Data Transformation: Remote Sensing
Contents
TOC o “1-3” h z u HYPERLINK l “_Toc323462844” CHAPTER TWO PAGEREF _Toc323462844 h 2
HYPERLINK l “_Toc323462845” LITERATURE REVIEW PAGEREF _Toc323462845 h 2
HYPERLINK l “_Toc323462846” Introduction PAGEREF _Toc323462846 h 2
HYPERLINK l “_Toc323462847” Mapping and Classification of Forest Species PAGEREF _Toc323462847 h 3
HYPERLINK l “_Toc323462848” Continuum-Removed Absorption PAGEREF _Toc323462848 h 4
HYPERLINK l “_Toc323462849” Data Transformation through Discrete Wavelet PAGEREF _Toc323462849 h 8
HYPERLINK l “_Toc323462850” Data Transformation through First derivative PAGEREF _Toc323462850 h 10
HYPERLINK l “_Toc323462851” Conclusion PAGEREF _Toc323462851 h 11
HYPERLINK l “_Toc323462852” Bibliography PAGEREF _Toc323462852 h 14
CHAPTER TWOLITERATURE REVIEWIntroductionRemote sensing methods have found their application in evaluating the absorption of foliar macro elements for various grasses and plants even though it has mostly been done under laboratory conditions, which are generally controlled in nature. The increased use of remote sensing in management of forest and natural resources is as a result of substantial advancements in spectral resolutions coupled with advancement in data processing techniques over recent years. These have enabled production or generation of meticulous maps for cataloging forest communities, specific plant species, groups and sub-groups of species, in addition to forest species can be generated to provide a better source of information for an array of administrative resolutions and environmental applications.
Nevertheless, remote sensing is a vital technique that helps in boosting the understanding of flora and fauna in terms of feeding patterns and other living patterns. It is therefore imperative that the quality of data collected be of high quality to improve the quality of judgment from the information gathered. Therefore, in order to improve on discrimination and classification of forest species, it is essential that the data acquired via remote sensing be as helpful and high quality as possible. Data transformation is one of the techniques that have been used to remove the noise contained in the hyperspectral reflectance data to obtain accurate measurements of biochemicals and macronutrients in the forest species. In light of this, this literature review focuses on three data transformation methods that help in noise removal in the hyperspectral data obtained through remote sensing of Eucalyptus. The methods focused on are Continuum removal, discrete wavelet and First derivative.
Mapping and Classification of Forest SpeciesMapping and classification of the spatial distribution of individual Eucalyptus species or any other forest species is a significant ecological subject that calls for sustained study to correspond with developments in remote sensing equipments. The use of high spatial resolution (80cm) or hyperspectral remote sensing imager data has been scrutinized and examined by various studies for mapping tree biophysical aspects in different places around the world. For instance, Goodwin et al (2010) conducted a study in which they used hyperspectral data from the Compact Airborne Spectrographic Imager 2 (CASI-2) for discrimination, classification and development of appropriate mapping for the Australian eucalypt forest. In the study, CASI-2 provided an operative dataset to enable effectual discrimination and allow for generation of maps for spectrally multifaceted species thereby allowing for successful sub-genus grouping. The use of high spatial resolution dataset enabled proper discrimination of individual tree crowns of the Eucalyptus their crown aspects like sunlit aspects and shaded aspects.
On the other hand, spectral data from ten narrow bands ranging from 400nm to 700 nm in the visible range and data from ranges from 700nm to 1300nm within the near-infrared wavelengths provided comprehensive data for thin foliage absorption and reflective aspects (for instance, the green reflectance climax at 550 nm). The findings of the study indicated that economically significant Syncarpia glomulifera (Turpentine) can be successfully discriminated within communities that have assorted species. While the study employed a multi-stage analysis, strong spectral resemblance were exhibited by the foremost two phases of analysis for specific eucalypt tree species that were once more replicated by low CASI-2 categorization accuracies. Spatial resolution relates to the size of the pixel and the acquisition of higher spatial resolutions not only bolsters the ability to detect targets besides allowing the assessment of spatial correlations between pixels within particular tree crowns (Franklin et al. 2000). In contrast, Spectral resolution indicates the bandwidth response for a specific band. Constricted spectral bands (<10nm) can intensify the number of bands documented for a given optical region and help in targeting particular absorption attributes such as chlorophyll absorption of the plant leaves.
Continuum-Removed AbsorptionVarious studies have established that continuum-removed absorption is a useful data transformation method that can be used to compare the predicted measurements of mineral distribution against the mapping band shapes of the remotely sensed data (e.g. Roberts et al 2011; Datt 2000; Muya and Oguge 2000). Kokaly and Clark (1999) conducted a study in which they applied the continuum-removed absorption method alongside a refined method of band depth analysis of biochemical absorption features in studying dried plant material of Eucalyptus species and found that when these two methods are used three problems are overcome. In a follow-up study by Kokaly (2001), the study found that this method can appropriately be used in vegetation science.
