SOILS AND MACHINE LEARNING OF ANCIENT MAYA WETLAND FIELD COMPLEXES

massive stretches of ancient Maya wetland raised fields in Northwest Belize (Beach et al., 2019). The largest area of these fields lies along the Rio Bravo and has never been studied before, only being recognized through airborne LiDAR survey in 2016. This paper outlines the first evidence for the chronology and use of these newly identified fields, possibly the largest known in the Maya Lowlands. First, we develop a machine learning model to extract the canals and raised fields from the LiDAR data to quantitatively estimate their size and shapes. Next, we present initial chronologies from AMS radiocarbon dating of charcoal and geochemical analysis of multiple raised fields across the floodplain. These excavations revealed two buried soils – one ~1.2 m below the surface and the other ~3.5 m below the surface – which suggest multiple significant changes in floodplain conditions through the time of ancient Maya occupation. The lowest buried soil, dating to Preclassic Maya, is covered by over a meter of erosional clay before returning to typical floodplain layered stratigraphy. This sequence suggests early Maya occupation caused significant erosion in the watershed and aggradation of clay in the floodplain, before erosion slows and annual flooding begins layering sediments again. Later, during the Classic to Postclassic periods, the second buried soil was covered by the raised fields’ construction. The timing and geochemistry suggest the fields were likely an adaptation to environmental conditions (high sulfate in groundwater and soils, too much or too little water, or insufficient soil nutrients) and high resource demands. Lastly, we use stable carbon isotopes to show that maize was likely grown intensively in this floodplain both before the construction of the raised fields as well as on them. This study presents a multi-method approach to understanding the chronology and lasting hydrologic legacy or these newly rediscovered ancient Maya wetland fields, and what we can learn about long-term floodplain management and sustainability.