Global-scale improvement of the estimation of terrestrial gross primary productivity by integrating optical and microwave remote sensing with meteorological data

Abstract

Photosynthesis (a key ecological process) is measured based on gross primary productivity (GPP), emphasizing the criticality of accurate GPP estimation to climate change research. The extant remote sensing-based approaches for GPP estimation were typically based on optical remote sensing data, neglecting the potential supplementary information from microwave remote sensing data. Thus, based on the random forest algorithm, we developed a GPP model through the integration of optical and microwave remote sensing with meteorological data (OMM-GPP). The software and tools used for data processing, modeling, and analysis are the standard and third-party libraries based on the Python Language. Our OMM-GPP model was trained and validated using GPP data (referred to as “observed GPP”) retrieved from carbon dioxide fluxes measured at 137 flux towers. The results indicated that GPP estimation by integrating optical and microwave data with meteorological data was more accurate than GPP estimation by integrating single-source remote sensing data (optical or microwave data) with meteorological data across eight vegetation types. The model performed well on daily and monthly scales, with determination coefficients (R2) (root-mean-square errors, RMSE) of 0.85 (1.52 gC m−2 d−1) and 0.83 (1.49 gC m−2 d−1), respectively, which increased (decreased) by 0.17–0.03 (0.58–0.12 gC m−2 d−1) and 0.11–0.02 (0.39–0.10 gC m−2 d−1) compared with the R2 (RMSE) values obtained by integrating single-source remote sensing data with meteorological variables. Further, the OMM-GPP model effectively captured the seasonal variations in daily and monthly GPP across most vegetation types. The eight-day-scale comparison of the model with the VODCA2GPP dataset revealed its enhanced performances, increasing R2 by 0.33 and decreasing RMSE by 0.97 gC m−2 d−1. Overall, the integration of microwave and optical remote sensing data with meteorological data can enhance GPP estimation accuracy, as demonstrated by the established OMM-GPP, across different vegetation types.

Keywords

• Gross primary productivity
• Microwave vegetation index
• Time series
• Machine learning