remote sensing
Article Vegetation Phenology Influenced by Rapid Urbanization of The Yangtze Delta Region
Haiyong Ding 1,*, Luming Xu 2, Andrew J. Elmore 3,4 and Yuli Shi 1
1 School of Remote Sensing and Geomatics Engineering, Nanjing University of Information Science and Technology, Nanjing 210044, China; [email protected] 2 School of Geography Science, Nanjing University of Information Science and Technology, Nanjing 210044, China; [email protected] 3 Appalachian Laboratory, University of Maryland Center for Environmental Science, 301 Braddock Road, Frostburg, MD 21532, USA; [email protected] 4 National Socio-Environmental Synthesis Center (SESYNC), 1 Park PI, Suite 300, Annapolis, MD 21401, USA * Correspondence: [email protected]
Received: 6 May 2020; Accepted: 29 May 2020; Published: 1 June 2020
Abstract: Impacts of urbanization and climate change on ecosystems are widely studied, but these drivers of change are often difficult to isolate from each other and interactions are complicated. Ecosystem responses to each of these drivers are perhaps most clearly seen in phenology changes due to global climate change (warming climate) and urbanization (heat island effect). The phenology of vegetation can influence many important ecological processes, including primary production, evapotranspiration, and plant fitness. Therefore, evaluating the interacting effects of urbanization and climate change on vegetation phenology has the potential to provide information about the long-term impact of global change. Using remotely sensed time series of vegetation on the Yangtze River Delta in China, this study evaluated the impacts of rapid urbanization and climate change on vegetation phenology along an urban to rural gradient over time. Phenology markers were extracted annually from an 18-year time series by fitting the asymmetric Gaussian function model. Thermal remote sensing acquired at daytime and nighttime was used to explore the relationship between land surface temperature and vegetation phenology. On average, the spring phenology marker was 9.6 days earlier and the autumn marker was 6.63 days later in urban areas compared with rural areas. The spring phenology of urban areas advanced and the autumn phenology delayed over time. Across space and time, warmer spring daytime and nighttime land surface temperatures were related to earlier spring, while autumn daytime and nighttime land surface temperatures were related to later autumn phenology. These results suggest that urbanization, through surface warming, compounds the effect of climate change on vegetation phenology.
Keywords: vegetation phenology; urbanization; climate change; remote sensing; the Yangtze River Delta
1. Introduction Vegetation phenology refers to annual reoccurring cycles of plant activity, such as leaf onset and offset in spring and autumn, respectively. These cycles are driven by climatic cues [1] and are therefore an important indicator of trends in annual weather conditions [2,3]. For example, in Europe it was found that a spring air temperature increase of 1 ◦C has been associated with an advance in the beginning of the growing season by 7 days [4]. Similar results are found in the eastern United States with spring arriving 4–6 days earlier since the mid-1960s [5–10]. In China, spring has advanced on average 2.88 days per decade in response to spring warming [1]. Therefore, trends in vegetation
Remote Sens. 2020, 12, 1783; doi:10.3390/rs12111783 www.mdpi.com/journal/remotesensing Remote Sens. 2020, 12, 1783 2 of 18 phenology have emerged as one of the most obvious and easily measured vegetation responses to climate change [1,4–6,11–15]. Rapid urbanization observed in developing countries such as China and elsewhere [6,16–20] is accepted as an important factor leading to ecosystem change [6,15,21]. Notably, the increasing area of impervious surfaces (roads, buildings, parking lots, etc.) has changed the energy exchange process between the land surface and the atmosphere. These changes have intensified urban temperatures and contribute to the urban heat island phenomena [17,19,22,23]. Consequently, it has been widely observed that due to the Urban Heat Island (UHI) effect, urban areas typically exhibit an earlier spring phenology and later autumn phenology than their rural counterparts [6,11,16,24–26]. In some cases, urban landscapes have been invoked as examples of how climate change might be expected to impact vegetation phenology. It was found in research of China’s 32 major cities that the growing season started 11.9 days earlier and ended 5.4 days later in urban zones compared to rural counterparts, and the phenology change was closely related to the land surface temperature (LST) [26]. It was also found that UHI reduced the sensitivity of vegetation to climate