Volume 76 (2026) Issue: 2026 No#2

Deep learning-based forecasting of heat stress events and daily milk yield depreciation in dairy cattle using meteorological data: an in-silico study

Author(s): Ibrahima Mahamane Abdourhamane

Keywords:Precision Livestock Farming; Heat Stress; LSTM; Digital Twin; Milk Yield Prediction; Dairy Cattle.

Climate change is a serious threat to food security as heat stress compronises animal welfare and production. Traditional statistical models struggle to capture these lagged and non-linear dynamics, limiting proactive herd management. Therefore, the aim of this study was to develop an early warning prediction system for milk yield losses due to heat stress in dairy cows by explicitly modeling physiological delay effects using Long Short-Term Memory (LSTM) based deep learning in a Digital Twin environment. For a in silico approch, a high-fidelity digital twin dataset was created to simulate a herd of 500 Holstein-Friesian cattle over a three-year period (1,095 days). The present study involved calculating the Temperature–Humidity Index (THI) by integrating biologically based response functions with time-series meteorological data, and simulated the associated physiological stress responses. A recurrent neural network model based on the LSTM architecture was trained on 80% of the time-series dataset to model the nonlinear temporal relationships between ambient conditions and milk yield. Model performance evaluation demonstrated strong predictive capability, with a root mean squared error (RMSE) of 1.48 kg/day and a coefficient of determination (R²) of 0.81. Correlation analysis further revealed a strong negative association between THI and milk production (Pearson’s r = −0.76, p-value < 0.001). The model also successfully detected the early onset of heat stress and captured the biological lag effect, as evidenced by its accurate prediction of seasonal declines in productivity. Overall, these findings, derived from a controlled simulation environment, support the potential applicability of LSTM-based frameworks as early warning systems to guide proactive mitigation strategies and reduce heat stress–related milk yield losses in dairy cattle. However, further validation under real farm conditions is necessary before practical implementation.