Landsat arrives
In October 1975, the Department of Scientific and Industrial Research’s Physics and Engineering Laboratory in Lower Hutt received a Landsat II scene of New Zealand for the organised national investigation programme. was the principal investigator, with the Remote Sensing Section handling much of the technical work. The programme reports treat this as the start of the formal Landsat II investigation, although an earlier Landsat-1 scene of Canterbury, recorded on 7 December 1973, was later forwarded to DSIR. Satellite imagery had already been collected before the national programme began. By late 1975, however, PEL had a continuing programme for receiving, processing and distributing Landsat data.
The material arrived as computer-compatible magnetic tape rather than as a finished map. Landsat’s multispectral scanner recorded reflected energy in several wavelength bands and stored the measurements as a regular grid. Each cell represented an area on the ground and carried numerical values for the sensor bands. The data could be processed before any image was printed or displayed. This differed from conventional aerial photography, where the photograph itself was normally the primary object and later measurements were derived from it.
PEL staff therefore worked with raster data before raster GIS became routine software. The grid structure allowed programs to alter contrast, compare bands, remove known sensor effects, group similar values and register one scene against another. The output could still be a map or colour image, but the processing was applied to the stored measurements rather than to a photographic print. Forestry, agriculture, snow studies, land evaluation and other applications could use different combinations of the same source bands. The same scene could produce several derived products without changing the original tape.
Processing the tapes
The PEL processing environment was spread across several computers, and the data physically travelled with the workflow. Contemporary reports describe access to an IBM 370/168 in Wellington, together with in-house HP 2100 and Varian 620-series systems. Programs on the IBM were written in PL/I. NASA tapes were decoded on the central system, known artefacts were corrected, and selected image areas could then be transferred to nine-track tape for further work on local machines. The workflow depended on both the computing capacity of the larger IBM and the availability of smaller systems within PEL. A satellite had already crossed the country in minutes; getting a useful piece of its imagery through the computers was still a decidedly terrestrial exercise.
The Remote Sensing Section’s staff are unusually well documented for an early New Zealand digital-mapping programme. worked on IBM tape decoding, rectification, image cataloguing and distribution. worked on digital image enhancement and cluster analysis. was involved in colour processing, multispectral compositing and calibration, while worked on electronics and preparation of the aircraft multispectral system. Dr headed the Remote Sensing Section. The programme required scientific interpretation, software, electronics, data handling and cartographic correction.
The processing routines included histogram equalisation, band ratios, scaling, line-error removal, enhancement and clustering. These operations were constrained by the size of the machines and by the amount of data that could be moved conveniently between them. Instead of repeatedly processing an entire scene, staff often worked with selected subsets. Intermediate products were written to tape and moved between systems. Storage and transfer were part of the technical workflow rather than background services that could be ignored.
Output was also limited by the available devices. Early work included small 128 by 128 image areas printed through an electrostatic line printer and then photographed to produce colour composites. The satellite was modern; the output chain could still end with somebody photographing a printout. Later reports describe an Optronics Colorwrite and the LANSYS image-processing package. The processing chain could therefore run from satellite sensor to magnetic tape, central computer, local computer, printer or colour-output device, and finally to a product that a scientist or mapping specialist could interpret. None of those steps was automatic, and several depended on locally written software.
Rectification and registration
Raw satellite measurements required correction before they could be aligned with New Zealand maps. The imagery contained geometric effects associated with the sensor, satellite motion and the way the Earth was viewed from orbit. PEL staff used rectification methods to correct the geometry and relate the raster to map coordinates. Once rectified, an image could be compared with mapped features and with imagery from another date. Registration placed repeated observations of the same area into a common geographic frame.
Researchers comparing change through time needed consistent acquisition dates and processing methods. If two scenes were slightly misaligned, a simple comparison could produce false differences along field edges, rivers or forest boundaries. Registered images reduced that problem by placing corresponding locations in the same part of the grid. The same approach was used with aircraft multispectral imagery and map control. Satellite, aircraft and mapped data could then be examined together instead of being treated as unrelated pictures.
