Share | 10/13/2020
The widespread of pests like bark beetle can cause catastrophic damage in forests, thus the importance of its early detection. Manual surveys are costly and limited in the number of acres that can be covered making it hard for foresters to detect and remove the infected trees on time. The limitations of traditional practices call for better ways to identify the infestations so foresters can act quicker and prevent irreversible damages.
Remote sensing methods, from satellite to drones, offer an easier and quicker way to map and monitor large forests. However, the precision of these applications and the amount of detail obtained in the outputs depend on the sensors used.
RGB sensors are the most common, providing context and a visual overview of the forest canopy, and depending on its spatial resolution, can also be used for stand count. However, its low-spectral resolution limits the phenomena that can be detected, since many early symptoms of pest infestation or disease can only be detected with multispectral sensors that can capture plant reflectance in both the visible and invisible bands of the light spectrum.
SKYLAB is a German company dedicated to the analysis of aerial data using machine learning algorithms and convolutional neural networks to provide precision forestry maps throughout the growth cycle – from seedling survival and weed mapping to timber and carbon stock inventories and change mapping.
Their customers and partners work with a wide range of sensors, so SKYLAB methodology is designed to process and analyze data from off-the-shelf DJI drones to customized drones with high-resolution RGB and multispectral cameras, plane-based photogrammetry and LiDAR, and satellite imagery.
Their experience analyzing all kinds of data have given them a better idea about the benefits and limitations of both RGB and multispectral sensors for forestry mapping. “The use of multispectral data provides an extra level of information which can be vital in some more difficult settings. For example, for differentiation of seedlings in areas of very high weed density and diversity or for more precise species differentiation in a mixed forest.”
On other applications like bark beetle detection, multispectral cameras like the MicaSense series RedEdge can aid in the detection of infection in spruce during the critical 6 weeks of active infection, when the trees are still green and don’t show any visual signs of discoloration. High-resolution RGB cameras, on the other hand, can also help detect infected trees by identifying a slight discoloration on the crown of trees that is not visible from the ground.
“The challenge of using multispectral cameras in the field is that it requires more experience and attention to detail to make sure calibrations and flight settings are optimal. If done incorrectly, the data quality can suffer.”
SKYLAB’s analysis process starts with the creation of orthomosaics and 3D point-cloud maps using standard software like Agisoft or Pix4D. The following analysis is done with SKYLAB’s proprietary methodologies and algorithms, including advanced machine learning in convolutional neural networks. The typical turnaround for projects varies, from 24 hours for standard seedling survival mapping to 4-6 weeks for a fully comprehensive timber stock inventory for a large forestry estate. The amount of detail provided will depend on the quality of the captured imagery.
Montes del Plata is a Uruguayan industrial-forestry company dedicated to the production of eucalyptus cellulose pulp. Their headquarters are located near the town of Conchillas in the department of Colonia, while forest plantations are spread throughout the country, which makes it challenging for them to supervise all locations. Thus, Montes del Plata hires third-party companies for seeding and relies on drone mapping companies to audit the process.
One of their fields located in Durazno was flown with a MicaSense series RedEdge mounted on a fixed-wing drone from Foto Aérea, 90 days after seeding to determine the number of trees planted and to detect weeds. This information is important since seeding companies only get paid if the percentage of tree survival is above 95%.
In this particular case, the multispectral data collected with the MicaSense series RedEdge and SKYLAB’s proprietary algorithms provided Montes del Plata with precise information on tree count and tree health.
The use of aerial imagery can save foresters time and money, according to a research run by SKYLAB the cost of the terrestrial survey, with a search area of 30 to 40 ha per day, lies between €5 and €40 per hectare, with a rather low success rate of 30 to 65%. By comparison, drones can cover up to 1,000ha per day, aircraft can fly 10,000ha and satellites can produce images over hundreds of thousands of hectares per day.
By using aerial data the terrestrial disease search can be carried out much more efficiently, with a time-saving of about 70%. Compared with a purely terrestrial search, the ground coverage area is significantly increased and it is likely that more infestation hotspots can be identified. For tasks like seedling survival rate monitoring, terrestrial monitoring is no longer necessary.
In addition to time savings and increased accuracy, using remote sensing can lead to significant cost savings. Satellite-based analysis can be done for as little as €0.25 per ha, but even drone-based, high-resolution analysis can start at €2 per ha.
Like the cost, the needs for specific data vary as well. Satellite imagery can be enough to map deforestation, a simple high-resolution RGB camera can cover applications like stand count or even some stress detection, but for early detection multispectral data is key and can also add information needed for precise seeding and weed mapping, species differentiation, and stress and disease analysis.
For more information on forestry applications, download our forestry whitepaper.
Sign up to receive updates directly to your inbox.
