Validating remotely sensed land surface phenology with leaf out records from a citizen science network

Abstract

Vegetation phenology indices derived from multispectral remote sensing data are used to estimate primary productivity, track impacts of climate change and predict fire seasons. Such indices may, however, lack accuracy due to effects of snow and water, different vegetation types, and parameter choices for determining green-up and green-down. Here, we compare remotely sensed green-up dates with an extensive database of 57,000 leaf out and flowering observations from the Alberta PlantWatch citizen science network. We evaluate older global 5 km resolution VIP-NDVI and VIP-EVI2 v4 and v5 products, a regional 250 m resolution MOD09Q1-NDVI v6 product specifically designed for Alberta, and a recent 500 m resolution MCD12Q2-EVI2 v6 product. Overall, we find that MCD12Q2-EVI2 had the highest precision and least bias relative to ground observations, representing a significant advance over earlier phenology products. Different vegetation types showed a staged remotely sensed phenology in Alberta, with deciduous forest green-up first, followed by grasslands about 5 days later, and conifer forests green-up with a 10-day delay, allowing for corrections for different vegetation types. All products showed reduced interannual variability compared to ground observations, which may also lead to underestimating impacts of directional climate change. However, also in this respect MCD12Q2-EVI2 was superior, maintaining approximately 60% of the interannual variability. Nevertheless, the analysis shows that remotely sensed time series estimations of advances in leaf out may benefit from bias correction.

Keywords

• Remote sensing
• Ground truthing
• Plant phenology
• Citizen science
• NDVI
• EVI2