Since the remote sensing of macronutrients helps in determination of plant quality in terms of determination of plant growth and development or health status, studies have employed the continuum-removed absorption method to help in understanding the plant quality of Eucalyptus especially in the tropical rangelands (e.g. Datt 2000). In their study, Muya and Oguge (2000) found that the use of continuum-removal minimizes the problem of over-fitting when near-infrared spectroscopy laboratory methods are used to estimate macronutrients at the canopy level of the tropical rainforests.
First, the problem of inconsistency that has been noted when methods such as multiple linear regression analysis are used across different vegetation types. In addition, these regression methods suffer problems of over-fitting and when the number of wavebands used is more than the samples, there is a higher likelihood of getting higher spectral variability (Kokaly & Clark 1999). Since spectral variability is independent of biochemical concentration, spectral variability is another problem that is solved when continuum removal is used in data transformation given that by using continuum removal the known chemical absorption features of the Eucalyptus are standardized (Zhang et al 2006).
Another problem that necessitates noise removal in the spectral data is that when dealing with canopies in large forest species, water that may be present in the canopies masks absorption features thereby complicating the remote sensing of the biochemicals more so at the field level (Cheng et al 2011). This problem becomes worse when soil background features together with atmospheric absorption effects are considered. Thus, the use of continuum removal standardizes the data by overcoming these problems and removing undesirable noise from the spectral data. When Kokaly and Clark (1999) used the band depth analysis with continuum removal in their study, they established that there is a strong correlation between nitrogen concentration and absorption (r2 = 0.95). They used dried plant material of Eucalyptus sub-genera. In another laboratory experiment, Curran et al., (2001) applied the methodology used by Kokaly and Clark on 12 macronutrients and achieved higher accuracy. It is however notable that most studies have been conducted under laboratory conditions rather than field level. In addition, not many studies have aimed at exploring or estimating foliar nutrient status of certain nutrients such as calcium, potassium, and magnesium.
The following diagram shows a graph of reflectance plotted against wavelength.
Figure SEQ Figure * ARABIC 1: Mean canopy spectrum and whole fresh leaf spectrum.
In the diagram, HYMAP 3-m spatial resolution data for 60 mature eucalypt trees was obtained. The researchers, Huang et al (2004) were interested in getting comparatively pure spectrum hence they located each of the individual eucalypt using the false color image of the HYMAP. They achieved this via field inspection. Since the sensor receives reflectance from both vertical mixture of the foreground and the background, separating tree pixels from the adjacent pixels is considerably easy given that tree pixels are different color to the background (Huang et al 2004). When the researchers applied the continuum removal data transform method, they obtained the following spectral profile.
Figure SEQ Figure * ARABIC 2: Continuum Removed Spectral Profile of eucalypt
For the study, Huang et al (2004) used the continuum removal method calculated as the band depth normalized as a ratio of the band depth at the centre of the absorption aspect and they used the following formula:
Where:
R is the reflectance at the waveband under consideration Ri represents the reflectance of the continuum line at the waveband being considered, Rc represents the reflectance at the centre of the absorption feature and Ric represents the reflectance of the continuum line at the center of the absorption feature.
In another study, Mutanga and Skidmore (2003) carried out a study in which they aimed at developing further and extending the band depth analysis method to estimate the concentration of the above macronutrients. The researchers enhanced the accuracy and validity of their study by combining the short wave infrared absorption effects that had earlier been used by Kokaly and Clark with two other key absorption features situated in the visible region. The effect of water is minimal in this region. Mutanga and Skidmore (2003) further developed and tested a modified first derivative reflectance approach to enhance the objective of continuum removal in data transformation.
Data Transformation through Discrete WaveletWavelet transform has in recent times become a very popular data transformation method when it comes to analysis, noise removal and compression of signals and images. Various research studies have been carried out exploring the latent benefits of combining active and passive remotely sensed data for assessment of forest structures (e.g. Banskota et al 2011). Image fusion has been applied as a way maintaining the incongruent data features that might be relevant to mapping of the forest structures under consideration. In the study carried out by Jan et al (2011), Eucalyptus plantations in the midlands of South Africa were studied using the near-infrared and the visible bands of ASTER and a fine beam Radarsat-1 images. ASTER is the Advanced Spaceborne Thermal Emission and Reflection Radiometer. The researchers obtained the data and modified it using the discrete wavelet transformation. In addition, the researchers obtained spatially documented data sets for the 38 plantations for the sake of comparisons between the measured data and the referenced data. In order to test whether fused data sets could produce better statistical models, the researchers applied ordinary least squares regression and multiple regression analyses to obtain empirical relationships. In their findings, it was established that single bands from both data sets did not provide sufficient adeptness for modeling basal area or even merchantable volume of timber. The adjusted R2 produced values that ranged below 0.3. when they used an optimized multiple regression, they got improved results in terms of mean and standard deviation when they compared the results to those generated from single bands (also in Zhang et al 2006). Nevertheless, these were still found to be unsuitable for application or mapping of forest species (Gong et al 2001). Williams and Amaratunga (1995) used Discrete Wavelet transform in their study and obtained the following results after data transformation.
