Kunihiko Yoshino

Spectrally segmented principal component analysis of hyperspectral imagery for mapping invasive plant species

Principal component analysis (PCA) is one of the most commonly adopted feature reduction techniques in remote sensing image analysis. However, it may overlook subtle but useful information if applied directly to the analysis of hyperspectral data, especially for discriminating between different vegetation types. In order to accurately map an invasive plant species (horse tamarind, Leucaena leucocephala) in southern Taiwan using Hyperion hyperspectral imagery, this study developed a spectrally segmented PCA based on the spectral characteristics of vegetation over different wavelength regions. The developed algorithm can not only reduce the dimensionality of hyperspectral imagery but also extracts helpful information for differentiating more effectively the target plant species from other vegetation types. Experiments conducted in this study demonstrated that the developed algorithm performs better than correlation‐based segmented principal component transformation (SPCT) and conventional PCA (overall accuracy: 86%, 76%, 66%; kappa value: 0.81, 0.69, 0.57) in detecting the target plant species, as well as mapping other vegetation covers.

Guidelines for soil conservation towards integrated basin management for sustainable development: A new approach based on the assessment of soil loss risk using remote sensing and GIS

Soil suspension in the Cidanau River of western Java, Indonesia, has increased recently, perhaps because of rapid environmental change in this watershed region. The objectives of this research are to assess soil loss risk using remote sensing and GIS and to develop effective guidelines for soil conservation in this watershed. To assess soil loss risk, a new soil loss model based on Universal Soil Loss Equation (USLE) was applied, in which C factor (crop management factor) was estimated using the perpendicular vegetation index (PVI); this was computed with satellite remote sensing data and used to simulate soil loss risk. The simulation showed that areas with highest risk of soil loss are on northern- and southern-facing hillsides with poor vegetation. Guidelines for soil conservation in the watershed were proposed: under these guidelines soil-loss risk is managed by evaluating the effectiveness of contour farming, belt farming, and mulch farming. Some recommended measures for soil conservation are as follows: (1) Green vegetation cover should be preserved as much as possible. (2) Vegetation coverage must be increased by forestation in steep sloped areas. (3) Belt farming and contour farming are recommended in areas with slopes under 100%, and mulch farming (more than 50% ground cover mulching is recommended) is desirable in areas with slopes over 100% and without green cover.

Characterizing temporal vegetation dynamics of land use in regional scale of Java Island, Indonesia

Improving the understanding of land use and land cover is a major research challenge for the human-environmental sciences and is essential for many aspects of global environmental research. Considering seasonal vegetation dynamics or phenological dynamics in multi-year series leads to a broader view of land use and land cover. This study is based on the hypothesis that a pixel representing a complex but consistent land use has a typical, distinct and repeated temporal pattern of vegetation index inter-annually, which can be used as characteristic signatures for land use classification. Considering the seasonal events and climatic variability in Indonesia, we characterized the temporal vegetation dynamics of long-term land use by using multi-temporal Moderate Resolution Imaging Spectroradiometer (MODIS) enhanced vegetation index 16-day composite data from 2001 to 2007, and then generated a land use map using those characteristics. Accuracy assessment of the results showed the need to evaluate such methods for land use types that do not have a consistent yearly pattern. On the other hand, the identification of the intensive agriculture lands, such as paddy rice and upland, was satisfactory. Although the mixed pixel issue is quite problematic when using MODIS data, the results indicate that MODIS data offer great promise for characterizing seasonal as well as multi-year variation at large scales. Indeed, the methodology proposed in this research distinguished among many specific land use classes based on temporal land cover information properties. Characterization of temporal vegetation dynamics patterns would provide sufficient, significant and useful information regarding the patterns of land use; consequently it should be possible to consider the actual, subtle nature of inter-annual land use change as well as overall land use.

