Integrating Artificial Neural Network-PROSAIL With Sentinel-2 to Monitor Crop Traits Dynamics and Nitrogen Status

In precision agriculture, remote sensing imagery plays a vital role in assessing the spatio-temporal variability of crop physiological components, including chlorophyll content (Cab). Determining such components allows evaluating crop nitrogen (N) status to provide fertilizer recommendations to ensure food security. This study investigates the efficacy of Sentinel-2 in estimating Cab and leaf area index (LAI) within a two-year winter wheat (Triticum aestivum L.) field experiment carried out in Aranjuez, Spain, encompassing various N and water regimes. We employed a hybrid artificial neural network coupling the PROSAIL model with Sentinel-2 images to retrieve Cab and LAI. A multiple linear regression model was constructed with these plant traits to estimate crop N status and grain N output (kg N ha⁻¹). The accuracy of the plant trait retrievals was validated using field data collected with various sensors and destructive samples across different growth stages. Furthermore, we assessed the predictions of Cab and LAI using Sentinel-2 time series. The accuracy of the proposed hybrid approach, and its relevance for precision agriculture were evident in its accurate estimation of key parameters: Cab (RMSE = 7.6 μg cm⁻²), LAI (RMSE = 0.97 m² m⁻²), nitrogen nutrition index (RMSE = 0.13) and grain N output (RMSE = 17.34 kg N ha⁻¹). Furthermore, it correctly tracked temporal fluctuations in Cab and LAI using Sentinel-2 images, meaning it can be used to optimize N fertilization rates. This method allows for the timely identification of N-sufficient and N-stressed wheat plots, particularly evident starting 100 days after the sowing date.