Multi-temporal data fusion techniques for archaeological prospection in Poland (Sławsko case study

Multi-temporal data fusion techniques for archaeological prospection in Poland (Sławsko case study) Dominik Ruciński1, Łukasz Banaszek2,3, Sławomir Królewicz4, Włodzimierz Rączkowski2 1Wasat

sp. z o.o., ul. Żurawia 22, 00-515 Warszawa, Poland 2Institute of Prehistory, Adam Mickiewicz University in Poznań, ul. Umultowska 89D, 61-614 Poznań, Poland 3Ludwig Boltzmann Institute for Archaeological Prospection and Virtual Archaeology, Hohe Warte 38, A-1190 Wien, Austria 4Institute of Physical Geography and Environmental Planning, Adam Mickiewicz University in Poznań, Dzięgielowa 27, 61-680 Poznań, Poland

Research

Introduction

Test site

Enhancing the effectiveness of satellite images for archaeological heritage detection, understanding and protection is one of aims of the ESA-funded project ArchEO – Archaeological applications of Earth Observation techniques.

The method has been tested on Early Medieval stronghold in Sławsko, Pomerania (Poland).

The purpose of the data fusion methods testing was to increase the readability of grassmarks showing up the remains of an Early Medieval stronghold in Sławsko as well as its vicinity (including natural and cultural features) for better understanding of the past landscape.

For the purpose of the ArchEO project, thirteen scenes of Sławsko stronghold acquired by four different satellite sensors were obtained. For further analysis eight scenes with visible marks of the stronghold were selected (Table 1).

The stronghold played an important role in the 11th and 12th centuries as a seat of regional power. It was located on the bottom of Wieprza river valley. Currently, the embankments are almost completely levelled and the place is used as meadow.

Next, multispectral scenes were fused with the use of panchromatic images acquired in different time periods. The following algorithms were analysed: Gram-Schmidt, Nearest Neighbour Diffusion, Principal Components, HSV, Brovey, Pendock, Paris, Radiometric Control Set, Local Mean and Variance Matching, Simple Mean and ESRI.

According to our experience gained within the project, the ground resolution seems to be the most important limitation of successful application of satellite data for archaeological purposes (at least in temperate climatic and vegetation conditions in Central Europe). In order to overtake this limitation we moved to the concept of data fusion.

Compilations of results given by different data fusion algorithms were prepared for each multispectral image (see example on Figure 2). The visibility of the stronghold’s structure on individual fused datasets was analysed and compared by archaeologists involved in the project.

In most cases, the processing is limited to increasing resolution of multispectral channels with information contained in a panchromatic image of the same satellite scene (pansharpening). In this paper, we modified that approach by using various data fusion algorithms for data acquired in different time periods by different satellite sensors.

Sensor

Date

Figure 1. Location of the Sławsko stronghold and cropmarks visible on the pansharpened Pleiades image (29/03/2014).

Results and conclusions • The fusion of RapidEye scenes with panchromatic channel from different images significantly improves readability of the former.

Spatial resolution

• The use of RapidEye’s red edge channel leads to the most accurate identification of archaeological features.

Panchromatic - 0.5 m Multispectral - 2 m Multispectral - 5 m Panchromatic - 0.5 m Multispectral - 2 m

GeoEye-1

05/07/2011

RapidEye

17/03/2012

Pleiades

29/03/2014

RapidEye

24/06/2014

Multispectral - 5 m

RapidEye

04/07/2014

WorldView-2

09/08/2014

RapidEye

03/08/2015

Pleiades

14/08/2015

Multispectral - 5 m Panchromatic - 0.5 m Multispectral - 2 m Multispectral - 5 m Panchromatic - 0.5 m Multispectral - 2 m

• Multi-temporal data fusion, based on matching panchromatic and multispectral channels from different dates, increases readability of archaeological features on panchromatic image, however it does not contribute to recognition of new objects. • Gram-Schmidt and Principal Components algorithms implemented in ENVI gave the most readable images. • Nearest Neighbour Diffusion algorithm requires precise line up of rasters, which is difficult in case of fusion of data from different dates and sensors. • A very precise rasters’ matching is crucial for identification of small archaeological objects, which are characteristic and most common in Poland.

Table 1. Satellite data used in the research.

• Ground conditions have the strongest impact on archaeological features visibility. 2

3

4

5

6

7

8

9

Acknowledgements The project Archaeological application of Earth Observation techniques (ArchEO) is financed by European Space Agency under the Polish Industry Incentive Scheme.

Figure 2. 1 – RapidEye (MS; 4/7/2014); 2 GeoEye-1 (Pan; 5/7/2011); 3 - Pleiades (Pan; 29/3/2014); 4 - WorldView-2 (Pan; 9/8/2014); 5 - Pleiades (Pan; 3/8/2015); 6, 7, 8 and 9 – results of data fusion with use of Gram-Shmidt algorithm.

1