Ecosystem Modeling using Artificial Neural Networks: An archaeological tool

L. S. BURRY; M. B. MARCONETTO; M. SOMOZA; P. PALACIO; M. TRIVI; H. D´ANTONI

Journal of Archaeological Science Report

Academic Press

Año: 2017

2352-409X

Prediction of past Normalized Difference Vegetation Index (paleo NDVI) in Valle de Ambato (Catamarca, Argentina) in the periods of 550-650 and 1550-1650 AD was carried out to test the efficacy of Artificial Neural Network (ANN) to predict past environments for Archaeology. This work shows that subtropical Yunga vegetation and xerophytic Monte vegetation respond in contrasting fashion to changes in climate forcings. To predicting the past an ANN perceptron multilayer model was used. Modern NDVI data and Tree-Ring data were obtained from NOAA-Paleoclimate, and other public sources. These data were used to train the model. Real data and predictions were close (Pearson correlation 0.83 ? 0.90) and warranted the following step, hindcasting. Important paleo-NDVI fluctuations lasting 15 to 20 years were identified in both periods under study. The paleo-NDVI fluctuations in the earlier period were probably related to the unidentified eruption of 583. The fluctuations in the later period appear related to the eruption of 1604 to 1620 the Huaynaputina volcano (SW Peru). These findings suggest that the model accurately identified vegetation fluctuations in response to changes in the volcanic forcing. Hence, the ANNs may be considered as apt tools for modeling past environments in support of archaeology.