The coordinate framework used by mapping agencies therefore remained part of remote-sensing work. Landsat supplied numerical measurements, but analysts still needed control points, map coordinates and geometric correction before the data could be combined reliably with other spatial information. PEL’s technical work included both radiometric processing and cartographic correction. The programme combined image processing with mapping.
RILIS
As the image collection grew, PEL also needed a way to find the right scene. and documented the Remote-sensing Image Library Information System, RILIS, in an April 1976 user guide. RILIS ran in Time Shared BASIC on the HP 2100. It stored information about available imagery and allowed users to search the collection using geographic and other identifying information. The system dealt with the catalogue rather than performing the image analysis itself.
RILIS is an early example of a problem that later became routine in spatial-data management. A growing archive required metadata describing location, date, sensor and other characteristics. Without that information, researchers could have a shelf of useful tapes and still spend considerable time working out which one covered the area they needed. RILIS gave the Remote Sensing Section a computerised index to the imagery it was receiving and processing. The underlying image archive and the information used to discover it were maintained as separate but related resources.
The distinction between source data and metadata also applied to processed products. A single Landsat scene could generate corrected subsets, enhanced images, colour composites and classifications. Each derivative needed enough documentation for later users to know what had been done to it. By 1976 PEL was already managing both the image data and the descriptive information required to locate and interpret it. Later national spatial-data catalogues would operate at much larger scale, but the administrative problem was already present.
Darfield and ground truth
Satellite classification depended on observations from the ground. PEL used Darfield as one of its recurring test areas and supported the Landsat programme with aircraft and field work. The aircraft system used four Hasselblad cameras arranged to record different spectral ranges chosen to approximate Landsat bands. Radiometers and cassette-tape recording were also used in the survey programme. Farmers in the study area supplied information about crops and ground conditions so that spectral patterns could be compared with known land cover.
of DSIR Crop Research is credited in the programme reports with proposing an in-camera radiometer approach. worked on the multispectral-survey equipment and electronics. The aircraft observations provided measurements at a different scale from Landsat and could be matched with field information. Researchers could then test whether patterns in the satellite data corresponded with the land-cover classes they were trying to identify. This reduced the risk of treating a mathematically tidy cluster as a known crop or surface type without field evidence.
Ground truth also exposed seasonal and environmental variation. The same crop could produce different spectral values at different stages of growth, and moisture, soil background and atmospheric conditions could alter the recorded signal. Classification therefore required more than a fixed lookup table. Staff compared sensor values with observations, adjusted processing methods and assessed whether the resulting classes were useful for the intended application. The work combined field knowledge with numerical analysis.
Classification
Cluster analysis was one of the methods used to group pixels with similar spectral characteristics. Other processing combined or compared bands to make particular land-cover differences easier to see. A classified output could then be plotted as a thematic map rather than as a conventional photographic image. The boundaries were generated from the numerical data and the classification method, although analysts still had to decide how the resulting groups should be interpreted and labelled.
This created a different production sequence from conventional thematic cartography. In a traditional mapping project, an interpreter might identify categories from field observations, photographs and other records, then draw or compile the boundaries. In digital image classification, the computer could help derive candidate groups across a large raster before the final categories were assigned. The method worked well where spectral differences corresponded closely with the classes of interest and less well where several land covers had similar responses. The resulting map still depended on classification rules and checking against known conditions.
PEL researchers applied these methods to several sectors. Forestry work used Landsat to examine vegetation and forest patterns. Agricultural studies investigated crop and land-cover discrimination. Snow and hydrology research used repeated imagery to measure broad changes in snow extent, while land-resource work explored ways of combining remote observations with existing mapping. By the final Landsat II programme report, PEL had received dozens of New Zealand scenes and had built a substantial processing and distribution operation around them.
Snow, forests and large areas
Snow studies demonstrate why relatively coarse satellite imagery could be useful. A field measurement provided detailed information at one location, while an aircraft survey could cover a larger area when a flight was available. Landsat repeatedly covered broad areas using the same sensor system. Researchers could register successive scenes and compare the mapped extent of snow across a catchment or mountain region.
