OpTeC Colloquium

OpTeC sponsors a colloquium throughout the academic year. This is a forum where a wide range of optical science and engineering topics are discussed at a level that can be understood by a broad spectrum of optics students, staff, and faculty. The primary purposes are to train students to give effective scientific talks and to promote cross-disciplinary interactions of students, staff, faculty, and local industry employees. Speakers include MSU students, staff, and faculty, and visitors from external universities, research labs, and companies.

November 20, 2025
4:10 pm
Norm Asbjornson Hall
Room 149

on the Montana State Univeristy Campus

Click here to join via WebEx

Download the flyer here

From Satellites to Soil: ML-driven Daily Crop Water Predictions

Abstract

Current farming practices increasingly rely on a variety of remote sensing platforms, including satellites, drones, and UAVs, to collect detailed information about crop health, weed detection, plant growth, and nutrient status. These observations support precision management applications such as targeted irrigation, fertilization, and pest control. However, such data are often collected intermittently, leaving gaps that limit applications requiring continuous monitoring, including variable-rate irrigation, water management, and environmental modeling.

This talk presents DFUSIONET, a machine learning framework that integrates multi-source data (satellite imagery, infield sensors, and meteorological measurements) to generate daily, high-resolution maps of plant evapotranspiration (ET), an indicator of crop water needs. A key component, Proportional-Offset Interpolation (POI), bridges temporal gaps between observations, producing a complete daily dataset for predictive modeling. The gap-filled data, combined with corresponding daily weather data, are then used to train deep neural networks to make predictions for future dates.

By combining satellite and infield optical measurements, DFUSIONET enables continuous, field-relevant monitoring, informs sustainable farming practices, and demonstrates the potential for integrating advanced data-driven approaches into operational decision-making. Future directions include spatially continuous variable-rate irrigation zone design and reinforcement learning-based optimization of field-scale water use.

Biographical Sketch

Farshina Nazrul Shimim is a Postdoctoral Researcher at the Nugent Precision Ag Lab, Montana State University (MSU), specializing in machine learning and AI for precision agriculture. During her Ph.D. in Electrical Engineering at the same university, she designed DFUSIONET, a multimodal machine learning (ML) framework that integrates satellite imagery, IoT sensor data, and weather forecasts for adaptive, decentralized irrigation and field management. She is now enhancing it with state-of-the-art architectures, including transformers, to enable AI-driven multi-field, variable-rate irrigation systems. She also works with the Montana Headwaters Tech Hub, coordinating with AgTech companies to assess their suitability for the MSU Testbeds and facilitating field trials that bring research innovations into practical use.

Past Speakers