Gregory S. Biging

Individual Tree-Crown Delineation and Treetop Detection in High-Spatial-Resolution Aerial Imagery

The cost of forest sampling can be reduced substantially by the ability to estimate forest and tree parameters directly from aerial photographs. However, in order to do so it is necessary to be able to accurately identify individual treetops and then to define the region in the vicinity of the treetop that encompasses the crown extent. These two steps commonly have been treated independently. In this paper, we derive individual tree-crown boundaries and treetop locations under a unified framework. We applied a two-stage approach with edge detection followed by markercontrolled watershed segmentation. A Laplacian of Gaussian edge detection method at the smallest effective scale was employed to mask out the background. An eight-connectivity scheme was used to label the remaining tree objects in the edge map. Subsequently, treetops are modeled based on both radiometry and geometry. More specifically, treetops are assumed to be represented by local radiation maxima and also to be located near the center of the tree-crown. As a result, a marker image was created from the derived treetop to guide a watershed segmentation to further differentiate touching and clumping trees and to produce a segmented image comprised of individual tree crowns. Our methods were developed on a 256- by 256-pixel CASI image of a commercially thinned trial forest. A promising agreement between our automatic methods and manual delineation results was achieved in counting the number of trees as well as in delineating tree crowns.

Modeling Conifer Tree Crown Radius and Estimating Canopy Cover

Models of tree crown radius were developed for several conifer species of California. Typical forest inventory variables (DBH, height, height-to-crown base, crown class, basal area per hectare, and trees per hectare) were considered as independent variables in model development. Models were fitted using both ordinary and weighted least squares methods. It was found that for the species studied, an ordinary least squares linear regression with DBH as the only independent variable was appropriate. For some species studied, the addition of other independent variables provided minor improvements over the model with only DBH. These models of crown radius could be summed to give an estimation of canopy cover. Using crown mapped data, it was possible to test and calibrate these models to predict non-overlapping canopy cover. Linear and non-linear models were considered for calibration. A non-linear model with an upper asymptote seemed to be the best calibration. These models enable an efficient and unbiased method of estimation of canopy cover as an alternative to photointerpretation estimation of cover.

Comparison of single-stage and multi-stage classification approaches for cover type mapping with TM and SPOT data

Abstract This article presents a comparison of the performance of TM and SPOT data for cover type mapping on the Central Sierra of Spain. A novel multi-stage iterative classification, and four single-step classifications are performed for each type of data. The single-stage classifications differ from one another in the band selection process, the use or not of prior probabilities, and/or the supervised or unsupervised nature of the classification. The accuracy of each classification method and data type (TM vs. SPOT) is expressed as an error matrix from which K statistics and their large sample variances are derived. The values of the K statistics are used to compare the performance of the classification methods, two at time, by means of a Z statistic. Results from this research show that the iterative classification approach is superior to any other classification for both types of remotely sensed data. TM data proves superior to SPOT data only when the iterative classification approach is used. Overall comparison of the performance of TM and SPOT data shows that there is not a statistically significant difference between these types of data for large-scale cover type mapping.

EO‐1 Hyperion, ALI and Landsat 7 ETM+ data comparison for estimating forest crown closure and leaf area index

In this study, mixed coniferous forest crown closure (CC) and leaf area index (LAI) were measured at the Blodgett Forest Research Station, University of California at Berkeley, USA. Data from EO‐1 Hyperspectral Imager (Hyperion) and Advanced Land Imager (ALI) acquired on 9 October 2001, and from Landsat 7 Enhanced Thematic Mapper Plus (ETM+) on 25 October 2001 were used for estimation of CC and LAI. A total of 38 forest CC and LAI measurements were used in this correlation analysis. The analysis procedure consists of (1) atmospheric correction to retrieve surface reflectance from Hyperion, ALI and ETM+ data, (2) a total of 38 patches, corresponding to ground CC and LAI measurement plots, extracted from data from the three sensors, and (3) calculating univariate/multivariate correlation coefficient (R 2) and root mean square error (RMSE) using CC and LAI measurements and retrieved surface reflectance data of the three sensors. The experimental results indicate: (1) higher individual band correlations with CC and LAI appear in visible and short wave infrared (SWIR) regions due to spectral absorption features (pigments in visible and water and other biochemicals in SWIR); (2) based on ALI individual band wavelengths, the R 2/RMSE produced with Hyperion bands are all better than those with ALI, except ALI band 1, due to atmospheric scattering of Hyperion bands in the visible region; (3) based on ETM+ individual band wavelengths, Hyperion is better than ALI, which is better than ETM+, especially for the NIR band group of Hyperion; (4) based on spectral region, Hyperion, again, is better than ALI which is better than ETM+, and optimal results appear in the visible region for ALI and in SWIR for Hyperion; and (5) if considering just six bands or six features (six principal components) in estimating CC and LAI, optimal results are obtained with six bands selected from the 167 Hyperion bands. In general, for estimation of forest CC and LAI in this study, the Hyperion sensor has outperformed the ALI and ETM+ sensors, whereas ALI is better than ETM+. The best spectral region for Hyperion is SWIR, but for ALI and ETM+, the visible region should be considered instead.

An iterative classification approach for mapping natural resources from satellite imagery

Abstract This project explores an iterative classification process as an alternative to two-stage classifications. In the iterative classification approach cover types are classified, one or two at a time, and the band selection process is repeated in each iteration, so that the combination of bands that provides the best separability among the classes that remain to be classified is selected. The optimum number of bands to perform the classification is also determined for each iteration, so that the classification of the area that is masked in each iteration achieves the highest possible accuracy. Spectral Pattern Analysis, and Spectral Separability Indices are used in the band selection process. GIS analysis is used to obtain prior probabilities, and to determine if variables such as elevation, slope, and aspect can result in a source of information for segmentation of the study area into more homogeneous strata. The results of this study show that: (1) The proposed iterative classification approach is ...

