This paper is concerned with the recovery of Leaf Area Index (LAI) time series in intense agriculture areas from moderate resolution remote sensing data (MODIS or SENTINEL). Although their resolution limits an analysis at a parcel level, their high temporal rate enables to monitor land use/land cover through the temporal evolution of key biophysical parameters as LAI. However in practice, frame-by-frame estimation is unsatisfactory since the quality of each single data is subjected to undesirable effects due to atmosphere disturbance, sun geometry, viewing geometry, etc. These effects lead to a lack of temporal consistency of resulting time series. The reconstruction of such time series is delicate using conventional interpolation methods since underlying physical processes are not taken into account. In this paper, we tackle this issue by exploiting the prior information of a plant growth model, namely GreenLab, using stochastic data assimilation techniques. Our experiments on challenging situations, such as few data and fragmented landscapes, demonstrate the approach is robust on various challenging situations and enables to extract additional information about observed fields..