基于Landsat-MODIS图像填充与时空融合的红树林物候优化提取

[Objective]Vegetation phenology pertains to the study of seasonal life cycle events in plants.Mangroves,a type of forest grows in intertidal zones where seawater and freshwater converge,require precise phenological monitoring to understand their growth dynamics and sustain coastal ecosystems.However,the cloudy conditions in coastal regions and the timing of satellite passes often necessitate the use of low temporal resolution,cloud-free image sequences for extracting phenological features.The Combination of image gap filling and spatio-temporal fusion algorithms can enhance the accuracy of mangrove phenology extraction,but this approach requires extensive data and computational resources.Consequently,remote sensing phenology extraction of mangroves remains challenging.[Method]Mangroves in Zhangjiangkou(Fujian),Qi'ao Island(Guangdong),Beilun(Guangxi),and Dongzhaigang(Hainan),China,were selected as research subjects using 30m Landsat,250m MODIS,and 500m MODIS data from 2020.The image filling algorithm and spatio-temporal fusion algorithm were combined on Google Earth Engine to optimize mangrove phenology extraction.Firstly,500m MODIS data was processed using the Gap Filling and Savitzky-Golay filtering(GF-SG)algorithm to synthesize the Landsat NDVI data series,eliminating local null values.Secondly,linear interpolation addressed missing values in the 250m MODIS NDVI daily data,followed by Savitzky-Golay(S-G)filtering to smooth the time series.An ESTARFM-like algorithm was then applied to fit and compute high-resolution time series from the combined Landsat and MODIS data.Finally,phenological parameters were ultimately estimated using the maximum separation method,which relies on the differential ratio of observations exceeding the dynamic threshold before and after the central date.[Result](1)The GF-SG algorithm showed effective gap filling and image reconstruction for both Landsat and MODIS images.Compared with the original images,the RMSE of the reconstructed Landsat NDVI images were 0.088,0.131,0.100 and 0.083,and the RMSE of the reconstructed MODIS NDVI images were 0.088,0.131,0.100 and 0.083,respectively.(2)The ESTARFM-like algorithm effectively fits 30m daily NDVI data,demonstrating high consistency with GF-SG fitted NDVI imagery across various mangrove scenes,with R2 values of 0.991,0.710,0.997,and 0.999,respectively.(3)The NDVI of mangroves in the four regions exhibited distinct seasonal variations.From January to April,the NDVI series displayed a downward trend or remained at a low value;after April,the NDVI series initially increased and then decreased.(4)Significant differences exist in the start of growth(SoS)and length of the growth period(LoS)of mangroves across different regions,with an earlier SoS observed at lower latitudes.However,Dongzhaigang's SoS is later than that of Zhangjiangkou,likely due to climate-induced phenological anomalies.No consistent differences are observed in the end of growth(EoS)and maximum greenness(MaxGreen).[Conclusion]The study confirmed the potential of image gap filling and spatio-temporal fusion algorithms for extracting mangrove phenology.The proposed method offers a valuable reference for more comprehensive and convenient monitoring of mangrove phenology.