Synthetic Landsat data through data assimilation for winter wheat yield estimation

Abstract

Emergency Response Mechanism (ERM) for remote sensed image sequential deficiency is extremely crucial and the concerned researches are in urgent need. STARFM (Spatial and Temporal Adaptive Reflectance Fusion Model, STARFM) is such a fusion method for image sequences prediction, which is a sound theoretical and valid applicable approach. Gao et al. have successfully produced synthetic Landsat imagery with the corresponding approach, and Hiker et al. have applied it in coniferous study. However, STARFM algorithm has limit modeling power, for it has not fully considered disturbance events occurring within the observation process, meanwhile this paper does not supply an accuracy of synthetic images with variance indicators. Currently, it is still a pending issue addressing the stated estimation and uncertainty existence of sensor measurement values. In this paper, it is assumed that multiple sources data cross-use could minimize the observation error by effective blending technology. Under this assumption, we propose to modify the existing STARFM model, to optimize the predicted imagery through data assimilation theory, and the resulting synthesis image after iterations will be close to the true image. This paper proposes a practical approach of dynamic assimilated STARFM algorithm (DASTARFM), and supplies an application of crop yield estimation by remote sensing technology. In this paper, we mainly: (1)discuss the limits of the current concerned image synthetic researches; and (2) give a description of the practical approach which is featured as adaptive to the dynamic optimization, namely the dynamic optimized STARFM. Besides, (3) based on the new model-DASTARFM, we give the concrete implementation example to present the optimal estimated achievement of this model and illustrate its benefits over the typical algorithm according to the prior and post variances.