Intercomparison and evaluation of spring phenology products using National Phenology Network and AmeriFlux observations in the contiguous United States
Introduction
Plant phenology is widely used as an independent measure and powerful indicator of how ecosystems are responding to climate change (Badeck et al., 2004, Friedl et al., 2006, Parmesan, 2006, White et al., 2009, Ma et al., 2013). It is linked to primary productivity, crop yields, insect emergence, and bird migration (Parmesan, 2006, Rosenzweig et al., 2007, Ault et al., 2015). Shifts in the timing of spring arrival can change the time of seeding in agricultural systems (Knudson, 2012), disrupt ecological interactions between plants and animals (Burkle et al., 2013, Kellermann et al., 2015), and contribute to changes in species distributions (Chuine and Beaubien, 2001). Earlier spring onset, in particular, can alter the surface energy balance, accelerate transpiration (Wilson and Baldocchi, 2001, Piao et al., 2015), enhance mid-summer droughts, and increase the incidence of severe wildfires (White and Nemani, 2003, Angert et al., 2005, Westerling et al., 2006).
Land surface phenology (LSP), the study of the timing of recurring seasonal pattern of variation in vegetated land surfaces observed from satellite sensors, has been widely used in detecting and understanding climate changes and ecosystem dynamics (Gonsamo et al., 2012, Friedl et al., 2006, Parmesan, 2006). Various studies have tended to find that spring phenology has occurred earlier in recent decades, especially across the Northern Hemisphere, as a result of rising global temperatures (McCabe et al., 2012, Allstadt et al., 2015). However, not all documented trends in spring phenology have been consistent, whether detected by satellite-based sensors, ground-based sensors, or human observations (Menzel et al., 2006, Reed, 2006, Piao et al., 2007, White et al., 2009). Because the differences in phenological trends influence the forecasts of climate change, crop development, and ecological dynamics, it is critical to understand the nature of the inconsistencies.
Over the last decades, a number of LSP products have been developed from satellite data, which have different spatial and temporal characteristics. The products include (Table 1): (1) the Moderate Resolution Imaging Spectroradiometer (MODIS) Land Cover Dynamics Product (MCD12Q2) (Zhang et al., 2006a, Zhang et al., 2006b, Ganguly et al., 2010); (2) Global Vegetation Index and Phenology Multi-sensor Phenology (VIPPHENEVI2) (White et al., 2009, Didan, 2010, Didan et al., 2016), which uses data from the advanced very high resolution radiometer (AVHRR) and MODIS data based on daily two-band enhanced vegetation index (EVI2); (3) Global Long-Term Climate Modeling Grid Land Surface Phenology (CMGLSP) (Zhang et al., 2014, Zhang, 2015); (4 and 5) the MODIS-based products (MOD09Q1PEVI and MOD15PHN) in support of the North American Carbon Program (NACP), which were produced over the contiguous United States (CONUS) using MOD09 reflectance and MOD15 leaf area index (LAI), respectively (Morisette et al., 2009, Tan et al., 2008, Tan et al., 2011); and (6) the United States Geological Survey/Earth Resources Observation Systems (USGS/EROS) Phenology Product (AVHRRP) over the CONUS (Reed et al., 1994). In addition, agencies and researchers have developed many other specialized and regional products (Verstraete et al., 2008, Hargrove et al., 2009, Dash et al., 2010, Lacaze et al., 2011, Gonsamo and Chen, 2016).
Current LSP products could contain considerable discrepancies and uncertainties because there are no standardized methods and criteria to quantify the LSP metrics (de Beurs and Henebry, 2010, White et al., 2009). After comparing 10 methods for estimating green-up onset dates (GUD) from the same satellite data source, White et al. (2009) found that estimates differed by more than one month depending on the method used. Unfortunately, we know of no studies that have compared and validated a range of LSP products, particularly which use different satellite data sources, vegetation indices, and analytical methods to estimate spring phenology. Thus, direct validation and intercomparison of existing LSP products are essential for further applications that rely on accurate estimates of phenological metrics. These applications include climate models, agricultural models, and estimates of shifts in species performance and distribution.
