Predictive modeling of target spot disease in cotton using environmental sensors and machine learning spore quantification

Overview

This solution addresses target spot disease caused by the fungal pathogen Corynespora cassiicola, which affects cotton and other crops. The platform integrates in-field environmental sensors with spore traps and applies machine learning image processing to rapidly and accurately quantify captured spores. By modeling spore density over time against temperature, moisture, and leaf wetness data, the system identifies environmental conditions favorable for sporulation and dispersal. The resulting predictive model can inform optimal fungicide timing and reduce unnecessary chemical applications, supporting more efficient and sustainable disease management.

Technical specifications

Core approach:

• Environmental sensors deployed within cotton fields capture temperature, moisture, and leaf wetness data continuously throughout the growing season
• Spore traps collect airborne C. cassiicola spores across multiple field locations and time points
• Microscope imaging of collected spores feeds into machine learning models for automated spore quantification, replacing labor-intensive manual counts
• Spore density is correlated with environmental variables to build predictive models of sporulation and dispersal favorability
• Validation combines visual spore counts and qPCR (a molecular technique that detects and measures pathogen DNA) confirmation to ensure quantification accuracy
• Once the model is established, field application may not require ongoing spore trapping, since environmental variables alone can drive predictions

Key benefits:

• Early detection of pathogen activity before visible symptoms appear, overcoming limitations of visual scouting during early epidemic stages
• Reduced fungicide use through data-driven application timing
• Standardized, scalable data collection across multiple environments

Technology readiness level

The program is at an early-to-intermediate research stage. Multi-location field experiments have already evaluated yield impacts of target spot, and controlled environment studies have identified optimal conditions for lesion development. Spore traps have been deployed in field studies and have detected variable spore densities over the season. Image analysis methods using machine learning have been implemented but have not yet been applied to spore counting. Next steps include deploying spore traps in ongoing field studies, validating machine learning spore quantification against visual counts and qPCR, and modeling spore density with environmental sensor data. This work represents the first stage toward a real-time spore detection and disease forecasting system.