Abstract Canga ecosystems are iron-rich habitats and pose a challenge for conservation and environmental governance in Brazil. They support high levels of biodiversity and endemism and, at the same time, have suffered intense losses and degradation due to large-scale iron ore mining. The Peixe Bravo River Valley in the Brazilian savanna is one of the last natural canga areas that has yet to face the irreversible impacts of mining. However, there are vast gaps in data on the vegetation cover, location, spatial distribution, and area of occurrence of this ecosystem. Therefore, more information is needed on the appropriate scale, without which it is difficult to establish conservation planning and strategies to prevent, mitigate or compensate for impacts on canga ecosystems. In this study, we provide the first map of canga ecosystems in Brazil using the U-Net deep learning model and Sentinel-2 images. In addition, we estimate the degree of direct threat faced by ecosystems due to the spatial overlap of the mapped cangas and the location of mining concession areas for iron ore exploitation. The deep learning algorithm identified and segmented 762 canga patches (overall accuracy of 98.5%) in an area of 30,000 ha in the Peixe Bravo River Valley, demonstrating the high predictive power of the mapping approach. We conclude that the direct threat to canga ecosystems is high since 99.6% of the observed canga patches are included in mining concession areas. We also highlight that the knowledge acquired about the distribution of cangas through the application of an effective method of artificial intelligence and the use of open-source satellite images is especially important for supporting conservation strategies and environmental public policies.

Resumo Os ecossistemas de Canga, habitats com elevadas concentrações de ferro, são um desafio para conservação e governança ambiental no Brasil. Eles sustentam uma alta biodiversidade e endemismo, e sofreram intensas perdas e degradações de áreas naturais devido à mineração de ferro em larga escala. O Vale do Rio Peixe Bravo, localizado no Cerrado brasileiro, é uma das últimas regiões com ecossistemas de canga que ainda não sofreu impactos irreversíveis da mineração. Mas ainda há ausência de dados sobre a cobertura vegetal, localização, distribuição geográfica e a área de ocorrência desse ecossistema. Portanto, a ausência de informações em escala adequada dificulta o planejamento em conservação e as estratégias para prevenir, mitigar ou compensar os impactos nos ecossistemas de canga. Neste estudo, nós fornecemos o primeiro mapa de ecossistemas de canga no Brasil elaborado a partir de deep learning segmentação U-Net e imagens de satélite Sentinel-2. Além disso, nós estimamos o grau de ameaça direta dos ecossistemas devido a sobreposição espacial das manchas de cangas preditas e a localização dos títulos de concessão minerária para exploração do minério de ferro. O algoritmo de aprendizado profundo identificou 762 manchas de canga (acurácia acima de 98,5%) em uma área de 30.000 ha no Vale do Rio Peixe Bravo, demonstrando o alto poder preditivo do método de mapeamento. Nós estimamos que há um alto grau de ameaça direta aos ecossistemas de canga, uma vez que 99,6% das manchas de cangas preditas estão incluídas em áreas de concessão de mineração. Nós também destacamos que o conhecimento adquirido sobre a distribuição das cangas por meio da aplicação de um método eficaz de inteligência artificial e do uso de imagens de satélite de código aberto é especialmente importante para apoiar estratégias de conservação e políticas públicas ambientais.

ABSTRACT Species distribution modeling (SDM) studies of aquatic macrophytes are still attached to methodological paradigms focused on terrestrial plants, such as the use of bioclimatic layers. Our goal was to evaluate the applicability of this paradigm based on a SDM study of Egeria densa, Pontederia crassipes, and Salvinia auriculata in the São Francisco river, Brazil. We compared performances of optimizations of computed models using AUC and TSS with distribution records of these species and bioclimatic layers, or limnological layers generated from the interpolation of data obtained in the field. We calculated models using six algorithms. The models calculated using layers of limnological variables had higher performances than did those calculated using layers of bioclimatic variables, except when the Maximum Entropy Default algorithm was used. We attribute these results to the specificities of the data obtained to develop the limnological layers, such as observations obtained in different habitats of the river and during different hydrologic periods. We conclude that the use of bioclimatic layers, a methodological paradigm traditionally used for SDM of aquatic macrophytes, can be questionable for some situations, such as in investigations at local and regional scales.

The ferruginous rupestrian grasslands (FRG) in the Iron Quadrangle (IQ) are ecosystems characterized by rocky soils with reduced availability of water and nutrients, but high levels of metals. In order to comprehend the interference of microorganisms on the adaptive process of endemic plant Mimosa calodendrum (Fabaceae), bacteria associated with its roots and rhizosphere were isolated. Fourteen isolates were obtained and subsequently grown in the presence of different concentrations of arsenic (As) species. The isolate Mc250, an Alcaligenes faecalis strain, resisted to 10 mM of As (III) and 800 mM of As (V). In the presence of this strain, atomic spectrometer detected a reduction of 55% for As (III) and 72% for As (V) respectively in 10 mM and 500 mM solution. Scanning electron microscopy of this isolate demonstrated morphological modification and EDX spectroscopy revealed the presence of both As species adsorbed on the membrane, justifying the removal observed in the in vitro assays. To validate this potential removal of As in vivo, tomato plants were used as grown model in the presence and absence of A. faecalis in soil previously contaminated with 5 mM of As (III). After 14 days, plants from contaminated soil had their growth improved when compared to untreated control plants. All these results suggest for the first time that plant-associated bacteria from FRG-IQ present potential for soil rhizoremediation and may benefit the adaptive processes of plants in extreme environments including application in recovering degraded areas.