Reconstruction of Missing Meteorological Data Using Wavelet Transform

Land surface heterogeneity is effective on required sensor population of the regional or national telemetry networks used for agricultural information services. Temporary or permanently missing data sometime may occur in such large scale sensor networks due to type of failure. The agro-informatics measurement network (www.tarit.org) in Turkey is planned to have more than 30.000 sensors at 1200 stations. Agricultural risks, irrigation management and many other online services require continuity of the reliable data in real time. It is necessary to complete the missing data as close to its actual value until recovery in a fault tolerant acquisition system. A method is developed for recovery of missing data by using the correlation of wavelet coefficients of neighboring measurement stations. As different from similar solutions we reconstruct missing data through inverse wavelet transformation accompanied with regression model together. Mean square error (MSE) and mean absolute error (MAE) between the measured and reconstructed temperature and humidity patterns are used as performance measure. Mean square error (MSE) and mean absolute error (MAE) are seen to be reduced more than 26% for temperature data reconstruction with respect to pure linear regression case.