Mamoru Kubo

Watershed segmentation and classification of tree species using high resolution forest imagery

This paper proposes a procedure for classifying tree species in high spatial resolution aerial imagery. In this study, the images Mere observed by video camera mounted on a helicopter. The spatial resolution of these images is about from 7 cm to 10 cm. Since this resolution is higher than one of satellite, tree species can be recognized in details. Tree species for classification are three classes. One class is a broad-leaved tree, and other classes are needle-leaved trees. Each class have different spatial patterns of gray-level and spectral signatures. Although they are the effective features, the various size and shape of tree and shadow make complicated and randomly textured composition in the aerial images. For this reason, we performed a segmentation before classification. The segmentation method is based on watershed algorithm using a gradient of brightness effectively. In classification, the features were extracted from each segmented region. We used gray level co-occurrence matrix as the textural feature and two kinds of spectral features. Supervised classification using maximum likelihood decision rules was performed. We achieved in about 80 to 90 percent of accuracy

Delineation of tree crown in high resolution satellite image using circle expression and watershed algorithm

Forests play an important role as environment for living things. For controlling forests, it is useful to know condition of tree efficiently. Recently, high spatial resolution images are taken using satellite and detail of trees in forest can be identified by visual inspection of satellite image. Shape of tree crown is one of important parameters that can be identified using satellite image. In this study, we developed a method to delineate crown and applied to high resolution satellite image. Then result of crown delineation was compared with result of visual inspection. At first, we assumed that shape of crown resembled circle and calculated circle radiuses showing crown size. Calculated radius was larger as similar brightness region became wide. The radiuses at each pixel position were calculated and the result was used as the radius distribution image (RDI). In the RDI, the radius was larger as the position became close to center part of crown. If radius value is considered as elevation, RDI seems to be DEM (Digital Elevation Model). The value of each pixel of RDI was inverted and inverse radius distribution image (IRDI) was calculated. In the IRDI, it is expected that a cone-shaped geographical feature is on the crown region and each crown region make each watershed. Then crown regions were calculated applying watershed algorithm to IRDI. In the results, shapes of calculated regions were close to crown shape identified with visual inspection

A Hydrometeor Classification Method for X-Band Polarimetric Radar: Construction and Validation Focusing on Solid Hydrometeors under Moist Environments

AbstractA fuzzy-logic-based hydrometeor classification (HC) method for X-band polarimetric radar (X-pol), which is suitable for observation of solid hydrometeors under moist environments producing little or no hail, is constructed and validated. This HC method identifies the most likely hydrometeor at each radar sampling volume from eight categories: 1) drizzle, 2) rain, 3) wet snow aggregates, 4) dry snow aggregates, 5) ice crystals, 6) dry graupel, 7) wet graupel, and 8) rain–hail mixture. Membership functions are defined on the basis of previous studies. The HC method uses radar reflectivity Zh, differential reflectivity Zdr, specific differential phase Kdp, and correlation coefficient ρhv as its main inputs, and temperature with some consideration of relative humidity as supplemental information. The method is validated against ground and in situ observations of solid hydrometeors (dry graupel, dry snow aggregates, and ice crystals) under a moist environment. Observational data from a ground-based ima...

Z-R relation for snowfall using two small doppler radars and snow particle images

Snowfall data was simultaneously recorded by two small Doppler radars, two high sensitive snow gages and an image processing system with high accuracy at short time interval. The snowfall rate R was measured with two gauges and radar reflectivity factor Z was measured using small bistatic X-band radar and monostatic K-band radar. The images of falling snow particles were used to obtain size distribution. Since all the measurements were located in a small area, it can be said that the obtained data corresponds well to others, and it is possible to analyze Z-R relation in detail. The relationships between two radar reflectivity factors and snowfall rate were investigated and compared to the characteristics of snow particles.

Tree crown detection and classification using forest imagery by IKONOS

The purpose of this study is detection and classifi- cation of tree crowns using forest imagery taken by IKONOS. A forest image contains many tree crowns of different sizes and shape that are touching each other. By using IKONOS pan-sharpened data, discernment of tree crown and species is possible. To detect tree crowns in the image, we used Watershed segmentation. If an image is viewed as a surface, with mountains and valleys, the Watershed segmentation finds intensity valleys in an image. In this study, a gradient of intensity in an image was used in order to find valleys separating tree crowns from shadows. To classify tree species, the spatial features of each segmented region were calculated. Image features for the classification were extracted by texture analysis using gray level co-occurrence matrix. Image texture is produced by an aggregation of unit features, such as tree leaves and leaf shadows. Variations in crown texture are important in the identification of species. Supervised classification using maximum likelihood decision rules with these features was performed. Classification accuracies on the order 80% were achieved.

Snow particle automatic classification with texture operators

The backscattered data recorded by the meteorological radar is exploited for rainfall/snowfall rate calculation according to the Z-R relation. This relation is governed by parameters, which are influenced by the size and shape of the falling particles. The variety of snowflake types as well as the in class shape and size differences make this problem very difficult.

