Land change and carbon dynamics in the Colombian Amazon

Tropical deforestation is a significant source of CO2 emissions to the atmosphere. Quantifying land use changes and associated emissions is critical for reporting and reducing emissions of greenhouse gases. In the Colombian Amazon, areas of forest conversion estimated at biennial intervals using a combination of dense time series of Landsat observations and statistical estimators based on reference data indicate that deforestation is modest (87 kha year-1) relative to surrounding countries and regions. Other land cover and change areas can also be estimated at biennial intervals, including a land cover class representing regrowing secondary forest, which is on average five times larger than the forest-to-pasture conversion. Areas of gain and loss of secondary forest are very small for this region relative to deforestation. Errors in the detection of change negatively impact the precision of the land change area estimates. New methods estimate the uncertainty associated with maps of land change, represented as probability maps of omission and commission of change. These probabilities are higher in the deforestation frontier of the study area, where the fine spatial scale of the disturbances and the low temporal data density make it challenging to detect the changes accurately. The presented methods improve our ability to integrate uncertainty into applications that make use land change maps, such as spatial carbon models. Methods to estimate emissions based on bias-adjusted areas of land change show that net carbon emissions average 10 Tg year-1 (0.22 Mg ha-1 year-1) in the entire study area, and can be further disaggregated by the land cover contributing to the emissions or removals. This dissertation shows that the conversion from forest to pastures has been the largest forest loss pathway in the Colombian Amazon for almost two decades. While there is a small carbon offset due to sequestration by regrowing forests, conversion to pasture is also the main source of carbon emissions associated with land change. The methods and results presented in this dissertation demonstrate the potential of the Landsat archive to enable the quantification of land changes, their uncertainty, and their associated carbon emissions, even in areas with relatively infrequent cloud-free observations