Keywords
Land cover change, satellite remote sensing, machine learning, cloud computing
Start Date
15-9-2020 6:20 PM
End Date
15-9-2020 6:40 PM
Abstract
Satellite remote sensing is an essential source of data on forest landscape change that is necessary for estimating carbon stocks, assessing biodiversity, and developing conservation plans to sustain ecosystem services. The Upper Guinean Forest region of West Africa, a global biodiversity hotspot, currently lacks timely, consistent, and accurate long-term forest monitoring systems suitable for decision support. Major limitations include data gaps due to frequent cloud cover and limited satellite coverage. To address these limitations, we developed the West African Forest Degradation Data System (WAForDD) to use all available Landsat imagery to monitor forest change in the tropical forest zone of Ghana. The approach used a novel combination of spectral mixture analysis (SMA), machine learning classification, and time-series modelling implemented on the Google Earth Engine (GEE) cloud computing platform. Outputs included annual maps of canopy cover for 2000-2019 at 30 m resolution along with annual change estimates of forest loss, degradation, and recovery. We found that random forest predictions of canopy cover based on SMA fractions resulted in more accurate predictions than those based on other types of spectral indices. Postprocessing the canopy cover maps with the LandTrendr algorithm for segmented time series regression further improved map accuracy. The final maps had an observed-predicted correlation of 0.87 and a mean absolute error of 12.7%. Examination of change trajectories found that decreases in canopy cover coincided with known historical disturbances such as wildfires, logging, land clearing for agriculture, and mining activities. By leveraging the GEE platform and co-developing the software with partners in Ghana, we were able to help the Forestry Commission of Ghana meet their measurement, reporting, and verification (MRV) needs for REDD+ implementation and ensure that use of the software can be sustained into the future.
Software for Monitoring Forest Change in Tropical West Africa Using Satellite Remote Sensing
Satellite remote sensing is an essential source of data on forest landscape change that is necessary for estimating carbon stocks, assessing biodiversity, and developing conservation plans to sustain ecosystem services. The Upper Guinean Forest region of West Africa, a global biodiversity hotspot, currently lacks timely, consistent, and accurate long-term forest monitoring systems suitable for decision support. Major limitations include data gaps due to frequent cloud cover and limited satellite coverage. To address these limitations, we developed the West African Forest Degradation Data System (WAForDD) to use all available Landsat imagery to monitor forest change in the tropical forest zone of Ghana. The approach used a novel combination of spectral mixture analysis (SMA), machine learning classification, and time-series modelling implemented on the Google Earth Engine (GEE) cloud computing platform. Outputs included annual maps of canopy cover for 2000-2019 at 30 m resolution along with annual change estimates of forest loss, degradation, and recovery. We found that random forest predictions of canopy cover based on SMA fractions resulted in more accurate predictions than those based on other types of spectral indices. Postprocessing the canopy cover maps with the LandTrendr algorithm for segmented time series regression further improved map accuracy. The final maps had an observed-predicted correlation of 0.87 and a mean absolute error of 12.7%. Examination of change trajectories found that decreases in canopy cover coincided with known historical disturbances such as wildfires, logging, land clearing for agriculture, and mining activities. By leveraging the GEE platform and co-developing the software with partners in Ghana, we were able to help the Forestry Commission of Ghana meet their measurement, reporting, and verification (MRV) needs for REDD+ implementation and ensure that use of the software can be sustained into the future.
Stream and Session
false