Open access publications in the SLU publication database

One large difference between sensor measurements done in the laboratory and measurements done in-field is that in-field measurements are influenced by soil moisture. The soil moisture will vary between measurement times but also depending on where in the filed the measurement is done. Influence of soil moisture and variations in soil moisture are part of the reason for the often slightly worse predictions of e.g. soil texture or heavy metal content from in-field sensor measurements compared to measurements on dried soil in the laboratory.

The aim of the short project was therefor to study the possibility to use near infrared reflectance (NIR) spectroscopy for in-filed soil moisture predictions. In-filed predictions of soil moisture by a portable NIR instrument in connection with other sensor measurements could then be used to correct for soil moisture in other sensor measurements and thereby improve the predictions of soil texture, organic matter and heavy metal content without having to collect a lot of soil samples.

According to the results it is possible to use in-filed measurements with NIR spectroscopy to predict soil moisture. But texture has a large influence on both soil moisture and the NIR spectra and it is important to cover the variation in both soil moisture and texture in the calibrations.

Best predictions were achieved by using a model including only a small number of sampling sites, two, that covered the extremes in soil texture. At both sites 10 different soil moisture levels were included in the model. This is possible in theory, but not in practice if the models prove to be as specific as the results in this study suggests. This needs to be tested on more fields.

A more practical strategy could be to do local (field or farm wise) calibrations in connection with each senor measurement campaign. This means that a small number of soil samples would have to be collected for soil moisture analysis. However, since other sensors also often require some calibration with local samples this does not necessarily mean a lot of extra work. In this project 10 and 16 samples were used in this type of soil moisture calibrations. That was almost too few, but the three fields included in this study had a very large variation in soil texture with clay content from a couple of percent to over 60 percent and sand content form less than three percent to over 90 percent sand. With smaller variations it should be possible to keep the number of calibration samples low.