Page tree

    Loading...


Vegetation Height Estimation

Download Product Sheet


Average vegetation height for border zone of electrical power lines in Finland using Worldview-2&3 (0.5m) (Source: GMV)

Product Category

  • Land use
  • Natural Disaster
  • Climate Change
  • Coast Management 
  • Marine

Financial Domains

  • Investment management 
  • Insurance management 
  • Green finance

User requirements 

UN37: Projection of risk to portfolio assets into future

Description

Vegetation height estimation is important for many aspects of the financial management sector.  In sectors like construction and infrastructure development, accurate vegetation height estimation is vital for assessing potential risks related to buildings, power lines, and transportation projects. Overgrown vegetation near critical infrastructure can lead to safety hazards and increased maintenance costs. For industries focusing on renewable energy, such as solar and wind farms, knowing vegetation height is crucial. Tall vegetation around these facilities can obstruct sunlight and wind flow, affecting energy production and efficiency. In forestry and natural resource management, understanding vegetation height aids in estimating timber volume, planning harvest rotations, and assessing forest health. These factors directly influence revenue generation and sustainable resource utilization. Vegetation height can be estimated using VHR satellite images and machine learning models. To train these models, it is essential to have ground truth data such as in-situ or LIDAR data.

Spatial Coverage Target

Asset level

Data Throughput

Rapid tasking 

Data availability

  • High
  • Low

PRODUCT SPECIFICATIONS

Main processing steps

The vegetation height machine learning model can be a regression-based deep learning approach that utilizes a Convolutional Neural Network (CNN), particularly an auto-encoder architecture such as DenseNet, ResNet, and SENet. Its primary objective is to predict the height of vegetation in a high-resolution satellite image that contains RGB channels.  The model aims to create a canopy height map based on this single input image. When the ground truth data is LIDAR, the initial steps involve converting LIDAR point clouds into canopy height models. Then, the vegetation in the VHR satellite image is masked using vegetation indices and supervised machine learning models. The deep learning model is subsequently trained using VHR images timely aligned with the LIDAR data (ground truth data). After successful training and validation, the model can be deployed to estimate vegetation height in any desired image.

Input data sources

Optical: VHR based on the availability like Pleiades 1A/1B & NEO, WorldView2&3, and SPOT6/7

Radar: N.A

Satellite-based products: N.A

Supporting data:  Ground truth data such as LIDAR

Accessibility

Optical VHR imagery: commercially available on demand from EO service providers.

Spatial resolution

≤ 1m

Frequency (Temporal resolution)

Daily

Latency

Daily

Geographical scale coverage

Globally

Delivery/ output format

Data type: Raster

File format: GeoTIFF

Accuracies

Thematic accuracy: 80-85%

Spatial accuracy: 1.5-2 pixels of input data

Constraints and limitations

  • Lack of ground truth data (LIDAR)
  • The cost of the VHR satellite images
  • Cloud presence
  • The machine learning model is limited to regions with similar vegetation characteristics where it was trained.

User's level of knowledge and skills to extract information and perform further analysis on the EO products.

Skills: Essential

Knowledge: Essential


Edit


P19: Vegetation height estimation

Download the product sheet gap analysis 

Maturity score

Mean: 2.5

STD: 0.65

Constraints and limitations

·  Cloud presence

·  High cost of VHR satellite imagery

·  The machine learning models are limited to regions with similar vegetation characteristics where it was trained.

·  Uncertainty related to machine learning models

Relevant user needs

UN37: Projection of risk to portfolio assets into the future.

R&D gaps

·  Lack of time series ground truth data (Light Detection and Ranging (LIDAR))

Potential improvements drivers

·  Provide training datasets for different vegetation types over different regions in the world

Utilisation level review

Utilisation score

Mean: 2.40

STD: 0.49

No utilisation

Low utilisation

·  The use of the Global Ecosystem Dynamics Investigation (GEDI) sensor to assess carbon capture in standing/planted forests which are part of an offset mechanism.

·  Unawareness of the existence of commercial EO products with better specifications

Medium utilisation

Higher cost of using the best available commercial EO product

High utilisation

Critical gaps related to relevant user needs

Guideline gap

UN37: Projection of risk to portfolio assets into the future

Edit
  • No labels

This page has no comments.