Organically grown databases

The creation and continued re-creation of the institutional archaeological databases of the regional archaeological museums in Norway has been described as ‘organic’ by some (Axelsen, Reference Axelsen2021: 57). This, in part, reflects how their development followed the design implemented when the databases were established in 1988 (Matsumoto & Uleberg, Reference Matsumoto, Uleberg, Piotrowska and Konieczny2015: 159). After two digitization projects in the 1990s, the joint university museums’ IT organization (MUSIT) was established in 2007, including both natural and cultural-historical museums, and ended in 2021 (Matsumoto & Uleberg, Reference Matsumoto and Uleberg2021). It was the result of a long-held and shared belief in a common national standard for the storing and sharing of scientific records, making them accessible and usable by researchers and the public alike (Uleberg & Matsumoto, Reference Uleberg and Matsumoto2009: 1–2; Matsumoto & Uleberg, Reference Matsumoto, Uleberg, Piotrowska and Konieczny2015: 159). Building on this work, the operation and maintenance of the cultural-historical university museums’ various collection databases is, for now, ensured through UniMus:Kultur (Uleberg et al., Reference Uleberg, Matsumoto, Pantos and Bonelli2023: section 1).

Data fields are largely identical in all five databases. Yet, despite a common standard being the goal, normalization and standardization of the individual records across and within museum regions have not been a priority (Uleberg & Matsumoto, Reference Uleberg and Matsumoto2009: 3; Uleberg et al., Reference Uleberg, Matsumoto, Pantos and Bonelli2023: sections 2 and 3). Many fields are free-text entries, with as little or as much text as a data recorder wants to enter and in a manner that suits them best. Consequently, fields that should be standardized, e.g. information such as the year a find was made or its coordinates, is not. Researchers who want to use the data thus have to spend time on extensive data cleaning after exporting what they need from the artefact databases.

Digital data floods of ‘human’ big data

Archaeological research increasingly relies on digital data and methods (e.g. Huggett et al., Reference Huggett, Reilly and Lock2018; Huggett, Reference Huggett2020: 15). Databases are rapidly growing, and seem even less likely to de-accession potentially useful digital records than their physical counterparts. While most physical and digital archaeological records have so far been collected by professionals, private finders have also contributed to these collections since the creation of the Norwegian archaeological university museums (Shetelig, Reference Shetelig1944). The situation is similar in many European countries, and the proportion of finds reported and handed in by non-professionals is increasing as the popularity of hobby metal detecting is growing.

There have been at least two clear and steep increases in the quantity of recorded finds in Norway (Matsumoto & Uleberg, Reference Matsumoto and Uleberg2021: fig. 3). The first is the result of extensive burial mound excavations from 1870 to 1880. The second is a consequence of an increase in large and mandatory excavations after 2000 and the dramatic influx of reported hobby metal-detected finds after 2014 (see Axelsen, Reference Axelsen2021). This ‘flood’ of digital data is not unique to Norway, and has been referred to as a ‘data deluge’ in England (e.g. Bevan, Reference Bevan2015; Cooper & Green, Reference Cooper and Green2017) and overlaps with a long-standing concern over a ‘curation crisis’ within the field (see, e.g. for Sweden, Friberg & Huvila, Reference Friberg and Huvila2019: 364–66, with references). This is making existing issues concerning archaeology's data management practices even more pressing. Among these are a lack of standardization, both within a country and between them, and a common ontology that would make it easier to aggregate the many datasets that are available through infrastructure systems such as ARIADNEplus (Green, Reference Green, Gillings, Hacigüzeller and Lock2020: 433; see also Dobat et al., Reference Dobat, Deckers, Heeren, Lewis, Thomas and Wessman2020: 276).

