Research Programme

Timely and accurate data on land cover are critical for improved management of our land environment, providing a basis for better resource management decisions and allowing more effective use of the natural resource while not compromising environmental outcomes. The New Zealand Land Cover Database (LCDB) is a digital map of land surface cover based on satellite imagery. It contains detailed information on categories of land cover and their boundaries and is a record of land-cover changes over time. The first edition was mapped from 1996/97 data and a second edition update was mapped from 2001/02 data. The LCDB has become a key dataset widely used by central and regional government agencies, research institutes, and industry to answer questions on resource state or to underpin environmental models. For example, the LCDB is used for national environmental reporting, identifying native land cover, habitats of pest species, and vulnerability to erosion or fire; assigning biodiversity priorities; and monitoring changes in land use (MfE website 2012).

The central focus of this programme is to create a stable home for the LCDB – both to generate new editions and conduct the necessary research and development to improve mapping quality, efficiency, and accessibility. The major achievement for 2011/12 has been the generation of LCDB v3.0, which incorporates a third time-step based on 2007/08 satellite imagery. LCDB v3.0 was publically released on 29th June 2012 and, in addition to the extra time-step, significant improvements have been made to its classification accuracy, its line work quality, and its consistency with other New Zealand map datasets. As an indication, approximately 64,000 polygons were manually modified in this revision, with about 36% of these being real change between 2001/02 and 2007/08 and the remainder being corrections on previous mapping.

Over the past year research has focused on automatic delineation of features within SPOT5 satellite imagery mosaics. We have developed a new segmentation algorithm (Shepherd et al. 2012) to identify image boundaries so that these no longer need to be digitized by hand. The technique has proved to be computationally efficient and has been recently trialled over paired SPOT5 imagery from 2008 and 2012. We have made the new algorithm freely available within an open source software package, the Remote Sensing and GIS software library (RSGISLib; The method produces a suite of polygons, all greater than the minimum mapping unit (1 ha), which represent features observable within the satellite imagery over very large regions. These candidate polygons now form the basis for the 2012/13 research into ‘classification’ – how to attach thematic attributes accurately to each zone, and ‘smart editing’ – how to incorporate new useful polygons back into the existing LCDB map to generate LCDB v4 in a more automated manner.

Shepherd, J.D., Bunting, P., Dymond, J.R., 2012. Segmentation of imagery based on iterative elimination, Submitted to Remote Sensing.

Research Themes

•Image understanding and automatic feature extraction (colour, context, time)

A key goal of this research programme is to make the best use of historic, current and future remotely sensed image data for land cover mapping to benefit NZ. While there is is starting to be a relatively rich sequence of satellite imagery of NZ, the imagery is from different dates, sensors, resolutions and views. Historically this information has been assessed and classified manually, and while manual interpretation by trained experts is powerful is it also expensive, slow and can suffer from inconsistencies between operators. This research area aims to reduce the need for manual interpretation by using computer interpretation of standardised image sequences. Over the initial year of the LCDB research programme development was focused on image segmentation (Shepherd et al. 2012). This development has produced a method to automatically process an image sequence into a candidate set of image features (or polygons) representing both land cover and land cover change. The 2012/13 research plan for this work area focuses on segment classification, both change identification and classification at a given date. The intent is to relate spectral characteristics of local similar segments to existing map information to inform the classification using exisiting LCDB context. A scientific paper documenting this segment classification process will be written and methods to include the classified segmented with the LCDB4 mapping stream developed in 2012/13.

•Smart editing, using semi-automatic vector processes

Automated delineation and classification of features from a satellite imagery sequence is only the first part in the process of automating the LCDB mapping process. The current LCDB dataset is vector and is edited and updated within an ARCmap editing environment. Development is required to desgn a process to select appropriate polygons from the automated mapping and accept and insert these within the vector dataset in an efficient  and clean way. By clean we are referring to ensuring we do not produce "slivers" less than the MMU and enforse sensible topology. This smart editing idea is not trivial as we expecting to merge automatically generated polygons with an existing manually digitised dataset and preserving its integrity while improving the potential to add information from the underlying satellite imagery. Software tools will be developed to add capability to ARCMap to efficiently insert automatically extracted features in 2012/13 and these tools will be used in the LCDB4 mapping process.

•Image calibration to standardised reflectance

Further basic research into removing the effected of view, illumination and slope angle within calibrated satellite mosaics is an underpinning theme for the LCDB3 research programme. A UAV capable of flying and capturing imagery from a pared set of visible and near-infrared DSLR cameras is planned for 2012/13. The imagery will be from calibrated fish-eye images and will enable further parameterisation of both topographic and BRDF correction processes that are applied to the satellite imagery mosaics that are input to feature extraction and classification. UAV equipment will be tested and imagery acquired 2012/13. Improved correction models will be developed and a scientific paper written on the improved correction process in 2013/14.

•Specialist layers for specific applications (LCDB+)

No development in 2012/13. Extra "plugin-in" layers such as further indigenous forest classes, improved tussock grassland mapping and wetland delineation is planned post LCDB4.

•Mapping accuracy, polygon precision

•Web based delivery & feedback

•Use of future sensors (Lidar)