17:15 〜 19:15
[ACG46-P10] Improving remote estimation of autumn phenology based on GCOM-C/SGLI satellite data and a modified logistic function
キーワード：Autumn phenology、Modified logstic function、GCOM-C/SGLI
Understanding autumn phenology is crucial for assessing ecosystem responses to climate change. While the mechanisms driving the green-up onset date (GOD) using satellite data are well established, autumn phenology remains less understood due to the complex interactions governing vegetation senescence and the challenges in extracting phenological transition dates from satellite data. Unlike spring, autumn phenology is influenced by multiple factors, including decreasing temperature, shortened photoperiod, and variable precipitation, making its estimation more complex. Moreover, current satellite-based methods often fail to capture the two-stage decline in vegetation index time series, leading to biases in estimating key autumn phenological metrics such as the senescence onset date (SOD) and dormancy onset date (DOD).
In this study, we introduce the Modified Logistic Function (MLF) to improve autumn phenology estimation from remote sensing data. The MLF retains the advantages of the DLF while better accommodating two-stage declines in vegetation indices. By incorporating three logistic functions, the MLF enhances curve-fitting performance for complex seasonal transitions, leading to more precise estimations of SOD and DOD. The MLF was validated through comparisons with near-surface observations, demonstrating significant reductions in root mean square error (RMSE) and bias for both spring and autumn phenological metrics. While the MLF performed similarly to the DLF for spring phenology, it substantially improved autumn estimates, particularly in regions exhibiting two-stage declines. Comparisons among satellite datasets (SGLI, MODIS, and VIIRS) confirmed the robustness of the MLF. Applying the MLF to MODIS and SGLI datasets produced consistent results, with SGLI’s 250 m resolution further improving phenological accuracy by reducing surface heterogeneity effects. Long-term trend analyses at the Harvard site revealed that MLF-derived trends were more consistent with near-surface camera observations than MODIS MCD12Q2 data. The MLF also more accurately captured delayed senescence and dormancy onset trends, demonstrating its potential for detecting long-term phenological shifts under climate change. Additionally, we analyzed two-stage autumn declines in the NDGI time series and identified grasslands as the most affected vegetation type. Grasslands were highly sensitive to climatic factors such as water stress and temperature extremes, leading to asynchronous senescence patterns. Croplands also exhibited two-stage declines due to harvesting and post-harvest regrowth, while forests displayed this phenomenon less frequently, with deciduous broadleaf forests being the most prominent contributors. Overall, the development of the MLF represents a significant advancement in phenological studies, particularly for autumn phenology estimation.
In this study, we introduce the Modified Logistic Function (MLF) to improve autumn phenology estimation from remote sensing data. The MLF retains the advantages of the DLF while better accommodating two-stage declines in vegetation indices. By incorporating three logistic functions, the MLF enhances curve-fitting performance for complex seasonal transitions, leading to more precise estimations of SOD and DOD. The MLF was validated through comparisons with near-surface observations, demonstrating significant reductions in root mean square error (RMSE) and bias for both spring and autumn phenological metrics. While the MLF performed similarly to the DLF for spring phenology, it substantially improved autumn estimates, particularly in regions exhibiting two-stage declines. Comparisons among satellite datasets (SGLI, MODIS, and VIIRS) confirmed the robustness of the MLF. Applying the MLF to MODIS and SGLI datasets produced consistent results, with SGLI’s 250 m resolution further improving phenological accuracy by reducing surface heterogeneity effects. Long-term trend analyses at the Harvard site revealed that MLF-derived trends were more consistent with near-surface camera observations than MODIS MCD12Q2 data. The MLF also more accurately captured delayed senescence and dormancy onset trends, demonstrating its potential for detecting long-term phenological shifts under climate change. Additionally, we analyzed two-stage autumn declines in the NDGI time series and identified grasslands as the most affected vegetation type. Grasslands were highly sensitive to climatic factors such as water stress and temperature extremes, leading to asynchronous senescence patterns. Croplands also exhibited two-stage declines due to harvesting and post-harvest regrowth, while forests displayed this phenomenon less frequently, with deciduous broadleaf forests being the most prominent contributors. Overall, the development of the MLF represents a significant advancement in phenological studies, particularly for autumn phenology estimation.