B230 - Development of image analysis techniques to enable tropical rainforests to be monitored from aircraft

Lead Research Organisation: University of Cambridge
Department Name: Plant Sciences


I wish to undertake the project entitled "Development of image analysis techniques to enable tropical rainforests to be monitored from aircraft". I wish to
take part in research more generally because I found this to be the most enjoyable part of my course and I am hoping to take further steps in research, in
particular over a longer period. I particularly like this project as it combines both my strengths and passions.This comes for the fact that a large part of the
work in the project will be in image analysis and development of an end-user interface to further the research. This appeals to my strengths from
mathematics as my original degree, as well as the data processing and reproducible research pipeline that I worked on in my main research project. Not
only this, but the project has a focus on conservationism, which is something that I am passionate about and will really help me to stay motivated and
engaged with the project. I also really like the fact that the project has the scope for me to get involved with the data collection process. Whilst I don't have
experience with UAVs myself, this is always something I've been interested in getting involved with. I will certainly be very keen to be pro-active in this
aspect of the project and this is a strong factor in attracting me to the project.


10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
NE/N008952/1 01/10/2016 30/03/2021
1799562 Studentship NE/N008952/1 01/10/2016 31/03/2021 Jonathan Williams
Description We have shown the utility of UAVs for tracking tropical rainforest recovery. In particular we have shown it is possible to automatically detect key tree species for the recovery process from a drone. We have also looked at how extreme climate events affect recovering rainforests, showing that young raniforest is particularly badly hit, but is also able to rebound strongly and quickly once the event is over.
Exploitation Route The ability to map trees with UAVs will allow others to map and monitor the trajectories of tropical forests for restoration. Similarly, understanding how severe weather events affect young recovering forest will enable better modelling and mitigation to protect these precious ecosystems.
Sectors Environment

