Per-Pixel Forest Attribute Mapping and Error Estimation: The Google Earth Engine and R dataDriven Tool

Area of Science:

• Remote Sensing
• Forestry
• Geospatial Analysis

Background:

• Remote sensing products often lack per-pixel accuracy assessments, limiting their utility.
• Significant accuracy variations exist across different map areas and classes.
• Estimating per-pixel uncertainty is crucial for improving remote sensing data usability.

Purpose of the Study:

• Introduce the dataDriven open-access tool for estimating per-pixel uncertainty in remote sensing maps.
• Address the challenge of variable accuracy in remote sensing products.
• Enhance the usability and potential of remote sensing data through uncertainty estimation.

Main Methods:

• Utilized a statistical design-based approach with bootstrap resampling.
• Leveraged Sentinel-2 remote sensing data and Google Earth Engine.
• Applied the R programming language for data analysis.

Main Results:

• Tested the dataDriven tool in a forest estate in Tuscany, Italy, focusing on volume density.
• Estimated pixel errors for volume density ranged from 93 to 979 m³ per hectare, averaging 285 m³ per hectare.
• Demonstrated the tool's capability to provide spatially explicit error estimates.

Conclusions:

• The ability to produce per-pixel error estimates is novel for remote sensing and forest monitoring.
• The dataDriven tool supports forest management by identifying reliable and unreliable map areas.
• Aims to facilitate the spatially exhaustive use and validation of remote sensing-derived products.