Figure SEQ Figure * ARABIC 3: Discrete Wavelet Transform using Low Pass (a) and High Pass (b) (Williams & Amaratunga 1995)
Studies have shown that since more vigorous statistical methods are requisite for investigating phenological time series due to their characteristic of being noisy and non-stationary, wavelet transform analytic methods have been found to handle such data easily (e.g. Zhang et al 2006). Hudson et al (2011) conducted a study in which they delved to characterize flowering of eucalypt and the climate influences this flowering. In the study, the researchers used wavelet transform to remove noise from the remote sensing data. They used maximal overlap discrete wavelet transform to analyze the flowering records of four Eucalyptus subgenera. The flowering records were for the period between 1940 and 1970 and identified four sub-constituents in each flowering sequence. The subcomponents were the non-flowering phase, the duration cycle, the annual cycle and intensity cycle. A diminishing overall tendency in flowering was recognized by the maximal overlap discrete wavelet transform when the series were smoothed. Similar results were achieved by Cheng et al (2011), in which the researchers observed that wavelet correlation found the same simultaneous effects of climate on flowering for all the four Eucalyptus subgenera. When the researchers carried a wavelet cross-correlation analysis, they found that rainfall and temperature have a cyclical effect on the peak flowering intensity of Eucalyptus (P < 0.0001). For every species of Eucalyptus, there are 6 months of the yearly cycle in which any particular climate variable affects flowering intensity positively. In the same cycle, there are 6 months that any specific climate variable influences flowering peak negatively. The study established that for all the Eucalyptus species, rainfall wields a negative impact as long as the temperature is positive.
In another study, Curran et al (1995) used wavelet data transform method to investigate the relationship between reflectance of near infrared or visible beam and the chlorophyll content in Eucalyptus leaves. In the study, the reflectance properties of near infrared and the visible beam for the leaves of several Eucalyptus were analyzed to establish suitable indicators for remotely sensing the chlorophyll content in the leaves (Hudson et al 2011). The study employed the use of a scatter correction method to the reflectance spectra to decrease the additive and multiplicative scattering consequences of foliar surface and interior structure. The study further established that with an improved calibration of the chlorophyll content, reflectance near 710nm wavelength demonstrated greatest response to chlorophyll content. Moreover, reflectance near 550nm showed a less sensitivity to chlorophyll content in the Eucalyptus leaves. Generally, there are two levels of discrete wavelet transform that can be used to transform hyperspectral data (Hudson et al 2011). There is the single level technique and the multiple level decomposition technique.
Data Transformation through First derivativeVisible and near infrared areas of spectrum frequently exhibit spectral differences, which are often used to describe various vegetation classes. Reflectance spectrum and first derivative spectrum that stretch from 350 to 700 nm are used to enhance the shape disparities between the spectral signatures for every tree species used or included in a sample. By precisely encapsulating these spectral differences, it is possible to improve vegetation classification by allowing for investigation of band ratios and vegetation indices. In the study done by Datt (2000), the best performing reflectance index ratio was the (R850-R710)/ (R850-R680) and hence it was proposed as the new index for estimating chlorophyll content in higher plants such eucalypt.
While Huang et al (2004) conducted a study in which they employed continuum removal to transform the hyperspectral data for the eucalypt tree, they also used standard derivative method to transform the same data and compare the outcomes. However, the researchers used the standard derivative data transform method to estimate nitrogen concentration in the eucalypt trees. They used the standard derivative method to transform the log (1/R) spectra data where R is the reflectance at the waveband under consideration. In order to reach the best possible combination for the average spectra as well as the maximum spectra, the researchers tried various scatter correction methods. They however established that Modified Partial Least Squares method resulted to a higher coefficient of determination when maximum spectra were used as compared to when the mean spectra were used. The results of the study by Huang et al (2004) agree with the findings by Mutanga and Skidmore (2003) where the latter carried out a study to investigate the correlation between nitrogen absorption and the chlorophyll level in eucalypt. For the study by Huang et al (2004), when the standard derivative method was used for nitrogen absorption in eucalypt, the following diagram illustrates the results.
