Search

The decreased instantaneous field of view (IFOV) is one of the principal advances noted for the Thematic Mapper (TM) sensor. The 42.5 microradian IFOV of TM and the 710 km nominal orbit altitude result in a 30 m nominal spatial resolution at the earth surface. This is a considerable decrease in the projected pixel area when compared to the 79 m nominal spatial resolution of the Landsat Multispectral Scanner (MSS). An experiment was conducted which allowed a rigorous test of the influence of classifier design, with data spatial resolution of TM (30 m) and approximately that of the Landsat MSS (90 m), on classification performance for a particular TM scene. The experiment involved evaluation of the results for the per-point Gaussian maximum likelihood (GML) classifier and the supervised ECHO (Extraction and Classification of Homogeneous Objects) classifier.

A fixed effect, three factor (two levels per factor) analysis of variance was used to quantitatively assess the significance of the improved spectral, spatial and radiometric resolution capabilities of the LANDSAT-4 thematic mapper sensor relative to the familiar MSS sensor. TM data acquired over the Washington, D.C. area were progressively degraded in spectral, spatial and radiometric characteristics to simulate the MSS, and classification accuracies were derived in a consistent manner for all eight treatments in the ANOVA design. Statistical testing of the significance of differences in classification accuracies between treatments indicated that the increased number of spectral bands and the improved quantization capabilities afforded by the TM sensor design would lead to significant improvements in classification accuracies attainable relative to MSS. In contrast, however, the improved spatial resolution provided by the TM sensor did not enhance classification accuracy. This latter result was felt to be more a function of the type of classification algorithms available.

Classification capabilities with TM data result from the interactive effects of all of the sensor's attributes which complicates a more quantitative evaluation of the effects of individual sensor improvements. An experiment conducted to quantify the effect of individual sensor parameters (e.g., spectral, spatial, and radiometric resolution) on classification accuracy is described on classification accuracy. Preliminary results obtained using TM data acquired over the Washington, D.C., area indicate that the additional number of spectral bands and quantization levels of the TM relative to the MSS increase capabilities for the recognition and discrimination of land cover/use categories by per-pixel maximum likelihood classification. The refinement of spatial resolution, however, seems to hinder classification.

A fixed factor analysis-of-variance (ANOVA) approach was used to quantify the impact of each TM sensor characteristic (spectral, spatial, and radiometric resolutions) on classification accuracy. All assessments were made relative to MSS sensor characteristics, and the impact of each characteristic was assessed individually, and in all possible combinations. Thematic Mapper data acquired over the Washington, D.C. area on November 2, 1982 were utilized to conduct the experiment. The following results were obtained: the reduction of quantization level from eight bits to six bits caused a decrease in overall accuracy (7%); the use of only three bands (TM 2, 3 and 4) covering the isible and near infrared portion of the spectrum caused a decrease in overall accuracy (7%); and the decrease in spatial resolution resulted in an increase in overall accuracy (4%). Results indicate that the increased radiometric and spectral resolution of the TM instrument do provide increased information content. The result of the spatial resolution degradation is somewhat misleading, in that the result is more a function of spatial resolution.

A three factor (spectral, spatial, and radiometric resolution), two level (TM and MSS) analysis of variance (ANOVA) approach allowed evaluation of the effects of each factor individually and in all possible combinations. Digital classification accuracy was used as the figure of merit. Nine study sites in Washington, D.C. each of approximately 256 x 256 TM pixels, were randomly selected from the full scene for analysis. These results strongly suggest that the quantization level improvements and the addition of new spectral bands in the visible and middle IR regions (both afforded by the TM sensor design) can result in improved capabilities to accurately delineate land cover categories using a per point Gaussian maximum likelihood classifier. On the other hand, results indicate that the increase in spatial resolution to 30m does not significantly enhance classification accuracy. The spatial result points to an inherent limitation of a per point classifier and to the need to improve data analysis techniques to handle high spatial resolution data.

On July 16, 1982, the second decade of land remote sensing from space was inaugurated with the successful launch of Landsat-4. This satellite carries the Multispectral Scanner (MSS) and a new sensor, the Thematic Mapper (TM). The TM represents the result of an effort in which all of the major improvements in remote-sensing capability were simultaneously integrated into one system. An experiment was developed and conducted to quantify the effect of each TM sensor parameter on classification accuracy. This paper discusses the experimental design and summarizes the results obtained using TM data acquired over the Washington, DC area on November 2, 1982. Attention is given to a study site/data description, the experimental design, photointerpretation and digitization, spectral simulation, radiometric simulation, and spatial simulation.

Thematic Mapper Simulator data were collected over a 23,200 hectare forested area near Baxter State Park in north-central Maine. Photointerpreted ground reference information was used to drive a stratified random sampling procedure for waveband discriminant analyses and to generate training statistics and test pixel accuracies. Stepwise discriminant analyses indicated that the following bands best differentiated the thirteen level II - III cover types (in order of entry): near infrared (0.77 to 0.90 micron), blue (0.46 0.52 micron), first middle infrared (1.53 to 1.73 microns), second middle infrared (2.06 to 2.33 microsn), red (0.63 to 0.69 micron), thermal (10.32 to 12.33 microns). Classification accuracies peaked at 58 percent for thirteen level II-III land-cover classes and at 65 percent for ten level II classes.

