Automated Forest Restoration

Although forest restoration science has advanced considerably in recent years, the technologies used to perform restoration tasks remain prehistoric. Typical restoration projects involve large numbers of people, carrying baskets of seedlings, equipment and materials, often over long distances, across rough, steep terrain to remote restoration sites. Weeds are slashed with machetes and planting holes dug with hoes, in much the same way as our iron-age ancestors would have done.

Lack of access is often a problem. Flat sites, close to roads, are already occupied with agriculture, so they are not available for forest restoration. So, most restoration sites are remote, often on steep slopes with infertile soils. Hauling trees, materials and equipment into such sites, and returning to carry out weeding, fertilizer application and monitoring, are extremely laborious and stakeholders are often reluctant to commit to such arduous work. Automating restor-ation tasks would, therefore, make forest restoration, more feasible, especially on large scales and in remote areas.

Recent advances in technologies, such as unmanned aerial vehicles (drones) and imaging, are now making automation of some forest restoration task possible such as: -

• Pre-restoration site assessments and post restoration monitoring - both plants and animals 
• Locating seed trees and seed collection 
• Aerial seeding by drone, to replace tree planting 
• Automated maintenance - weeding and fertilizer application

Aerial surveys, to assess site conditions and plan restoration interventions, and to track tree growth, and the return of forest cover, can now be done routinely, using off-the-shelf consumer drones. These 4 videos provide a very basic step-by-step guide - how to start using a DJI Phantom drone, to obtain an overview of a restoration site. They are aimed at first timers - student researchers getting started with using a drone to collect project data.

3D forest models can be made using a consumer drone and its on-board RBG camera. The heights of trees in such models can be measured - so you can calculate growth rates over repeated flights. You can also see how forest structure changes as restoration progresses.

The drone must be flown at a constant height above the ground (using Litchi Flight Planner) in a grid that achieves 80% overlap between adjacent photos. Pix4D (free version) can then be used to construct a 3D model from the flat images using parallax. Every pixel in the model is tagged with its altitude (above sea level). Tree-height can then be calculated by subtracting the difference in altitude, between the tree top and the drone, from the constant flight-path height of the drone above the ground (as set in Litchi) - as explained in detail in the slide-show below:

Here's an example of a 3D model of restored forest vs. non-restored control plot. 

This is a simulated "fly-over" of a model of forest restored by the framework species method, planted in 1998, aged 24 years. The model was constructed using Pix4D from 270 photos, overlapping by 80%, captured during autonomous flight, at a constant 70 m above ground level (controlled by Litchi flight planning software), over the BMSM98.2 plot on July 26th 2022. The model shows dense canopy cover, with highly developed structure, over the area planted with framework tree species and distinct "canopy shyness". The grassy patch at the far end, where the fly-over reverses direction is the CONTROL plot, where no tree planting occurred. It has undergone poor natural regeneration. Even after 24 years, it remains dominated by Thysanolaena grass, bracken fern and introduced exotic herbaceous weeds such as Eupatorium and Mikania. The few trees establishing there come mostly from seeds of trees planted in the adjacent plot. The control plot is fire prove (it burnt twice since the experiment started) due to weeds providing fire fuel, and it remains unattractive to seed-dispersing animals. This probably accounts for the persistent lack of tree cover there. The drone flight was included in a short training workshop for Bangladeshi forest officers, organized by AIT, and hosted by FORRU-CMU, 25-26th July 2022.

Locating seed trees , in photographs taken by drones, can be done by eye for distinctive species. Current artificial intelligence (AI) systems struggle to do it autonomously. However such systems are rapidly advancing and we expect AI-identification of target tree species to become more reliable and accessible in the next few years.

Seed collection by drone is not yet feasible. Technologies that combine computerized vision systems with robot arms are already used to harvest fruits in horticulture, but the technology has not yet progressed to robot arms on drones, capable of collecting fruits from forest trees.

Aerial seeding by drone, as a replacement for conventional tree-planting, is already being practiced. Several commercial companies now offer this service: 

• Drone Seed
• Dendra
• Flash Forestry
• Dronecoria 

In 2015, FORRU-CMU ran a workshop on Automated Forest Restoration, which brought together experts in forest ecology, seed technology, aeronautics, computer science etc. to brainstorm how new technologies might make forest restoration more practicable. The result - 14 original papers on how technologies could help with site surveys, seed collection, aerial seeding, weeding, monitoring etc. and a research agenda to inspire graduate students. Click on the book to find out more about how technologies could help you implement forest restoration or get ideas for a MSc/PhD project.

The research agenda was presented at the international conference: Reforestation for Biodiversity, Carbon Capture and Livelihoods - an online event, organized by the UK's Royal Botanic Gardens, Kew in February 2021.