Soybean yield prediction is essential for optimizing agricultural management and supporting decision-making in the agribusiness sector. The lack of yield data at the farm level hinders the application of machine learning models for accurate estimation. This work proposes the use of domain adaptation techniques as a strategy to overcome this limitation, transferring knowledge from a model trained at the municipal level, where yield labels are publicly available, to farm-level predictions. The methodology employs time series of vegetation indices and climate variables derived from satellite data, processed by a deep learning model to capture patterns relevant to soybean yield. For evaluation, the results are compared with yields from previous harvests of farms with known data, allowing assessment of the estimation accuracy. This approach is expected to contribute to more efficient precision agriculture practices and support strategic decisions in agricultural planning.