Mapping productive forests farming and farmers in the Brazilian Amazon

Keywords: Agroforestry Systems, Açaí fruit, Euterpe oleracea, Land use, Remote sensing, Deep learning, Amazon estuary
Session Type: Virtual Paper Abstract
Day: Sunday
Session Start / End Time: 2/27/2022 11:20 AM (Eastern Time (US & Canada)) - 2/27/2022 12:40 PM (Eastern Time (US & Canada))
Room: Virtual 14

Abstract

On many accounts, the açaí fruit economy is today the Amazon’s most important production system. Despite its importance, the extent and geography of its production systems are poorly measured. Estimates rely mainly on key informant surveys and decadal agricultural censuses. Accurate and scalable estimates of açaizais based on remote sensing are limited by the availability of suitable data and techniques to identify specific plant species, especially given the structural complexity of açai’s production system. As a result, we do not know the actual scale of açai production through agroforestry and forest management and the contributions of this production system to landscape change. The recent availability of very high-resolution images, the advance of deep learning algorithms, and an increased computational capacity opened new possibilities to map açaizais. Here, we use a deep learning algorithm called U-net and high-resolution image composites from Google Earth to identify, segment, and map açaizais over Abaetetuba, Pará, in the Brazilian Amazon. We first visually interpreted açaizais over 200 sq. km of Abaetetuba territory. The visually interpreted samples were used as a reference to train the U-net algorithm and validate the accuracy of the results. We estimate ~10,000 ha of açaizais in total with an accuracy of 87.3%. We contrasted our estimates with official surveys, highlighting significant inconsistencies with the data reported. We discuss the implications of the work for estimating the economic and environmental contributions of agroforestry and forest-based production systems and the potential for scaling up and applying this approach to other agroforestry-based land uses.