Estimation of soil properties using a combination of spectral and scalar sensor data

Soil is a critical component of a production agriculture system and must possess favorable physical properties and fertility in order to be productive. Due to the importance of soil attributes, sophisticated methods and systems have been developed for their measurement. Furthermore, the availability of affordable global positioning systems (GPS) has extended the characterization of soil attributes to a site-specific basis. This management concept, referred to as precision agriculture, has demonstrated the high spatial variability of soil attributes and the need for field-mobilized sensors and systems for their measurement. The research reported here tests the effectiveness of an on-the-go near infrared (NIR) reflectance sensor for the measurement of a number of soil attributes. Two experiments that include data from seven different fields are used to demonstrate accuracy in predicting total carbon, total nitrogen, calcium, CEC, and moisture. Accurate predictions of pH and magnesium are also demonstrated when an improved reflectance system was used. Furthermore, the results indicate that the augmentation of NIR spectra with auxiliary sensors including electrical conductivity, temperature, and ion-selective pH electrode improved the prediction accuracy of many of the above attributes and was most effective when the calibrations span multiple fields. In addition, in order to create the calibrations, an automated method was developed to test the following approaches: multiple linear regression (MLR), multiple linear regression after principal components compression (MLRPC), partial least squares regression (PLS), and locally-weighted PLS (LWPLS). LWPLS is shown to create the most accurate calibrations in 9 of the 11 successful multi-field calibrations. Lastly, the calibrations were applied to the sensor data to create attribute maps for each of the seven fields investigated.