Hyperspectral imaging of hill country farms : a thesis presented in partial fulfilment of the requirements of the degree of Doctor of Philosophy of Agriculture and Horticulture at Massey University, Manawatu, New Zealand

Thumbnail Image
Open Access Location
Journal Title
Journal ISSN
Volume Title
Massey University
The Author
This thesis uses hyperspectral aerial imagery, processed and classified using a Support Vector Machine (SVM) approach applied to categorise the New Zealand hill farming environment. The analysis of hyperspectral imagery presented in this thesis provides information on land use and land cover that can assist land management decision-making for hill country farming. The ability of the approach to provide a mechanism to examine complex and inaccessible environments and capture information in fine detail makes it relevant to the management of other heterogeneous environments and marginal farming systems worldwide. Precision farming techniques, used regularly in other farming sectors, hold the promise to better understand the hill farming landscape and therefore improve strategic management decisions. Pasture is the primary resource on the farm but due to the heterogenous nature of the hill farm landscape, the pasture area is currently only estimated. Aerially applied fertiliser applications represent the largest single input for these farms and are also a major source of nutrient contamination in waterways so finding ways to reduce costs and environmental damage are important. The definition of the area and various pasture groups is critical information needed to improve fertiliser efficiency via use of Variable Rate Application Technology systems. This research was able to classify pasture area to 99.59% (Kappa 0.991). Accurate base landscape information can improve management decisions, the accuracy of valuations, income expectations from lending organisations and the overall prosperity of the hill farming sector. Currently farmers and external groups must make major financial and strategic decisions with local expert opinion which is difficult to validate or question. Therefore, information derived from the hyperspectral classification is also shown to have benefits for strategic farm management decision-making and the wider farming community. This research was able to classify a number of economically valuable resources to high accuracies including;water bodies (99.97%), Thistle (98.51%), Pine (99.44%), Kanuka (89.03%) and Manuka (97.71%). By applying SVM to hyperspectral imagery the classification of pasture could be enhanced by the use of plant functional groups. The classes of High Fertility Responsive (HFR) represented sown rye varieties and had a classification accuracy of 89.06%. Low Fertility Tolerant (LFT) represented mixed swards dominated by browntop with a classification accuracy of 89.81%. The highest accuracy achieved for the legume class was 99.81%. The findings from this study represent a notable advance in our understanding of hill country farm and remote sensing research relevant to hill country farming. This is the first study to classify several key landscape components that are economically or environmentally important to the hill country farming community and this study created the most detailed map of hill farm pasture quality using plant functional groups so far. The ability to use a single hyperspectral aerial survey, to provide such a wide variety of information, useful to many industry actors, improves the potential return on investment and viability of the survey operation.
Land cover, New Zealand, Remote sensing, Hill farming, Management, Pastures, Fertilizers, Agricultural informatics, Hyperspectral imaging