AI-Based Sugarcane Yield Prediction and Smart Irrigation Optimization Using Sensor and Weather Data

This paper presented an integrated system for sugarcane yield prediction and autonomous farm management, distin-guished by a LangGraph AI agent with a rule-based validator node, a Marathi-first PWA, and an end-to-end IoT pipeline verified through live deployment. The system demonstrated 28% water savings under furrow irrigation, reduced daily farmer management time from 2.5–3 hours to under 45 min-utes, and produced actionable yield projections (XGBoost CV R2 = 0.72 ± 0.064, RMSE = 7.96 t/ha). Planned future work includes: •  Physical valve actuation - connecting the relay module to a solenoid valve to complete the physical control loop. •  LoRa transport - replacing WiFi HTTP with a LoRa/LoRaWAN link for reliable coverage through dense sugarcane canopy and multi-field deployment. •  ESP32 NTP - replacing server-side timestamp override with on-device NTP synchronization for true edge times-tamping. •  Multi-crop and multi-farm deployment - extending the PWA’s crop selection screen and retraining the yield model on diverse Maharashtra varieties and field data. •  Soil nutrient and multispectral sensors - adding EC, pH, and NPK probes for direct nutrient measurement; integrating low-cost multispectral cameras for canopy health assessment. •  Web Push notifications - replacing 30-second polling with server-sent push for lower latency and battery-friendly background alerts. •  On-device LLM (Gemma) - running a lightweight LLM on the Pi for offline fallback when the Cloudflare tunnel is unavailable. •  Market price APIs - integrating APMC mandi data for harvest timing recommendations that account for current sugarcane prices. •  Marathi/Hindi voice-first redesign - extending the cur-rent voice input/output to a full conversational interface for farmers with limited smartphone literacy. The vision is an AI agent that serves as a tireless, trans-parent farm partner: monitoring around the clock, anticipating problems, acting decisively, logging every decision with its reasoning, and communicating clearly in the farmer’s own language.

[1]     S. S. Kale and P. S. Patil, “A machine learning approach to predict crop yield and success rate,” in Proc. IEEE PuneCon, 2019, pp. 1–5. [2]     S. Kumar, M. Pant, and A. Nagar, “Forecasting the sugarcane yields based on meteorological data through ensemble learning,” IEEE Access, vol. 12, pp. 176539–176553, 2024. [3]     J. M. Lemus, D. Mera, J. L. Queme´ de Leo´n, and J. M. Cotos, “Imple-mentation of long-term forecasting models in sugarcane for agricultural planning and yield goals setting,” IEEE Access, vol. 13, pp. 213499–213507, 2025. [4]     W. Pituckwanich et al., “The implementation of a prediction system for sugarcane’s destruction rate from sugarcane stem borer via hybrid machine learning,” IEEE Access, vol. 13, pp. 45594–45608, 2025. [5]     P. Elkoury, R. Awad, T. Atallah, and S. Kesrouani, “IoT-enabled smart drip irrigation system using ESP32,” IoT, vol. 4, no. 3, pp. 236–253, 2023. [6]     M. K. Rao, M. S. Kumar, M. Jaijaivenkataramana, and Ch. S. Sowjanya, “ESP32 based irrigation system,” in LNNS, vol. 545, Springer, 2023, pp. 471–479. [7]     S. R. Prathibha, A. Hongal, and M. P. Jyothi, “IoT based monitoring system in smart agriculture,” in Proc. ICRAECT, IEEE, 2017, pp. 81–84. [8]     M. Chapale, P. Bhosale, Y. Gaikwad, and A. Maurya, “Data-driven system for sugarcane yield optimization in Maharashtra,” MIT ADT Univ., Tech. Rep., 2026. [9]     M. F. Manzoor, “A review of machine learning techniques for precision agriculture and crop yield prediction,” Premier J. Plant Biology, vol. 1, p. 100005, 2024.