Multiseasonal analysis of rice crop yield prediction with Sentinel-2 time series and UAV imagery in Lambayeque (Peru)
Keywords: Time series, multispectral imaging, NDVI, yield prediction, precision agriculture
Abstract. Accurate crop yield prediction is crucial for efficient agricultural and socio-economic management. Remote sensing, using satellite imagery and unmanned aerial vehicles (UAVs), provides an effective approach for yield prediction and crop monitoring, allowing time series of vegetation indices such as NDVI to be obtained and facilitating detailed analysis of crop phenology. In this study, multi-seasonality in rice yield prediction is analysed using NDVI time series obtained from Sentinel-2 (S2) and UAVs in the Lambayeque region, Peru. NDVI from S2 was extracted by applying scene classification map (SCM) masks to remove clouds and shadows. A total of 7 and 11 UAV flights were conducted during the growing season for 2022 and 2023, and yield was collected mechanically in 35 rice-producing plots. The results showed an overestimation of NDVI values obtained by UAV compared to Sentinel-2 values, as well as a significant difference in yield prediction between 2022 and 2023. In 2022, by integrating S2 and UAV NDVI series, a coefficient of determination (R2) of 0.66 was obtained for the combination of UAV and S2, higher value than those obtained with UAV or S2 independently, with a root mean square error (RMSE) of 1.096 t/ha and a %RMSE of 10.36. In 2023, a R2 of 0.32, a RMSE of 0.85 and a %RMSE of 9.21 were achieved. This difference is interpreted as a consequence of the cyclone Yaku, which caused rainfall and damage to the irrigation infrastructure, and fungi disease, leading to water stress and a decrease in yield, highlighting the importance of considering the meteorological conditions in the development of yield predictions based on NDVI series metrics obtained along the campaign.