- F. Poncet von (1)
- M. Schlund (1)
- C. Schmullius (1)
- A.K. Skidmore (2)
- C. Vaiphasa (2)
- T. Vaiphasa (1)
Importance of bistatic SAR features from TanDEM-X for forest mapping and monitoring
Schlund, M. ; Poncet, F. von; Hoekman, D.H. ; Kuntz, S. ; Schmullius, C. - \ 2014
Remote Sensing of Environment 151 (2014)sp. issue. - ISSN 0034-4257 - p. 16 - 26.
land-cover - southeast-asia - feature-selection - polarimetric sar - tropical-forest - decision tree - alos palsar - rain-forest - sir-c - classification
Deforestation and forest degradation are one of the important sources for human induced carbon dioxide emissions and their rates are highest in tropical forests. For man-kind, it is of great importance to track land-use conversions like deforestation, e.g. for sustainable forest management and land use planning, for carbon balancing and to support the implementation of international initiatives like REDD + (Reducing Emissions from Deforestation and Degradation). SAR (synthetic aperture radar) sensors are suitable to reliably and frequently monitor tropical forests due to their weather independence. The TanDEM-X mission (which is mainly aimed to create a unique global high resolution digital elevation model) currently operates two X-band SAR satellites, acquiring interferometric SAR data for the Earth's entire land surface multiple times. The operational mission provides interferometric data as well as mono- and bistatic scattering coefficients. These datasets are homogenous, globally consistent and are acquired in high spatial resolution. Hence, they may offer a unique basic dataset which could be useful in land cover monitoring. Based on first datasets available from the TanDEM-X mission, the main goal of this research is to investigate the information content of TanDEM-X data for mapping forests and other land cover classes in a tropical peatland area. More specifically, the study explores the utility of bistatic features for distinguishing between open and closed forest canopies, which is of relevance in the context of deforestation and forest degradation monitoring. To assess the predominant information content of TanDEM-X data, the importance of information derived from the bistatic system is compared against the monostatic case, usually available from SAR systems. The usefulness of the TanDEM-X mission data, i.e. scattering coefficients, derived textural information and interferometric coherence is investigated via a feature selection process. The resulting optimal feature sets representing a monostatic and a bistatic SAR dataset were used in a subsequent classification to assess the added value of the bistatic TanDEM-X features in the separability of land cover classes. The results obtained indicated that especially the interferometric coherence significantly improved the separability of thematic classes compared to a dataset of monostatic acquisition. The bistatic coherence was mainly governed by volume decorrelation of forest canopy constituents and carries information about the canopy structure which is related to canopy cover. In contrast, the bistatic scattering coefficient had no significant contribution to class separability. The classification with coherence and textural information outperformed the classification with the monostatic scattering coefficient and texture by more than 10% and achieved an overall accuracy of 85%. These results indicate that TanDEM-X can serve as a valuable and consistent source for mapping and monitoring tropical forests.
Estimating and mapping grass phosphorus concentration in an African savanna using hyperspectral image data
Mutanga, O. ; Kumar, L. - \ 2007
International Journal of Remote Sensing 28 (2007)21. - ISSN 0143-1161 - p. 4897 - 4911.
kruger-national-park - neural-networks - reflectance spectroscopy - absorption features - feature-selection - south-africa - aviris data - nitrogen - classification - quality
We tested the utility of imaging spectroscopy and neural networks to map phosphorus concentration in savanna grass using airborne HyMAP image data. We also sought to ascertain the key wavelengths for phosphorus prediction using hyperspectral remote sensing. The remote sensing of foliar phosphorus has received very little attention as compared to nitrogen, yet it plays an equally important role in explaining the distribution and feeding patterns of herbivores. Band depths from two continuum-removed absorption features as well as the red edge position (REP) were input into a backpropagation neural network. Following a series of experiments to ascertain the optimum wavelengths, the best trained neural network was used to predict and ultimately to map grass phosphorus concentration in the Kruger National Park. The results indicate that the best trained neural network could predict phosphorus distribution with a coefficient of determination of 0.63 and a root mean square error (RMSE) of 0.07 (28% of the mean observed phosphorus concentration) on an independent test data set. Our results also show that the absorption feature located in the shortwave infrared (R 2015-2199) contains more information on phosphorus distribution, a region that has hardly been explored before in most spectroscopic experiments for phosphorus as compared to the visible bands. Overall, the study demonstrates the potential of imaging spectroscopy in mapping grass phosphorus concentration in savanna rangelands.
