Scientific mapping

New Zealand scientists had been making maps long before GIS arrived. Geologists compiled formations and structures from field observations, samples, aerial photographs and earlier surveys. Ecologists divided the country into regions and districts based on recurring combinations of landform, climate, vegetation and biological patterns. Hydrologists worked with catchments, river networks, rainfall stations and flow records. The map was already part of scientific method because many of the questions being asked depended on where an observation had been made and how it related to surrounding places.

Computers entered that work in stages. By 1981 the New Zealand Geological Survey had published material on computerised map plotting. Remote sensing had its own digital lineage from the mid-1970s, while land-resource mapping was already being processed as vector data. The scientific problem in the 1980s was therefore not how to invent spatial thinking. Researchers needed to convert existing maps and observations into information that could be revised, combined and reused.

Scientific mapping also carried a different kind of uncertainty from cadastral work. A geological contact is an interpretation of where one mapped unit gives way to another, based on available exposure, landform, samples and the geologist’s reading of the evidence. An ecological boundary similarly represents a classification applied to a landscape rather than a surveyed legal line. GIS could store those boundaries with great numerical precision, but the number of decimal places in the coordinates did not increase the certainty of the underlying science.

That distinction became more visible as scientists began putting several datasets on the same screen. A geological unit, vegetation class, ecological district, catchment and property boundary could all be represented as polygons or lines, even though each had been created for a different purpose and at a different scale. GIS made comparison easier and therefore increased the need to understand provenance. A clean overlay could reveal a useful relationship, or it could simply place unlike information in the same coordinate system.

Mervyn C. Probine coordinated the DSIR Physics and Engineering Laboratory’s Landsat II investigation. The September 1977 final report records the computing and image-processing work behind the programme, connecting satellite observations with the national effort to develop practical digital remote-sensing capability.

Source WEB-PROBINE-PEOPLE-2026 · DSIR Landsat II final report, September 1977

The Biological Resources Centre began this work in the mid-1980s. In late 1985 the centre was considering how ecological-region and ecological-district boundaries should be digitised rather than continued only through traditional cartographic methods. At about the same time, DSIR’s Science Mapping Unit was beginning to use an Intergraph GIS. Mary McEwen later documented the work in detail while considering how the new Department of Conservation might organise its own geographic information.

The scientific framework already existed. Ecological regions and districts divided New Zealand into areas intended to reflect recurring ecological character. The boundaries could then provide a common geography for other biological information. A protected area, wetland, wildlife site or other inventory record could be related to the ecological district in which it occurred, while geological and soil information could be brought into the same system when required.

A surviving 1986 mock-up of a proposed Biological Resources Information System for the North Taranaki Ecological District records how that idea was being tested. The example contained graphical and textual output and combined material that had originally been held in different map projections. The task was therefore partly scientific and partly technical. Before different inventories could be compared, their coordinates, map frameworks and classifications had to be reconciled well enough for the overlay to mean what the scientists thought it meant.

Biologists used the system alongside field observation and ecological classification. It provided a structure in which observations and mapped classifications could be linked by location. A wildlife record could be selected because it fell inside a particular district. A wetland inventory could be viewed beside protected areas. The geographic framework allowed several scientific records to refer to the same part of the country without requiring a new combined paper map to be drawn for every question.

McEwen’s report also records the maintenance burden that followed. A biological information system would only remain useful if its boundaries, classifications and source records were maintained after the initial conversion. Some data changed because scientific understanding improved, while other information changed because reserves were created, sites were reclassified or new observations were made. The digitised boundaries required continuing maintenance.

Geoff Park, then manager of the Biological Resources Centre, worked with Mary McEwen on the possibilities of GIS for biological information. Her 1989 report recalls their interest when ecological boundaries were being digitised in late 1985. Park also convened the technical advisory group that developed the Protected Natural Areas Programme’s survey method. The connection was practical: field assessment would supply information that could be organised within the proposed Biological Resources Information System.

