Solar induced chlorophyll fluorescence (SIF) is an important substitute for photosynthesis. The TROPOSphere Monitoring Instrument (TROPOMI) carried by the Copernicus Sentinel-5P mission provides almost global coverage of fine spectral resolution, enabling reliable SIF retrieval. However, the SIF dataset currently obtained by satellites only has a relatively coarse spatial resolution, which limits its application at fine scales. Here, we used the weighted superposition algorithm to generate the Chinese High Spatial Resolution SIF dataset (500 meters, 8 days) (HCSIF) from TROPOMI satellite from 2000 to 2022, with a spatial resolution of 3.5 kilometers by 5.6-7 kilometers. Our algorithm demonstrates high accuracy on the validation dataset (R2=0.87, RMSE=0.057 mW/m2/nm/sr). The HCSIF dataset was evaluated against OCO-2 SIF, tower based SIF measurements, and total primary productivity (GPP) of flux towers. Our dataset can facilitate the understanding of fine scale terrestrial ecological processes, enabling the monitoring of biodiversity in ecosystems and accurate long-term assessments of crop health, productivity, and stress levels. The scaling factor is 0.0001.
Sentinel-5P
Using the weighted superposition set algorithm, TROPOMI SIF (HCSIF) with a resolution of 500 meters over China was downscaled for a long period of time (2000-2022) at 8-day intervals. China has the largest number of small farmers with less than 2 hectares of land (230 million). The HCSIF data comes from TROPOMI SIF data 11 from the California Institute of Technology, MODIS vegetation indices 38,39, fPAR data 40, digital elevation model (DEM) data 41, and ERA5 Land data 42. The three objectives of this study are: (1) to develop a stack based downscaling model and combine it with terrain related variables; (2) Develop a long-term high-resolution SIF dataset (HCSIF) for China; (3) Evaluate HCSIF using SIF measurements from flux tower, OCO-2 SIF, GOME-2 SIF, and total primary productivity (GPP) tower base. We expect that this new dataset has the potential to monitor ecological processes, carbon cycling, and assess the impact of climate variability on agriculture and forestry at a fine scale.
The HCSIF dataset is generated using multiple features related to SR, LST, fPAR, meteorological factors, and terrain. We use a weighted ensemble learning algorithm consisting of three basic learners (CatBoost, RF, and GBDT) to construct a scaled down model. The HCSIF dataset significantly improves the spatial resolution of TROPOMI SIF, which helps to better represent spatial details. This enhancement leads to high accuracy in SIF and GPP validation, both at the site level and satellite observations.
This work is licensed under
CC BY 4.0 (Creative Commons Attribution 4.0 International License).
https://cstr.cn/31253.11.sciencedb.16910
10.57760/sciencedb.16910