The same scale suited some forestry and land-cover work. Landsat did not provide the detail of low-altitude aerial photography, but it covered large areas consistently and could be processed using the same numerical methods from one scene to the next. Resolution had to match the question being asked. A regional vegetation pattern or snow-covered catchment did not require individual trees or small property features to be visible.
Repeated acquisition also produced an archive rather than a single survey. Later scenes could be compared with earlier ones, provided the geometry and processing were controlled well enough. That gave researchers a way to examine change using observations collected by the same broad sensing system. The archive increased steadily as more scenes were acquired, which also increased the need for cataloguing, storage and consistent processing records.
From imagery to mapped data
Remote sensing and orthophotography were developing at the same time but followed different technical paths. Lands and Survey bought a computer system for orthophoto production in 1977, and orthophoto mapping of the Clutha-Alexandra area followed in 1978 and 1979 for Ministry of Works hydro-development planning. Orthophotography started with aerial photographs and corrected their geometry so the imagery could be used in map space. Landsat started with a numerical raster and required geometric correction before it could be aligned with maps and other datasets.
Both workflows depended on coordinates and control. Once an image had been rectified, it could be overlaid with mapped information rather than used only as a visual reference. This made imagery easier to combine with cadastral, topographic and resource data. The later spread of GIS would place these sources inside the same software environments, but the geometric work required to make them compatible was already being done in the 1970s.
The two image traditions also produced different kinds of records. Orthophotos retained a photograph-like appearance at relatively high local detail. Landsat stored broader multispectral measurements that could be analysed numerically before being rendered for display. Both later became ordinary GIS layers, but their production histories, resolutions and analytical uses remained different.
Raster work before commercial GIS
By the end of the 1970s, New Zealand organisations were maintaining several types of machine-readable geographic information. PEL had digital satellite rasters, image-processing software, rectification methods and a computerised imagery catalogue. Lands and Survey was using computers in orthophoto and cadastral mapping. Land-resource staff were digitising mapped environmental units into polygon databases. These systems were developed by different organisations for different operational needs and were not yet parts of one standard national GIS environment.
Raster and vector processing also remained technically distinct. Satellite imagery was organised as grids of measured values. Cadastral and land-resource systems represented discrete features through coordinates and polygons. Each model suited different source material and different questions. Commercial GIS software later combined both forms within common interfaces and data-management systems, but New Zealand practitioners had already accumulated experience with each before that consolidation.
Landsat research continued beyond the first investigation period. Later satellites provided more bands, different resolutions and larger archives, while personal computers and specialist image-processing packages reduced some of the early hardware constraints. The 1975–77 PEL work remains a clear early record of New Zealand organisations processing the landscape directly as digital raster data.
During the same period, land-resource staff were creating the New Zealand Land Resource Inventory from field survey, aerial photography and thematic information. Between 1977 and 1980 they digitised the mapped units and their descriptions into a national database managed through LADEDA. Chapter 5 follows that work with polygons and attributes in an explicitly described geographic information system.
Chapter source notes
1. The principal Landsat source is the DSIR Physics and Engineering Laboratory final report for Landsat II Investigation Programme No. 28230, available through NASA NTRS as 19770026620, together with associated PEL documentation. The organised New Zealand Landsat II programme began in 1975; a Canterbury Landsat-1 scene dated 7 December 1973 prevents the chapter from treating October 1975 as the first New Zealand Landsat image.
2. The PEL evidence supports the use of IBM 370/168, HP 2100 and Varian computing environments, PL/I and specialist image-processing routines only to the extent documented in the technical reports. The chapter does not infer a single integrated modern remote-sensing workstation from those separate systems.
3. The 1976 RILIS material supports early New Zealand interactive raster/image work and the use of time-shared BASIC where documented. The source trail is held in the Chapter 4 development packet and Master Research Register.
4. Ground-truth and classification examples, including the Darfield work, are documented in the relevant PEL/Landsat project publications rather than to later summaries alone. These sources support supervised interpretation and classification, not autonomous modern machine learning.