A Selection of Experiments for Understanding the Strengths of Time Series SAR Data Analysis for Finding the Drivers Causing Phenological Changes in Paphos Forest, Cyprus

Abstract

Observing phenological changes are important for evaluating the natural regeneration process of forests, especially in Mediterranean areas where the regeneration of coniferous forests depends on seeds and the changes in blossoming time are influenced by climate change. The high temporal resolution of Sentinel-1 data allows the time series analysis of synthetic aperture radar (SAR) data, but it is still unknown how these data could be utilised for better understanding forest phenology and climate-related alternations. This study investigates the phenological cycle of Paphos forest, Cyprus using SAR data from 1992 to 2021, acquired by ERS-1/2, Envisat and Sentinel-1. An average phenological diagram was created for each space mission and a more detailed analysis was performed from October 2014 to November 2021, using the higher temporal resolution of Sentinel-1 data. Meteorological data were used to better understand the drivers of blooming alternations. Using the interquartile range (IQR), outliers were detected and replaced using the Kalman filter imputation. Forecasting trend lines were used to estimate the amplitude of the summer peaks and the annual mean. The observation of the average phenology from each satellite mission showed that there were two main blooming peaks each year: the winter and the summer peak. We argue that the winter peak relates to increased foliage, water content and/or increased soil moisture. The winter peak was followed by a fall in February reaching the lower point around March, due to the act of pine processionary (Thaumetopoea pityocampa). The summer peak should relate to the annual regeneration of needles and the drop of the old ones. A delay in the summer peak—in August 2018—was associated with increased high temperatures in May 2018. Simultaneously, the appearance of one peak instead of two in the σVH time series during the period November 2014–October 2015 may be linked to a reduced act of the pine processionary associated with low November temperatures. Furthermore, there was an outlier in February 2016 with very low backscattering coefficients and it was associated with a drought year. Finally, predicting the amplitude of July 2020 returned high relevant Root Mean Square Error (rRMSE). Seven years of time series data are limiting for predicting using trend lines and many parameters need to be taken into consideration, including the increased rainfall between November 2018 and March 2020.