Validation of spatiotemporally dense springtime land surface phenology with intensive and upscale in situ

Authors: LIANG, LIANG - University Of Kentucky; SCHWARTZ, MARK - University Of Wisconsin; WANG, ZHUOSEN - University Of Massachusetts; Gao, Feng; SCHAAF, CRYSTAL - University Of Massachusetts; TAN, BIN - National Aeronautics And Space Administration (NASA); MORISETTE, JEFF - Collaborator; ZHANG, XIAOYANG - National Oceanic & Atmospheric Administration (NOAA)

Submitted to: Remote Sensing of Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/15/2014
Publication Date: 4/17/2014
Publication URL: http://handle.nal.usda.gov/10113/59918
Citation: Liang, L., Schwartz, M., Wang, Z., Gao, F.N., Schaaf, C., Tan, B., Morisette, J., Zhang, X. 2014. Validation of spatiotemporally dense springtime land surface phenology with intensive and upscale in situ. Remote Sensing of Environment. 52:7513-7526.

Interpretive Summary: Vegetation phenology refers to the timing of seasonal developmental stages in plant life cycles which affect the rate of photosynthesis and plant water use. Changes in vegetation phenology at large scales will consequently affect regional carbon and water cycles and are critical for regional and global climate change and carbon balance studies. Land surface phenology (LSP) can be detected using a dense time series of remote sensing data at regional to global scales. However these LSP products need to be validated using compatible in situ plant phenology data. This paper computes and validates the remotely sensed LSP from the 500 meter resolution Moderate Resolution Imaging Spectroradiometer (MODIS) and 30 meter resolution Landsat data. Results show that the LSP green-up time computed from daily MODIS data product matches to within five days the timing of deciduous tree full bud burst at the landscape scale level. The fused Landsat based LSP captures green-up dates at a finer plant community level with a seven day difference on average, but does not show enough sensitivity to distinguish inter-community phenological variations. The paper demonstrates an effective approach to extract LSP at the Landsat scale which is important for detecting changes in water and carbon cycles at the plant community level, and particularly for monitoring crop conditions at the field scale required by the National Agricultural Statistics Service and Foreign Agricultural Service for more accurate yield assessments and predictions.

Technical Abstract: Land surface phenology (LSP) developed using temporally and spatially optimized remote sensing data, is particularly promising for use in detailed ecosystem monitoring and modeling efforts. Validating spatiotemporally dense LSP using compatible (intensively collected) in situ phenological data is therefore important to ensure the relevance of derived remote sensing information to the underlying biophysical processes. In this study, we validated springtime LSP time series and start of growing season (greenup) derived from daily Moderate Resolution Imaging Spectroradiometer (MODIS) Nadir BRDF (Bidirectional Reflectance Distribution Function)-Adjusted Reflectance (NBAR) vegetation indices (VI) and Landsat fused data over a temperate mixed forest site. The ground reference is from intensively observed (high spatial density and temporal frequency) and upscaled multi-year (2006-2009) springtime tree leaf phenology from a northern forest in the state of Wisconsin, USA. Results showed that the daily MODIS NBAR VI based LSP greenup occurred near the time of deciduous tree full bud burst over the landscape with approximately five days of mean absolute error. The spatial and temporal variations of LSP agreed with that of landscape phenology at the study area level. In addition, though LSP from daily MODIS NBAR fused with Landsat did not show enough sensitivity to distinguish intercommunity phenological variations at our study sites; it captured greenup dates at the community level with seven days of overall mean absolute error. Such agreements between ground and satellite measures indicate the potential of using spatiotemporally dense LSP for more detailed vegetation dynamic characterization, phenological monitoring and ecosystem modeling. On the other hand, our study demonstrated that in situ phenology data collected intensively to match the spatial and temporal properties of satellite data provides improved accuracy in LSP validation.