Spectral diversity allows remote detection of the rehabilitation status in an Amazonian iron mining complex

Markus Gastauer,Wilson R Nascimento,Cecílio Frois Caldeira,Silvio Junio Ramos,Pedro Walfir M Souza-Filho,Jean-Baptiste Féret

doi:10.1016/j.jag.2021.102653

Abstract

Mineland rehabilitation is intended to reduce the overall impacts of mining on biodiversity and ecosystem services and requires periodic monitoring to guarantee institutional tractability and to refine rehabilitation practices. As time- and money-consuming field surveys challenge this monitoring, the aim of this study was to develop a remote sensing framework to assess the environmental quality of minelands undergoing rehabilitation based on free and open-access multispectral Sentinel-2 images. For this purpose, we linked spectral diversity, i.e., measures of spectral variation among neighboring pixels, to the field-surveyed environmental quality of a rehabilitation chronosequence covering five waste piles in the Carajás National Forest, Eastern Amazon. Field data were separated into a training data set from 2017 (54 plots) and a testing data set from 2019 (66 plots). Based on the training data set, we optimized the computational parameters of spectral diversity (separation of vegetated from nonvegetated pixels by the normalized difference vegetation index [NDVI], size of the buffer zones, number of clusters representing distinct spectral species and applied diversity metrics), maximizing the accuracy of the remote monitoring approach. Then, we validated the procedure with the testing data set and compared the number of areas undergoing rehabilitation and their quality from 2017 and 2019. The overall accuracy of our methodology was 83%, and user and producer accuracies exceeded 60% for all rehabilitation classes, enabling the remote sensing of successional advancement of rehabilitating minelands. Despite punctual losses in spectral diversity, we detected comprehensive gains in spectral diversity within the target structures. This is due to an increase in the overall rehabilitation area from 282.39 to 364.54 ha, but enhancements in the spectral diversity of already vegetated areas indicate the successional advancement of these areas with time, reducing the overall impact of mining activities on biodiversity and ecosystem services. We conclude that the remote sensing framework presented here is promising for mapping environmental quality in minelands undergoing rehabilitation, and we encourage its integration in current monitoring protocols to reduce costs and increase the transparency of a company's rehabilitation activities, as freely available satellite images and open-source software are applied.

Highlights

By restituting species communities, vegetation structure and ecological processes as far as possible to predisturbance levels (Gastauer et al, 2018; Wortley et al, 2013), mineland rehabilitation is intended to reduce the overall impacts of mining on biodiversity and ecosystem services (Perring et al, 2015; Gann et al, 2019; Guerra et al, 2020)
This study was carried out using five mining waste piles from the N4N5 iron mining complex and in the neighborhood of a small quarry called Arenito situated in the Carajas National Forest, Eastern Amazon, Para, Brazil (Fig. 1)
Spectral diversity computed for the training data set was positively associated with field-detected environmental quality, but the accuracy of the remote recognition of rehabilitation status differed among opti mization runs (Fig. 3)

Summary

Introduction

Vegetation structure and ecological processes as far as possible to predisturbance levels (Gastauer et al, 2018; Wortley et al, 2013), mineland rehabilitation is intended to reduce the overall impacts of mining on biodiversity and ecosystem services (Perring et al, 2015; Gann et al, 2019; Guerra et al, 2020). The environmental quality of rehabilitating minelands, i.e., the degree to which ecological attributes recover toward predisturbance levels, is commonly measured as en hancements in vegetation structure, community diversity and ecological processes (Wortley et al, 2013; Ruiz-Jaen and Aide, 2005) It is esti mated as the proportion to which rehabilitating sites converge to the reference sites after ordinating degraded areas, rehabilitating sites and reference ecosystems based on field-surveyed ecosystem characteristics using multivariate approaches (Fig. 2A) (Gastauer et al, 2020a). The increasing availability of high-resolution satellite imagery, supported by technological, computational, and modeling advances, provides data on the optical properties of the Earth’s surface over un precedented spatial and temporal scales (Martin, 2020; de Almeida et al, 2020) This information opens perspectives for a wide range of applications in ecology (Kerr and Ostrovsky, 2003; Lechner et al, 2020), including environmental monitoring activities (Johansen et al, 2019). Different vegetation indices, such as the normalized difference vegetation index (NDVI) (Jackson and Huete, 1991), allow the estimation of biomass accumulation (Gonzalez-Alonso et al, 2006; Prabhakara et al, 2015), but the remote detection of further ecological attributes such as diversity, ecosystem functioning or environmental quality in rehabilitating minelands is still challenging (McKenna et al, 2020)

Objectives

Methods

Results

Discussion

Conclusion