given plot is not always representative since it is calculated for varied pixel values representing e.g. openings or an admixture of other species included in a stand. Following the examination of standard deviation values and visual assessment, 109 spectrally homogenous stands were chosen for further analysis.

The results of univariate analyses of variance helped to define the bands that allow to distinguish specific defoliation classes as well as their mean values for a given range, confidence intervals, standard errors and top and bottom spectral reflectance boundaries.

Spectral analysis of Landsat TM images made it possible to distinguish homogenous defoliation classes in all spectral bands and band ratios. Band ratioing was applied in order to calculate the biomass, damage, brightness and vegetation indices. The results of univariate analysis of variance (table 1) show that: 1) for TM1 band, which registers the blue region of the spectrum, two groups can be distinguished: a class aggregating stands with 0-39.9% assimilatory apparatus loss (class 10+30) and a class with defoliation ranging above 40% (class 50+70+80); 2) for TM2 band (green radiation) the mean value for class 80 falls between the classes 10 and 30 and the classes 50 and 70, therefore the decrease of registered value is observed. Due to this, not only is it difficult to distinguish class 80 within this spectra range but also it makes it harder to distinguish classes 10+30 and 50+70; 3) for TM3 band (red region of the spectrum) every subsequent defoliation class was registered with a greater reflectance value but nevertheless the value ranges for separate classes overlap and it is possible only to distinguish two groups of assimilatory apparatus loss (just as for TM1 band): class 10+30 and class 50+70+80; 4) similarly, two groups can be distinguished within TM4 band (near infrared radiation), but the first group includes stands from the classes 10, 30, 50, 70 and the second stands from class 80, that is those with large extent of defoliation. Within class 80 the mean spectral reflectance value greatly decreases; 5) like for TM3 band, also for TM5 band (near infrared radiation) every subsequent defoliation class was assigned a greater mean reflectance value. However, the classes are much more easily distinguishable and it is possible to recognize 4 classes: class 10, class 30, class 50 and class 50+80; 6) TM6 band (far infrared), despite the fact that it registers 120x120 pixels, makes it possible to distinguish 3 groups: class 10+30, class 50+70 and class 80; 7) Band TM7 (middle infrared radiation) assigned a greater mean reflectance value to every subsequent class. A test for group homogeneity showed each class as a separate group, however, mean values for classes 10 and 30 are close to each other and therefore these classes can only be distinguished under conditions favorable from the point of view of spectral class discernment; 8) for the band ratio which is the biomass index only class 80 was distinguished,

although a certain regularity can be observed: classes with less than 40% defoliation have similar, greater values, whereas the following classes, stands with a greater assimilatory apparatus loss have lower biomass index values; 9) the band where upper defoliation classes could be clearly distinguished was the band ratio called the damage index. Classes 10 and 30 are grouped together but every next class (50, 70, 80) was classified as a separate group; 10) the vegetation index, like the biomass index, distinguished the class with defoliation above 80% as a separate group. Also the distribution of mean index values in defoliation classes is similar to mean biomass index values: index values for classes 10 and 30 are high and similar (there is even a slight increase in mean value between class 10 and 30), later there are a decrease in index value for every following class; 11) brightness index, similarly to the one above, distinguishes only stands with defoliation above 80% as a separate group.

Table 1. The results of the univariate analysis of variance.

 The results of the univariate analysis of variance

Defoliation classes can be distinguished most clearly for the following bands: TM7, TM5 (both are middle infrared bands) and damage index - a band ratio, which has also been selected as a stand defoliation index in studies dealing with the assessment of assimilatory apparatus loss in spruce stands.

3. Assessment of the relationship between spectral reflectance and defoliation of stands. In order to establish whether there is a relation between spectral reflectance registered with Landsat Thematic Mapper scanner and defoliation, and how strong this relation is, a regression and correlation analysis was carried out. Simple regression of defoliation was carried out for each raw band and band ratio (table 2).

Table 2. Simple regression results.

 Simple regression results

The strongest relation to pine stand defoliation occurs for: band TM7 (68.78%), damage index (60.05%) and band TM5 (59.56%). For the bands mentioned above correlation coefficients are higher than 0.77, whereas standard estimation error values are the lowest. The simple regression analysis results for these three bands with the greatest correlation to stand defoliation are presented below.