Source P5-S05 · Mary McEwen 1989 DOC report

The institutional setting changed again in 1992 when DSIR was disestablished and scientific work was redistributed across ten new Crown Research Institutes, drawing also on MAF Technology, the Forest Research Institute, part of the Meteorological Service and the Health Service Laboratories. The reorganisation had already begun inside DSIR: on 1 April 1990 the Science Mapping Unit, Ecology Division, Botany Division and Division of Land and Soil Sciences were merged as DSIR Land Resources. When the CRIs began on 1 July 1992, Land Resources did not pass intact into one successor. Its functions were split between Landcare Research and the Institute of Geological and Nuclear Sciences, later GNS Science, while Landcare also absorbed other groups including the Physics and Engineering Laboratory's Remote Sensing Group. The change moved people, maps, laboratory records, databases, software and technical knowledge into different organisations. Atmospheric, freshwater and marine research went into NIWA, forestry into the New Zealand Forest Research Institute, and agricultural and horticultural work across AgResearch, Crop & Food Research and HortResearch. ESR inherited national environmental and health-science functions. Not every CRI developed a major GIS strand, and the short-lived Institute for Social Research and Development, which closed in 1995, has produced no source-secure GIS story in the bounded research completed for this book.

For spatial science, the transfer raised practical questions about custody. A database could survive an organisational restructuring only if the files, documentation, software and staff knowledge moved with it or were deliberately reconstructed. Chapter 8 recorded a forestry case in which a digital inventory disappeared while a paper printout survived. The scientific institutes inherited many datasets that had been built over decades, so continuity depended as much on information management as on new GIS software.

The Regional Geological Map Archive and Database provides one documented example of deliberate continuity. It was recognised as a nationally significant database during the formation of the Crown Research Institutes. The archive linked the new institution to generations of geological mapping rather than treating the 1990s as a blank start. Field sheets, published maps and geological interpretations remained part of the evidence base from which new digital mapping could be produced.

Landcare Research inherited a different kind of continuity problem in the National Soils Database. The soil records had been entered into Datatrieve on a MicroVAX II from 1980. In 1993, just after the CRI restructuring, FRST funded their transfer to Borland Paradox on Windows. In 2000 the database moved again, this time to Microsoft SQL Server with access through Soil Explorer. Maintaining the data through successive software changes required continuing work. A national point database representing about 1,500 soil profiles survived institutional and computing changes because the data were repeatedly moved into maintainable systems rather than being left behind with the machine that first held them.

The new institutes used spatial computing for different research tasks. CLIMPACTS, developed from 1993 with Richard Warrick and Gavin Kenny among its researchers, brought NIWA, AgResearch, Crop & Food Research, HortResearch, Landcare Research and university researchers into a common climate-impact assessment programme. Its national and regional work combined climate information with sector models to examine spatial changes in agricultural and horticultural suitability. That was a different model from QMAP or a corporate GIS: specialist science remained distributed across organisations, while a shared spatial framework allowed the components to be analysed together.

Smaller projects compared older information with new imagery and field observations. In Gisborne, John Dymond, M. J. Page and L. J. Brown used Landsat TM imagery to prepare 1:100,000 vegetation mapping during debate over a forestry incentive scheme. The classification was compared with vegetation information already held in GIS, much of it roughly twenty years old, and the largest discrepancies were taken back into the field. Some represented real landscape change and some were classification error. After field correction the reported overall accuracy rose from 84 to 90 per cent. Remote sensing had not removed the need to look out the window, and in this case the old GIS data were useful partly because they were old.

Fire modelling supplied a more dramatic test. George Perry's 1996 University of Canterbury work joined Arc/Info to the Rothermel fire-spread model in a system called PYROCART, then tested it against a fire that had already burnt about 580 hectares of the Cass Basin on 27 and 28 May 1995. Differential GPS, aerial photography, terrain and vegetation information helped reconstruct the landscape through which the fire had moved. Perry, Ashley Sparrow and Ian Owens later reported overall predictive accuracy of about 80 per cent, with high-wind conditions among the places where the model struggled. The research demonstrated useful predictive power but also concluded that the data preparation and parameterisation were too demanding for use at the fire front. Researchers could use GIS to analyse the fire afterwards, while assembling the same information during an active fire remained difficult.

