Fusing Landsat and SAR time series to detect deforestation in the tropics

Résumé

Fusion of optical and SAR time series imagery has the potential to improve forest monitoring in tropical regions, where cloud cover limits optical satellite time series observations. We present a novel pixel-based Multi-sensor Time-series correlation and Fusion approach (MulTiFuse) that exploits the full observation density of optical and SAR time series. We model the relationship of two overlapping univariate time series using an optimized weighted correlation. The resulting optimized regression model is used to predict and fuse two time series. Using the MulTiFuse approach we fused Landsat NDVI and ALOS PALSAR L-band backscatter time series. We subsequently used the fused time series in a multi-sensor change detection framework to detect deforestation between 01/2008 - 09/2010 at a managed forest plantation in the tropics (Pinus caribea; 2859 ha). 3-monthly reference data covering the entire study area was used to validate and assess spatial and temporal accuracy. We tested the impact of persistent cloud cover by increasing the per-pixel missing data percentage of the NDVI time series stepwise from ~ 53% (~ 6 observations/year) up to 95% (~ 0.5 observation/year) while fusing with a consistent PALSAR time series of ~ 2 observations/year. A significant linear correlation was found between the Landsat NDVI and ALOS PALSAR L-band SAR time series observables for logged forest. The multi-temporal filtered PALSAR HVHH backscatter ratio time series (HVHHmt) was most strongly correlated with the NDVI time series. While for Landsat-only the spatial and temporal accuracy of detected deforestation decreased significantly with increasing missing data, the accuracies for the fused NDVI-PALSAR case remained high and were observed to be above the NDVI- and PALSAR-only cases for all missing data percentages. For the fused NDVI-HVHHmt time series the overall accuracy was 95.5% with a 1.59 month mean time lag of detected changes. The MulTiFuse approach is robust and automated, and it provides the opportunity to use the upcoming data streams of free-of charge, medium resolution optical and SAR satellite imagery in a beneficial way for improved tropical forest monitoring.