Statistical estimate of the hourly near‐surface air humidity in eastern Canada in merging NOAA/AVHRR and GOES/IMAGER observations

Abstract

Estimates of the relative humidity near the ground are frequently requested by scientific communities concerned about weather forecasting, disease prediction, and agriculture. To face the dearth of meteorological observations provided by synoptic networks, remote sensing measurements are particularly useful, specifically because they can provide coherent information at a regional representative scale. The present investigation gives an update on the potential for using satellite data to estimate the near‐surface relative humidity. The IMAGER sensor on board the Geostationary Operational Environmental Satellite (GOES) is used to obtain the hourly infrared datasets. In addition, data from the Advanced Very High Resolution Radiometer (AVHRR) flown on the National Oceanic and Atmospheric Administration (NOAA) Sun‐synchronous satellite series is used to calculate the daily normalized difference vegetation index (NDVI). Estimates of the relative humidity are assessed using various variables like the surface temperature, NDVI, the precipitable water, the digital elevation model, the date and local time. The study approach combines empirically these variables into third‐order polynomial multiple regressions with stepwise functions. The data are split in two parts: the algorithm development dataset and the validation dataset. The estimation model is developed by a stepwise function, which selects independent variables and decides corresponding coefficients. The model validity is further assessed by employing a comparison with the results obtained from the model output using a validation dataset. The accuracy achieved using the validation dataset is in a good agreement with development dataset accuracies. The relative humidity accuracy derived from the present method is within 10% compared to field measurements. The largest discrepancies between model and measurements were observed over forested areas. One outcome from this study is that the difference in results between forested and non‐forested targets is enhanced with the topography.

Acknowledgments

This study was supported by funds from SOPFEU. Satellite data were kindly provided by Environment Canada. Ground‐based data were kindly provided by SOPFEU, Environment Canada, and the POMMIER network. Digital elevation model data were provided by Geomatic Canada. Finally, the authors would like to thank anonymous reviewers of this paper. Their comments and suggestions were appreciated to improve the paper quality.