PROGRESS IN GEOGRAPHY ›› 2018, Vol. 37 ›› Issue (8): 1031-1044.doi: 10.18306/dlkxjz.

• Articles • Previous Articles     Next Articles

Comparative study of vegetation phenology extraction methods based on digital images

Yuke ZHOU()   

  1. Key Laboratory of Ecosystem Network Observation and Modeling, Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China
  • Received:2017-11-21 Revised:2018-05-10 Online:2018-09-04 Published:2018-09-04
  • Supported by:
    National Key R&D Program of China, No.2016YFC0500103;National Natural Science Foundation of China, No.41601478;LREIS Open Fund, No.2016

Abstract:

Vegetation phenology is an effective index that reflects the growth condition of vegetation and their response to climate change. Analyzing vegetation phenology at landscape or smaller scale will be a powerful supplement to remote-sensing based and artificial observed phenological studies. In this study, we applied the time series digital images from typical observation stations in camera-based phenology network (PhenoCam) to fit the seasonal growth curves at region of interest (ROI) and pixel scales. Key vegetation phenological parameters on the growth curve were extracted using multiple methods. First, we realized the custom definition of ROI by arbitrarily drawing polygons on the image and growth curve fitting based on the greenness index in the ROI. The results indicate that double logistic method is adapted for modeling the vegetation growing process at middle-high latitude that have a single growth peak. The spline method shows a good performance for fitting growth curves of vegetation that has multiple growth seasons. Second, based on the fitted curve, we used four models (Klosterman, Gu, TRS, and Derivatives) to derive the key phenological parameters for vegetation having single growth season. MODIS EVI-based phenological parameters were extracted as a comparison with camera based phenological parameters. We found that their capabilities in detecting phenological parameters in forest and cropland were consistent. For the vegetation with more than one growth season, the growth curve was fitted with spline method, then the change point detection method was applied to determine growth seasons on the curve and extract the key phenological parameters. The uncertainty of the fitting methods and phenology methods were estimated, with all of the RMSE less than 0.005. Klosterman method is identified as the most robust. Vegetation greenup date was extracted for each pixel in the ROI that can be useful for distinguishing the phonological diversity between different species. Pixel level vegetation phenology analysis could be used to identify biodiversity at landscape scale in the future.

Key words: vegetation phenology, phenology camera, ROI(region of interesting), SOS(start of growing season), EOS(end of growing season), double Logistic curve, threshold method, derivatives method