Archaeological collections in general contain a multitude of different materials, such as various forms of grey literature, samples, photographs, field diaries, and physical remains. The same is true for the records from hobby metal detecting. We suggest that the digitized information on these finds constitute what Green (Reference Green, Gillings, Hacigüzeller and Lock2020: 432–33) has referred to as ‘human’ big data. In contrast to ‘big data’, human big data is not defined by its size, but rather by the opportunity it provides to aggregate multiple and varied datasets and search across them (see also Boyd & Crawford, Reference Boyd and Crawford2012: 663). Referring to Gattiglia (Reference Gattiglia2015: 114), who sees ‘big data’ as a concept, Green considers that, while ‘big data’ is high in volume, velocity, and variety, these factors are relative rather than absolute.

Building on this, Green (Reference Green, Gillings, Hacigüzeller and Lock2020: 433) offers the following useful definition of ‘human’ big data as ‘datasets that are too complex and/or large to process without the use of computer algorithms/scripts’. He adds that these datasets can also be analysed ‘in an exploratory manner’, with hypotheses deriving from the analyses rather than the other way around. This is similar to McCoy's description of the process of ‘knowledge discovery’, a term from the field of data science; the aim is to use quantitative and/or computational methods to find, for example, ‘new patterns’ (McCoy, Reference McCoy2017: 76). Working with this type of data requires embracing a certain level of ‘messiness’ (Gattiglia, Reference Gattiglia2015: 114). The degree of ‘mess’ can, however, be lessened by taking several precautionary steps, as outlined by Green (Reference Green, Gillings, Hacigüzeller and Lock2020: 433). These are: 1) plan ahead; 2) explore the history of the existing datasets; 3) document the quality of the datasets; 4) establish common ontologies; 5) use big data analytics to identify anomalies.

Although hobby metal-detected finds in Norway are, for now, hardly ‘big data’, strictly defined by a set quantity of tera- or petabytes (McCoy, Reference McCoy2017: 76), they do fit the definition of ‘human’ big data. To make it possible to gain quantitative and qualitative information across a growing number (and sizes) of datasets on a national, European, and even global scale, the precautionary measures suggested by Green are, we believe, a necessity (see also McCoy, Reference McCoy2017). Doing this in close cooperation with others—including knowledge exchanges and discussions of the process—is clearly beneficial, allowing data to be ‘things made’ rather than ‘things given’ (Huggett, Reference Huggett2020: 9).

Collecting and analysing the national metal detecting record

Our dataset consists of records collected on three separate occasions: 12 August 2019, 27 September 2021, and 14 March 2022. These dates are random, unconnected to the flow of incoming reported metal-detected finds. They are nonetheless given, so that any significant changes in the datasets connected to these dates can be scrutinized. The two first searches were conducted individually and joined in a combined database. The searches undertaken in 2021 and 2022 were more targeted than the first, as the recording of metal-detected finds seems to have become fairly standardized across the museums in the last five years (Table 1). We had both conducted several extensive and separate searches in the five archaeological artefact databases, leading up to the three main searches that form the foundation of our combined Access database and the analyses presented in this article. The search criteria used to identify the metal-detected finds are summarized in Table 1 (for how the search criteria were determined, see Axelsen, Reference Axelsen2021: 172–73; Fredriksen, Reference Fredriksen2023: 61–69).

Table 1. Combined results of database searches.

When working with digitized (or ‘born-digital’) archaeological data, considering and treating the information as ‘human’ big data is potentially fruitful. It can be difficult to find and use different types of archaeological data in rapidly growing databases where the information they contain is not always standardized or normalized (Axelsen, Reference Axelsen2021: 54–55, 212). Open and exploratory qualitative and quantitative searches and analyses, we suggest, are a practical method for identifying, gathering, and making use of the material. Additionally, such exploratory searches are especially useful when undertaken by different people, with the aim of later combining the results and achieving ‘knowledge discoveries’. It requires time and effort, but, when well documented, the process can help improve both the current and future data quality and the potential of archaeological databases, and thus the potential for aggregating them.

Assumptions from archaeological theory always precede data collection and analysis, and indeed analysis will be constrained by the theoretical constructs applied during the recognition, categorization, and collection of the data (Huggett, Reference Huggett2020: 14). Being open and clear about the many factors that influence the chosen data are vital (e.g. Lock, Reference Lock2009: 82).

