**1. Introduction**

Land surface phenology (LSP) has been recognized as one of the most effective indicators of climate change [1–4] and is closely related to animal migration, gross primary production, and crop productivity [5–7]. Methods for measuring phenology include ground observations (i.e., PhenoCam network and phenology network) and satellite observations [8–10]. Ground observations usually only reflect the phenological information of

local ecological communities [11–13]. However, satellite remote sensing has the potential to continuously observe the variation in vegetation phenology at multiple scales [14–17].

Phenology has been widely monitored in different types of remote sensing data in attempts to understand the interactions between vegetation and climate change during the past few decades [18–22]. A variety of vegetation indexes were used to monitor vegetation phenology in previous studies [23–25]. The two most used vegetation indices are the enhanced vegetation index (EVI) and the normalized difference vegetation index (NDVI) [26–29]. Meanwhile, several sets of freely accessible remote sensing products with different resolutions were released, such as the third generation GIMMS (GIMMS3g) NDVI with a spatial resolution of 1/12 degree that is derived from the Advanced Very-High-Resolution Radiometer (AVHRR) series satellites [30,31]; the Systeme Probatoire d'Observation de la Tarre (SPOT) NDVI with a 1 km spatial resolution [32]; and the moderate-resolution imaging spectroradiometer (MODIS) NDVI with spatial resolutions of 250, 500, or 1000 m [33,34]. Among them, GIMMS3g NDVI is the latest and longest-used product, and MODIS NDVI has several different spatial resolutions. These products have been widely used for studies involving phenology extraction [15]. Recent studies showed that the temporal and spatial variation trends of vegetation phenology that were observed in some areas by GIMMS3g NDVI and MODIS NDVI were consistent, but the conclusions that were reached in other regions were opposite [35,36]. The topography has a significant impact on the phenology of different product identifications [37]. Furthermore, the spatial phenological heterogeneity of data with a different resolution increases with the increase in landscape fragmentation [5]. However, the impacts of satellite products with different spatial resolutions on LSP extraction over regions with a heterogeneous topography have not been well clarified.

In general, data with a finer spatial resolution possess more information about the seasonality and phenology properties of vegetation [38–40]. Meanwhile, data with a fine spatial resolution have the problem of providing a larger amount of data, having a slow computation speed, and being time-consuming [41]. The coarse spatial resolution data is more suitable for monitoring phenology at a landscape scale [42]. Moreover, there will be some differences in vegetation phenology that are estimated from the coarse spatial resolution data to the fine spatial resolution data [5]. Thus, it is important to select remote sensing products with an appropriate spatial resolution in order to investigate vegetation phenology changes.

In this study, we explored the applicability of 250, 500, and 1000 m MODIS NDVI and GIMMS3g NDVI across the entire Loess Plateau, which is a typically ecological fragile region with a heterogeneous topography. The aims of this study were to (1) investigate the spatial and temporal patterns of vegetation phenology in the Loess Plateau, (2) analyze the applicability of different types of satellite data in complex terrain regions, and (3) explore the factors that influence the differences in the phenology of multiple datasets.

#### **2. Materials and Methods**
