Simple regressions of single band thematic mapper simulator (TMS) reflectance data against stand basal area and leaf biomass show that the spectral feature most strongly related to vegetation amount is visible reflectance (TM bands 1, 2 and 3), which decreases as conifer basal area increases. Principal components analysis of TMS spectral data for sample stands, and stepwise multiple regression of transformed axes, showed the first principal component, interpreted as scene brightness, to be the composite spectral feature most correlated to vegetation amount. TMS data showed some ability to discriminate spectrally between coniferous forest stands dominated by different species. Dense stands of large timber dominated by red fir (Abies magnified var. shastemis) had a different average spectral reflectance from white fir (Abies concolor) dominated stands, particularly in TM bands 4 (0.76-0.90μm) and 6 (10-4-12-5μm).

Thematic mapper analysis of coniferous forest structure and composition

Seeds are very diverse in their shape and size. In the mature state they contain an embryo, the next generation of plant, surrounded by a protective structure (the seed and/or fruit coat) and, in species in which the nutritive reserves are not stored within the cotyledons, by an alternative storage tissue (endosperm, perisperm, or megagametophyte). Most seeds contain large and characteristic quantities of polymeric reserves. The major ones are carbohydrates, oils, and proteins, with minor amounts of phosphate-rich phytin. Starch, a polymer of glucose, is the most common form of stored carbohydrate, contained within cytoplasmic starch granules; less common are the hemicelluloses, stored in secondary cell walls, usually as mannan polymers. Oils are triacylglycerols, each composed of glycerol and three fatty acids that are specific to the oil; these are present within oil bodies. Storage proteins, of which there are three types, albumins, globulins, and prolamins, are sequestered in protein storage vacuoles. These reserves are vital components of human and animal diets, and their production in crops is a basis of agriculture.

Structure and Composition

In recent years significant effort has gone into collecting, processing, and analysing normalized difference vegetation index (NDVI) data derived from the National Oceanic and Atmospheric Administration's (NOAA) 1-km advanced very high resolution radiometer (AVHRR) for environmental applications. The ISODATA clustering procedure has been employed using time-series AVHRR maximum NDVI composites to produce a land cover characteristics data base. This study examines the sensitivity of the ISODATA clustering results to changes in sampling procedures and processing parameters when applied to NOAA NDVI time-series data. The effects of varying the sampling strategy, sampling intensity, clustering parameters, and number of iterations have been evaluated. The ISODATA clustering procedure was most sensitive to variations in sampling intensity and clustering parameters.

Sensitivity of ISODATA to changes in sampling procedures and processing parameters when applied to AVHRR time-series NDV1 data

These two papers deal with a new method of data transformation. By analysing grey level curves (broken lines) of various ground features in image bands of different satellites, we have found that, inherent in 3-or 4-band satellite images (SPOT, IKONOS, Quick Bird, OrbView, FORMOSAT and MSS) there are three basic remote-sensing characteristics as follows: (1) the general radiance level L; (2) the visible-infrared radiation balance B; and (3) the band radiance variation vector (direction and speed) V. However, inherent in 5-or 7-band satellite images (NOAA, TM), besides the above three, there is an extra basic remote-sensing characteristic, i.e. the thermal radiation intensity I. This is denoted by thermal bands, i.e. the TM band 6 or NOAA (AVHRR) bands 4 and 5, which are relatively independent and can be used directly, and hence are unnecessary for information extraction or data transformation. Therefore, the data transformation only lies in extracting the L, B and V from original satellite images. Furthermore, we have also found that there are three basic ground-cover elements on the Earth's surface: i.e. the bare land (in a broad sense), the vegetation and the water body, which, in different proportions, constitute all ground cover. Moreover, there are three basic (primitive) colours on colour image (including colour composite of satellite images): i.e. red, green and blue, which generate all colours on the colour image. Further research has revealed that the three basic remote-sensing characteristics, the three basic ground-cover elements and the three basic colours on the composite can conceptually constitute a three-to-three corresponding regular triangle scheme. Perhaps a good method of data transformation should make the scheme realistic, i.e. make the three 'threes' all correspond to each other. The research presented here has completed this task by regression calculations and selection of specific variables. First, the methodology and transformation equations for TM images are discussed. The transformed L, B and V images have relatively independent and equally distributed information as well as clear and definite physical, mathematical and geographical significance. They can be used effectively for generating high-quality colour composites, on which the red, green, blue, yellow, pink, cyan and other colours of various kinds are all generated and all pure, saturated, equilibrated, meaning-definite and close to the colours of ground features in nature. As a result, interpretations and discriminations of ground features can be easier and conducted not only by experience, but also by logic. The L, B and V images can also be used effectively for classification and digital analysis of ground features. With regard to the transformation equations for SPOT, NOAA, IKONOS, Quick Bird, OrbView, FORMOSAT and MSS images and the method application will be dealt with in the second paper.

A new method of data transformation for satellite images: I. Methodology and transformation equations for TM images

The changing patterns of dry land salinisation in the Eastern Wimmern area in Southeastern Australia are closely related to changes in regional water balance regimes. Predictive mapping of dry land salinisation in this area requires distributed water balance modelling, for which, timely and accurate information on regional land cover patterns and dynamics from computer‐assisted processing of remotely sensed digital data, although not yet routinely adoptable, is regarded as one of the key inputs. With this purpose in mind, a multispectral classification procedure based on the mosaic model (Jupp 1988) is designed, and applied to a Landsat 5 TM subimage (1024 x 960 pixels) captured on Feb. 21 1988 from the Charlton area of the Eastern Wimmera, using microBRlAN, a PC‐based digital image processing system. Results show that application of the mosaic model to TM data for the Eastern Wimmera can yield 14‐class land cover maps with classification accuracy of 83 ± 7.52 % at the 95% confidence limits.

Land cover mapping in a Salinised Dryland farming area in southeastern Australia using landsat TM data

Three feature extraction methods, canonical analysis (CA), principal component analysis (PCA), and band selection, have been applied to Thematic Mapper Simulator (TMS) data in order to evaluate the relative performance of the methods. The results obtained show that CA is capable of providing a transformation of TMS data which leads to better classification results than provided by all seven bands, by PCA, or by band selection. A second conclusion drawn from the study is that TMS bands 2, 3, 4, and 7 (thermal) are most important for landcover classification.

A comparison of the usefulness of canonical analysis, principal components analysis, and band selection for extraction of features from TMS data for landcover analysis

The conditions under which a hybrid of clustering and canonical analysis for image classification produce optimum results were analyzed. The approach involves generation of classes by clustering for input to canonical analysis. The importance of the number of clusters input and the effect of other parameters of the clustering algorithm (ISOCLS) were examined. The approach derives its final result by clustering the canonically transformed data. Therefore the importance of number of clusters requested in this final stage was also examined. The effect of these variables were studied in terms of the average separability (as measured by transformed divergence) of the final clusters, the transformation matrices resulting from different numbers of input classes, and the accuracy of the final classifications. The research was performed with LANDSAT MSS data over the Hazleton/Berwick Pennsylvania area. Final classifications were compared pixel by pixel with an existing geographic information system to provide an indication of their accuracy.

/pdf/optimization-of-a-non-traditional-unsupervised-2l9in2prya.pdf

R. K. Boyd

Papers

A comparison of the usefulness of canonical analysis, principal components analysis, and band selection for extraction of features from TMS data for landcover analysis

Optimization of a Non-traditional Unsupervised Classification Approach for Land Cover Analysis