On the detection and monitoring of invasive exotic conifers in New Zealand using remote sensing.
| dc.contributor.author | Dash, Jonathan P. | |
| dc.date.accessioned | 2020-12-02T01:38:13Z | |
| dc.date.available | 2020-12-02T01:38:13Z | |
| dc.date.issued | 2020 | en |
| dc.description.abstract | Conifers are amongst the most economically and culturally valued trees on Earth. They provide significant ecosystem services both within and outside of their natural range. For this reason, many northern hemisphere conifer species have been planted extensively in southern hemisphere countries. The evolutionary history of many conifer species means that they are frequently invasive in indigenous ecosystems. In New Zealand large areas of northern hemisphere conifers have been planted. These areas include commercial plantations managed for timber production and areas established for erosion control, hydrology management, agricultural, and amenity purposes. Many historical introductions were inappropriately located and included conifer species that are now known as invasive. This resulted in exotic trees with vigourous growth, rapid maturity, and abundant seed production occurring in remote and mountainous areas from where they have spread. The land area now invaded by exotic conifers is estimated to total around 2,000,000 ha and is increasing even under current management. The ecological and social impact of this situation is great, and this has led New Zealand government agencies, and other organisations, to undertake substantial control programmes. Without these efforts a much larger area will be invaded leaving a substantial detrimental legacy. Invasive conifer control relies on herbicide application or mechanical removal. These methods are expensive and challenging particularly when conducted across large areas of frequently mountainous terrain. Accurate information on the location of invasive conifers is either not currently available or is inadequate. Remote sensing surveys potentially offer a useful solution as they can provide valuable information across large areas of challenging terrain and can deliver detailed information about targeted areas. With appropriate research remote sensing methods could provide valuable information to help improve the efficacy and efficiency of invasive conifer control programmes. There has been substantial research into the remote detection of invasive plants originating from many parts of the world. Space-borne and both piloted and unpiloted airborne platforms have been used. In many instances this has provided useful and practical information for managers and policy makers but, to date, research into remote detection of invasive conifers has been limited in scope, incomplete, or restricted to a single environment. Furthermore, a comprehensive review of the research literature revealed several gaps that the experimental research contained within this thesis addressed. Two of the research topics used the emerging technology of unpiloted aerial vehicles (UAVs). These flexible and reliable platforms provide a new data source that can provide ultra-high-resolution data over small to medium sized areas. With appropriate method development this represents a new source of critical information on the early stages of invasive conifer spread where detecting small plants is critical. The technology has delivered substantial efficiency compared to conventional ground- based measurements but has ushered in a new sampling challenge where the operator must decide how to deploy the UAV for data collection. This is particularly relevant where efficient re-use of UAV-based models is desirable to save model development and reference data labelling effort. The challenge of monitoring historic trends in invasive conifer spread was also investigated through analysis of the Landsat archive. No previous research had addressed this topic although automated methods developed for other applications had considerable promise. Single-date analysis can only provide information on the current distribution of invasive conifers and so the impacts of historical policy and management activities remain anecdotal without considering the temporal dimension. Given the knowledge gaps identified in the published literature three broad research questions were developed and addressed through experimentation. 1. Can ultra-high-resolution data be used to detect the presence of invasive exotic conifers prior to the onset of early coning in a highly vulnerable environment? 2. Can UAV-based models of invasive conifer distribution be transferred between sites and does site complexity have a significant effect on model portability? 3. Given the size and availability of the Landsat archive can automated methods be developed that allow tracking of the historical spread and management of invasive conifers? Three experiments were implemented to address these three research questions. To address research question one a detailed field study was installed in a vulnerable environment with a simplistic grassland non-target vegetation structure. The study site was subject to a first order conifer invasion event spreading from a shelter-belt on agricultural land. Very high-resolution multispectral and laser scanning data was collected over the site using both UAV and conventional piloted aircraft. These data were used to develop statistical models suitable for detecting invasive conifers based on their spectral and structural properties. A large field dataset (ca. 17,000 trees) was also collected and used to validate the accuracy of the remote detection methods. Detection errors were characterised with reference to the size, age, and coning status of trees measured at the study site. To address research question two a multi-site case study that encompassed an ecological and site complexity gradient of the vulnerable ecosystems within New Zealand was installed. Across all sites UAV-based models of invasive conifer distributions were developed. These were transferred to all other areas of interest (AOI) within the experimental framework. The experimental design meant that the portability of models both within, and between, sites could be tested. The influence of the complexity of both the donor and receiver site on model predictive accuracy after transfer was also explored. In all cases model accuracy was assessed using cross-validation. To address research question three a methodology was developed based on an implementation of a land cover change tracking algorithm within Google Earth Engine. This technique was applied to a vast and heterogeneous mountainous area of New Zealand’s South Island. The algorithm was carefully configured to identify changes in the pixel-level spectral trajectory within the Landsat archive that might be associated with invasive conifer spread or management control. These change segments were treated as base learners in a subsequent random forest attribution model that defined the causal agent of identified changes in the landscape. The main findings of the research were as follows. 1. Ultra-high-resolution data was extremely useful for the early detection of invasive conifer invasions including small trees. Critically relatively simple statistical methods were suitable for the detection of 99% of all trees found to be coning. 2. UAV-based invasive conifer distribution models were found to be robust to transfer to different AOIs within the same site without a decrease in accuracy. UAV-based invasive conifer distribution models could be transferred to sites with similar or lower complexity than the sites used for model development without a significant decrease in accuracy. However, models transferred to more complex sites could not produce viable results. Invasive conifer models based on spectral data were found to be more robust to transfer than those based on ALS data. 3. The methodology implemented offered a viable means of detecting vegetation changes over time through the Landsat archive. The attribution models developed had moderate accuracy but substantial class confusion remained. Nevertheless, maps of invasive conifer spread and control for the period 2000 - 2019 could be produced and were accurate enough to be useful for assessing the impact of historical management and land use policy within the expansive study area. 4. This research has shown that through exploiting a range of different platforms remote sensing can provide critical information for the detection and monitoring of invasive conifers across a wide range of vulnerable environments. Through matching the platform and sensor to the desired application practitioners can acquire accurate information that operates from the sub-tree scale up to the regional or national scale. New knowledge on the portability of UAV-based invasive conifer distribution models has been produced with clear findings that will increase sampling efficiency, and offer guidance for practitioners on when models can be reused and when new training and model development data is required. The most comprehensive systematic review on the use of an emerging remote sensing technology (UAVs) for invasive alien plant detection was also produced. This review has highlighted trends in the current research, research gaps, and offered guidance on future development pathways that must be followed to increase the value that can be extracted from these datasets. | en |
| dc.identifier.uri | https://hdl.handle.net/10092/101329 | |
| dc.identifier.uri | http://dx.doi.org/10.26021/10392 | |
| dc.language | English | |
| dc.language.iso | en | |
| dc.publisher | University of Canterbury | en |
| dc.rights | All Right Reserved | en |
| dc.rights.uri | https://canterbury.libguides.com/rights/theses | en |
| dc.title | On the detection and monitoring of invasive exotic conifers in New Zealand using remote sensing. | en |
| dc.type | Theses / Dissertations | en |
| thesis.degree.discipline | Forestry | en |
| thesis.degree.grantor | University of Canterbury | en |
| thesis.degree.level | Doctoral | en |
| thesis.degree.name | Doctor of Philosophy | en |
| uc.bibnumber | 2957879 | |
| uc.college | Faculty of Engineering | en |