A computer algorithm to segment Landsat Thematic Mapper (TM) images into areas representing surface features is described. The algorithm is based on a region growing approach and uses edge elements and edge element orientation to define the limits of the surface features. Adjacent regions which are not separated by edges are linked to form larger regions. Some of the advantages of scene segmentation over conventional TM image extraction algorithms are discussed, including surface feature analysis on a pixel-by-pixel basis, and faster identification of the pixels in each region. A detailed flow diagram of region growing algorithm is provided.

A three factor (spectral, spatial, and radiometric resolution), two level (TM and MSS) analysis of variance (ANOVA) approach allowed evaluation of the effects of each factor individually and in all possible combinations. Digital classification accuracy was used as the figure of merit. Nine study sites in Washington, DC, each of approximately 256 x 256 TM pixels, were randomly selected from the full scene for analysis. These results strongly suggest that the quantization level improvements and the addition of new spectral bands in the visible and middle IR regions (both afforded by the TM sensor design) can result in improved capabilities to accurately delineate land cover categories using a per point Gaussian maximum likelihood classifier. On the other hand, results indicate that the increase in spatial resolution to 30 m does not significantly enhance classification accuracy. The spatial result points to an inherent limitation of a per point classifier and to the need to improve data analysis techniques to handle high spatial resolution data.

Preliminary experimental results of airborne thematic mapper (TM) data taken to quantify the effect of three major TM sensor parameters, spectral, spatial, and radiometric resolution, six months after launch of Landsat-4 are reported. The flight took place on Nov. 2, 1982 over Washington, D.C., and data gathered were compared with ground reference data from color airborne photography on a 1:40,000 scale. Analyses proceeded by deleting one band from each of four data sets, thus making the data equivalent to MSS data. Attention was directed to land cover/use classes in a quick-look format. A per-pixel maximum likelihood scheme was found to increase the recognition and dicrimination categorization capabilities. Finer spatial resolution, however, impeded classification due to increased within-class variability of the field-center pixels, which also incresed class overlap in the spectral data base. Improved data analyses techniques are therefore needed to exploit the available higher spatial resolution of the TM.

Kershaw County, South Carolina was selected as the study site for analyzing simulated thematic mapper MSS data and dual-polarized X-band synthetic aperture radar (SAR) data. The impact of the improved spatial and spectral characteristics of the LANDSAT D thematic mapper data on computer aided analysis for forest cover type mapping was examined as well as the value of synthetic aperture radar data for differentiating forest and other cover types. The utility of pattern recognition techniques for analyzing SAR data was assessed. Topics covered include: (1) collection and of TMS and reference data; (2) reformatting, geometric and radiometric rectification, and spatial resolution degradation of TMS data; (3) development of training statistics and test data sets; (4) evaluation of different numbers and combinations of wavelength bands on classification performance; (5) comparison among three classification algorithms; and (6) the effectiveness of the principal component transformation in data analysis. The collection, digitization, reformatting, and geometric adjustment of SAR data are also discussed. Image interpretation results and classification results are presented.

Data sets simulating three different spatial resolutions (SR's) are computed from data with a 15-m nominal SR that were obtained with NASA's Thermatic Mapper Simulator from an altitude of about 6 km. The classification accuracies (CA's) achieved with the data of each of the four different SR's using a per-point Gaussian maximum likelihood classifier (GMLC) are intercompared. The CA's obtained using simulated 30-m SR data with the per-point GMLC are compared with the CA's achieved with a per-field classifier. It is found that: (1) the use of successively higher SR data resulted in lower overall CA's for classifications with the per-point GMLC, especially in cover classes associated with relatively higher spectral variability across adjacent pixels; (2) higher CA's were achieved using the per-field classifier with 30-m SR data than were achieved with the per-point GMLC; and (3) the largest increases in CA's were achieved with the per-field classifier in cover classes associated with relatively high levels of spectral variability across adjacent pixels.

This study involved the evaluation of the characteristics of multispectral scanner data relative to forest cover type mapping, using NASA's NS-001 multispectral scanner to simulate the proposed Thematic Mapper (TM). The objectives were to determine: (1) the optimum number of wavebands to utilize in computer classifications of TM data; (2) which channel combinations provide the highest expected classification accuracy; and (3) the relative merit of each channel in the context of the cover classes examined. Transformed divergence was used as a measure of statistical distance between spectral class densities associated with each of twelve cover classes. The maximum overall mean pair-wise transformed divergence was used as the basis for evaluating all possible waveband combinations available for use in computer-assisted forest cover classifications.

Separate holograms of horizontally (HH) and vertically (HV) polarized responses obtained by the APQ-102 side-looking radar were processed through an optical correlator and the resulting image was recorded on positive film from which black and white negative and positive prints were made. Visual comparison of the HH and HV images reveals a distinct dark band in the imagery which covers about 30% of the radar strip. Preliminary evaluaton of the flight line 1 date indicates that various features on the HH and HV images seem to have different response levels. The amount of sidelap due to the look angle between flight lines 1 and 2 is negligible. NASA mission #425 to obtain flightlines of NS-001 MSS data and supporting aerial photography was successfully flown. Flight line 3 data are of very good quality and virtually cloud-free. Results of data analysis for selection of test fields and for evaluation of waveband combination and spatial resolution are presented.