The goal is to design, develop, and deploy new digital twin systems in regards to crop growth modeling for smart irrigation management. The motivation lies in our observation that, existing agriculture digital twin works rely on merely simulators with only coarse-grained field measurements, which suffer non-trivial model-to-reality gap. To fulfill this unique research need, we propose the CropTwin system with three core research tasks. First, we develop a cost-efficient IoT system to deploy limited sensors but predict more environment and crop parameters by designing new deep learning models. Second, we design an automatic digital twining with the assured attribute of fidelity (i.e., minimized model-to-reality gap) and synchronicity (i.e., day-to-day update). Third, we design new smart irrigation solutions enabled only by CropTwin, towards water conservation and yield improvement. We deploy and evaluate CropTwin in the real-world research farm at UNL, throughout a full soybean growth session. The long-term goal is to design applied digital twin systems for a wide range of agriculture scenarios, including but not limited to crop growth, insect control, vertical farming and livestock management. This interdisciplinary team is devoted to address real-world challenges in agriculture digital twins, in terms of computation complexity, cost efficiency and robustness, toward near-future farm deployment. The success of this project will enable us to take the first step in investigating agriculture digital twins in Nebraska, Midwest, and Beyond.