Propagation of errors associated with scaling foliage biomass from field measurements to remote sensing data over a northern Canadian national park

If a change detection result based on time series of remote sensing data indicates that there was a 10% increase in an ecosystem property between two years over a specific land area, does it mean there was a real change in the ecosystem property, or could it be merely an estimation error? This question must be addressed before ecosystem managers or policy makers can use the result with confidence for addressing related environmental or natural resource management issues. One means of answering this question is through systematic error propagation analysis. In this study, we analyzed error propagation for detecting inter-annual changes in foliage biomass over Wapusk National Park, Canada. Specifically, we first estimated uncertainties in all input data, including sampling errors in foliage and random errors in AVHRR and Landsat data. Secondly, we evaluated the error propagation from inputs to the remote sensing-derived foliage biomass estimates (including the Landsat-based foliage biomass, AVHRR-derived foliage biomass, and the inter-annual changes in foliage biomass), and determined the threshold of detectable change in foliage biomass. Finally, we investigated approaches that can reduce the threshold. Our results indicated that over Wapusk National Park during 1985-2006, the threshold for a clear-sky AVHRR pixel between two single years was ~ 40% with a confidence level of 84%, and can be reduced to 10% for a land cover class with more than 10 clear-sky AVHRR pixels between two 5-year “State of Park” reporting periods.

► Comprehensive error propagation analysis for remotely sensed foliage biomass changes ► Detectable foliage biomass change is quantified by standard estimation error. ► Substantial estimation error is reduced through temporal and spatial averaging.