3.4. Change Detection Metrics
The annual change detection metrics were designed to highlight inter-annual changes of spectral reflectance while reducing false detections due to reflectance fluctuations and inconsistent clear-sky observation availability. Metrics represent a set of statistics extracted from (a) the normalized surface reflectance time-series of the current and preceding years, independently, (b) the array of differences in 16-day composite values between the current and preceding years, and (c) the entire time interval. These statistics are stored as a set of 16-bit unsigned raster files in the same tile system as 16-day composites (see Landsat ARD Structure for details).
The GLAD system is designed to provide different sets of change detection metrics for various applications. Some of the metric sets are designed specifically to highlight the changes that happened in the last year, while others are sensitive to changes that happened within a longer interval. The latter has higher sensitivity for change detection but should only be implemented when the mask of earlier detected changes is available. Based on our ongoing research projects and feedback from the user community, our team will develop new metric types. We will add new metric types to this section and provide the corresponding codes for metric generation and application. Data users are encouraged to request new types of metrics for their specific applications (contact Peter Potapov firstname.lastname@example.org). Each metric set has a specific letter code. Before using a new metric set, do the following:
- Read metrics methodology and understand naming convention.
- Make sure that the computer resources (disk volume, RAM) are suitable for the selected metric type.
- Download 16-day ARD data following specific requirements.
- Download and install all the required codes.
Supported metric types:
Change detection type A (change_A)
Purpose: Annual forest loss mapping. May only be applicable when change data for earlier years exists.
ARD requirement: Four years of data are recommended, at least 2 years required.
Computer requirement: At least 13GB RAM required for metric calculation. The metrics dataset size for one tile / one year is 9GB.
Change detection type B (change_B)
Purpose: Annual forest cover loss mapping. Specifically designed for low capacity PC.
ARD requirement: Two years of data are required (current and preceding).
Computer requirement: At least 8GB RAM required for metric calculation. The metrics dataset size for one tile / one year is 2.2GB.