Background and Motivation
CODED is an algorithm developed to monitor for forest conversion and degradation using time series analysis of Landsat data. The algorithm is based upon previous developments in continuous land cover monitoring [1] and tropical degradation mapping using spectral unmixing models [2] and is built upon the Google Earth Engine processing and data storage system. The algorithm was originally implemented in Python but required large data storage and computing resources for processing the vast amounts of data. Therefore, CODED was ported to the Javascript language for easier use over large areas.
The motivation behind the CODED project was to develop a methodology that could successfully identify low-magnitude forest disturbances in addition to differentiate between deforestation and degradation. While recent scientific and technological advances have greatly improved our ability to map high-magnitude forest clearings [3], no such methodology has proven as reliable for identifying degradation [4]. Despite this, degradation has been estimated to account for 40-212% the carbon emissions of deforestation [5-7]. Degradation events can often be small, subtle, and temporary, making them tricky to identifying with traditional approaches to change detection [8]. CODED attempts to overcome some of these limitations through sub-pixel spectral mixture analysis performed continuously through time. Timeseries analysis allows for subtle changes to be differentiated from ephemeral variations due to clouds or cloud shadows. Multi-temporal information is then used for land cover classification. The result is an approach that has proven successful at being able to differentiate disturbances that do not result in a change in land cover, or degradation, from forest conversion.
While CODED is able to produce disturbance maps that can be used for a variety of purposes, the subtle and temporary nature of degradation events means that errors of omission can potentially be large. Therefore, CODED has been shown to be most successful in the past for stratifying a landscape, which can then be used to derive sample-based estimates of area of degradation.