3.1. Simple regression of defoliation and TM5. The equation of simple regression for defoliation as a dependent variable (y) and TM5 spectral reflectance value as an independent variable (x) has the following form:

DEF = -78,1734 + 2,6131*TM5

At 99% confidence level the value of determination coefficient R2 shows that the regression equation with TM5 as independent variable at 59.56% accounts for the variability of dependent variable DEF (defoliation), whereas the correlation coefficient equal to 0.77 indicates a moderately strong relation between the variables. According to standard estimation error SE, the standard deviation of defoliation distribution equals 8.78. This expression defines the dispersion of defoliation in relation to the regression line. Thus, it is the measure of random factor variability. Knowing its value, the significance level and the number of degrees of freedom, it is possible to establish confidence intervals for every defoliation value and for new TM5 observation values.

3.2. Simple regression of defoliation and TM7. The equation of simple regression for defoliation as a dependent variable (y) and TM7 spectral reflectance value as an independent variable (x) has the following form:

DEF = -43,0745 + 4,99709*TM7

The value of determination coefficient R2 (at 99% confidence level) shows that the regression equation with TM7 as an independent variable at 68.78% accounts for the variability of dependent variable DEF (defoliation), whereas the correlation coefficient equal to 0.83 indicates a moderately strong relation between the variables. According to standard estimation error SE, the standard deviation of defoliation variable distribution equals 7.72.

3.3. Simple regression of defoliation and damage index. The equation of simple regression for defoliation as a dependent variable (y) and damage index as an independent variable (x) has the following form:

DEF = -47,0289 + 93,0731* damage index

The value of determination coefficient R2 shows that the regression equation with damage index as an independent variable at 60.05% accounts for the variability of dependent variable DEF (defoliation), whereas the correlation coefficient equal to 0.77 indicates a moderately strong relation between the variables. According to standard estimation error SE, the standard deviation of defoliation variable distribution equals 8.73.

4. Assessment of defoliation on the basis of multi-channel satellite image analysis. Multiple regression, which investigates the relation between more than two measurable variables was also analyzed. The aim of this analysis was to use the regression function to correlate the dependent variable (defoliation) simultaneously with several other measurable variables (spectral reflectance values). Apart from the multiple regression equation, the square of multiple correlation coefficient R2 was also calculated taking into account the degrees of freedom. This coefficient defines the magnitude of relation between the dependent variable and the independent variables. Besides, significance levels of correlation coefficients for each independent variable were calculated.

The equation of multiple regression between defoliation as a dependent variable and 7 TM bands as independent variables has the following form:

DEF = 161,626 - 0,885156*TM1 + 3,59979*TM2 - 3,21512*TM3 - 1,57094*TM4 + 3,1199*TM5 - 1,06739*TM6 + 0,658071*TM7

At 99% confidence level the value of determination coefficient R2 shows that the regression equation accounts for 73.08% of defoliation variability, whereas the value of determination coefficient R2 corrected according to the number of degrees of freedom (more suitable to compare with models with a different amount of independent variables) is 71.20%. According to standard estimation error SE, standard deviation of defoliation variable distribution is 7.38%. This expression defines defoliation dispersion in relation to regression line, and thus it is the measure of random factor. Knowing its value, significance level and number of degrees of freedom, it is possible to calculate, for different defoliation values, confidence levels for predicted values.

Mean absolute error MAE, calculated for above equation defines the mean difference between measured defoliation value and the value calculated on the basis of multiple regression equation, where spectral reflectance TM values were independent variables.

In the regression equation where 7 TM bands were independent variables MAE error is 5.73, so the defoliation values measured and those calculated differ on average by 5.7% of assimilatory apparatus loss.

Multiple regression analysis was also carried out for various band combinations. It turned out that the determination coefficient R2 corrected according to the number of degrees of freedom is 71.56% in a multiple regression of defoliation where bands TM4, TM5, TM6 were used as independent variables.

The equation of multiple regression between defoliation as a dependent variable and TM4, TM5, TM6 bands as independent variables has the following form:

DEF = 123,717 - 1,41305*TM4 + 3,2391*TM5 - 1,07399*TM6

The value of determination coefficient R2 shows that the regression equation accounts for 72.36% of defoliation variability, whereas the value of determination coefficient R2 corrected according to the number of degrees of freedom (more suitable to compare with models with a different amount of independent variables) is 71.56%. According to standard estimation error SE, standard deviation of defoliation variable distribution is 7.33%.

In the regression equation where TM4, TM5 and TM6 bands were independent variables MAE error is 5.81, so the defoliation values measured and those calculated differ on average by 5.8% of assimilatory apparatus loss.