This study compares GUD in the six major LSP products (Table 1) that are widely used and regularly updated by their developers. The main goals are to answer several crucial questions as follows: (1) what are the consistencies and discrepancies of GUD estimates across the products? (2) which product provides the best consistency when compared to ground-based observations? (3) what is the temporal and spatial retrieval efficiency in each product?
Section snippets
Methods
Our study focused on the intercomparison and validation of GUD in LSP products across the CONUS. It was due to the fact that the well-organized ground reference measures were available, which were spatially distributed across the entire CONUS and included a number of ecosystems and land cover types. These data allowed us to compare the performance of land surface phenology products across a range of conditions.
Spatial consistency of green-up onset dates among LSP products
Fig. 2 shows the geographical distributions of GUD (day-of-year, DOY) averaged for 2001–2007 from each LSP product. All products showed relatively early GUD in southern locations and later GUD in northern locations, as expected. In the eastern CONUS, all products showed comparable GUD (∼100 DOY) (Fig. 2).
The LSP products differed substantially in their estimates of GUD. MCD12Q2 GUD was generally earlier (Fig. 2a) while CMGLSP GUD was later (Fig. 2d). The products also provided spatially
Evaluating LSP GUD estimates using ground observations
Reconciling ground and satellite-based phenological observations is very challenging because of the mismatch in scales. Ground observations often describe phenological events in individual trees or stands, whereas satellites capture landscape-level phenology (Zhang and Goldberg, 2011, Ganguly et al., 2010, Liang and Schwartz, 2009, Morisette et al., 2009, Schwartz and Hanes, 2009, White et al., 2009, Zhang et al., 2006a, Zhang et al., 2006b). Thus, the point observations of the first leaf dates
Conclusions
Validation of LSP products has been highly expected to quantify the quality or accuracy (Delbart et al., 2015, White et al., 2009, Zhang et al., 2006a, Zhang et al., 2006b, Liang et al., 2011). However, it is currently challenging because in-situ data that spatially match satellite footprints are barely available. On the other hand, widely acceptable criteria to define the LSP GUD were not available. Thus, a high quality LSP product should be self-consistent spatially and temporally.
In this
Acknowledgements
The authors acknowledge Jake Weltzin and Theresa Crimmins (USA National Phenology Network) for their useful comments and the following data support: Lilac/honeysuckle data were provided by the USA National Phenology Network, Joseph M. Caprio, Mark D. Schwartz, and all contributors to past U.S. Dept. of Agriculture regional phenology projects; MCD12Q2 from Land Processes Distributed Active Archive Center (LP DAAC); Long Term Land Surface Phenology Record from //bruin.sdstate.edu/ftp/XYZ/AVHRR_EVI2/in
References (78)
- et al.
Assessing eddy-covariance flux tower location bias across the Fluxnet-Canada Research Network based on remote sensing and footprint modelling
Agric. For. Meteorol.
(2011) - et al.
The use of MERIS Terrestrial Chlorophyll Index to study spatio-temporal variation in vegetation phenology over India
Remote Sens. Environ.
(2010) - et al.
Comparing land surface phenology with leafing and flowering observations from the PlantWatch citizen network
Remote Sens. Environ.
(2015) - et al.
MODIS collection 5 global land cover: algorithm refinements and characterization of new datasets
Remote Sens. Environ.
(2010) - et al.
Autumn, the neglected season in climate change research
Trends Ecol. Evol.
(2015) - et al.
Land surface phenology from MODIS: characterization of the Collection 5 global land cover dynamics product
Remote Sens. Environ.
(2010) - et al.
A comparison of multiple phenology data sources for estimating seasonal transitions in deciduous forest carbon exchange
Agric. For. Meteorol.
(2011) - et al.
Circumpolar vegetation dynamics product for global change studies
Remote Sens. Environ.
(2016) - et al.
Predicting deciduous forest carbon uptake phenology by upscaling FLUXNET measurements using remote sensing data
Agric. For. Meteorol.
(2012) - et al.
Monitoring elevation variations in leaf phenology of deciduous broadleaf forests from SPOT/VEGETATION time-series
Remote Sens. Environ.
(2011)