Consideration of smoothing techniques for hyperspectral remote sensing
Vaiphasa, C. - \ 2006
ISPRS Journal of Photogrammetry and Remote Sensing 60 (2006)2. - ISSN 0924-2716 - p. 91 - 99.
feature-selection - derivative analysis - vegetation - spectra - model - differentiation - discrimination - inversion - canopies - accuracy
Spectral smoothing filters are popularly used in a large number of modern hyperspectral remote sensing studies for removing noise from the data. However, most of these studies subjectively apply ad hoc measures to select filter types and their parameters. We argue that this subjectively minded approach is not appropriate for choosing smoothing methods for hyperspectral applications. In our case study, it is proved that smoothing filters can cause undesirable changes to statistical characteristics of the spectral data; thereby, affecting the results of the analyses that are based on statistical class models. If preserving statistical properties of the original hyperspectral data is desired, smoothing filters should then be used, if necessary, after careful consideration of which smoothing techniques will minimize disturbances to the statistical properties of the original data. A comparative t-test is proposed as a method for choosing a smoothing filter suitable for hyperspectral data at hand.
Tropical mangrove species discrimination using hyperspectral data: A laboratory study
Vaiphasa, C. ; Ongsomwang, S. ; Vaiphasa, T. ; Skidmore, A.K. - \ 2005
Estuarine Coastal and Shelf Science 65 (2005)1-2. - ISSN 0272-7714 - p. 371 - 379.
feature-selection - reflectance variability - chlorophyll - vegetation - canopy - classification - ecosystems - airborne - wetland - leaves
The aim of this study is to test whether spectra of crown canopy leaves of various tropical mangrove species measured under laboratory conditions contain sufficient spectral information for discriminating mangroves at the species level. This laboratory-level study is one of the most important prerequisites to the future use of airborne and satellite hyperspectral sensors for mangrove studies. First, spectral responses of 16 Thai tropical mangrove species (2151 spectral bands between 350 nm and 2500 nm) were recorded from the leaves, using a spectrometer under laboratory conditions. Next, the mangrove spectra were statistically tested using one-way ANOVA to see whether they significantly differ at every spectral location. Finally, the spectral separability between each pair of mangrove species was quantified using the Jeffries¿Matusita (J¿M) distance measure. It turned out that the 16 mangrove species under study were statistically different at most spectral locations, with a 95% confidence level (p <0.05). The total number of spectral bands that had p-values less than 0.05 was 1941, of which 477 bands had a 99% confidence level (p <0.01). Moreover, the J¿M distance indices calculated for all pairs of the mangrove species illustrated that the mangroves were spectrally separable except the pairs that comprised the members of Rhizophoraceae. Although the difficulties of discriminating the members of Rhizophoraceae are expected, the overall result encourages further investigations into the use of on-board hyperspectral sensors to see whether mangrove species can be separated when the difficulties of the field conditions are taken into account.
Integrating Imaging spectrometry and Neural Networks to map tropical grass quality in the Kruger National Park, South Africa
Mutanga, O. ; Skidmore, A.K. - \ 2004
Remote Sensing of Environment 90 (2004)1. - ISSN 0034-4257 - p. 104 - 115.
land-cover classification - chlorophyll content - red edge - absorption features - feature-selection - vegetation types - nitrogen - leaf - savanna - spectrometry
A new integrated approach, involving continuum-removed absorption features, the red edge position and neural networks, is developed and applied to map grass nitrogen concentration in an African savanna rangeland. Nitrogen, which largely determines the nutritional quality of grasslands, is commonly the most limiting nutrient for grazers. Therefore, the remote sensing of foliar nitrogen concentration in savanna rangelands is important for an improved understanding of the distribution and feeding patterns of wildlife. Continuum removal was applied on two absorption features located in the visible (R550-757) and the SWIR (R2015-2199) from an atmospherically corrected HYMAP MKI image. A feature selection algorithm was used to select wavelength variables from the absorption features. Selected band depths from the absorption features as well as the red edge position (REP) were input into a backpropagation neural network. The best-trained neural network was used to map nitrogen concentration over the whole study area. Results indicate that the new integrated approach could explain 60% of the variation in savanna grass nitrogen concentration on an independent test data set, with a root mean square error (rmse) of 0.13 (+/- 8.30% of the mean observed nitrogen concentration). This result is better compared to the result obtained using multiple linear regression, which yielded an R-2 of 38%, with a RMSE of 0.16 (+/- 10.30% of the mean observed nitrogen concentration) on an independent test data set. The study demonstrates the potential of airborne hyperspectral data and neural networks to estimate and ultimately to map nitrogen concentration in the mixed species environments of Southern Africa. (C) 2004 Elsevier Inc. All rights reserved.