Archaeological remains generated by human activities in prehistorical and historical times result in anthropogenic spatial components of modern landscapes. The study of these features in geographical areas of interest can be advantageous to assist in both cultural heritage protection policies and the efficient management of the archaeological risk impacts inherent in infrastructural planning. As such, predictive distribution models are becoming an investigation method for archaeological landscapes. Moreover, the use of remote sensing data in this type of model, both in archaeological and ecological contexts, is gaining momentum. Based on presence-only data, two inductive predictive modelling approaches, i.e., a geographical information system (GIS)-based multiparametric spatial analysis (MPSA) and maximum entropy model (MaxEnt), were tested, integrated and evaluated for archaeological applications at different scales over the case study area of the Tavoliere Plain (Southern Italy). Presence data included sites identified via both remote sensing and archaeological survey techniques. Environmental variables included topographic, geomorphological and remote sensing derived attributes. Two contrasting repeatable criteria (i.e., correlation coefficients plus MaxEnt statistics and spatial principal component analysis (PCA) were used to select non-redundant environmental variables and Akaike's information criterion (AIC) to select the most parsimonious model configurations. A threshold analysis of the presence data was performed on the best model to define the minimum size of the presence data set with respect to the extent of the study area as well as the relative amounts of remote sensing derived response data to reach a stable performance of the model. The results show a higher performance of the MaxEnt models with respect to the GIS_MSPA models regardless of scale, indicating that the two approaches cannot be considered alternatives. The most parsimonious configurations of these models indicate that, depending on their scale, both spatial PCA and MaxEnt statistics can be used to select non-redundant input variables. The threshold analysis suggests that for the illustrated case study, a density of 0.2 presence sites/km2and 45% of remotely sensed sites are necessary to improve and stabilise the performance of the model. Critical issues and opportunities for advances in the performance of the MaxEnt approach in an operational archaeological context are highlighted. These results suggest stakeholders should deal with predictive modelling prior to the undertaking of both land development and scientific field surveys and offer advice for informed decisions at this regard.

Investigating the distribution of archaeological sites: Multiparametric vs probability models and potentials for remote sensing data

Cafarelli, Barbara;
2018-01-01

Abstract

Archaeological remains generated by human activities in prehistorical and historical times result in anthropogenic spatial components of modern landscapes. The study of these features in geographical areas of interest can be advantageous to assist in both cultural heritage protection policies and the efficient management of the archaeological risk impacts inherent in infrastructural planning. As such, predictive distribution models are becoming an investigation method for archaeological landscapes. Moreover, the use of remote sensing data in this type of model, both in archaeological and ecological contexts, is gaining momentum. Based on presence-only data, two inductive predictive modelling approaches, i.e., a geographical information system (GIS)-based multiparametric spatial analysis (MPSA) and maximum entropy model (MaxEnt), were tested, integrated and evaluated for archaeological applications at different scales over the case study area of the Tavoliere Plain (Southern Italy). Presence data included sites identified via both remote sensing and archaeological survey techniques. Environmental variables included topographic, geomorphological and remote sensing derived attributes. Two contrasting repeatable criteria (i.e., correlation coefficients plus MaxEnt statistics and spatial principal component analysis (PCA) were used to select non-redundant environmental variables and Akaike's information criterion (AIC) to select the most parsimonious model configurations. A threshold analysis of the presence data was performed on the best model to define the minimum size of the presence data set with respect to the extent of the study area as well as the relative amounts of remote sensing derived response data to reach a stable performance of the model. The results show a higher performance of the MaxEnt models with respect to the GIS_MSPA models regardless of scale, indicating that the two approaches cannot be considered alternatives. The most parsimonious configurations of these models indicate that, depending on their scale, both spatial PCA and MaxEnt statistics can be used to select non-redundant input variables. The threshold analysis suggests that for the illustrated case study, a density of 0.2 presence sites/km2and 45% of remotely sensed sites are necessary to improve and stabilise the performance of the model. Critical issues and opportunities for advances in the performance of the MaxEnt approach in an operational archaeological context are highlighted. These results suggest stakeholders should deal with predictive modelling prior to the undertaking of both land development and scientific field surveys and offer advice for informed decisions at this regard.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11369/374415
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 13
  • ???jsp.display-item.citation.isi??? 12
social impact