Prototype Bayesian Belief Networks for Validating Soil Moisture Telemetry In Sensor-Based Irrigation Management for ‘Beauregard' Sweetpotato

Information from remotely accessible sensors is becoming an integral component in knowledge-based agricultural decision-making. Management decisions, such as the timing and duration of irrigation, are important from the point of view of efficiency and in matching soil moisture availability with the phenological requirements of the crop. Hence the validity of sensor readings is important in making informed decisions. We present prototypic Bayesian belief network (BBN) models for validating remotely-accessed sensor-based soil moisture information used for irrigation management decision-making in sweetpotato cv. ‘Beauregard.’ The prototype validation models were learned from on-site agroclimatic data collected in Chase, LA, during the 2010 growing season. This reduced data set represented air and soil temperature, volumetric water content, relative humidity, and rainfall data collected during the growing season where the decision to add supplemental irrigation was considered a priority if the soil moisture approached 12% VWC at the 15 cm depth. To simulate the ability of the models to detect presumptive soil moisture sensor faults, VWC readings were changed to 25% and 50% of its original values. Two measures were used to evaluate the suitability of the validation models. First, we evaluated the model’s ability to detect if a decision to irrigate was based on erroneous VWC reading considering evidence from other sensors. Second, we evaluated the model’s failure to detect information needed to trigger supplemental irrigation.  We will outline how the candidate BBN models were developed, present a summary of the validation results, and outline the potential usefulness and limitations of this validation approach within the context of a model-based sweetpotato production management.