A previously unknown grave field containing 16 grave mounds has recently been discovered by automatic pattern recognition software developed by the Norwegian Computing Center. Covered by spruce forest, the grave field is impossible to discover from aerial photography.

The automatic method scans systematically for heaps in airborne laser scanning (ALS) data. The ALS data contains three-dimensional points, each labeled as either vegetation, ground, building, or other. By only keeping the ground points, a very detailed digital elevation model (DEM) is obtained, in which the forest vegetation has been removed. This makes it possible to see the grave mounds even in dense forest. The automatic method slides a model heap over all positions in the DEM, and assigns scores to anything resembling a heap. By using heaps with gradually increasing diameters from 4 to 15 meters, the method is able to detect all heaps in this range. The integer scores range from 0 to 6.

The method developed by the Norwegian Computing Center has searched for grave mounds in large parts of Larvik municipality, Vestfold County, Norway. This has included many well known iron age grave fields, and the grave mounds are now more precisely geo-referenced by the new method. But when the method made a series of detection with high score values within a 40 by 200 meters area, archaeologist Christer Tonning of Vestfold County Administration was puzzled. This was not an area in which he would have gone looking for grave mounds. But the automatic detections led Tonning to 14 grave mounds, the largest discovery of grave mounds in Norway in many years. 150 meters further north was another pair of grave mounds, which were covered by dense forest, and difficult to spot when walking in the forest, so they would not have been discovered without the automatic method.

The automatic method was originally developed by the Norwegian Computing Center for ALS data with at least five emitted pulses per square meter on average, and has been verified to work on known grave fields with such high resolution data. By accident, it was then attempted on a dataset with only one emmited pulse per square meter, which led to the surprising detection. With this encouraging result, Tonning has urged the Norwegian Computing Center to run the automatic detection method on more datasets with similar resolution.