Tuning and Effectiveness in Heritage Models

Abstract

Modelling can explore heritage responses to environmental pressures over wide spatial and temporal scales, testing both theory and process. However, compared to other fields, modelling approaches are not yet as common in heritage management. Some heritage models have become well known, though they struggle to have an impact beyond academia, with limited practical applications. Successful models appear to be adaptable to multiple sites or objects, intuitive to use, run using widely available software and produce output translatable into practical actions. Model tuning is also vital for the model to be effective. A specific purpose should be determined from the outset to enable tuning in the earliest design stages. Heritage models can be developed to explore theories or processes that affect or interact with heritage. Input should also be tuned to relevant temporal and spatial scales and consider duration and location. Additionally, it is important to account for materials and elements specific to heritage. Models need to be useful and usable if they are to be effective. User-friendly programs and interfaces help practical use. However, success can create problems, as input and output could become socially or commercially sensitive. The wider use of models may require broader discussion among heritage professionals and the provision of training.

1. Introduction

Heritage sites and objects possess a uniqueness that makes them irreplaceable [1]. However, heritage faces many threats, which can result in a loss of artistic worth and evidential value [2,3], although others might see this as an acceptable process of change [4]. Heritage practitioners and scientists need novel approaches to investigate processes that pose a threat to heritage to manage the pressures at heritage sites [5]. The methods used in field experiments can be highly constrained due to the potential for causing damage to objects or sites, and laboratory experiments take place in highly controlled conditions. Modelling methods allow for a safe mode of experimentation that can incorporate multiple processes over a wide range of spatial and temporal scales and enable the testing of both theory and process regimes.

Within the heritage field, several models have become well known, and their findings are frequently cited. Examples include the model output of future heritage climate developed under EU-funded projects, such as NOAHs ARK [6] and Climate for Culture [7]. However, these models, along with ones published in the recent academic literature, have often struggled to have a substantial impact on practice, predominantly remaining within the academic sphere [8]. The use of models in heritage science has been relatively uncommon when compared with other practical fields [8].

When models are used in practice, such applications are often recorded in the grey literature, which can limit a widening readership. Models that have gained broader adoption in practical contexts and influenced heritage management have tended to be: (i) adaptable to multiple sites or objects, (ii) intuitive to use, with, for example, a graphical user interface, (iii) run using widely available software and (iv) produce outputs that are translatable into practical actions [8]. Furthermore, for models to be effective in heritage science, “the modelled spatial and temporal scales need to be relevant to the impact of processes of change relevant to the heritage value and utility associated with sites and objects” [8].

As with any scientific approach, models and their underlying methods should be continually refined. This requires feedback between model developers and model users, “such that: (i) developers understand the challenges and benefits that users have with using the model, and (ii) users can share ideas and experience” [8]. There also needs to be considerable attention paid to the effect of uncertainty in model output. All models inherently contain uncertainty, but understanding how such uncertainties could result in errors being propagated through to actual decision making [9] requires further research in terms of heritage management. Recent work has suggested that while uncertainties can cause a range in output magnitude, when looking at change over time, the direction of change is an important indicator for developing management strategies [10,11].

Just as fieldwork or lab experiments are broad concepts that capture a range of methods, there are an array of models developed for a heritage context that we can use to explore the requirements and challenges in developing effective models. The Special Issue on Effective Models in Heritage Science aims to bring together modelling-based studies, and it acts as a basis to explore the idea of tuning heritage models that have emerged in developing this Special Issue.

2. Tuning Heritage Models

Models inherently require a system to be simplified. Therefore, model builders have to decide which elements and interactions within a given system to include. These decisions are context-dependent for the model to have the necessary specificity and precision to make the model and its outputs effective. In a heritage context, models should be tuned to: (i) the management requirements and purpose, (ii) relevant theory or process, (iii) issues of scale, spatial and temporal, and the location and time period under consideration and (iv) the relevance to specific heritage materials or elements of construction.

2.1. Purpose

When models are built to address a specific heritage question, ideas of model tuning can be included from the earliest design stages. This can be advantageous as the model can be readily tailored to research and management needs. When models are fit for purpose and have clear practical application, they can have far-reaching implications. For example, models developed for the National Trust were able to assess the economics of dust [12] and housekeeping [13]. Lloyd’s [13] research demonstrates how an easy-to-use, adaptable model in Microsoft Excel can aid with housekeeping and conservation decisions across a portfolio of historic homes. However, many new models fail to gain traction in the wider heritage community, often due to a lack of awareness that the model exists and the type of information it could provide. Furthermore, specific technical expertise or computational resources likely required to run the model can act as a barrier to uptake [8].

