Objectives  Accurate measurement of vegetation phenology in high-altitude mountainous areas is critical in understanding the response of sensitive ecosystems to global climate change. The extraction and comparison of the phenological information using phenological cameras (PhenoCams) and remote sens‍ing technology can help evaluate the performance of PhenoCams in vegetation phenology extraction, which provides an important reference for the accuracy of remote sensing phenological data in mountainous areas.

  Methods  Firstly, the green chromatic coordinates (GCC) and normalization difference vegetation index (NDVI) of the vegetation are extracted for characterizing the profile of vegetation annual change based on the observed data from four PhenoCams stations and the remote sensing data in the Bayanbulak region of the Xinjiang Uygur Autonomous Region, China. Secondly, the denoising performance of seven filters for green index signals is comprehensively investigated. The phenological parameters extracted by 20 combinations of five curve fitting meth‍ods and four phenological parameter extraction methods are compared and analyzed.

  Results  (1) Vegetation PhenoCams can accurately provide high temporal resolution variation of GCC information of grasslands (including sparse vegetation types) in Tianshan mountainous areas, China, and they are effective means to observe mountain phenology and verify remote sensing phenology data. (2) Weather conditions such as rain and snow have a strong impact on GCC, and therefore it is necessary to select appropriate filters for denois‍ing. (3) ‍Curve fitting methods and phenological extraction methods have an important impact on the val‍ues of phenological parameters. Moreover, obvious differences exist between the extraction meth‍ods. The phenological values extracted by Threshold and Derivatives methods are similar, and the extraction start and stop peri‍ods can well match the artificially observed periods of rejuvenation and withering respectively. The phenological values extracted by the Klosterman method and Gu method are similar, which are consistent with the observations. (4)The effectiveness of 20 combinations in extract‍ing phenological information from remote sensing data in mountainous areas is only 48%. The most effective extraction method for moderate resolution imaging spectroradiometer (MODIS) data is the combination of Beck+Derivatives, and the best extraction methods for visible infrared imaging radiometer suite (VIIRS) data are the combination of Beck + Threshold and that of Elmore + Derivatives.

  Conclusions  PhenoCams data and remote sensing data have obvious differences in spatial and temporal scale, and the PhenoCams data provide a higher temporal resolution than the remote sensing data. Moreover, the PhenoCams data are less affected by weath‍er conditions, and thus the signal pollution caused by weather conditions can be reduced. Regulating the operational PhenoCams observation and expanding the spectral observation range of PhenoCams will improve the extraction of phenological information and help validate remote sensing phenological information. All of this certainly can help build a stable and long-term scientific data set for vegetation observations.