Two Novel Cloud-Masking Algorithms Tested in a Tropical Forest Setting Using High-Resolution NICFI-Planet Basemaps

Area of Science:

• Remote Sensing
• Environmental Monitoring
• Geographic Information Systems

Background:

• High-resolution satellite imagery, such as the NICFI-Planet collection on Google Earth Engine (GEE), is crucial for fine-scale tropical forest monitoring.
• Persistent cloud cover, shadows, and haze in satellite data significantly reduce its utility for accurate environmental analysis.
• Existing cloud-masking methods may not be optimized for specific datasets like NICFI-Planet or may require extensive labeled data.

Purpose of the Study:

• To develop and present two simple, fully reproducible cloud-masking algorithms tailored for the NICFI-Planet image collection on Google Earth Engine.
• To evaluate the performance and applicability of these algorithms across diverse tropical deltaic mangrove environments.
• To provide operational tools that can be readily applied without the need for labeled samples.

Main Methods:

• Algorithm A: A thresholding approach utilizing Blue and Near-Infrared spectral bands.
• Algorithm B: A Sentinel-2-derived statistical thresholding method with per-band cutoffs.
• Both algorithms were implemented end-to-end within Google Earth Engine (GEE) for operational use and tested in the Sundarbans, Bidyadhari Delta, and Ayeyarwady Delta.

Main Results:

• Algorithm B effectively removed the most cloud and bright-water pixels but showed a tendency to over-mask haze and low-contrast features.
• Algorithm A retained more usable pixels, though its performance varied regionally, being more conservative in the Sundarbans and over-inclusive in India and Myanmar.
• Both algorithms offer a practical alternative to methods requiring labeled data, with GEE implementation ensuring reproducibility and consistency.

Conclusions:

• The developed cloud-masking algorithms provide a pragmatic, first-pass filtering solution for NICFI-Planet data in tropical forest monitoring.
• The algorithms' effectiveness can vary across different environmental conditions, necessitating careful consideration of regional characteristics.
• Their simple, shareable GEE code facilitates consistent application across diverse geographic regions, enhancing the usability of high-resolution satellite imagery.