Finding woodlands in drylands: Bases for the monitoring of xeric open forests in a cloud computing platform

Abstract

Conservation and sustainable management of woodlands in drylands is a priority at the global level considering the numerous benefits and ecosystem services they offer for local people. Sustainable use of forest resources requires the planning of forest conservation, restoration, and use on a global scale through National Forest Plans. It is imperative to generate information about their location and main characteristics to develop such plans. In this context, satellite imagery presents invaluable advantages related to the possibility of covering greater extensions in less time using fewer resources. However, there is limited knowledge about dryland woodlands spatial distribution due to the challenges associated with their identification. Particular features of these forests may conceal them from remote sensors. In Argentina's drylands, the detection of Monte biogeographical province's woodlands is hindered due to the spectral confusion that takes place between them and the surrounding shrubland. The objective of this study was to assess and compare different methodologies in a web-based image processing platform to develop a monitoring methodology easy to apply by public offices. We selected two study areas in the Monte region where we performed and assessed 16 types of supervised image classifications. In drier sites characterized by high contrast between woodlands and surrounding areas, forest detectability improved. In both study areas, we found better results using higher resolution images. The improvement associated with the inclusion of a vegetation index, as well as the classification algorithm and scheme implemented was case-dependent. In both study areas, classifications detected non-forest areas more accurately, indicating the possible application of the assessed procedures for the monitoring of forest loss. The cloud computing platform proved very useful and included multiple advantages. Bases for the detection of woodlands in drylands emerge from this study, considering improvements in accuracy and suitable alternatives: the use of Sentinel images, the addition to the mosaic of the SATVI index, the use of the CART algorithm, and a detailed classification scheme. These do not intend to be rigid instructions, but baseline recommendations to orient the design of monitoring approaches of these hidden forests.