# Discrepancy in Phenological Indicators from CO2 Flux, MODIS Image and Ground Observation in a Temperate Mixed Forest and an Alpine Shrub Ecosystem

## TL;DR

This study compares different methods for tracking plant growth cycles in two ecosystems and finds that results vary depending on the data and method used.

## Contribution

The study reveals discrepancies in phenological indicators across data types and extraction methods in different ecosystems.

## Key findings

- PheNDVI and PheGPP showed earlier SOS and delayed EOS compared to ground observations.
- The dynamic threshold method provided more consistent estimates of plant phenology.
- Vegetation complexity influenced the discrepancy between PheNDVI and PheGPP more in the mixed forest.

## Abstract

Different approaches have been developed to assess the phenological dynamics of ecosystems. However, diverse data sources and extraction methods for assessing ecosystem phenology can result in discrepant and inaccurate results, especially across different types of vegetation under various climate classifications. Based on the phenology of dominant plant species (Pheplant) obtained from ground monitoring in an alpine shrub meadow at Haibei Station (HBS) on the Qinghai–Tibetan Plateau and in a broad-leaved Korean pine forest at Changbai Mountain (CBF) in Northeastern China, we extracted vegetation phenology from the Normalized Difference Vegetation Index (PheNDVI) and photosynthetic phenology from gross primary productivity (PheGPP) using five common methods. These methods included Gaussian fitting, single logistic function fitting, double logistic function fitting, and smoothing techniques combined with fixed threshold and derivative-based determination approaches. There was no consistent interannual trend in either plant phenology or environmental factors at the two sites. Among the three types of plant phenology, a similar interannual pattern in the start of the growing season (SOS) was observed, whereas the interannual patterns for the end of the growing season (EOS) and the growing season length (GSL) were asynchronous. Compared to Pheplant, both PheNDVI and PheGPP exhibited an earlier SOS, a delayed EOS, and consequently an extended GSL. The SOS derived from both PheNDVI and PheGPP was advanced by increasing spring temperatures at both sites, while the relationship between EOS and air temperature was relatively weak. The discrepancy between PheNDVI and PheGPP was more pronounced at CBF than at HBS, likely due to the complex vegetation composition and structure of the mixed forest. The different extraction methods produced more consistent and less variable estimates of SOS compared to EOS and GSL at both sites. Among the five methods, the dynamic threshold approach showed a relatively small difference between PheNDVI and PheGPP, suggesting that it could provide a more consistent estimate of plant phenology across the two sites. This study clearly reveals the inherent discrepancies associated with using different types of phenological data and the influence of extraction methods on phenology across different plant functional types. More attention should be given to improving the accuracy of EOS and understanding the influence of vegetation composition on phenological variation in future studies.