Title Multi-Class Graph Cut (MCGC) for individual tree detection 
Description We have developed a novel approach to detecting individual trees form typical three-dimensional remote sensing datasets. This approach applies methods from computer vision and mathematics to automatically produce 3D models of each tree in the dataset. The results of this algorithm are in a prepared manuscript to be very shortly submitted as a journal article. This algorithm was also included in a benchmarking study as set out under my collaborations. Once we submit the article introducing this method we will also share the code for this algorithm publicly for the use of the remote sensing community. As noted, this means the output is not yet currently openly accessible, but will be in the coming months once introduced by means of a journal article 
Type Of Material Computer model/algorithm 
Year Produced 2019 
Provided To Others? No  
Impact At present the work is in-house software. However, we have included this in a benchmarking study which is expected to raise the profile of such approaches in tropical ecology. We hope this involvement, combined with publishing our manuscript, will lead to increased use of remote sensing data in assessing change in tropical forests. 
Title UAV data collected at Harapan Rainforest 
Description In 2008, Burung Indonesia (BI), the Royal Society for the Protection of Birds (RSPB) and BirdLife International (BirdLife) received a 100-year license to manage the Hutan Harapan lowland rainforest restoration project in central Sumatra. This project contains just short of 100,000 ha of previously logged forest. As part of my funded project I have over seen 3 months of data collection at Hutan Harapan. Here multiple UAVs and cameras have been used to collect detailed aerial imagery over the project at Hutan Harapan Rainforest. These data can then be used to build models of the forest which can be used for subsequent change and ecologicial analyses. We also collected field data on the locations of approximately 400 trees of 4 key species to help build and validate models and algorithms for automatically detecting these species from UAV data. The collected data is currently being processed to produce 3D models and detailed geo-referenced maps of the areas that were flown. These will then be held in a data repository to be shared to facilitate future studies on the restoration process on-going at Harapan. We haven't yet hared the raw data as this consists of many tens of thousands of geo-referenced images which need to be processed to correct for lens effects, exposure effects and similar issues. We do intend to share the processed data and would be willing to arrange sharing of the raw images for regions if partners were interested in these data. 
Type Of Material Database/Collection of data 
Year Produced 2018 
Provided To Others? No  
Impact We expect this dataset to allow larger scale and higher resolution maps of restoration progress at Harapan than has so far been afforded by manual survey or satellite imagery respectively. The data remains in the processing phase so the analysis that will give rise to these impacts is as of now incomplete. 
Title UAV image analysis for indicative species mapping in tropical forest recovery 
Description My latest work has developed a pipeline for image analysis using UAV data. This pipeline enables mapping and the automated detection across management units of key species indicative of forest degradation and recovery in tropical forest. The software for this is currently in the final stages of preparation and will be made publicly available once we submit the associated manuscript for this work (which is itself in preparation). 
Type Of Material Computer model/algorithm 
Year Produced 2020 
Provided To Others? No  
Impact None yet, but we intend to submit this methodology as a methods paper and publish both the code and dataset alongside this. 
Description A comparative assessment of the performance of Individual Tree Crowns delineation algorithms from ALS data in tropical forests 
Organisation University of Montpellier
Country France 
Sector Academic/University 
PI Contribution We have worked with a team from the University of Montpellier on a benchmarking study of algorithms to detect trees in remote sensing data from Tropical Forests. We submitted our results from the work arising out of our project, applying our approach to the benchmark data. The complete study is now in the process of preparation for submission as a journal article.
Collaborator Contribution The partners collected the remote sensing data as well as the field collected data used to assess the input from the project partners. The partners have built the suite of software to assess the results form all methods submitted to the benchmark. The partners facilitated the benchmark and are also the primary authors of the resulting manuscript in development. We have worked with the partners to prepare the manuscript, but as we note, the partners are the main driving-force of the project. Other researchers also worked with the data for this project and submitted their results to the study, but the collaboration primarily worked through multiple bi-lateral partnerships with the organisers at Montpellier.
Impact This project is in the process of being written up into a manuscript for submission as a journal article. This will be the primary output from the project. The primary fields of this project are forest ecology and remote sensing though each individual partner who submitted their results to the benchmark may work across related fields.
Start Year 2017
Description Partnership for fieldwork in Indonesia 
Organisation Jambi University
Country Indonesia 
Sector Academic/University 
PI Contribution I partnered with the University of Jambi (UNJA) to plan and facilitate my field data collection in Indonesia in 2018. Here I oversaw and completed the design of and actual collection of data. I provided the equipment and ideas for the data collection, as it was to be used primarily for my PhD.
Collaborator Contribution UNJA regularly met with me and helped me to plan my fieldwork design. On top of this I am still in communication with the team at UNJA to further advise on my project and using the data we collected. The team at UNJA also provided a lot of help and advice on the steps and paperwork for me to get a research visa for the work.
Impact Here I was able to collect the data from the project at Hutan Harapan as laid out in my detailed outcome on the data.
Start Year 2018
Title Multi-Class Graph Cut (MCGC) for individual tree detection 
Description This is the software for detecting individual trees from remotely-sensed data we developed as part of our new method as laid out in the methods section of our reporting. This has been cleaned and simplified to make a user friendly package to use in MATLAB. Once we submit our manuscript we intend to publish the full code-base in a publicly available repository at which point we will adopt and appropriate license. it is expected that this software will be made avaailable for use on the condition that attribution is given to our project members and reference is made to the article introducing our work. 
Type Of Technology Software 
Year Produced 2019 
Impact As of yet the software is only available to our group or when shared on request. Once shared we hope this software will be used in project working with tropical forests as will be justified by the results in our manuscript and the benchmarking study we have been partners to. 
Description Presenting at Science Festival 
Form Of Engagement Activity Participation in an open day or visit at my research institution
Part Of Official Scheme? No
Geographic Reach Regional
Primary Audience Public/other audiences
Results and Impact We took part in the annual University of Cambridge science festival. Here we prepared a stall on the work of my research group. The focus was on palm oil and it's implications for conservation. I contributed remote sensing data collected with my UAV showing the encroachment of oil palm on tropical forest. Here we were show-casing the sort of data we are able to collect with the technologies I am working on and we also explained about how these data can be used to help mitigate effects from oil palm encroachment.
Year(s) Of Engagement Activity 2018