change [27]. Urban areas stayed warmer in the winter, which reduces the exposure of vegetation to winter chilling and, therefore, spring warming does not have as strong an effect on vegetation phenology [27]. Although urban heat islands are generally associated with the extent of hardened surfaces, urban population was also found to be related to earlier spring budburst (by up to 13 days) and delayed leaf coloring (up to 15 days) in the Seoul Capital Area, South Korea, through a study of the spring budburst dates of six species and the leaf coloring date for two species [11]. However, most of these studies explored the difference in phenology between urban centers and their rural counterparts. Therefore, they do not account for the possibility that vegetation is compositionally different in cities and rural areas. Research into vegetation phenology spatial gradients and temporal trends is facilitated through the analysis of time series of satellite remote sensing data. Remote sensing of phenology is typically accomplished using broad-band, moderate-resolution instruments such as the Moderate Resolution Imaging Spectra-radiometer (MODIS), which can provide annual estimates of spring and autumn vegetation phenology. While there are several different approaches for extracting phenology markers from these time- series [10], the most reliable methods fit functions to the time series data that down-weight cloud-affected or otherwise unreliable observations [28]. Here we adopted such an approach to analyze the influence of urbanization on vegetation phenology from 2001 to 2018 in a rapidly urbanizing area, the Yangtze River Delta. Our analysis includes the relationship between vegetation phenology and land surface temperature in both urban and rural areas, which experienced different rates of development during the analysis period. Specifically, the aims of this research are: (1) to quantify and compare the influences of urbanization and time on vegetation phenology, and (2) to understand the relationship between LST and vegetation phenology among urban and rural regions.
2. Materials
2.1. Study Area Our study focused on vegetation phenology during the period of 2001–2018 on the Yangtze River Delta (Figure1), which is within the range of 29 ◦020–33◦250 N and 118◦200–122◦570 E, and was chosen because it is one of the most developed regions in China. In the past 20 years, the Yangtze River Delta, together with two other regions, the Beijing–Tianjin–Hebei region and the Pearl River Delta, have rapidly urbanized. Many metropolitan cities, such as Shanghai City, Nanjing City, Hangzhou City, and Suzhou City, are located on the Yangtze River Delta. These cities have experienced rapid urban expansion and increasing population, currently reaching 76 million people [24]. Remote Sens. 2020, 12, 1783 3 of 18 Remote Sens. 2020, 12, x FOR PEER REVIEW 3 of 18
Figure 1. Figure 1. The location of the Yangtze River DeltaDelta inin China.China. 2.2. Data 2.2. Data The data used in this research includes information on the spatial and temporal extent of urban landThe cover, data spring used and in this autumn research vegetation includes phenology, information and on land the surface spatial temperature.and temporal The extent urban of urban areas wereland cover, identified spring from and the autumn Suomi vegetation National Polar-orbiting phenology, and Partnership land surface (SNPP) temperature. Visible Infrared The urban Imaging areas Radiometerwere identified Suite from (VIIRS) the Suomi nighttime National light Polar-orbi data (Figureting2a) Partnership by applying (SNPP) an optimal Visible thresholds Infrared methodImaging basedRadiometer on the Suite urban (VIIRS) land-use nighttime product light generated data (Figure by the 2a) Resource by applying and an Environment optimal thresholds Data Center method of Chinesebased on Academy the urban of land-use Sciences product [18,29,30 generated]. Strong spatialby the heterogeneityResource and inEnvironment the spectral Data characteristics Center of ofChinese urban Academy areas, including of Sciences many [18,29,3 mixed0]. Strong pixels atspatial moderate heterogeneity resolution, in the increase spectral uncertainty characteristics in the of urban areas, including many mixed pixels at moderate resolution, increase uncertainty in the location location of the urban boundary. Nighttime light data, however, produce reliable delineation of urban of the urban boundary. Nighttime light data, however, produce reliable delineation of urban clusters, clusters, and clearly distinguish urban areas from the surrounding suburban and rural areas that lack and clearly distinguish urban areas from the surrounding suburban and rural areas that lack nighttime lighting [6]. In this study, the urban areas derived from nightlight data from 2000, 2005, 2010, nighttime lighting [6]. In this study, the urban areas derived from nightlight data from 2000, 2005, and 2015 were used to represent the urban area in 2001–2002, 2003–2007, 2008–2012, and 2013–2018, 2010, and 2015 were used to represent the urban area in 2001–2002, 2003–2007, 2008–2012, and 2013– respectively [25]. 