The value of determination coefficient R2 corrected according to the number of degrees of freedom allows us to compare how precise different regression equations are. In table 3 values of coefficient R2 together with the regression coefficients are given for both regressions presented earlier. For the presented variants the square of multiple correlation coefficient R2 is higher than 70%.

Table 3. Multiple regression results

 Multiple regression results

5. Results. The analyses presented above show that spectral reflectance registered in raw bands as well as band ratios are highly correlated with defoliation of stands. As it can be concluded from the analyses, satellite images can be applied to define defoliation classes of pine stands, for which the most useful bands are TM 4, 5 and 7 and the damage index.

Theoretical considerations showed that TM7 makes it possible to distinguish classes with defoliation 0-19%, 20-39%, 40-59%, 60-79%, 80-100%. However, because of the fact that not only the condition of conifer needles but also other stand elements influence the mean spectral value registered by the satellite's scanner, this theoretical approach has to be modified. In practice, as shown in the analyses presented, it is possible (under optimum conditions) to distinguish three defoliation classes: 0-50%, 50-70% and above 70%.

The knowledge of spectral characteristics of each defoliation class as well as of bands highly correlated with defoliation, allowed for satellite image classification aimed at distinguishing unaffected pine stands and defoliated ones. As it has been mentioned before, as a result of processing, pine stands were distinguished from the remaining part of Kozienice Forest. Pine stands were divided into unaffected pine stands, pine stands with spectral characteristics placing them in the 50-70% defoliation class and pine stands with defoliation higher than 70%.

As a result of classification, 540 stands out of 6626 (stands in the analyzed area of Kozienice Forest according to inventory data) were found to include pixels (one or more) with spectral values characteristic of defoliation. Out of these 540 stands 522 included pixels pointing to defoliation from 50 to 70% and 98 stands to more than 70%. The stands with defoliation from 50 to 70% were defined as impaired forest stands and those representing a class of defoliation above 70% were described as heavily impaired stands.

Impaired stands appeared mostly in the subdivisions of Jedlnia (Radom division), Kozienice (Kozienice division), Zwolen and Garbatka (Zwolen division). The biggest area of impaired stands (damage by conifer sawfly) was Jedlnia subdivision. The next subdivision with an equally large area of impaired stands is Kozienice, the subdivision closest to Kozienice power plant.

Heavily impaired stands, like the impaired ones, appeared mainly in Jedlnia and Kozienice subdivisions, and also in Zwolen subdivision. Within Garbatka subdivision heavily impaired stands have been classified in a small area (1.48 ha). Such areas were not noted in Pionki and Zagozdzon subdivisions.

As a result of suitable satellite image processing based on statistical analyses presented in this paper, it became possible to distinguish stands at different stages of defoliation.

The application of a spatial analysis system helped to produce a cartographic presentation of stand impairment distribution on the forest map, as well as to carry out spatial analysis in order to characterize impaired and heavily impaired stands which occupied an area of 319.05 ha.

The parameters of impaired and heavily impaired pine stands distinguished on satellite images are presented in the table 4, with their average values and average values for all Kozienice Forest stands. As it can be concluded from this table, average impaired stand parameters, apart from crown closure, do not greatly differ from average values for all Kozienice Forest stands.

Multi-temporal analyses applying images acquired by Landsat TM (1994) and Ikonos (2000) satellites showed that during the 6 year period (1994-2000) forest condition improved. On the most recent satellite images defoliation due to insect infestation was not observed.

Table 4. Average parameter values for all Kozienice Forest stands and for impaired stands

 Average parameter values for all Kozienice Forest stands and for impaired stands

6. Conclusions. Research results presented in this paper show that remote sensing techniques can form an operating tool for monitoring post-disaster areas in forests. The applicability of remote sensing techniques does not have to be restricted to aerial photographs, already quite popular in environment monitoring. As high resolution satellite images and specialized software for digital processing of images become more accessible, their application can become more and more widespread. As illustrated by the research presented in the paper, satellite images can be used to locate impaired/defoliated forests (the damage being due to forest degradation caused by industrial air pollution, forest chain disease, fire or infestation).

The results show that, when monitoring pine stand defoliation, it is possible to distinguish forest condition classes with defoliation: 0-50%, 50-70% and above 70%.

Stratification of forests' condition through their defoliation assessment based on satellite images is possible only when middle and near infrared range images are applied. Thematic Mapper scanner installed in Landsat satellites and HRVIR scanner installed in the newest SPOT satellites fulfill this requirement.




На головну


Hosted by uCoz