AI-Driven Yield Forecasting & Predictive Analytics for Smart Greenhouses in Nepal

IoT-enabled microclimate monitoring combined with LSTM neural networks to forecast bell pepper and strawberry yields under variable Himalayan conditions.
Experimental Methodology & Laboratory Protocol
ESP32 microcontrollers measuring soil electrical conductivity, ambient humidity, and photosynthetically active radiation (PAR), streaming telemetry to a FastAPI backend.
Key Discoveries & Research Outcomes
- Predicted harvest yield dates within a 48-hour error window 14 days in advance.
- Automated drip-fertigation cut water consumption by 28% without biomass loss.
- Deployed functional prototype currently operating in Bhaktapur agricultural nursery.
Full Publication Analysis
Precision agriculture presents a transformative opportunity for smallholder greenhouse farmers across Nepal. By combining affordable IoT sensor nodes with recurrent neural networks, growers can mitigate adverse climate fluctuations and synchronize marketing with peak demand.
Architecture
Sensors record soil moisture, ambient vapor pressure deficit (VPD), and temperature every 5 minutes. Telemetry is ingested into a lightweight edge container that calculates daily water deficits and triggers automated micro-drip cycles.
Er. Subash Karki
Faculty of Information Technology
Er. Subash Karki, Sandeep Karki, Manish Rijal
https://doi.org/10.1016/j.compag.2026.10771