Veterinary epidemiology used spatial modelling for a different kind of moving hazard. At the 1998 SIRC colloquium, J. S. McKenzie, D. U. Pfeiffer and R. S. Morris described ArcView 3.0a and Spatial Analyst models for tuberculosis risk in possums and cattle. One model combined vegetation and slope information on a 20-metre grid, while another related farm geography to disease risk. The results were incorporated into the EpiMAN(TB) decision-support system so control effort could be directed spatially. The analysis examined factors associated with the disease distribution. The map was becoming part of the argument about where limited control resources should go next.

Freshwater research provided another example in 2002. David Rowe, Ude Shankar and M. James used ArcInfo to build a three-dimensional model of Lake Taupō's shallow littoral zone from roughly 300 shoreline transects, aerial photography, direct observations, echosounding and differential GPS. They were interested in a practical management question: how would different permitted lake levels alter the clean-sand habitat used by smelt for spawning at depths between about half a metre and two and a half metres? Five lake levels could be tested against the model rather than treated as an abstract change in water elevation. The GIS connected a management setting at the lake surface to fish habitat around kilometres of shoreline.

By 2004 NIWA was also turning several long-running freshwater datasets into infrastructure rather than leaving them as separate research collections. Don Robertson and Mary de Winton described the Freshwater Biodata Information System, FBIS, as a web-based system joining records for fish, invertebrates, submerged macrophytes and other freshwater biota, some of them extending back decades. Users could search by organism, sampling attributes or location, and a web map allowed sites to be selected geographically. The contemporary article even preserved a screen image of the system, a small but valuable survival from an era in which the data often outlasted the interfaces built to reach them.

AgResearch shows a third pattern. At SIRC in December 2007, Bruce McLennan described GIS uptake inside the institute as piecemeal: individual researchers had built useful project-level capability, but access to software, core datasets and support was uneven across its four campuses. AgResearch was therefore preparing a company-wide geographic information service intended to provide common software access, core data, specialist support and in-house training. The CRI began organising GIS as a shared service for its research teams.

The same transition created an opportunity to redesign how national science mapping was organised. Instead of keeping each new geological sheet largely as an isolated cartographic product, GIS allowed the mapped units and structures to be managed as data that could be checked, updated and carried into later products. The organisation still needed field geology and cartographic judgement. The difference was that more of the interpretation could persist in a reusable digital form between editions and between map sheets.

CLIMPACTS linked assessments at national, regional and site scales. NIWA supplied climate expertise, while agricultural, horticultural and land-resource researchers contributed models suited to their own sectors. The University of Waikato participated alongside NIWA, AgResearch, Crop & Food Research, HortResearch and Landcare Research. A common assessment framework allowed researchers to examine how a climate scenario could affect different land uses in different places. The work depended on the relationships between the component models and their spatial inputs, as well as the maps used to present the results.

QMAP began in 1994 with compilation of the Dunedin sheet. By the time the twenty-first sheet completed the series, the project had run for nearly eighteen years and cost about NZ$24 million. GIS was used for data capture, data management and map production, so the result was an attribute-rich national spatial database as well as a replacement map series. It was an attribute-rich, nationally consistent spatial database built from new fieldwork, earlier maps, unpublished research and revised geological interpretation.

The production team is unusually well documented. Mark Rattenbury and Mike Isaac's project history names Philip Carthew, Greg Drummond, David Heron, Biljana Lukovic, Jeff Lyall, Ben Morrison, Penny Murray and Belinda Smith Lyttle among those providing text production, GIS and cartography. Universities supplied thesis and unpublished mapping, while NIWA supplied offshore data. Digital image products were being released from 2004 and vector GIS products from 2006. By 2012 QMAP data could also be consumed through web map services, so a programme conceived as a national map series had become a reusable national data service.