USE OF BALLOON AERIAL PHOTOGRAPHY FOR CLASSIFICATION OF KUSHIRO WETLAND VEGETATION, NORTHEASTERN JAPAN

Kushiro wetland in northeastern Japan is a Ramsar-designated wetland of international importance (1980) that is characterized by high biodiversity and spatial heterogeneity. These characteristics of the wetland also present innumerable challenges for mapping and monitoring such unique ecosystems. Recent advances in remote sensing technology have provided many sensors with different spatial and spectral scales and resolutions. However, they are still inadequate for mapping wetland vegetation at a large scale for various reasons, such as inadequate resolution and high costs. This study was designed to evaluate the potential of balloon aerial photography to acquire high resolution (15 cm pixel size) imagery for mapping wetland vegetation in the Akanuma marsh. We used a standard 28-mm non-metric camera (Nikon-F-801), which seven specific categories (species mixes) were successfully delineated. It was possible to classify small shrubs mixed with herbaceous plants; moss bogs with pools; dwarf shrubs with sedges; and moss with alpine plants. From this research, it seems that balloon aerial photography is a powerful tool for mapping temperate wetland vegetation, allowing classification of specific and typical vegetation types to the genus and species level.

Biomass estimation of Sonneratia caseolaris (l.) Engler at a coastal area of Hai Phong city (Vietnam) using ALOS-2 PALSAR imagery and GIS-based multi-layer perceptron neural networks

This study tested the use of machine learning techniques for the estimation of above-ground biomass (AGB) of Sonneratia caseolaris in a coastal area of Hai Phong city, Vietnam. We employed a GIS database and multi-layer perceptron neural networks (MLPNN) to build and verify an AGB model, drawing upon data from a survey of 1508 mangrove trees in 18 sampling plots and ALOS-2 PALSAR imagery. We assessed the model’s performance using root-mean-square error, mean absolute error, coefficient of determination (R2), and leave-one-out cross-validation. We also compared the model’s usability with four machine learning techniques: support vector regression, radial basis function neural networks, Gaussian process, and random forest. The MLPNN model performed well and outperformed the machine learning techniques. The MLPNN model-estimated AGB ranged between 2.78 and 298.95 Mg ha−1 (average = 55.8 Mg ha−1); below-ground biomass ranged between 4.06 and 436.47 Mg ha−1 (average = 81.47 Mg ha−1), and total carbon stock ranged between 3.22 and 345.65 Mg C ha−1 (average = 64.52 Mg C ha−1). We conclude that ALOS-2 PALSAR data can be accurately used with MLPNN models for estimating mangrove forest biomass in tropical areas.

Detecting land-use change from seasonal vegetation dynamics on regional scale with MODIS EVI 250-m time-series imagery

Simultaneous analysis of land surface attributes and their seasonal changes provides a broader view of land-use and land-cover change. This study attempted to detect the change in inter-annual temporal vegetation dynamics, which reflects a change in land surface attributes. We explored 250-m multi-temporal MODIS EVI 16-day composite data from 2001 to 2007 to characterize a change in vegetation dynamics related to land-use change detection. The MODIS data was filtered in time-frequency space by wavelet function in order to identify and reduce the overall noise so as not to lose useful information from the time series data. The results show that by characterizing temporal vegetation dynamics, it is possible to distinguish actual land-use change based on land-cover dynamics. The result was evaluated using 18,626 reference pixels and showed an overall accuracy of 76.10%. In agricultural land use, such as upland and plantation, the weakest results were caused by mixed pixels from MODIS 250-m grid data as well as by temporal complexity related to the climate-driven change of land cover existing in the study area. On the other hand, for land use types which were not significantly affected by climate variability such as paddy rice fields with sufficient irrigation systems, natural forest and mangrove, the accuracy was satisfactory. Although the mixed pixel issue is quite problematic in this study, the results show that the characterization of temporal vegetation dynamics is an alternative when considering the accuracy of land-use change, which is not necessarily coincident with the temporary change in land cover. The methodology proposed in this research provides sufficient and useful information regarding land-use change, such as location, area, time and pathways of the change; consequently, it should be possible to provide broad scale data on the terrestrial environmental change.

