Drone ecology: using UAVs to monitor and map forest phenology
Lead Research Organisation:
Imperial College London
Department Name: Life Sciences
Abstract
Phenology is a key driver of ecosystem processes in tropical forests. The timing of leaf flush, flowering and fruiting events can drive the movements of animals (Curran and Leighton 2000), and may potentially explain observations dating back as far as Alfred Russell Wallace (1869) as to why tropical rainforests sometimes appear deserted and at other times are buzzing with life. In this project, we will combine aerial observation of a forest canopy with field measurements of invertebrate biomass to test this hypothesis, and use our results to examine the likely consequences of forest fragmentation on biodiversity. This project represents a novel extension for the use of drones in tropical ecology and conservation (Koh and Wich 2012), and fieldwork will be conducted at the site of the SAFE Project in Sabah, Malaysia (Ewers et al. 2011).
We will develop and implement an aerial UAV platform for capturing light and/or hyperspectral photographs of a forest canopy, and use commercially available software to produce geo-referenced, 3-dimensional photo mosaics for analysis. We will develop image analysis techniques to identify and map the boundaries of individual tree canopies, to check and align the spatial location of tree canopies among photo mosaics taken at different times, and to detect systematic changes in the colour composition of tree canopies that could indicate phenological events. Phenological events will, in turn, be automatically categorised into leaf flush, fruiting and where possible flowering events, with tree mortality and/or treefall events also likely to be identified and mapped. Phenology across space and time will be monitored and mapped through the implementation of routine, bi-weekly UAV flights across standard flight paths. We will rely heavily on statistical methods originally developed to determine extinction dates of species (Roberts and Solow 2003), and more recently extended to determine the timing of the onset and cessation of phenological events (William Pearse, pers. comm.), to interpolate our fixed time observations into continuous date estimates of phenological timings.
To determine the link between tree phenology and invertebrate movements, we will establish grids of activity-based invertebrate traps in the forest canopy. Traps will be monitored bi-weekly for at least one 12 month period, with collected invertebrates dried and weighed to determine invertebrate biomass. For each tree in which there is a trap, we will correlate the time series of invertebrate biomass in that tree with the binary events of presence/absence of leaf flush and/or flowering of that tree. We will use time lags in the analysis to examine the effect of fruiting, which is expected to follow flowering in a predictable procession. If our hypothesis is correct, we will detect an increase in the biomass of canopy invertebrates during phenological events.
We will use our results to determine the scaling relationship between forest area and 'phenology density', represented as the proportion of area undergoing a phenological event at any given time. This relationship will be replicated in a spatially explicit simulation and used to determine the minimum size a forest fragment might need to be to ensure a continuous supply of the phenological events that underpin the supply of resources to invertebrate communities.
We will develop and implement an aerial UAV platform for capturing light and/or hyperspectral photographs of a forest canopy, and use commercially available software to produce geo-referenced, 3-dimensional photo mosaics for analysis. We will develop image analysis techniques to identify and map the boundaries of individual tree canopies, to check and align the spatial location of tree canopies among photo mosaics taken at different times, and to detect systematic changes in the colour composition of tree canopies that could indicate phenological events. Phenological events will, in turn, be automatically categorised into leaf flush, fruiting and where possible flowering events, with tree mortality and/or treefall events also likely to be identified and mapped. Phenology across space and time will be monitored and mapped through the implementation of routine, bi-weekly UAV flights across standard flight paths. We will rely heavily on statistical methods originally developed to determine extinction dates of species (Roberts and Solow 2003), and more recently extended to determine the timing of the onset and cessation of phenological events (William Pearse, pers. comm.), to interpolate our fixed time observations into continuous date estimates of phenological timings.
To determine the link between tree phenology and invertebrate movements, we will establish grids of activity-based invertebrate traps in the forest canopy. Traps will be monitored bi-weekly for at least one 12 month period, with collected invertebrates dried and weighed to determine invertebrate biomass. For each tree in which there is a trap, we will correlate the time series of invertebrate biomass in that tree with the binary events of presence/absence of leaf flush and/or flowering of that tree. We will use time lags in the analysis to examine the effect of fruiting, which is expected to follow flowering in a predictable procession. If our hypothesis is correct, we will detect an increase in the biomass of canopy invertebrates during phenological events.
We will use our results to determine the scaling relationship between forest area and 'phenology density', represented as the proportion of area undergoing a phenological event at any given time. This relationship will be replicated in a spatially explicit simulation and used to determine the minimum size a forest fragment might need to be to ensure a continuous supply of the phenological events that underpin the supply of resources to invertebrate communities.
Organisations
People |
ORCID iD |
| Robert Ewers (Primary Supervisor) | |
| Ross Gray (Student) |
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| NE/P012345/1 | 01/10/2017 | 30/09/2027 | |||
| 1945378 | Studentship | NE/P012345/1 | 01/10/2017 | 30/06/2021 | Ross Gray |
| Description | Tropical forest phenology links with biodiversity and ecosystem process but knowledge gaps remain surrounding the dynamics and drivers of all phenological types, its influence on invertebrate distribution, and how fragmentation may change phenological resources available to higher taxa. Over 1 year, we monitored a 16 ha area of logged tropical forest in Malaysian Borneo. Using UAV imagery within a machine learning pipeline we showed the potential of UAVs to identify individual tree crowns and track their phenology through time but highlighted limitations in classifying spectral outliers into phenological events. In parallel, we ground-truthed the UAV phenology data for 135 of these trees, representing 45 taxa and 17 plant families, demonstrating phenology was taxon specific and highly asynchronous across the landscape, with no one environmental factor determining phenological change. Furthermore, using a grid of canopy-set traps we monitored invertebrate biomass weekly, showing they maintain a low-level, patchy distribution across the landscape but are not influenced by phenology. Finally, in a 10-year spatially explicit simulation of forest fragmentation, we showed fragmentation significantly reduced and de-stabilised temporal phenological resource availability in a 7,200 ha area of forest. The asynchrony displayed by tropical forest phenology demonstrates it is a complex process to monitor and despite leveraging the high spatiotemporal resolution of UAV data and the power of machine learning, there are still challenges that need to be addressed. However, our detailed overview of forest phenology demonstrates it can be susceptible to fragmentation and future climatic change, and even though invertebrates were not affected here, this could have reverberating consequences for higher taxa and ecosystem function. Therefore, we would do well to successfully tackle the challenges of UAV monitoring in the tropics, to rapidly and automatedly assess phenological change and help understand the impacts of these environmental changes on phenology. |
| Exploitation Route | Further research would do well to successfully tackle the challenges of UAV monitoring in the tropics, to rapidly and automatedly assess phenological change and help understand the impacts of these environmental changes on phenology. |
| Sectors | Environment |
| Description | British Ecological Society Conference Presentation |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | Presentation centred around preliminary results of my phd research. This sparked questions and discussion with several colleagues from other scientific institutions, who found the talk and results really interesting. This should hopefully result in future collaborations with those colleagues with relation to my phd research. |
| Year(s) Of Engagement Activity | 2019 |