Scaling up spring phenology derived from remote sensing images

Land surface phenology, especially spring phenology, has been reported as a powerful indicator of ecosystem responses to climate change. It also exerts strong control on the carbon, water and energy balances and, hence, climatic feedbacks. Researchers have produced numerous spring phenology products from various coarse-resolution remote sensing data at regional or global scales. Scaling up observations of spring phenology from plot-level (or finer resolution) to coarser resolution is important for the validation, synthesis, and evaluation of those products. The best method for scaling up is unclear although coarse resolution data can be obtained by averaging across fine-scale pixels, or selecting the start of spring phenology (SOS) date associated with the earliest 30% (or another percentile) of fine-scale pixels within a coarse-scale pixel. In this study, we tested different methods that were average and percentile approaches to aggregate SOS as measured at 250 m (SOS (250 m)) resolution to 8 km (SOS (8 km)) resolution pixels, and then to ecosystems and national scales for the continental United States. The results indicated that the average absolute difference (AAD) between SOS (250 m) and SOS (8 km) from the average approach was close to that achieved by the percentile approach. Relatively large AAD values occurred in the western and southern regions of the continental United States. The distribution of AAD was positively related to landscape heterogeneity. The percentile approach generally yielded smaller AADs than the average approach did, but these two approaches performed similarly. Across landscapes and ecosystems, the optimal percentile usually ranged from 30–45th instead of a single value. Our findings indicated that the percentile approach may be best for finer scale areas, but that the average approach is an adequate alternative for scaling up SOS in most circumstances. In addition, the detailed error distributions of scaling up spring phenology across scales are helpful to identify the appropriate method of scaling up for validating the coarse SOS products derived from remote sensing images