Uncertainty of Vegetation Green-Up Date Estimated from Vegetation Indices Due to Snowmelt at Northern Middle and High Latitudes

Abstract

Vegetation green-up date (GUD), an important phenological characteristic, is usually estimated from time-series of satellite-based normalized difference vegetation index (NDVI) data at regional and global scales. However, GUD estimates in seasonally snow-covered areas suffer from the effect of spring snowmelt on the NDVI signal, hampering our realistic understanding of phenological responses to climate change. Recently, two snow-free vegetation indices were developed for GUD detection: the normalized difference phenology index (NDPI) and normalized difference greenness index (NDGI). Both were found to improve GUD detection in the presence of spring snowmelt. However, these indices were tested at several field phenological camera sites and carbon flux sites, and a detailed evaluation on their performances at the large spatial scale is still lacking, which limits their applications globally. In this study, we employed NDVI, NDPI, and NDGI to estimate GUD at northern middle and high latitudes (north of 40° N) and quantified the snowmelt-induced uncertainty of GUD estimations from the three vegetation indices (VIs) by considering the changes in VI values caused by snowmelt. Results showed that compared with NDVI, both NDPI and NDGI improve the accuracy of GUD estimation with smaller GUD uncertainty in the areas below 55° N, but at higher latitudes (55°N-70° N), all three indices exhibit substantially larger GUD uncertainty. Furthermore, selecting which vegetation index to use for GUD estimation depends on vegetation types. All three indices performed much better for deciduous forests, and NDPI performed especially well (5.1 days for GUD uncertainty). In the arid and semi-arid grasslands, GUD estimations from NDGI are more reliable (i.e., smaller uncertainty) than NDP-based GUD (e.g., GUD uncertainty values for NDGI vs. NDPI are 4.3 d vs. 7.2 d in Mongolia grassland and 6.7 d vs. 9.8 d in Central Asia grassland), whereas in American prairie, NDPI performs slightly better than NDGI (GUD uncertainty for NDPI vs. NDGI is 3.8 d vs. 4.7 d). In central and western Europe, reliable GUD estimations from NDPI and NDGI were acquired only in those years without snowfall before green-up. This study provides important insights into the application of, and uncertainty in, snow-free vegetation indices for GUD estimation at large spatial scales, particularly in areas with seasonal snow cover.

Evaluation of Vegetation Indexes and Green-Up Date Extraction Methods on the Tibetan Plateau

“…The GUD estimated by NDPI showed a greater negative bias than that estimated by NDGI, mainly in the southwest TP with arid alpine steppe as the main vegetation type. Similar results were also found in studies at grassland flux tower sites in the middle and high latitudes [23], and NDPI has been proved to have greater uncertainty in GUD extraction than NDGI in grasslands in Kazakhstan, Mongolia, and North Asia [25]. The probable reason is that NDGI accounts for dry grass-whose reflectance varies almost linearly from green to NIR-and the rationale behind NDGI gives dry grass a value of approximately zero [23].…”

“…Its application is limited to tundra and grassland ecosystems. In large spatial scales, no matter which VI is used, the estimated GUD may have uneven distribution of accuracy due to the spatial heterogeneity of latitudes and vegetation types [25,26]. It is still unclear how these VIs perform in GUD extraction on the TP.…”

Remote sensing of temperate and boreal forest phenology: A review of progress, challenges and opportunities in the intercomparison of in-situ and satellite phenological metrics

“…have shown to be effective in detecting seasonal patterns of vegetation in regions of high snowfall or with seasonal snow cover (Delbart et al, 2005;Delbart et al, 2015;Cao et al, 2020); also, VIs sensitive to water (e.g., land surface water index -LSWI) or less influenced by soil (e.g. optimized SAVI) showed better potential in tracking growing season phenology of evergreen forest ecosystems (Wu et al, 2014).…”

Adopting “Difference-in-Differences” Method to Monitor Crop Response to Agrometeorological Hazards with Satellite Data: A Case Study of Dry-Hot Wind

“…However, both NDVI and EVI do not use the shortwave infrared (SWIR) band, which responds strongly to the dry-hot wind because of its sensitivity to plant leaf water content. Conversely, the normalized difference phenology index (NDPI) uses the SWIR1 band in its calculation, which was originally developed to improve phenology monitoring for the deciduous ecosystem by minimizing the impact of snow and soil background on VI [40,41]. NDPI has a very similar form with NDVI by replacing the red band reflectance in NDVI with a weighted combination of the red and SWIR band reflectance (Equation (1)).…”