Relating to this, although our searches can be replicated by using the data provided in Table 1, it should be noted that some museums transfer partially catalogued records from their accession database to their artefact databases. Hence, the data in both databases may be deleted later when an artefact is deemed not to qualify as ‘official heritage’ and consequently is not included in the archaeological collections. Because of this, and the fact that some finds are handed in to the museums several years after they were discovered by hobbyists, the reported and recorded number of metal-detected finds will always be an estimation. Even though the number of finds in some areas means that many of the large-scale trends are unlikely to change significantly, others may look quite different within a relatively short time span.

On 12 August 2019, the database contained 10,277 unique records. On 27 September 2021 it was 15,274, and by 14 March 2022 it had reached 16,948. The increase between September 2021 and March 2022 is primarily owed to finds being discovered in 2020. There are, however, also minor increases in the number of recorded finds in most years between 2002 and 2019. These finds are likely to have been handed in some time after they were discovered by metal detectorists, and/or stored by the county archaeologists for a while before being passed on to the museums. Until 2021, it took all the museums at least two years to fully record the finds, but most of them seemed to have accessioned them as soon as they arrived. By 2023, only the Museum of Cultural History in Oslo spent more than two years on cataloguing a metal-detected find.

Regional and local differences in recording practices and metal detecting activity

Although the national guidelines for the private use of metal detectors have been in effect for a few years now, the same regional differences that were visible in the metal-detected material before 2017 (Axelsen, Reference Axelsen2021: 131–32, 139, 176–96) are still visible. There are, of course, large geographical and historical variations between the management areas of the archaeological museums. They also vary in size, population, population density, and the surface of cultivated land within and between the regions (Table 2). This must be taken into consideration when assessing where metal detectorists are able to operate. The Museum of Archaeology's area is the smallest, consisting of only one county with a land area of 8575 km². The Arctic University Museum of Norway's management area, on the other hand, which cover the country's two most northern counties, is twice the size of Denmark (Figure 1, Table 2).

Table 2. Population size and density, area size, and percentage of cultivated land in each museum region. Data and percentages were collected and calculated from data of the Statistics Norway's open access tables ‘11342: Population and area (M) 2007 – 2020’, ‘06462: Agricultural area for selected crops (decares) (M) 1969 – 2020’, and ‘11506: Agricultural area, by use (decares) (C) 1969 – 2021’. Only ‘fulldyrka mark’ (arable land) in the category ‘cultivated land’ is included. Surface-cultivated and infield pastures are excluded.

Considering the size of the area it serves, population density, and amount of cultivated land, it is not surprising that the Museum of Cultural History in Oslo has the largest number of detectorists who report their finds and the highest number of recorded metal-detected finds. Some seventy-five per cent of Norway's cultivated land and forty-seven per cent of the population is within the management area of the museum (Table 2). There are, however, also large differences between the counties served by the Oslo-based museum, with almost ‘empty’ areas such as Agder and very find-rich counties such as Vestfold and Telemark, Viken (predominately within the area of the former county of Østfold), and Innlandet (Figure 3).

Figure 3. Number of reported and recorded metal-detected finds per county.

One reason for the difference in the number of recorded detectorists and quantity of reported metal-detected finds may be in the way findspots are recorded in Askeladden. Four labels are currently used to classify the location of an archaeological object: 1) automatically protected; 2) unresolved; 3) not protected; 4) removed (see Figure 4). Data from Askeladden show that the country's (per 2023) eleven counties use different classifications. These categories affect whether it is legal for a detectorist to continue detecting on, for example, parts of or an entire field. Detecting on a listed site is not permitted without an exemption from the CHA by the relevant county. Additionally, the guidelines for the private use of metal detectors advises detectorists to avoid detecting within a 25 m radius from sites labelled as automatically protected or unresolved (Directorate for Cultural Heritage, 2017: 3).

Figure 4. Distribution of differentially recorded findspots in the Askeladden database. The data was downloaded 30.1.2020.