The effect of successive years of defoliation by the larch sawfly (Pristiphora erichsonii (Hartig)) on the foliage properties of the Japanese larch (Larix kaempferi (Lamb.) Carr.), with particular reference to the CN balance hypothesis

The carbon/nutrient balance hypothesis (CNBH) attempts to explain the mechanism of induced changes in plant properties. The responses of the Japanese larch (Larix kaempferi) to defoliation by the larch sawfly (Pristiphora erichsonii) were examined from the perspective of the CNBH. This study was conducted in seven Japanese larch plantations in central Hokkaido, Japan. The defoliation intensity was determined from canopy photos taken from 2009 to 2012. The chemical and physical properties of the foliage were determined from 2010 to 2012. Severe insect defoliation was found at two sites in 2009 and at all seven sites in 2010 and 2011. A decrease in foliar nitrogen and increases in phenolics, tannins, and the CN ratio were found in the years following severe defoliation and were significantly influenced by the 2009 defoliation intensity. The influence of defoliation in 2010 and 2011 was weaker. These results indicated that the past defoliation history additively affected the foliage properties in the 2xa0years following insect defoliation. In addition to the 2009 defoliation effects, site effects were found on phenolics, sugars, and the CN ratio. Relative to the other sites, the CN ratio was high at both sites where severe defoliation was found in 2009. Phenolics and sugars did not increase linearly with the CN ratio, indicating that limitations affected their synthesis. These results suggest that the induced changes in L. kaempferi properties are partially up-regulated under nitrogen limitation, but that secondary compound synthesis was, most likely, influenced by external site-dependent factors other than nitrogen limitation.

Identification of individual tree crowns from satellite image and image-to-map rectification

In forest area, there are few landmarks to be ground control points (GCPs) used for registration of satellite images or maps. Additionally, geographic information from the Global Positioning System (GPS) in field measurement survey is insufficient accuracy to identify individual tree crowns from satellite image. In this study, we propose the method of identifying individual tree crowns from satellite image using field measured data. First, in order to obtain the field measured data, we collected several information of individual trees in the test site. These are the tree stand locations, the distances between the tree trunk and outermost branch in eight directions, the diameter at breast height, and tree species. This survey was carried out on 20 September 2006. The area of this site is 160 meter by 80 meter, and there are about 60 canopy trees. Then, using the field measured data, we created the projected on-ground crown map which has the location and shape of individual trees. The each shape of tree crown is octagonal. Next, we detected the regions of tree crown from satellite image. In this study, we used an IKONOS panchromatic satellite image. The spatial resolution of analysis image is 1 meter per pixel. It can be recognized and identified an individual tree crown whose radius is more than 2 or 3 meter. Watershed algorithm was used for image segmentation, based on mathematical morphology considers gray-scale images to be sets of points in a three-dimensional space, the third dimension being the gray level. A gray scale landscape may be segmented according to the watersheds of the image. The segmented regions were classified to discriminate tree crown using the feature of spectral signature. Finally, we found out individual tree crowns related with field measured data from satellite image. Using a GCP by GPS equipment, we performed roughly registration of the satellite image to the projected on- ground crown map. For each tree crown in the map, we found out the same tree, which has the highest corresponding possibility to the tree crown in the map, among segmented regions obtained from satellite image. This tree-to-tree matching algorithm was performed using the fitness value of the location and octagonal shape of both tree crowns in image and map. We could obtain the optimum registration by afflne transformation of highest fitness value without ground control points. Consequently, we could identify individual tree crowns from satellite image by image-to-map rectification.

Identification of dead tree of Japanese oak wilt (JOW) using high spatial resolution satellite imagery

The incidence of Japanese oak wilt (JOW) has been increasing in Japan since late 1980s. The JOW is caused by the ambrosia fungus Raffaelea quercivorus vectored by an ambrosia beetle Platypus quercivorus. Detection of trees killed by JOW is important to prevent new incidence of JOW but difficult because most of JOW mortality occurs in mountainous area. Remote sensing with high spatial resolution imageries taken by IKONOS® or Quickbird® can be laborsaving technology to identify individual trees killed by JOW over wide areas. A purpose of this study is to develop a laborsaving image processing technique to identify individual tree crowns. The orthophoto imagery was transformed both to HSI (Hue, Saturation, and Intensity) and NDVI (Normalized Difference Vegetation Index) datasets. Individual dead tree crowns killed by JOW were identified by cluster analysis using datasets of H, S, and NDVI.

A hierarchical Bayesian model to estimate the unobservable predation rate on sawfly cocoons by small mammals.

Predation by small mammals has been reported as an important mortality factor for the cocoons of sawfly species. However, it is difficult to provide an accurate estimate of newly spun cocoons and subsequent predation rates by small mammals for several reasons. First, all larvae do not spin cocoons at the same time. Second, cocoons are exposed to small mammal predation immediately after being spun. Third, the cocoons of the current generation are indistinguishable from those of the previous generation. We developed a hierarchical Bayesian model to estimate these values from annual one-time soil sampling datasets. To apply this model to an actual data set, field surveys were conducted in eight stands of larch plantations in central Hokkaido (Japan) from 2009 to 2012. Ten 0.04-m2 soil samples were annually collected from each site in mid-October. The abundance of unopened cocoons (I), cocoons emptied by small-mammal predation (M), and empty cocoons caused by something other than small-mammal predation (H) were determined. The abundance of newly spun cocoons, the predation rate by small mammals before and after cocoon sampling, and the annual rate of empty cocoons that remained were estimated. A posterior predictive check yielded Bayesian P-values of 0.54, 0.48, and 0.07 for I, M, and H, respectively. Estimated predation rates showed a significant positive correlation with the number of trap captures of small mammals. Estimates of the number of newly spun cocoons had a significant positive correlation with defoliation intensity. These results indicate that our model showed an acceptable fit, with reasonable estimates. Our model is expected to be widely applicable to all hymenopteran and lepidopteran insects that spin cocoons in soil.