2026-10-07T09:47:02Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33850072 2026-09-22T07:55:52Z category_26872 item_type_3 month_year_09_2026

An improved Forest Phenology dataset of China from remote sensing and in-situ observations xiguang yang (24144980) Yixu Zhu (15993983) Jianfeng Zheng (1537786) Hasham Ahmad (9227249) Fan bu (24180624) ying yu (24144982) Ecological impacts of climate change and ecological adaptation Forest phenology remote sensing community Forests In China <p dir="ltr">This dataset is the China Forest Phenology dataset (CFP), which was generated mainly from MODIS MOD09A1 surface reflectance data for the period 2000–2022, together with ground phenological observations and GlobeLand30 land cover data. A dynamic threshold method was used to extract annual forest phenological metrics across China. During data processing, the quality-control band of MOD09A1 was first used to remove clouds, cloud shadows, and low-quality observations. The quality-controlled red, near-infrared, and blue bands were then used to calculate the Enhanced Vegetation Index (EVI). The original 8-day EVI time series were reconstructed into daily time series using piecewise cubic Hermite interpolation, followed by Savitzky–Golay filtering to reduce residual noise and abnormal fluctuations. Stable forest pixels were extracted from the GlobeLand30 land cover dataset, and fragmented or highly mixed pixels were removed using a neighborhood-based forest proportion threshold. Based on ground phenological observations, optimal thresholds for the start of season (SOS) and end of season (EOS) were calibrated at different sites. Threshold schemes were then established by climate zones and latitudinal bands, and SOS and EOS were extracted for each forest pixel. The complete workflow included MODIS EVI time-series construction, time-series interpolation and smoothing, stable forest mask generation, dynamic threshold calibration, pixel-wise phenological parameter extraction, and accuracy assessment. The processing was mainly conducted using the Google Earth Engine cloud-computing platform, together with common remote-sensing and GIS tools. Subsequent data organization, format conversion, and quality checking can be performed using Python, GDAL, ArcGIS, QGIS, or similar software.</p><p dir="ltr">The dataset covers the period from 2000 to 2022 and focuses on forested areas across China, including tropical, subtropical, warm temperate, temperate, and cold temperate forest regions. The temporal resolution is annual, meaning that one set of phenological metrics is provided for each year. The spatial resolution is 1000 m, and the spatial reference system is WGS84 geographic coordinates, corresponding to EPSG:4326. Each valid pixel value represents the phenological date of a forest pixel in a given year, expressed as day of year (DOY). Specifically, SOS represents the date of vegetation green-up or the beginning of the growing season, while EOS represents the date of vegetation senescence or the end of the growing season. The dataset is stored in GeoTIFF format. File names follow a unified naming convention, such as “year_phenological metric.tif”, for example, “2000_SOS.tif” and “2000_EOS.tif”. Since the dataset covers 23 years and includes two phenological metrics for each year, it contains 46 annual phenology raster files in total. GeoTIFF is a widely used raster format in remote sensing and geographic information science, and can be opened and processed using ArcGIS, QGIS, ENVI, Google Earth Engine, Python rasterio/GDAL, and other standard geospatial software tools.</p><p dir="ltr">Only valid phenological information for stable forest pixels is retained in this dataset. Non-forest areas, water bodies, built-up land, cropland, bare land, and pixels that failed the quality-control procedure are assigned as null or invalid values. Missing or uncertain values may occur in some areas due to cloud contamination, poor atmospheric conditions, mixed pixels near forest boundaries, weak seasonal EVI amplitudes in evergreen forests, smoothing-induced compression of very early or very late phenological dates, and scale differences between satellite observations and ground phenological records. Accuracy assessment showed that the CFP dataset had good agreement with ground observations. The coefficients of determination for SOS and EOS were both 0.90, with root mean square errors of 6.9 days and 8.21 days, respectively. A pixel-wise comparison with an independent ground-observation-based GP dataset further indicated that CFP can effectively capture the broad spatial patterns of forest phenology in China. However, some uncertainty remains in tropical and subtropical regions, especially for SOS, mainly because evergreen forests have weaker seasonal signals and more complex EVI curve shapes.</p> 2026-09-22T07:55:52Z Dataset Dataset 10.6084/m9.figshare.33850072.v1 https://figshare.com/articles/dataset/An_improved_Forest_Phenology_dataset_of_China_from_remote_sensing_and_in-situ_observations/33850072 CC BY 4.0