For the archaeologists who record and list the sites, it is not necessarily evident whether an area should be recognized as, for example, automatically protected. A letter of 2019 from the Directorate for Cultural Heritage to the country's archaeological museums and counties stated that when an unspecified number of finds indicates the presence of a site, it should be considered as automatically protected (Directorate for Cultural Heritage, 2019). Because perceptions and practices concerning hobby metal detecting vary (Axelsen, Reference Axelsen2021: 71–74; Fredriksen, Reference Fredriksen2021), detectorists can continue detecting on finds-rich fields in counties where find spots are usually classified as not protected or removed.

In counties where findspots are normally categorized as automatically protected or unresolved, detectorists must either apply to the county to continue detecting on the site or move at least 25 m away from its perimeter. This affects the number of objects that can be reported and handed in from fields within a county. When compared to the others, Innlandet stands out as the county with the highest frequency of unprotected find spots. Counties other than Trøndelag, Vestland, Troms, and to some extent Rogaland, prefer to classify finds in the unresolved category (Figure 4).

Prolific detectorists as ‘super users’

The most important factor influencing the representativity of the recorded metal-detected material is the activity level of individual detectorists. At one archaeological museum, two detectorists handed in around 44.3 per cent of the hobby metal-detected finds. Nationwide, the thirty most prolific finders, which includes one detecting club, have handed in about 39.7 per cent of the total of 16,948 metal-detected finds. In all, some 1063 different names were recorded as having reported protected objects before 14 March 2022 (Table 3). Records including a slash, indicating that one or more detectorists have detected together, were excluded. Most recorded detectorists in the artefact databases handed in few finds: about ninety-three per cent had, until 2020, reported fewer than fifty objects, with 837 people (79 per cent of the 1063 recorded finders) reporting ten or fewer finds. On the other hand, eleven different names, i.e. only one per cent of the total number of detectorists, had reported more than 200 protected objects. Those in the middle, who handed in between fifty and 200 artefacts, consist of sixty-one people, making up close to six per cent of the (recorded) hobby detecting group.

Table 3. Number of reported and recorded metal-detected finds and hobby metal detectorists, and their percentages per museum.

*Because forty names were recorded as having handed in finds at more than one museum, the total number of detectorists differs from the combined numbers from each museum.

These proportions are similar to what has been observed in other citizen science projects, where a so-called ‘one per cent’ or ‘90-9-1’ rule has been described (e.g. Haklay, Reference Haklay, Capineri, Haklay, Huang, Antoniou, Kettunen, Ostermann and Purves2016: 36). These projects show that it is common for a very small group representing about one per cent of the contributors or participants to be behind most of the activity, while the overwhelming majority only contribute occasionally. In other words, the ‘super users’—or ‘super detectorists’ in our case—are unlikely to be representative of the group in general. Although, being significant contributors and thus receiving much attention, they are, statistically speaking, outliers (Haklay, Reference Haklay, Capineri, Haklay, Huang, Antoniou, Kettunen, Ostermann and Purves2016: 36; see also Dobat et al., Reference Dobat, Christiansen, Jessen, Henriksen, Jensen and Laursen2019: 12–13).

The number of recorded detectorists in Norway may seem low when compared with other countries, such as the estimated 9600 metal detectorists in England and Wales (Robbins, Reference Robbins2014: 14). According to the Portable Antiquities Scheme annual reports (2018, 2019, 2020), individual contributors to the scheme numbered 4028 in 2018, 4143 in 2019, and 2846 in 2020. The 1063 recorded detectorists in Norway, in a population of c. 5,400,000, equals approximately twenty detectorists per 100,000 inhabitants. In terms of population size, the number of detectorists in Norway is similar to estimates of c. 15–30 detectorists per 100,000 inhabitants in other north-western European countries (Deckers Reference Deckers, Campbell, White and Thomas2019: 111).

