Adaptive Soil Moisture Estimation Using Performance‐triggered Model Reduction

This chapter develops an adaptive soil-moisture estimation framework that reduces the computational burden associated with Richards’ equation-based field models. It motivates the need for model reduction in large-scale agro-hydrological systems and highlights that a single reduced model may not adequately capture the range of conditions encountered during a growing season. To address changing field conditions, the chapter proposes a performance-triggered adaptive model reduction strategy in which prediction error is monitored and the reduced model is updated when the prediction error exceeds a specified threshold. Reduced-order models are constructed using a structure-preserving, clustering-based approach and are integrated into an adaptive reduced-order extended Kalman filter for soil moisture estimation. The framework is evaluated in a large-field case study and is compared against nonadaptive and time-triggered alternatives, where it demonstrates an improved balance between estimation accuracy and computational efficiency.