Abstract

Soil condition mapping is a manual, laborious and costly process which requires soil measurements to be taken at fixed, pre-defined locations, limiting the quality of the resulting information maps. For these reasons, we propose the use of an outdoor mobile robot equipped with an actuated soil probe for automatic mapping of soil condition, allowing for both, more efficient data collection and better soil models. The robot is building soil models on-line using standard geo-statistical methods such as kriging, and is using the quality of the model to drive the exploration. In this work, we take a closer look at the kriging process itself and how its parameters affect the exploration outcome. For this purpose, we employ soil compaction datasets collected from two real fields of varying characteristics and analyse how the parameters vary between fields and how they change during the exploration process. We particularly focus on the stability of the kriging parameters, their evolution over the exploration process and influence on the resulting soil maps.