Aboveground biomass estimation of mangrove species using ALOS-2 PALSAR imagery in Hai Phong City, Vietnam

Abstract. This study examined the potential of using the HH and HV backscatter from the Advanced Land Observing Satellite 2 (ALOS-2) with enhanced phased array L-band synthetic aperture radar (PALSAR) in high sensitive mode to estimate the above-ground biomass (AGB) of the two mangrove species of Hai Phong city, Vietnam. A positive correlation was observed between the mean backscattering coefficients of the dominant mangrove species at dual polarizations HH and HV and various biophysical parameters. In contrast, low correlations were observed between those coefficients and the tree densities for the two mangrove species. The AGB of the mangrove species were estimated at between 2.8 and 161.5  Mg ha−1 with an average of about 39  Mg ha−1 for Sonneratia caseolaris and between 27.6 and 209.2  Mg ha−1 with an average of ∼100  Mg ha−1 for Kandelia obovata. The main indicators used for the selection of the best potential models in estimating the AGB of different species were R2 and the root-mean-square error (RMSE). The results showed a satisfactory correlation between model estimation and field-based measurements with R2=0.51, RMSE=35.5  Mg ha−1 for S. caseolaris and R2=0.64, RMSE=41.3  Mg ha−1 for K. obovata. This research has illustrated the potential use of ALOS-2 PALSAR data in estimating the AGB of mangrove species in the tropics.

Mangrove analysis using ALOS imagery in Hai Phong City, Vietnam

Mangroves that appear in the inter-tidal zones along the coast in most tropical and semi-tropical countries play a vital role in coastal zones and can defend against the impacts of tsunamis. Nevertheless, these forests are under severe threat because of high population growth, weak governance, poor planning, as well as uncoordinated economic development. Hai Phong city is located on the Northern coast of Vietnam where the mangroves are distributed between zone I and zone II among the four mangrove zones in Vietnam. This city is vulnerable to rising sea levels and tropical cyclones, which are forecasted to become more severe in coming next decades. The objectives of this research were to analyze the current status of mangroves using different ALOS sensors in Hai Phong, Vietnam in 2010 and compare the accuracy of the post satellite image processing of ALOS imagery in mapping mangroves. A combination of object-based and supervised classification was used to generate the land cover maps. The results of this research indicate that the total area of mangrove was approximately 2,549 hectares and mangrove is present in the five coastal districts in Hai Phong. The findings of this research showed that ALOS AVIR-2 provides better accuracy than ALOS PALSAR. This research indicates the potential of utilizing image segmentation associated with supervised method for both optical and SAR images to map mangrove forests in coastal zones

Automated 3D Forest Surface Model Extraction from Balloon Stereo Photographs

We upgraded an automated forest digital surface model (DSM) extraction method from balloon stereo photographs of a tropical peat swamp forest in Narathiwat, Thailand by evaluating the image matching accuracy and forest surface height (FSH) estimation. We modified an image correlation matching method based on the characteristics of the tree crown shapes. The mismatched area was less than 3 percent of the total area. We estimated an FSH map in a 60 m 60 m plot by both photo estimation and field measurement, and set the unit area for FSH averaging at 10 m 10 m. The root mean square of the differences between the mean photo-estimated and mean field-measured FSH was 3.8 m, which was revised to 1.9 m when the forest gaps were extracted offline. These differences were within a reasonably practical range since the range of the mean field-measured FSH was 10.0 to 21.4 m.

Classification of wetland vegetation using aerial photographs by captive balloon cameras and aero NIR color video image, Kushiro northern wetland in Japan

The final goal is to develop an algorithm of satellite remote sensing for classification of specific vegetation patterns in wetland using balloon photograph interpretation and aerial Near Infrared (NIR) color video images. We focused on the following definite objectives. (1) Making vegetation maps by mosaic balloon photos with high resolution and discussing about the efficiency for the classification of wetland vegetation in Kushiro wetland, located in the eastern part of Hokkaido, Japan , (2) Investigation on applicability of the balloon mosaic photos as the training data for classification of aerial NIR color video images with low resolution, (3) Classification of wetland vegetation using aerial visible and NIR color video images taken from nadir angle and 45 degree zenith angle. We extracted training data for classification of wetland vegetation on NIR color video sequence image from aerial balloon photographs taken by two captive balloons at low altitude and we got higher accuracy classification: two vegetation types in low-moor, three types in transition-moor, and five types in high-moor. NIR color video sequence image is one of the most promising tools for wetland monitoring because it has fine resolution and NIR band that is sensitive to vegetation. In consequence of this research, it concluded that availability of the high resolution training data such as balloon mosaic photos were useful to classify NIR color video images or wetland vegetation in wide range.