Several heritage models, such as ECOS/RUNSALT [14], IMPACT [15] and HERIe [16], are used by both heritage academics and practitioners. These have a simple approach to input or graphical user interfaces. This enables researchers and practitioners to engage with a model developed to address a process that impacts heritage materials, such as salt crystallisation and changes in indoor environments [17,18]. It is highly important for experts to produce clear guidance for good practice, as well as avoiding common mistakes for these out-of-the-box models, as users can find the model limitations hard to identify, even though they were obvious to the developer. Godts et al. [19] provide an example of this for the widely used ECOS/RUNSALT model [20,21,22,23].

Model users must also grapple with model uncertainty. For example, in Lloyd’s management model [13], errors may lead to a housekeeping allocation that is too small in some properties, such that individual property managers might feel under-resourced or, at other times, too large leading to over-resourcing. Although mathematical errors in models can be assessed, it is often difficult to understand how these errors might affect decisions. It is particularly challenging in models where the underlying processes are hidden, such as with multi-layered neural networks [24,25,26,27]. When using models to plan for future conservation needs, capturing the direction of change with high confidence will be important for informing management plans. The direction of change can often be established, even when there is uncertainty over the exact magnitude of change [11]. An additional source of error can arise from biases within the model. Although Lloyd [13] tries to minimise the bias in inputs, outputs might be skewed towards overprotection rather than underprotection. However, some have argued that biases may be helpful, especially where the costs of action and inaction are very different [9].

Finally, feedback is crucial to ensure that a model fulfils, and continues to fulfil, its desired purpose. However, in heritage research, there are few meaningful feedback pathways between academic researchers and practitioners. This means that the experience of using the model in practice is commonly not fed back to the researcher [8], constraining improvement, thus limiting its overall usefulness.

2.2. Theory and Process

Models in a heritage context can be developed to explore relevant processes or theories. Theory-focused models are valuable tools for advancing concepts and ideas that underpin our understanding of process, as well as having the potential to develop interdisciplinary conversations [8]. More often, models focus on assessing process. For example, Bretti and Ceseri [28] developed a model to translate the specifics of the carbonation process into a future changing climate, and Hart et al. [29] constructed a linear model to assess rainfall impacts on adobe blocks with a range of conservation treatments. Such models can capture key processes while retaining versatility in treating physical, chemical and biological threats (e.g., the role insects play in the degradation of wood [30] or other organic materials [31]). However, when models are built for a specific site or material, there is a danger that these can be difficult to transfer to other heritage sites or objects, limiting model effectiveness. Such situations highlight the problems of over-tuning a model to the point that it becomes so specific it cannot be used in other settings.

In addition to building new models tuned to a heritage context, existing non-heritage-specific model outputs can also be tuned to make them relevant to heritage. A common example of this is tuning parameters to meet the requirements of dose–response functions [32]. This often means reworking the inputs, which might be meteorological parameters or air pollutant concentrations. For example, outputs from global climate models were re-processed to capture deterioration processes in timber [33], while Verney-Carron et al. [34] compared the ability of dose–response functions with kinetic laws to assess climate and pollution impacts on medieval stained glass.

Further work is needed to assess how models capture the effects of extreme events or the crossing of thresholds, as these are likely to become increasingly relevant in a future world experiencing extensive environmental changes, changing frequency of climate hazards and sea level rise [35], but they are infrequently explored in heritage models.

2.3. Scale: Time and Location

Heritage models also need to consider scale, with regard to both time and space. Modelling the effect of long-term pollution and climate change on sites and objects has been a consistent theme in heritage science. Declining levels of air pollution in recent years have been beneficial to heritage [36], while concern over climate change impacts has increased e.g., the Noah’s Ark Project [6]. Past exposure conditions for sites and objects can be difficult to determine, and although some have tried, e.g., assessment of historical weathering [37], examining the deposits on buildings [38,39,40], archive photographs [41,42] and economic records of repair [43], it can be difficult. Thus, there has been considerable interest in trying to model conditions in the past and relate them to observed damage [34,44,45].

There are usually questions about the relevance of time periods under investigation. A 75-year span is useful to model significant changes, which might occur by the end of the 21st century and is frequently used in long-term assessments [46,47,48]. However, planning for heritage management is usually on a shorter time scale. Even the 30-year periods adopted as part of the notion of climate normals can seem long to heritage planners [49]. Budgets may be set over three-year timescales, or models might need to be tuned to the lifetime of maintenance and repair [13]. Thus, shorter planning horizons or recommendations for immediately implementable actions [17] will see an increasing need for effective models to consider reduced time periods.

