IAP-24-002

Identification of native vegetation species by linking UAV imagery with terrestrial spectroradiometers

The bioecological richness of subtropical regions and their diverse vegetation are essential for maintaining their ecosystem balance and biodiversity. As environmental challenges increase, the demand for precise and efficient monitoring and classification tools becomes more urgent. Unfortunately, traditional methods, such as unsupervised satellite remote sensing classifications, focus primarily on common vegetation species (e.g., rice, corn, palm) [1] and often fail to capture the dynamic and intricate nature of subtropical ecosystems, particularly in smaller areas with native vegetation species [2]. This calls for developing advanced methodologies to monitor and understand the impact of climate change on vulnerable regions, such as subtropical areas.

This multidisciplinary project addresses the need for accurate and efficient methods to classify and monitor native vegetation species in subtropical regions. It focuses primarily on the northern area of Colombia, where indigenous communities rely on native crops as a primary income source. Subtropical ecosystems, characterized by diverse microclimates and vegetation, present complex monitoring challenges that demand innovative approaches beyond traditional remote sensing methods.

While multispectral satellite imagery has been widely used in large-scale agricultural assessments, its effectiveness is often limited by cloud cover, insufficient ground validation [3], and inadequate spectral resolution, especially in smaller, cloud-prone areas with diverse native vegetation [4]. Additionally, traditional spectral libraries frequently lack the environmental specificity required for these regions, as lab-based signatures are expensive and generally defined for standard crops, failing to account for the complex natural conditions in the wild.

However, advanced UAV technology, with more affordable and sophisticated drones equipped with multispectral sensors, more extended flight capabilities, and increased accessibility, presents promising solutions to these challenges [5]. Integrating UAV imagery with ground-based spectrometers offers an accessible and practical approach to capturing and classifying the intricate dynamics of native species in subtropical areas [6]. To address these gaps, this research proposes the following questions:

1. How can UAV multi-spectral imagery and field spectrometer data be integrated to enhance the classification accuracy of native vegetation species in subtropical regions?
2. What data fusion techniques can be developed to address cloud cover, spatial resolution needs, and environmental variability in native species monitoring?
3. To what extent could new land use/change classification methods contribute to understanding the socio-economic role of native species for indigenous communities in Colombia?

The PhD candidate will encounter several key challenges:

1. Acquiring accurate field data for native species, particularly in subtropical or cloud-prone areas, will require substantial coordination and support from biologist experts on vegetation classification.
2. Develop a novel data fusion methodology combining UAV multi-spectral imagery with field spectrometer spectral signatures for fine-grained classification of native vegetation species.
3. Create a comprehensive library of spectral signatures for native species, bridging the gap between controlled lab conditions and real-world subtropical environments.
4. Understanding and quantifying the socio-economic impact of this vegetation monitoring method on indigenous communities.
5. Apply and evaluate the method in a case study in northern Colombia, focusing on areas with high biodiversity and socio-economic dependence on native species.

Identification of environmental factors: Definition of the study areas for the selected native species in a controlled environment where we can study the environmental conditions and how they might affect the collection of data, then gather the spectral signatures using the on-field spectroradiometers to get the most precise measure of the reflectance for the study species.

Data collection and Quality: Using the UAV multi-spectral imagery, capture the imagery required as input for the data fusion method. This data collection process will be aligned with the environmental conditions presented in the study areas as well as the collection of multiple images for several periods of the year to have a heterogeneous sample of spectral UAV imagery for classification and calibration purposes.

Data Fusion and Processing: Create a data fusion method to integrate what has been collected with the drone (to be later used in large areas) and the spectral signatures collected in the study areas (that would provide a baseline or proxy for calibration and validation process). An extensive analysis of the reflectance response in both data collection processes will be considered in this step, likewise the integration of machine learning models to predict the values captured from the UAV imagery. Once the data has been properly integrated, we will use other UAV multispectral imagery to validate and calibrate a classification and identification of large areas. A collection of curated and available spectral signatures from field spectrometry as ground truth and UAV imagery as a use case will provide the confidence required to let other researchers or biologists classify and identify these native species in other multispectral imagery.

Socio-Economic Impact Assessment: Understanding and quantifying the socio-economic impact of this vegetation monitoring method on indigenous communities.

Year 1

Literature review, technical and methodological training, and establishing data collection protocols, including UAV flight plans and fieldwork strategies; this part will include significant collaboration with the co-supervisor expert on vegetation classification. Apply for necessary permits for field data collection in Colombia. Setting up sampling plots and collecting baseline spectral data for initial spectral signature library creation. Conduct UAV flights in selected test areas to capture preliminary multi-spectral imagery, focusing on distinct vegetation types. Data processing and pre-analysis. First National Conference Presentation

Year 2

Data fusion methodology design. Develop the framework for fusing UAV multi-spectral imagery with field spectrometer data. Outline initial models to combine spatial and spectral datasets. Run preliminary tests on data fusion algorithms, iterating based on performance in initial test plots and refining spectral signature library as necessary. Evaluation of gathering ground truth data from test areas to enhance model accuracy and compare with UAV-derived classifications. Adjusting for environmental variables (e.g., light changes). The initial version of the open-access spectral signature bank and libraries for automatic classification using machine learning models. First International Conference presentation.

Year 3

Apply refined data fusion methods across larger test areas, focusing on high-biodiversity zones and integrating multi-season data. Update the spectral library and refine fusion methods to improve data accuracy across diverse ecosystems. Begin collecting data on Indigenous community use of native species, gathering relevant socio-economic data with local institutions. Analyse the social and economic impacts of native species monitoring for local communities, linking findings with ecological data. Second National Conference presentation.

Year 3.5

Dissertation write-up and finalising the research findings.
Dissemination of research through academic publications (excepted at least 3), two national and one international conference presentations across the doctoral period, and knowledge sharing with relevant stakeholders in both Colombia and the UK.

In this project, core methodologies will focus on UAV data integration, spectral signature analysis, and spatial data fusion techniques to classify and monitor native vegetation. These methodologies will require extensive programming skills, ideally in Python or R, to process multispectral imagery, develop data fusion models, and create a comprehensive open-access spectral signature library. Geospatial analysis and remote sensing expertise will be essential to visualize and interpret research findings effectively.

This framework will leverage expertise from the University of St Andrews, the University of Durham, the University of Newcastle and the IAPETUS2 training network, providing foundational and advanced training in UAV data analysis, geospatial data integration, and ecological monitoring. Participation in internationally recognized summer schools focused on remote sensing, UAV applications, and spatial analysis will further strengthen the candidate’s proficiency in these cutting-edge techniques.

Further training in key transferable skills is available through the International Education and Lifelong Learning Institute (IELLI) at the University of St Andrews. This institute provides joined-up learning and teaching support by combining educational development, pedagogical workshops, courses in oral and written communication, and opportunities for media and public engagement.