Improving wheat crop yield estimation by assimilation of remote sensing biophysical variable in the Simple Algorithm for Yield Estimation (SAFY) model : a case study of Spain

Details

Supervisors

Defourny, Pierre

Faculty

Faculté des bioingénieurs

Degree label

Master [120] en sciences agronomiques et industries du vivant, à finalité spécialisée: Geo-Information Science and Earth Observation for Environmental Modelling and Management

Abstract

Reliable crop yield estimation is fundamental for effective agricultural management and food security. However, conventional methods such as crop-cutting trials and household surveys are expensive, time-consuming, and labour-intensive. In this master’s thesis, we adopt the use of biophysical variables retrieved from high spatial and temporal resolution remote sensing data, particularly Sentinel-2 (S2), coupled with the Simple Algorithm for Yield estimation (SAFY) model to improve the estimation of wheat yield at the field level in the regions of Castile and Leon and Castile-La Mancha in Spain. Initially, we used meteorological time-series data to estimate wheat yield for rainfed, irrigated, and combined cropfields. However, our findings revealed that the model could not adequately capture the wheat phenology, leading to an underestimation of the yield. Due to the lack of correlation between the estimated and field-measured wheat yield, the Root Mean Square Error (RMSE) values of 1325 kg/ha, 3289 kg/ha, and 1722 kg/ha were obtained for the three model simulation cases. To address this limitation, we improved the model performance by assimilating S2 derived leaf area index (LAI) into the SAFY model to optimize six free parameters using the Differential Evolution (DE) algorithm. The results showed a moderate correlation and improved wheat yield estimation for rainfed (R2 = 0.292 and RMSE = 744 kg/ha), irrigated (R2 = 0.152 and RMSE = 1786 kg/ha), and combined cropfields (R2 = 0.378 and RMSE = 952 kg/ha). Overall, the research results demonstrate the potential of high-resolution remote sensing data in conjunction with a simple crop model for improving crop yield estimation at the field level.