Time is also important with regard to the resolution of input and output data. Heritage science sometimes uses annual or monthly averages (e.g., [50,51]), but for models to capture processes that drive deterioration, much shorter time scales are often needed, commonly requiring daily [33] or sub-daily data [17,18]. In some cases, it can be possible to extract the higher-resolution data from monthly averages, but this is not always possible [52]. Thus, datasets at higher resolution can be of importance in a heritage context.

Heritage models are developed for a wide range of spatial scales from individual materials [28,34], objects and sites [17,29,30], countries and regions [53,54], as well as global-scale analyses [33]. The spatial scale of a heritage model will be determined by the purpose of the model, with strategic models tending to have larger spatial scales, while, for example, chemical changes within heritage materials will require smaller-scale models. Challenges can arise when transferring between spatial scales as the dominant processes can be dependent on the scale [30]. This can be particularly difficult when relating global climate to indoor conditions [55]. However, as heritage is influenced by both global and local environmental processes [56], effective modelling and heritage management are dependent on being able to combine findings across multiple scales.

Models should also be able to be tuned to a specific location. This might include modelling heritage in the context of its surrounding landscape or environmental conditions [57,58,59,60]. Flexibility in the set-up parameters enables models to represent local conditions, which is important in determining processes for specific objects or sites. It is important that these set-up parameters remain adjustable so that the model is not constrained to a fixed set of starting conditions.

2.4. Materials and Heritage Elements

Models assessing physical, chemical or biological change to heritage materials are predominantly developed for a specific material. For example, Hart et al.’s [29] model was developed for adobe blocks, while Bretti and Ceseri’s [28] model was for concrete. This means that models can consider the relevant processes that cause change in a material, as well as accounting for the varying rates at which different materials undergo change [18]. Other models have focused on capturing the conditions that drive the change, rather than the change itself [33], e.g., ECOS/RUNSALT [14,19]. Using a model to capture conditions might make it more easily alterable to fit a different material or context. For example, understanding the number of freeze–thaw cycles a site is exposed to will be relevant to many materials, including stone, earth and brick [54,61,62]. Some models, such as those that consider earthquakes, sea level rise, wind-driven sand or rain or flooding, frequently need to account for multiple materials that form heritage buildings or sites, more than focussing on isolated elements [59,60,63,64].

A clear aim is vital for models to be effective; models with a site- or object-specific purpose will require a much closer alignment to specific materials than more strategic models looking at country- and regional-scale patterns of change.

3. Effective Models

For a model to represent a system in a useful manner, the purpose of a heritage model needs to be clearly defined from the outset. By understanding the purpose, this can dictate other tuning factors, such as time, place and materials, and allows the parameterisation to be undertaken more efficiently.

These models also need to be useful and usable if they are to be effective. Usefulness requires that models can be used in practice, but engaging with models can be daunting due to: (i) complex underlying theory and (ii) specialised computer programming languages. It seems that approachable models include those written in Excel (as in Lloyd [13]) or as Java applets (e.g., IMPACT [15]). More complex models have to be written in more formal programming languages. Here, they benefit from Graphical User Interfaces that guide the user through the process of data entry and provide output that is easy to interpret. For models to remain applicable to other sites or objects, the underlying code should be flexible to enable such a transfer to be possible.

Elements of successful models include being able to: (i) address a given challenge and (ii) be used by relevant stakeholder groups. In academia, a proxy for success is typically measured using citations to a model, but such citations may not reflect practical use. The practical use of models may be much discussed among professionals at conferences and training events, but quantifiable metrics are harder to establish. For example, Lloyd’s [13] model of housekeeping has been used as a part of masters’ courses in Collections Care and Conservation Management [65], but such success is not captured by standard metrics.

When a model is used in practice, both the input and output may become sensitive as they can influence decision making regarding finance, staffing or management. Thus, success for a model might also mean that it becomes a valuable asset, which can consequently limit sharing it as an open resource.

The long-term useability of a model requires documentation and ongoing training if knowledge inherent to ensure the model is not confined to one person (or a small team of people). Thus, groups with a good understanding of previous model iterations are required to update a model.

4. Conclusions

Models in heritage science have the potential to be a useful tool for research and management but face limited uptake in practice. This may arise from a belief that they are theoretical constructs and do not confront the complex reality inherent to heritage, or that such models may introduce bias or errors in decision making. Tuning models with respect to purpose, process and theory, spatial and temporal scale and materials will help improve the effectiveness of heritage models. Many process-based models used in a heritage context engage with climate. While understanding the impacts of climate change on heritage is of great importance, heritage models need to widen their scope to include other aspects of the heritage environment. To improve the understanding of models, it is likely that engagement with professional societies and the development of training courses are needed to support a widening understanding of the contributions that can be made as a result of effective modelling in heritage science.