Data

By studying vegetation leafing phenology and its coupling with climate along the urban-rural gradient in Phoenix metropolitan region, USA, we elucidated the degree of urbanization-induced transformations of phenology and primary productivity. In this study we used time-series of Normalized Difference Vegetation Index (NDVI) from the Moderate Resolution Imaging Spectroradiometer (MODIS) and spatially interpolated rainfall. Our analyses were stratified by major land covers and dominant soil texture. We also assessed time scales at which NDVI responds most strongly to climatic factors. No distinctive patterns in phenology were found along the urban-rural gradient; however some important generalities were confirmed. Agricultural and urban developments introduce growth multimodality, which is not attributable to desert but customarily found in riparian ecosystems of the area. Despite the existence of summer flush of growth in the desert its signal is not detected by de-noised satellite data. Urban and agricultural vegetation is characterized by fast growth and senescence rates. While agriculture has the shortest growth length, most urban vegetation stays photosynthetically active for longer periods. Growth in the desert is controlled by precipitation accumulated for 2-5 months. Spatial patterns of NDVI are predicted by precipitation grids. Positive relationship between these two variables changes seasonally reaching the maximum near the peak of annual growth. Spring and summer NDVI grids are in better agreement with longer term accumulated precipitation, but the early autumn growth is correlated more with immediate rainfall. Spatial and temporal correlations of desert NDVI with temperature are negative confirming the role of temperature in stimulating water loss from the soil. Our results supported the hypothesis that coarse-textured soils limit evaporative losses of soil water and promote growth. Riparian NDVI are moderately positively correlated with temperature but only weakly with precipitation. NDVI dynamics in urban and agricultural land covers are completely unsynchronized with natural vegetation communities and decoupled with precipitation. They exhibit positive, yet low, correlation with temperature. Overall, urbanization adds a greater diversity of phenological patterns that are not determined by climatic variability. Instead, urban and agricultural vegetation dynamics is expected to be explained largely by socio-economic variables.

Meteorological data acquisition and preprocessing

Moderate Resolution Imaging Spectroradiometer (MODIS) instruments aboard TERRA/AQUA satellites have a revisit period of 1-2 days and collect data in 36 spectral channels at three spatial resolutions (250 m, 500 m, and 1 km) with geolocation accuracy of approximately 50 m at nadir (Wolfe et al. 2002). NDVI is computed from reflectance data corrected for molecular scattering, ozone absorption and aerosols, and adjusted to nadir and Sun angles with the use of Bidirectional Reflectance Distribution Function (BDRF) models (Huete et al. 1999, van Leeuwen et al. 1999). The index is based on the property of green vegetation to strongly absorb red and reflect near-infrared wavelengths, and is calculated as (Tucker 1979, Huete et al. 1999): NDVI = (NIR - RED) / (NIR + RED) where NIR is MODIS band 2 (841-846 nm) and RED is MODIS band 1 (620-670 nm). Final 16-day maximum composites are produced by combining multiple days and filtering out clouds and data with bad integrity. We obtained 250-m NDVI images (MOD13Q1 product) for 2000-2005 and examined data using quality flags from companion Quality Assurance (QA) images. Recognizing the existence of multiple methods for modeling phenology (Olsson and Eklundh 1994, Moulin et al. 1997, White et al. 1997, Moody and Johnson 2001, Reed et al. 2003, de Beurs and Henebry 2005, Bradley et al. 2007, Soudani et al. 2008), we chose to use TIMESAT (Jönsson and Eklundh 2006) to de-noise NDVI time-series and extract seasonality parameters. TIMESAT implements three processing methods based on least-squares fits to the upper envelope of NDVI time-series, accounting for negatively biased noise (Jönsson and Eklundh 2004). Data fitting with local polynomial functions classified as adaptive Savitsky-Golay filtering was preferred in our study. It most closely corresponded to raw NDVI data during calibration and method testing. Three adaptation runs were executed with search window set to 2, 3, and 4 consecutive NDVI values. Start and end of growth periods were defined as a 20% increase or decrease in NDVI values. The rate of growth is calculated as the ratio between NDVI values at the beginning of growth period those NDVI values at the 80% level of growth divided by the corresponding time difference (in days). The TIMESAT software is available online: http://www.nateko.lu.se/personal/Lars.Eklundh/TIMESAT/timesat.html