Several types of problems may occur in analysing observations of the sea surface temperature (SST) provided by different instruments. The measurements may be representations of different variables. In our case, the assumed real SST (in term of bulk temperature) is revealed by in situ measurements. Remotely-sensed images provide a biased SST contaminated by the atmosphere. Moreover the brightness temperature measurement of the sea surface is also questionable because of the temperature gradient in the skin layer. The lack of coincidence in the spatial and temporal locations of these two kinds of observations leads us to consider the problem from a geostatistical point of view. We present the method we used to monitor the NOAA satellite bias and merge in situ measurements and remotely-sensed observations of the SST. In order to perform a complete SST mixed chart, we use in situ data available from the Global Transmission System (GTS) and three NOAA-9 images observed during a time period of five days. First, we process a standard combination of infrared channels on the pixels recognized as cloud free. Secondly, we project in situ data on each image and assume no bias between the two data sets by adjusting a constant added to the multichannel algorithm. Thirdly, we merge the data in a unique file where the data, locations and origins are recorded. Finally, the SST is computed on a regular grid using linear interpolations. Such interpolations are provided by geostatistics and the technique is known as kriging.