Fieldwork remained at the beginning of the process. Geologists still examined rock exposures, structures, landforms and relationships between units. They used previous maps, reports, photographs and other observations to decide how the geology should be interpreted across areas where the rocks were not continuously visible. GIS did not decide where a fault ran through covered ground or which formation best explained a sequence of outcrops. It held the interpretation after that judgement had been made and allowed the linework and descriptions to be managed more systematically.

A geological map contains several linked kinds of information. A coloured polygon represents a geological unit, but the map also records contacts, faults, folds, point observations and other structural information. Descriptions identify age, rock type and relationships between units. In a paper series much of that information is compressed into the sheet, legend and accompanying text. In GIS the geometry and descriptive records can be maintained separately from the cartographic layout used for one publication.

That separation changed revision work. Correcting a geological boundary in a maintained dataset could feed later maps and analyses without redrawing every product built from the same information. A researcher interested in one formation could select its mapped extent directly. Another could combine the geology with elevation, geochemistry, mineral occurrences or other spatial evidence. The published sheet remained a useful scientific summary, but it was no longer the only durable form of the mapping.

QMAP also exposed the amount of work between field interpretation and final map. Field geologists produced observations and interpretations. Compilers reconciled material from different periods, scales and authors. Digital capture staff converted or edited linework and attributes. Cartographers then had to produce a readable map from a database that could contain far more detail than could sensibly be shown on one sheet. The finished sheet could look calm and authoritative. Producing the map required staff to reconcile field observations, earlier maps, database rules and the space available on the printed sheet.

Later GNS metadata makes some of that production work visible. David Heron is credited with map design, digital cartography and GIS development in the national geological-map lineage. Maureen Coomer, Aleksandra Lukovic, Belinda Smith Lyttle and Delia Strong are explicitly credited with digital capture in later national geology metadata. Those later credits should not be projected backwards onto the first QMAP sheets, but they document the kinds of specialist work required to maintain national digital geology.

Digital capture required interpretation and checking. The mapped line had to be connected to the correct geological unit and retain the intended topology and attributes. Adjacent polygons needed consistent shared boundaries. Faults and other linear features had to retain their own identities rather than becoming accidental edges between coloured areas. Where older source maps differed in scale or interpretation, somebody had to decide how the competing geometry should be handled.

The resulting database therefore carried a production history inside it. A polygon might incorporate a recent field observation, an older geological map, a revised stratigraphic interpretation and digital editing completed by several staff. The final map could look clean because much of that complexity had been resolved before publication. GIS made the result easier to maintain but did not remove the chain of judgement behind it.

A national geological series also exposed problems that were less obvious when individual sheets were treated largely on their own. Geological units did not stop at the edge of a map sheet, and the interpretation made by one compilation team had to remain compatible with work on the adjoining sheet. Names, codes and ages needed to be reconciled well enough that a unit crossing a sheet boundary did not become two unrelated database records simply because it had been mapped at different times.

GIS made those inconsistencies easier to detect. Adjacent datasets could be viewed together, edge geometry could be compared and attributes could be checked across the boundary. The correction still depended on geological judgement, especially where the source mapping differed or new field evidence had changed the interpretation. The computer could identify that two lines failed to meet or two codes differed; it could not decide which geological interpretation was better.

This national production work required a degree of standardisation that was less pressing in one-off research maps. Feature classes, naming conventions, map-unit descriptions and database structures had to be consistent enough for data to move between sheets and later be assembled nationally. Standardisation did not remove regional geological complexity. It provided a common structure in which that complexity could be stored without every new user having to reconstruct the meaning of the data from the printed legend alone.