It has been suggested that there is a conscious ‘selection’ of Norwegian medieval coins among metal detectorists in Norway, causing some to question the reliability of the recorded information for all metal-detected finds (Gullbekk et al., Reference Gullbekk, Sættem, Skogsfjord and Roland2019). The sale and circulation of Norwegian medieval coins without provenance (Gullbekk et al., Reference Gullbekk, Sættem, Skogsfjord and Roland2019: 178) certainly suggests that some people are illegally recovering these coins, with or without the aid of a metal detector (see also Mackenzie et al., Reference Mackenzie, Brodie, Yates and Tsirogiannis2020: 5). It should, however, be noted that the distribution of the Norwegian coins outlined by Gullbekk et al. (Reference Gullbekk, Sættem, Skogsfjord and Roland2019) seems to correlate with areas that have particularly prolific detectorists. Other possible explanations, or at least contributing factors, for the potential discrepancy in the coin material handed in by metal detectorists is that the hobbyists operating in these areas are:

1. 1) more skilled, recognizing the coins, which, due to their small size and low levels of silver, are indeed hard to notice in cultivated soil;

2. 2) more thorough than the less prolific finders—i.e. they investigate (more of) the so-called ‘poor’ signals (Axelsen, Reference Axelsen2021: 125).

Given what we know about other citizen science projects, it is plausible that prolific detectorists uncover and report not only more finds than the average detectorist, but also more different types of objects than the less prolific finders.

When looking at the distribution of the metal-detected finds in Norway (Figure 3), ‘hotspots’ appear in two geographical groupings, in south-eastern Norway and in northern Norway. In the two counties that are split between different museums, there are far more finds within one museum's area than the other. The recorded metal-detected finds in, for example, Nordland are concentrated in the northern municipalities, which lie within the Arctic University Museum of Norway's management area. The southern part of Nordland is at the outer limit of the NTNU University Museum's region.

Norway's concentrations of metal-detected finds are owed to a few individual detectorists and detecting clubs. Consequently, the productivity of a small minority, combined with how a handful of archaeologists choose to record findspots, may continue to influence the representativity of the collected material in the future. This can strengthen already visible patterns. Current trends could also change dramatically with an influx of prolific detectorists in areas which so far had very few mapped finds.

Coping with data floods from citizen scientists

One response to hobby metal detecting in countries lacking a national and common standard for recording finds has been to establish public finds recording schemes. This practical response to the growing popularity of hobby metal detecting is designed to make information about the metal-detected finds available and accessible to both the public and researchers (Dobat et al., Reference Dobat, Deckers, Heeren, Lewis, Thomas and Wessman2020).

Compounded with digital records that stem from archaeological projects, the rapidly escalating quantities of digital data from private individuals are causing Danish, English, and Welsh archaeology to experience their versions of big data floods (Bevan, Reference Bevan2012; Dobat et al., Reference Dobat, Deckers, Heeren, Lewis, Thomas and Wessman2020). Other countries with similar recording schemes, such as Flanders in Belgium (MEDEA), Finland (FindSampo), and the Netherlands (PAN), may experience the same data growth in years to come (e.g. Deckers et al., Reference Deckers, Bleumers, Ruelens, Lemmens, Vanderperren and Marchal2016; Vos et al., Reference Vos, Heeren, Ruler, Smallenbroek and Lassche2018; Wessman et al., Reference Wessman, Thomas, Rohiola, Koho, Tuominen, Jameson and Musteaţă2019). When data from these schemes are made accessible and adhere to a common standard and ontology, their scientific value and thereby the potential for new discoveries will increase as this will enable both qualitative and quantitative aggregation (Gattiglia, Reference Gattiglia2015: 113).

Compared to other European recording programmes, the Norwegian archaeological heritage system was among the earliest to make their data publicly available and reusable. This includes the metal-detected objects. Parts of the country have been overwhelmed with new finds, creating a backlog in fully recording the reported finds; although they are eventually fully recorded. Most of the stored information is made available to both the general public and researchers. Because of the lack of standardization in input data, it is, however, often not possible to identify finds has having been metal-detected, which is troublesome given the public interest in and origin of this finds group (Axelsen, Reference Axelsen2022).

