The operational global Numerical Weather Prediction (NWP) model of the Japan Meteorological Agency (JMA) is known to have such systematic errors that humidity in the middle troposphere is getting lower as the forecast time proceeds, and that there are less cloud fractions around the polar region compared to satellite observations and reanalysis data. In this study, we investigate which physical processes cause these systematic errors by using the data of special observations. In the T-PARC 2008 experiment, many dropsondes were deployed near tropical cyclones over the tropical and subtropical regions of the western North Pacific region, where atmospheric convections work a lot. In the Concordiasi project, many dropsondes were dropped around the Antarctica, where most of the clouds are in mixed and ice phases. These data are valuable, because in normal condition there are only a few observations at such places as surroundings of tropical cyclones, ocean, sea ice and the Antarctic. By using these data, we especially focus on the vertical profile of the boundary layer and the transport of heat and moisture from the boundary layer to the free atmosphere. Then, the specific issues of the physical processes in the model that lead to the systematic errors will be focused, and the remedies to cope with these problems will be discussed.