By the 2000s QMAP data were being used as inputs to analyses that went beyond geological map publication. Simon Cox and Mark Rattenbury documented one example in 2006 using the QMAP dataset for gold-prospectivity modelling. Their work used GIS and weights-of-evidence methods to combine geological information with other indicators associated with mineralisation. The national geological dataset was therefore being treated as an analytical layer rather than only as a collection of finished sheets.

That reuse depended on the data being structured consistently enough to query. If geological units were stored only as coloured artwork, an analyst would first have to reconstruct their extent and meaning. In a GIS dataset the geometry and attributes could be selected programmatically or through spatial queries. The same interpreted geology could support a regional map, a national comparison and a mineral-prospectivity model without being digitised again for each job.

The gain also came with constraints. A prospectivity model could identify areas whose mapped characteristics matched a set of criteria, but it did not create new field evidence by itself. The result inherited the strengths and weaknesses of the geology and other input layers. Sparse observations, old mapping or inconsistent classifications could still affect the output even when the model produced a precise-looking surface or ranking.

The same principle applied across scientific GIS. Digital reuse lowered the effort needed to combine datasets, so source quality and compatibility became more visible parts of research design. A scientist could spend less time redrawing boundaries and more time testing relationships, but only if the data had been documented well enough to understand what each layer represented.

Geological science also moved toward systems intended for use outside the research institution. In 1993 researchers from the University of Otago, the Institute of Geological and Nuclear Sciences and local-authority interests proposed the Dunedin Pilot Hazards Information System. The following year a trial used Arc/Info to bring natural-hazard, topographic and cadastral information into one spatial environment. The project belonged partly to science and partly to the emerging information requirements of councils.

The system allowed geological and hazard evidence to be related to the places where planning and building decisions were being made. A hazard zone could be viewed with property or topographic information rather than remaining inside a specialist report. The scientific interpretation still required geologists and other specialists, while GIS provided a way to organise and distribute the spatial result. Later chapters take up hazards and operational emergency mapping in much greater depth, so the Dunedin pilot is best read here as an example of scientific GIS crossing into administrative use.

This kind of transfer altered the audience for scientific maps. A geological map made primarily for geologists could contain conventions understood by trained readers. A hazards information system had to connect technical interpretation with people making land-use or building decisions. GIS made that connection easier technically while increasing the need for clear metadata, classification and explanation of uncertainty.

Hydrology presented a different spatial problem because the observations changed through time as well as space. Rainfall, streamflow, groundwater levels and other measurements could be attached to gauges or other locations, but the processes being studied operated across catchments and along river networks. A catchment model therefore needed a geographic structure as well as a time series.

Sohail Afzal Choudhry’s 1998 University of Waikato doctoral research on the Kaimai hydropower project catchment provides a detailed New Zealand example. The work combined surface hydrological modelling with GIS and spatial time-series databases. ARC/INFO and ArcView were used alongside object-oriented programming and dynamic segmentation techniques. The research integrated catchment mapping, modelling and spatial time-series data within the same workflow.

GIS provided a way to organise the geography used by the hydrological model. Catchment areas, channels and other spatial components could be related to the measurements and parameters associated with them. Time-series records could then be connected to locations or sections of the network rather than treated as anonymous columns in a table. The model could use spatially varying information instead of assuming that one value represented an entire catchment.

Dynamic segmentation is one example of the technical work involved. A river line could carry changing attributes along its length without requiring every section to be redrawn as a completely separate feature. That approach was useful where measurements or model conditions applied to particular reaches. The technique also illustrates how scientific GIS moved beyond static overlay into more structured representations of processes that occurred along networks.

The Kaimai project was university research; NIWA also used GIS in its operational work. In October 2006 NIWA's Brent Wood described PostGIS running behind about half a dozen systems, with some datasets containing hundreds of millions of features. In December 2008 he described PostGIS, Apache and UMN MapServer as a standards-based web-mapping environment supporting WMS and WFS for maritime, fisheries, ecological and oceanographic work. By 2009 the same practitioner record included GeoServer and OpenLayers, and Andrew Watkins was working with GeoNetwork in 2010.

