Detection and correction of defective relative humidity data collected from the greenhouse environment using nested Kalman filters with standard deviation analysis

Humidity sensors are critical in maintaining optimal conditions in greenhouse environments, but they are prone to data inaccuracies caused by sensor drift and severe environmental conditions. These inaccuracies, if left unaddressed, can lead to suboptimal crop growth and resource mismanagement. Detecting and correcting such faulty data is challenging, especially in the absence of labeled datasets to classify sensor readings as normal, defective, or severely defective. This research tackles this issue by proposing a novel method based on nested Kalman filtering to improve the accuracy of humidity sensor data in greenhouses.The proposed method integrates two layers of Kalman filters: an inner filter, which leverages data from neighboring sensors to generate predictions, and an outer filter, which refines these predictions using measurements from the evaluated sensor. To address severely defective cases, interpolated data from neighboring sensors is incorporated. A statistical analysis of the sensor data's standard deviation is employed to classify readings into three categories: normal, defective, and severely defective. For normal readings, the sensor data is directly used. For defective and severely defective readings, corrections are applied using the nested Kalman filters.Experimental results demonstrated the method's effectiveness in two scenarios: one where the sensor exhibited all three data cases and another with only normal and defective data. In these scenarios, the proposed method achieved substantial improvements, reducing mean absolute deviation (MAD) by 75.71% and 50.93%, and root mean square error (RMSE) by 66.41% and 41.80%, respectively, compared to the ground truth. These results highlight the method's capability to improve data accuracy even in challenging conditions where labeled data is unavailable.The findings of this research contribute to advancing sensor fault detection and correction methodologies by offering a robust and practical solution for improving data quality in greenhouse environments. The academic contribution lies in integrating nested Kalman filters with a statistical classification framework, which can be adapted for other sensor networks or domains requiring reliable environmental monitoring. This work has significant implications for precision agriculture, enabling more efficient resource management and optimized growing conditions, thereby enhancing productivity and sustainability.