Predicting CO2 emissions using UAV spectral and thermal data in urban lawns: A case study of buffalograss and St. Augustinegrass

Limited research has examined the use of remote sensing to estimate carbon emissions in small-scale urban landscapes, such as residential green yards, which this study aims to address. In this research, multispectral and thermal images captured by an Unmanned Aerial Vehicle (UAV) were employed to estimate CO₂ efflux in two turfgrass species (‘UC Verde’ Buffalograss and ‘A-G’ St. Augustinegrass) grown under three years of varied irrigation treatments at the University of California, Riverside Agriculture Experiment Station in Riverside, California. Field measurements of CO₂ efflux, soil moisture, and temperature were collected in conjunction with UAV imagery. A physics-based model, the Thermal-Optical Trapezoid Model (TOTRAM), was first applied to estimate soil moisture. Subsequently, models developed using both in situ soil moisture and UAV data were compared with those based solely on multispectral and thermal vegetation indices. The TOTRAM model was found to estimate soil moisture with a root mean squared error (RMSE) of 0.07 cm 3 cm -3 . The highest performance was observed in random forest models that incorporated soil moisture, NDRE, the near-infrared (NIR) band, and canopy temperature, with RMSE values of 3.05 µmol CO₂ m -2 s -1 for models using predicted soil moisture and 2.69 µmol CO₂ m -2 s -1 for models using measured soil moisture, corresponding to a 13% difference in RMSE between the two models. Lastly, no improvement in performance was observed when models were trained on species-specific data. These findings demonstrate that UAV-based models can be effectively used to predict carbon emissions in small urban green spaces.