The sources leave NIWA’s first GIS unresolved and provide no basis for treating one open-source stack as representative of every scientific group in the institute. They establish something more useful than a first: by the later 2000s spatial databases and web services were ordinary production infrastructure inside a national research organisation, handling datasets at scales that would have been awkward to treat as map files alone.

By 2000 GIS was also being used to create classifications from national spatial datasets. Lars Brabyn’s work on automatically classifying New Zealand landscapes used GIS to combine environmental variables and produce a national landscape classification. The method shifted some of the work from drawing regions manually toward applying explicit rules or procedures to digital data.

An automated classification still depended on choices made by the researcher. Variables had to be selected, transformed and weighted or grouped in ways that reflected the purpose of the study. Resolution affected the pattern that emerged. The resulting regions were produced consistently by the procedure, but consistency did not make the classification neutral or permanent.

The difference was repeatability. A manual national classification might depend heavily on an expert drawing boundaries after interpreting several maps. A GIS-based method could record the data and procedures used to produce the classes and then be rerun after a change in method or source data. That made comparison between alternative classifications easier and opened the work to forms of sensitivity testing that were difficult with hand-drawn boundaries.

The landscape-classification work also reflects the expanding supply of reusable spatial data by the end of the 1990s. Automated national analysis was only practical when elevation, land cover, climate or other environmental information existed in forms that could be combined systematically. Scientific GIS therefore depended on the national data infrastructure developing elsewhere in government and research organisations.

Across geology, ecology and hydrology the same change appears in different forms. Scientists were no longer producing only maps that happened to be made by computer. They were maintaining spatial evidence from which maps, tables, models and later analyses could be produced. The database became part of the scientific record.

That created new preservation requirements. A paper geological sheet could remain readable for decades without its original production system. A GIS dataset could contain more reusable information but depend on file formats, attribute definitions, database structures and documentation that were easier to lose. Institutional restructuring in 1992 and later software changes therefore made metadata and data stewardship part of scientific continuity.

It also changed authorship. A printed scientific map might carry the names of the principal geologists or editors, while the database behind it could depend on digital cartographers, programmers, capture staff and database managers whose work was less visible on the sheet. Later GNS metadata that names capture staff is useful precisely because much earlier scientific GIS production was documented mainly through the final map or senior authors.

The same issue applied to scientific models. A hydrological or prospectivity analysis might be published under a few authors’ names while depending on spatial datasets assembled by other teams over many years. GIS increased the amount of inherited information that could be folded into one analysis. Reuse therefore connected new research to older fieldwork and data-management decisions, even when those earlier contributors were not named in the final paper.

Geologists, ecologists and hydrologists used GIS in ways suited to their disciplines. Geology relied heavily on interpreted polygons, structural features and field compilation. Ecology used geographic frameworks to relate observations and inventories across the landscape. Hydrology needed networks, catchments and measurements that changed through time. Landscape classification used national environmental data to generate new spatial categories through repeatable procedures.

Their systems nevertheless shared a growing set of requirements. Data needed coordinates, documented classifications, maintainable attributes and a way to move between analytical software and cartographic output. Scientists needed to know scale, source, date and uncertainty. Organisations needed to preserve both the files and enough institutional knowledge to make them usable after staff or software changed.

Scientists used commercial GIS alongside their research methods during the 1990s. Arc/Info could store a geological unit or a river network because scientists had already decided what the feature represented and how it should be classified. ArcView could display model results because the modelling work had produced values tied to locations. The software made it easier to compare information held in maps, tables and reports.

By the early 2000s a national geological map series could be maintained as digital data, ecological frameworks could organise several scientific inventories, catchment research could join spatial and time-series information, and national landscapes could be classified through repeatable GIS procedures. Scientific mapping was becoming less dependent on one final sheet as the permanent home of the information. The next stage of the history follows the same tools into government organisations, where spatial data were increasingly used for administration, regulation, services and everyday operational work.