On the Use of Landsat Data to Detect Long-Term NDVI Trends in Canadian Boreal Forest

Abstract:

Recent studies suggest the boreal forests of Canada are responding to climate change. Time series of the Normalized Difference Vegetation Index (NDVI) from the Advanced Very High Resolution Radiometer (AVHRR) show declining trends in forested regions of boreal Canada over the period 1982-2012, which may reflect the responses of boreal forests to longer growing seasons, increased summer drought stress, and higher frequency of fire occurrence. In this paper, we describe research where we analyzed Landsat 5 TM and Landsat 7 ETM+ time series data for eleven scenes distributed across the Canadian boreal forest zone. To do this, we created annual cloud-free, peak-summer composites for 28 years of Landsat data (1984-2011) using a maximum NDVI compositing procedure. By isolating areas of the landscape that were undisturbed during the Landsat record, our results highlight the difficulties involved in distinguishing subtle trends from data artifacts caused by unscreened atmospheric effects, subtle changes in sensor view angles, and differences across sensors. Differences in sensor view geometry across adjacent, overlapping Landsat scenes cause the eastern scene to have higher reflectances in the red and NIR bands than the western scene, but does not significantly affect NDVI values. A significant difference was observed between the red bands of the Landsat 5 TM and Landsat 7 ETM+ sensors, with Landsat 5 TM data having a higher red reflectance on average. However, no such systematic difference was observed between the NIR bands of these two sensors. In contrast to the effect of view geometry, NDVI values derived from the TM and ETM+ sensors should not be used interchangeably because of the sensitivity of NDVI values to variations in the red band. More generally, the results from this work demonstrate that while the opening of the Landsat archive has provided unprecedented opportunities for studying changes to the Earth's biosphere over the last 30 years, care must be taken in analyzing these data because artifacts introduced to the time series from a variety of sources can compromise results from analyses of long-term trends.