Figures (2)  Tables (1)

    • Figure 1. 

      AI-empowered roadmap for sustainable agricultural nitrogen management. By integrating AI-driven perception, process-based modeling (PINNs/KGML), and actionable intelligence, the framework enables precision nitrogen management across scale, and transition from a decoupled crop-livestock system to a sustainable, circular N economy.

    • Figure 2. 

      Framework of the AI-Driven N management agent. The bottom layer continuously harvests multimodal, cross-scale data from crop, livestock, and waste nodes via in-situ sensors, UAVs, computer vision, and IoT (Internet of Things) networks. The middle layer serves as the mechanistic core, coupling traditional biogeochemical models with KGML to rapidly simulate N fluxes while strictly enforcing physical mass-balance constraints. Finally, the top layer acts as the cognitive engine. A master reasoning engine utilizes retrieval-augmented generation (RAG) and verified literature databases to synthesize model outputs and mitigate AI hallucinations.

    • Model name Core AI technology/modality Core applications in N management Key advantages
      AgroMind Multi-modal remote sensing + deep learning Regional crop N status mapping; canopy N concentration/deficiency detection High-resolution N sensing; links remote sensing data to N-cycle dynamics
      AgroLLM Text-based LLM + agricultural knowledge graph N fertilizer prescription Q&A; translates biogeochemical models to farm-level N management advice User-friendly natural language interface; professional N stewardship support
      AgroGPT Vision-language model (VLM) + agricultural computer vision Crop N-deficiency visual recognition; multi-modal agricultural scene interpretation for
      N stress      		 Accurate subtle N-deficiency detection; integrates visual/textual N knowledge
      Sinong (司农) Open-source agricultural LLM + federated learning Smallholder-friendly N fertilization guidance; democratized agricultural N knowledge access Low technical barrier; data privacy protection via federated learning
      Shennong 3.0
      (神农)      		 Text & reasoning LLM + 4R nutrient stewardship framework Site-specific N fertilization decision-making; dynamic optimization of N source/rate/time/placement Localized for Chinese agriculture; integrates soil/meteorological multi-source data
      EarthGPT Remote sensing foundation model + vision-language model (VLM) Agricultural landscape N mapping; soil nitrate level inference; large-scale N-cycle pattern extraction Strong object recognition; processes unstructured spatiotemporal remote sensing data
      Nutrient Expert Smart Fertilization LLM LLM (DeepSeek) + high-throughput soil sensing + knowledge distillation Personalized precision N fertilization; soil N deficiency diagnosis Higher recommendation efficiency; 18%–22% N fertilizer reduction (6%–14% yield increase)
      IMAP Comprehensive planting model + hybrid AI-optimization algorithm Whole-chain N nutrient management; N-fertilizer-crop matching for large-scale farming Improves NUE; integrates the entire agricultural production value chain
      PigGPT LLM + multi-source data fusion + 3D digital human interaction Pig feed N formulation optimization; manure N recycling; regional pig farming N emission monitoring/early warning; crop-pig N cycle integration Accurate N demand prediction; low operation threshold; boosts crop-livestock NUE
      Smart Layer Chicken Model S1 LLM + computer vision + IoT sensor fusion + precision feeding algorithm Layer feed N optimization; chicken house NH3 monitoring; poultry manure N utilization; orchard-chicken N cycle integration 12%–18% feed N reduction; 20%–25% N fertilizer reduction
      Suwu Smart Sheep Raising Model
      (苏武)      		 LLM + satellite remote sensing + grazing N-cycle model + precision supplementary feeding Grassland N monitoring; sheep feed N optimization; manure N return-to-grassland; grassland-sheep N balance 10%–15% supplementary feed N reduction; 18%–22% grassland N fertility restoration
      AI4DLLM Ruzi Niu Model (孺子牛) Domain-specific LLM + agricultural knowledge distillation + federated learning + edge computing Multi-source N data integration; lightweight N model development for edge devices; N management scheme generation Supports edge deployment; low threshold for N data analysis

      Table 1. 

      Summary of AI models in agricultural nitrogen management