Consequently, Norwegian databases, such as Kulturminnesøk and Unimusportalen, already meet the aims of the European Finds Recording Network, as summarized by Dobat and colleagues (Reference Dobat, Deckers, Heeren, Lewis, Thomas and Wessman2020). Work on a national digital finds recording scheme was initiated in summer 2023 by the Directorate for Cultural Heritage and the Museum of Cultural History in Oslo. This can ease the administrative burden for the national heritage management system, as well as the volunteer work of some of the country's most prolific citizen scientists (Axelsen, Reference Axelsen2022).

Trust and reuse of data

Many archaeologists do not trust the ability or willingness of some detectorists to offer accurate information about the finds they report (see Gullbekk et al., Reference Gullbekk, Sættem, Skogsfjord and Roland2019; Axelsen, Reference Axelsen2021: 122–26). Consequently, there is a lack of trust towards the data originators from those who would normally use the recorded data (Huggett, Reference Huggett, Wilson and Edwards2015: 18). Despite this, material collected by metal detectorists is frequently used in studies that include archaeological material obtained by conventional means (see Røstad, Reference Røstad2016 or Amundsen, Reference Amundsen2021 for recent examples).

Modern factors are clearly influencing the metal-detected material, in Norway and elsewhere. This, we contend, does not mean that the material is useless or ‘unfit’ for archaeologists who seek to understand the past. The biases affecting the distribution patterns of metal-detected finds, such as the productiveness of individual detectorists and the recording practices of archaeologists, is a growing field of research (e.g. Bevan, Reference Bevan2012; Cool & Baxter, Reference Cool and Baxter2016; Cooper & Green, Reference Cooper and Green2017; Oksanen & Lewis, Reference Oksanen and Lewis2020). By conducting exploratory, quantitative, and/or qualitative analyses of the material, we can better understand how we can and cannot use the finds discovered by detectorists.

Because non-professionals often lack the detailed knowledge and standards of the professionals, the data they gather can be met with suspicion and may lead to some researchers refusing to use the information at all. This can, as Balázs et al. (Reference Balázs, Mooney, Nováková, Bastin, Jokar Arsanjani, Vohland, Land-Zandstra, Ceccaroni, Lemmens, Perelló and Ponti2021: 147) point out, to some extent be remedied by ensuring that the relevant metadata and paradata is acquired and stored (see also Huggett, Reference Huggett2020: 12; Huvila, Reference Huvila2022). In Norway, the information about archaeological objects uncovered and reported by metal detectorists, although meant to follow a certain standard set by the national guidelines, varies in its accuracy. This is, in part, due to factors and practices that create biases in the collections and which are specific to hobby metal detecting (cf. Robbins, Reference Robbins2013, with references; see also Axelsen, Reference Axelsen2021: 205–08 for a summary of the situation in Norway). Moreover, the technical skills and digital literacy of the individuals using metal detectors to search for protected objects are, allegedly, affecting the precision of the reported data (Gundersen et al., Reference Gundersen, Rasmussen and Lie2016: 165–68; Axelsen, Reference Axelsen2021: 122–26). Worryingly, this information is not stored anywhere. Neither public finds recording schemes nor archaeological systems are, for now, taking this into account.

Although the many free-text fields in the Norwegian artefact databases allow for a substantial amount of contextual information to be recorded, the records are still ‘largely de-contextualised’ (as Huggett, Reference Huggett, Wilson and Edwards2015: 18 has argued; see also Huggett, Reference Huggett2020: 12). The information that has been given is recorded, usually by someone who is not the finder, and documented in a format that suits archaeological catalogues. What we lack is a systematic way of recording how the data was originated, how it has been treated, by whom, and how this and the digital systems themselves affect the material and how it can be used. To increase trust in hobby detecting records and thus the future use of it, that knowledge is vital. In other words, we need to ‘[find] the people within the systems’ (Huggett, Reference Huggett2020: 15, in reference to and paraphrasing Seaver, Reference Seaver2018: 382) to understand their ‘actions and decisions’. Hence a robust way of recording such information is necessary to ensure the integrity of the material.