Accuracy statistics for judging soft classification

In the literature one can find different accuracy measures that are built from the error matrix. However, standard accuracy assessment, which is based on the error matrix, is incomplete when dealing with fuzzy sets or when errors do not have the same importance. In this paper, we propose an extension of the error concept for soft (or crisp) classification that will be able to extend standard accuracy measures (e.g., overall, producers, users or Kappa statistic) that can be used in any framework: errors with different importance, soft classifier and crisp reference data (expert) or with a fuzzy expert. In particular, a weighted measure is built that takes into account the preferences of the decision maker in order to differentiate some errors that must not be considered equal.

Vineyard identification in an oak woodland landscape with airborne digital camera imagery

Using airborne multispectral digital camera imagery, we compared a number of feature combination techniques in image classification to distinguish vineyard from non-vineyard land-cover types in northern California. Image processing techniques were applied to raw images to generate feature images including grey level co-occurrence based texture measures, low pass and Laplacian filtering results, Gram-Schmidt orthogonalization, principal components, and normalized difference vegetation index (NDVI). We used the maximum likelihood classifier for image classification. Accuracy assessment is performed using digitized boundaries of the vineyard blocks. The most successful classification as determined by t-tests of the Kappa coefficients was achieved based on the use of a texture image of homogeneity obtained from the near infrared image band, NDVI and brightness generated through orthogonalization analysis. This method averaged an overall accuracy of 81 per cent for six frames of images tested. With post-classification morphological processing (clumping and sieving) the overall accuracy was significantly increased to 87 per cent (with a confidence level of 0.99).

Spectral fuzzy classification: an application

Geographical information (including remotely sensed data) is usually imprecise, meaning that the boundaries between different phenomena are fuzzy. In fact, many classes in nature show internal gradual differences in species, health, age, moisture, as well other factors. If our classification model does not acknowledge that those classes are heterogeneous, and crisp classes are artificially imposed, a final careful analysis should always search for the consequences of such an unrealistic assumption. We consider the unsupervised algorithm presented by A. del Amo et al. (2000), and its application to a real image in Sevilla province (south Spain). Results are compared with those obtained from the ERDAS ISO-DATA classification program on the same image, showing the accuracy of our fuzzy approach. As a conclusion, it is pointed out that whenever real classes are natural fuzzy classes, with gradual transition between classes, then its fuzzy representation will be more easily understood, and therefore accepted by users.

Tree rings show competition dynamics in abandoned Castanea sativa coppices after land-use changes

Abstract Questions: As a consequence of socio-economic changes, many Castanea sativa coppices have been abandoned and are now developing past their usual rotation length. Do we have to expect changes in stand structure and composition of abandoned Castanea sativa coppice invaded by other species? Is a tree ring-based approach adequate to early recognise changes in inter-specific competitive interaction? Location: Lowest alpine forest belt of the southern Swiss Alps. Methods: We selected a 60-year old abandoned Castanea sativa coppice stand with sporadic Fagus sylvatica and Quercus cerris mixed in. Using tree-ring based indices we analysed differences in the species-specific response to competition. Analyses were performed by comparing how subject dominant trees (10 Castanea, 5 Fagus, 5 Quercus) have differently faced competition from their immediate Castanea coppice neighbourhood, taking into account the changes over time and space. Results: Although no species appears yet to have made a difference in the surrounding coppice mortality, there are species-specific differences in growth dominance, which indicate potential successional processes. Castanea sativa growth dominated in the early stages of stand development. However, after approximately 30–35 yr Fagus sylvatica and Quercus cerris became much more dominant, indicating a change in competitive potential that does not favour Castanea sativa. Conclusions: Without interventions this coppice will develop into a mixed stand. A tree-ring based approach allows an early recognition of forthcoming changes in stand composition and structure and is likely to be an important tool for forest landscape management. Nomenclature: Aeschimann et al. (2004). Abbreviations: BAI = Basal area increment; TD = Tree density.

Simple calibration of AVIRIS data and LAI mapping of forest plantation in southern Argentina

During the 2001 EO-1 campaign in Argentina, two high spectral resolution image scenes of AVIRIS were acquired at two study sites in the Patagonia region of southern Argentina on 15 February 2001. A total of 70 LAI measurements were taken from different forest types in the same areas one month later, and some spectroradiometric measurements were also collected from the nearby highway and different forest stands in the areas. In this study, we compared the effectiveness of the three types of AVIRIS data used for estimating and mapping LAI. The three types of data correspond to AVIRIS original radiance (OR), corrected radiance (CR) and retrieved surface reflectance (SR). We first simulated the total at-sensor radiances using MODTRAN4, then used ground spectroradiometric measurements taken from different targets to improve the reflectances for each pixel on the image. The CR images were obtained by subtracting path radiance from the OR images. A 10-term LAI prediction model for each type of data was constructed to predict pixel-based LAI values. Finally, the pixel-based LAI value was sliced and mapped for all the three types of images. The results of mapping LAI using the three types of AVIRIS data (OR, CR and SR) indicate that mapping LAI by SR is the most realistic, followed by CR. The poorest result occurs when mapping LAI with OR data due to atmospheric effect. The SR data can lead to higher correlation with LAI in some bands and produce higher accuracy indices for the 10-term predictive model, although some indices from the test set for SR data have a somewhat lower correlation with LAI than those produced with OR data. Therefore, in general, it can be concluded that the retrieved surface reflectance data is more effective for mapping forest LAI compared to the other two types of data.