A new method based on physical patterns to impute aerobiological datasets.

Area of Science:

• Aerobiology
• Data Science
• Environmental Monitoring

Background:

• Limited research exists on imputation method accuracy for aerobiological datasets.
• Accurate data imputation is crucial for reliable aerobiological analysis.

Purpose of the Study:

• To evaluate the effectiveness of Gappy Singular Value Decomposition (GSVD) for aerobiological data imputation.
• To compare GSVD performance against the traditional moving mean interpolation method.
• To identify factors influencing imputation accuracy in pollen datasets.

Main Methods:

• A simulation study using complete pollen data from two monitoring stations in northeastern Italy (2022).
• Random generation of missing data with varying proportions (5-25%) and gap lengths (3-10 days).
• Imputation of 4800 time series using GSVD and moving mean algorithms; accuracy assessed via Root Mean Square Error (RMSE).

Main Results:

• GSVD demonstrated comparable imputation accuracy to the moving mean method.
• GSVD exhibited strong generalization capabilities across different data types.
• Pollen variability and monitoring station location were primary drivers of imputation error, irrespective of the method.
• High pollen concentration variability and missing data distribution negatively impacted accuracy.

Conclusions:

• A novel data-driven imputation method (GSVD) was successfully introduced and validated for aerobiological data.
• The findings support the use of GSVD as a viable alternative to statistical imputation methods.
• Further research is needed to enhance imputation methods for improved aerobiological data reconstruction.