Discussion
This study has shown that land surface phenology accuracy is affected both by data source and land cover type. Of the three sets of land surface phenology data used in this research, the NASA EVI data had the most significant correlations with ground-based phenology across all the landcover types, the Pickell NDVI data had the strongest correlations in each of the forested regions, and the NASA NDVI dataset was only useful for the Foothills Parkland region. The Pickell NDVI dataset was designed to be used in forested regions to estimate the end of the spring fire season, and it matches up quite well to the ground phenology in the forested landcovers. Thus, this dataset performed well for the landcover types it was designed for.
It’s unclear why the NASA EVI dataset performed considerably better than the NASA NDVI dataset. It would be interesting to compare these two in another region to see if the same difference in accuracy is observed, or whether the EVI is perhaps better for northern climates and landcover types while the NDVI might work better in different climate/landcover types. Another possibility is that the NDVI data were skewed towards earlier greenup dates when the data input changed from AVHHR (advanced very high resolution radiometer) to MODIS (moderate resolution imaging spectroradiometer) in the late 1990s. The NASA NDVI greenup dates show a rapid shift to earlier dates around this time, which cannot be seen in either of the other land surface phenology datasets (Figure 5).
The earlier dates of remote sensing greenup compared with aspen leaf-out suggests some other phenological event is driving the start of remote-sensed greenup at a landscape scale. The first two phenological events observed via PlantWatch in Alberta are consistently the first bloom of aspen and prairie crocus (Anemone patens) (Beaubien and Hamann 2011). This occurs slightly before aspen leaf-out, and roughly 20-30 days prior to saskatoon first bloom (Figure 4). It would be interesting to see if these phenological events are better matched up to the greenup dates from the NASA EVI and NDVI data. The greenup date in the Pickell data was determined by half-maximum NDVI (Pickell et al. 2017), whereas the NASA data determines it as the point where the vegetation index begins increasing (Didan et al. 2015). This explains why the Pickell greenup date was generally later than the NASA greenup date, and why the Pickell greenup date more closely matched the aspen leaf-out date.
It’s unclear why the NASA EVI dataset performed considerably better than the NASA NDVI dataset. It would be interesting to compare these two in another region to see if the same difference in accuracy is observed, or whether the EVI is perhaps better for northern climates and landcover types while the NDVI might work better in different climate/landcover types. Another possibility is that the NDVI data were skewed towards earlier greenup dates when the data input changed from AVHHR (advanced very high resolution radiometer) to MODIS (moderate resolution imaging spectroradiometer) in the late 1990s. The NASA NDVI greenup dates show a rapid shift to earlier dates around this time, which cannot be seen in either of the other land surface phenology datasets (Figure 5).
The earlier dates of remote sensing greenup compared with aspen leaf-out suggests some other phenological event is driving the start of remote-sensed greenup at a landscape scale. The first two phenological events observed via PlantWatch in Alberta are consistently the first bloom of aspen and prairie crocus (Anemone patens) (Beaubien and Hamann 2011). This occurs slightly before aspen leaf-out, and roughly 20-30 days prior to saskatoon first bloom (Figure 4). It would be interesting to see if these phenological events are better matched up to the greenup dates from the NASA EVI and NDVI data. The greenup date in the Pickell data was determined by half-maximum NDVI (Pickell et al. 2017), whereas the NASA data determines it as the point where the vegetation index begins increasing (Didan et al. 2015). This explains why the Pickell greenup date was generally later than the NASA greenup date, and why the Pickell greenup date more closely matched the aspen leaf-out date.
Conclusion
As the climate continues to change, measuring trends in phenology is becoming increasingly important. Ground phenological observations are generally clustered where people live and recreate, while remote sensing can extend the range of observations to a much larger area. The accuracy of remote sensing measurements of phenology appears to depend on both data source and landcover type at the pixel level. Studies like this help to identify regions where we can be confident in the land surface phenology greenup dates, and regions where we have little to no confidence. Maintaining both ground observation networks (such as PlantWatch) and publicly available land surface phenology records (such as the two NASA datasets) is important to the long-term monitoring of climate change and its associated impacts on ecosystems.