Detecting species-specific changes in spring phenology of Mt Medvednica forests using NDVI obtained from remote sensing with MODIS

Estimating climate-change-induced changes in biological systems, e.g., tree phenology, requires long, consistent and reliable datasets. Phenology observations within meteorological and other ground observation networks provide invaluable data but are of limited geographical distribution and observational range. Remote sensing observations surpass such problems and have been extensively used in phenology research, but suffer from other issues such as coarse spatial resolution, relatively long observation repeat cycles, issues due to clouds, short observational time series, etc. Those issues limit the number of remote sensing platforms which are suitable for the analysis of possible changes in the phenology of forest tree species at a local or regional level. NDVI derived from the MODIS daily measurements of reflectances at 250 m resolution spanning from 2000, in combination with tree species distribution maps from forest management plans, might be an optimal tool for investigating recent shifts in species-specific tree phenology at a local or regional level. Increased frequency of climate extremes, such as late spring cold spells, combined with the sensitivity of deciduous trees to frost damage during leaf unfolding and flowering, threatens the health and regeneration capacity of forests. The possible decoupling of tree and animal phenology additionally strains the entire forest ecosystem. These problems are exacerbated in urban and peri-urban forests, such as forests in Mount Medvednica Nature Park, Croatia, which are of major importance to the well-being of the people living in the Zagreb metropolitan area. Urban and peri-urban forests are exposed not only to the effects of climate change but also to the negative effects of pollution and the ‘urban heat island’ which all affect the timing and duration of leaf unfolding. 
Our research aims to quantify the changes in the spring phenology of Mt Medvednica’s forests during the period 2000–2020 using MODIS NDVI at 250 m, local forest management maps with tree species distribution information, and a digital elevation model. Our results show a delay in the Start of Season (SOS) with the overall trend of 0.12 days·yr^-1, but only due to SOS delay in common beech (F. sylvatica, 0.13 day·yr^-1), which might be a consequence of delay in meeting the chilling requirement needed to end the dormancy. Other deciduous tree species showed no significant trend in SOS. On the other hand, the End of Green-up (EOG) occurred sooner for all investigated tree species, with an average trend in EOG of –0.18 days·yr^-1 (F. sylvatica and Q. pubescens, –0.17; Q. cerris,  0.21; Q. petrea and Castanea sativa, –0.23 day·yr^-1). Consequently, during the 21 years of observation, the duration of leaf development of Mt Medvednica’s forests was on average shortened by 6.5 days.

Keywords: phenology, leaf unfolding, MODIS, NDVI, peri-urban forests, Mount Medvednica.

Acknowledgements:
The research has been supported by the Croatian Science Foundation project MODFLUX (HRZZ IP-2019-04-6325), by the Hungarian Scientific Research Fund (OTKA FK-146600) and by the TKP2021-NVA-29 project of the Hungarian National Research, Development and Innovation Fund. We thank to Croatian Forests Ltd. for granting access to their forest database and Croatian Meteorological and Hydrological Service for providing meteorological data.