Conclusion
Developing maps for spatial distribution of particular species is a significant ecological aspect that calls for sustained research to match the advances achieved in remote sensing technologies. Due to the characteristic of the hyper-spectral data of being noisy and non-stationary, it is imperative to use any of the wavelet transform analytic methods, which have been found to handle such data easily. The literature review has revealed that the of wavelet transform methods or techniques helps in removing noise from the hyperspectral data hence improve quality of the data for the sake of comparison with referenced data for individual plant species. The review has also revealed that the studies that used maximal overlap discrete wavelet transform to analyze the flowering records of forest species were effective in reducing inconsistencies and achieving increased accuracies. Out of many studies reviewed, many of them established that the use of data transform techniques for noise removal was not only a good approach to improving discrimination and classification among individual plant species but also increased discrimination and classification accuracies. The studies reviewed further show that reflectance spectrum and first derivative spectrum that stretch from 350 to 700 nm are effectively used to enhance the shape disparities between the spectral signatures for every tree species used or included in a sample. This is achieved by precisely encapsulating the spectral differences thereby improving vegetation classification by allowing for investigation of band ratios and vegetation indices.
The findings of the literature review also indicate that due to the economic viability of most of the eucalypt sub-species such as Syncarpia glomulifera (Turpentine) discrimination within communities that have mixed species can be achieved successfully using data transformation methods highlighted such as continuum removal, standard derivative and wavelet transformation. Lastly, the review has also showed that the math treatment method adopted also affects the consistency of the results hence it is important to choose method astutely. Most studies successfully used modified least squares regression method with increased accuracy.
BibliographyAsner, G. P., Wessman, C. A., Bateson, C. A., and Privette, J. L., (2000): “Impact of tissue, canopy and landscape factors on the hyperspectral reflectance variability of arid zone ecosystems” Remote Sensing of Environment” 74, pp. 69- 84.
Banskota, A., Wynne, R.H., & Kayastha, N. (2011). “Improving within-genus tree species discrimination using the discrete wavelet transform applied to airborne hyperspectral data” International Journal of Remote Sensing, 32, 3551-3563
Cheng, T., Rivard, B. and Sánchez-Azofeifa A. (2011) “Spectroscopic determination of leaf water content using continuous wavelet analysis” Remote Sensing of Environment 115 (2): 659–670
Clark, R. N., and Roush, T. L., (1984): “Reflectance spectroscopy: quantitative analysis techniques for remote sensing applications” Journal of Geophysical Research, 89, pp. 6329-6340.
Clevers, J. G. P. W., (1999): “The use of imaging spectrometry for agricultural applications. ISPRS” Journal of Photogrammetry and Remote Sensing, 54, pp. 299-304.
Clevers, J., and Buker, C., (1991) “Feasibility of the red edge index for the detection of nitrogen deficiency” Proceedings of the 5th International Colloquium – Remote sensing, 165 – 168.
Curran, P. J., (1989) “Remote sensing of foliar chemistry” Remote Sensing of Environment, 30, pp. 271-278
Curran, P. J., Dungan, J. L., and Peterson, L., (2001) “Estimating the foliar biochemical concentration of leaves with reflectance spectrometry: testing the Kokaly and Clark methodologies” Remote Sensing of Environment, 76, pp. 349-359.
Curran, P. J., Dungan, J. L., Macler, B. A., and Plummer, S. E., (1991) “The effect of a Red leaf pigment on the relationship between red edge and chlorophyll concentration” Remote Sensing of Environment, 35, pp. 69-76.
Curran, P. J., Windham, W. R., and Gholz, H. L., (1995) “Exploring the relationship between reflectance red edge and chlorophyll concentration in slash pine leaves” Tree Physiology, 15, pp. 203-206.
Datt, B. (2000). “Recognition of eucalyptus forest species using hyperspectral reflectance data” In T.I. Stein (Ed.), Igarss 2000: Ieee 2000 International Geoscience and Remote Sensing Symposium, Vol I – Vi, Proceedings (pp. 1405-1407). New York: Ieee
Dawson, T. P., and Curran, P. J., (1998) “A new technique for interpolating the reflectance red edge position” International Journal of Remote sensing, 19, pp. 2133-2139
Elvidge, C. D., (1990) “Visible and near infrared reflectance characteristics of dry plant materials” International Journal of Remote Sensing, 11, pp. 1775-1795
Fillella, I., and Penuelas, J., (1994) “The red edge position and shape as indicators of plant chlorophyll content, biomass and hydric status” International Journal of Remote sensing, 15, pp. 1459 – 1470.
Foley, B., Mcllwee, A., Lawler, I., Agragones, L., Woolnough, A. P., and Berding, N., (1998) “Ecological applications of near infrared spectroscopy – a tool for rapid, cost effective prediction of the composition of plant and animal tissues and aspects of animal performance” Oecologia, 116, pp. 293 – 305.