2018, respectively [25]. The vegetation types data (30 m resolution; Figure2b were acquired from the work of Gong et al. The vegetation types data (30 m resolution; Figure 2b were acquired from the work of Gong et 2012, and included five main categories: mixed forest (MF), evergreen forest (EF), shrubland (SB), al. 2012, and included five main categories: mixed forest (MF), evergreen forest (EF), shrubland (SB), grassland (GS), and cropland (CP) [31,32]. Land cover was resampled to 250 m to match the MODIS grassland (GS), and cropland (CP) [31,32]. Land cover was resampled to 250 m to match the MODIS enhanced vegetation index (EVI) data. The daytime LST at 10:30 a.m. and nighttime LST at 22:30 p.m. enhanced vegetation index (EVI) data. The daytime LST at 10:30 AM and nighttime LST at 22:30 PM local time were acquired from the MODIS LST product (MOD11A2, 8-day composite) with a spatial local time were acquired from the MODIS LST product (MOD11A2, 8-day composite) with a spatial resolution of 1 1 km. It has been confirmed that the accuracy of MODIS LST reaches 1K [33]. resolution of 1 ×× 1 km. It has been confirmed that the accuracy of MODIS LST reaches 1K [33]. Vegetation phenology information was extracted from Moderate Resolution Imaging Spectrometer (MODIS) enhanced vegetation index (EVI) product (MOD13Q1 EVI, 16-day composite, 250 m spatial resolution) for the period 2001–2018. These data reduce the impact of variation in atmospheric scattering compared with the Normalized Difference Vegetation Index (NDVI) [3,34], and have been demonstrated to be suitable for monitoring phenology information in urban areas [2,14,19,20,35,36]
Remote Sens. 2020, 12, 1783 4 of 18 Remote Sens. 2020, 12, x FOR PEER REVIEW 4 of 18
FigureFigure 2. Urban 2. Urban clusters clusters derived derived from from SNPP SNPP/ VIIRS/ VIIRS nighttime nighttime light data ( (aa) )and and land land cover cover data data (b) ( b) in the yearin the 2015. year 2015. 3. Methods Vegetation phenology information was extracted from Moderate Resolution Imaging TheSpectrometer methods (MODIS) used toanalyze enhanced the vegetation impacts ofindex urbanization (EVI) product on the(MOD13Q1 vegetation EVI, phenology 16-day composite, are given in Figure250m3 to givespatial a clearresolution) representation. for the period First, 2001–2018. vegetation These phenology data reduce markers the were impact derived of variation from MODIS in atmospheric scattering compared with the Normalized Difference Vegetation Index (NDVI) [3,34], EVI time series data, and urban areas as well as buffer zones were extracted from the nighttime light and have been demonstrated to be suitable for monitoring phenology information in urban areas data. Then, we analyzed the spatial-temporal variation of vegetation phenology in urban, suburban, [2,14,19,20,35,36] and rural areas from 2001 to 2018. Finally, the relationship between the urban–rural phenology difference3. Methods and LST difference was constructed using linear regression analysis, and the vegetation phenologyRemote Sens. responses 2020, 12, xto FOR urbanization PEER REVIEW across vegetation types were analyzed. 5 of 18 The methods used to analyze the impacts of urbanization on the vegetation phenology are given in Figure 3 to give a clear representation. First, vegetation phenology markers were derived from MODIS EVI time series data, and urban areas as well as buffer zones were extracted from the nighttime light data. Then, we analyzed the spatial-temporal variation of vegetation phenology in urban, suburban, and rural areas from 2001 to 2018. Finally, the relationship between the urban–rural phenology difference and LST difference was constructed using linear regression analysis, and the vegetation phenology responses to urbanization across vegetation types were analyzed.
Figure 3. The flow chart of the method used in this study. Figure 3. The flow chart of the method used in this study.
3.1. Asymmetric Gaussians Smoothing An asymmetric Gaussian function (A–G) was fitted to the MODIS EVI time series data to retrieve phenology markers for spring and autumn (Equation (1)) [10,37]. To fit EVI data accurately, MODIS pixel quality information was used to assess pixel reliability of EVI data. Pixels with higher reliability (higher quality) were given greater weight in the fitting algorithm (following Table 1), which reduced the impact of data noise [38,39]. The asymmetric Gaussian function applied was: