This Visiting Researcher Grant proposal aims to fund the visit of Dr. Matheus Pinheiro Ferreira, Professor of the Cartographic Engineering Section of the Military Institute of Engineering (IME) at the Department of Forestry Sciences of the Luiz de Queiroz College of Agriculture of the University of São Paulo (ESALQ/USP). During one year, Dr. Matheus Ferreira will work with data from the NewFor project supported by FAPESP (process nº 18/18416-2). Over three years, NewFor has been collecting inventory data about the structure, diversity, and function of "new forests" (restored forests, those undergoing natural regeneration and agroforests) with advanced remote sensing technologies such as light detection and ranging (LiDAR) and imaging spectroscopy. Recently, the project completed the construction of a unique database based on collecting leaf optical properties and functional traits of several tree species in different forest formations. Such a database provides a unique opportunity to investigate the relationship between functional and spectral diversity in restored tropical forests. The visiting researcher intends to explore this relationship at the leaf and canopy level using advanced artificial intelligence methods. Furthermore, the research will explore one of the main bottlenecks of the NewFor project and remote sensing as a whole, which is the remote measurement of biodiversity. This is an important frontier of knowledge that will be explored based on the combination of unique databases (>700 field plots and a spectral database of >400 tree species), sophisticated pieces of equipment (drone-lidar-hyperspectral system ), and a researcher of excellence in the subject, generating multiple academic benefits, with an impact on public policies (supporting the future development of monitoring systems for large-scale restoration) and on the training of human resources (through training and qualification of students from undergraduate and graduate courses involved in the research). It is hoped that the research results will contribute to NewFor achieving its central objective, which is related to obtaining information about the diversity, structure, and function of the "new forests" of São Paulo. As benefits of the visit, it is worth highlighting the expansion of the NewFor database through the acquisition of hyperspectral images, the intensification of scientific partnerships with other institutions in São Paulo with which the visiting researcher has partnerships and benefits related to guidance and participation in student committees of undergraduate and graduate courses. 

Funding:  São Paulo Research Foundation (FAPESP)

Urban forests provide essential ecosystem services to improve human well-being in cities.  Spatially explicit information on urban forest characteristics, such as the spatial distribution of tree species and functional traits, is crucial for green infrastructure management. This information is usually acquired with ground-based surveys, which are time-consuming and usually cover limited spatial extents. The combination of remote and field data allows urban forest monitoring at the city-scale level. In this project,  we combine multiscale remote sensing images and field inventory data to map tree species and retrieve their canopy functional traits in the city of Rio de Janeiro, Brasil. We aim to produce the first map of tree species at the individual tree crown (ITC) level for the entire city and estimate their functional traits from ITCs.

Related publications: Martins et al. (2021)

Funding:  Fundação Carlos Chagas Carlos Chagas Filho Foundation for Research Support of the State of Rio de Janeiro (FAPERJ) (grants 248496/2019 and 259727/2021)

Spatially explicit information on tree species distribution in tropical environments provides valuable insights for managing and conserving essential ecosystems such as forests, wetlands, and savannas. Traditionally, information on the spatial distribution of tree species is obtained by costly and labor-intensive field campaigns that cover small areas (< 1 ha). Remote sensing technology allows upscaling field-based assessments to large spatial extents, and recent advances enable us to depict Earth's terrestrial biosphere with unprecedented spatial and temporal detail. However, there is an urgent need for efficient and automatic approaches to exploit this data treasure, translating it into ecologically and economically relevant information. In this regard, convolutional neural networks (CNNs), a deep learning method, revolutionize our capabilities to harness remotely sensed data. The use of CNNs to map tree species in tropical environments is just beginning to be explored. There are many challenges to address, given these ecosystems' highly diverse nature. Our team has been working to deliver state-of-the-art deep learning methods and models to map tropical tree species from remotely sensed data. Some examples involve palm species mapping in Amazonian forests, accurate mapping of Brazil nut trees in Amazonia, and urban tree species mapping. Developing deep-learning-based approaches requires GPU computing power to reduce training and inference time. Specifically, our methods rely on CNN architectures for semantic segmentation, object detection, and instance segmentation. Such architectures involve many convolutional operations, mainly when applied to remote sensing images with more than three channels.

Related publications: Ferreira et al. (2020), Martins et al. (2021), Ferreira et al. (2021), Veras et al. (2022)

Funding:  NVIDIA Applied Research Accelerator Program, Brazilian National Council for Scientific and Technological Development (CNPq) grant  #306345/2020-0

Conservation and management of the Atlantic Forest (AF), one of the most threatened biomes of Brazil, depends on high quality information about its forest resources. Most of our knowledge about the AF comes from field measurements performed at the plot level (≤ 1ha), but data encompassing broad spatial extents are needed to better understand the structure and function of this important biome. Remote sensing holds great promise to generalize and extrapolate insights emerged from plot-based studies to whole landscapes. Particularly hyperspectral remote sensing has proved to be a pivotal technology to study tropical forests. Hyperspectral sensors are capable of measuring reflected light from the forest canopy over a large number of narrow spectral bands, thus allowing the detection of tree species and estimation of canopy foliar traits related to ecosystem processes. These applications are challenging in tropical environments in which the canopy is spectrally and structurally very complex, but they can be improved by the use of Radiative Transfer Models (RTMs). To fill a major knowledge gap in the retrieval of canopy foliar traits and information regarding biodiversity of tropical forests over landscape scales, the objectives of this project are: (i) to simulate the spectral response of tropical forest canopies with a RTM; (ii) to validate with field data, estimates of canopy functional traits obtained by inversion of our simulations and (iii) to improve tree species discrimination and mapping methods based on hyperspectral remote sensing. 

Related publications: Ferreira et al. (2018), Ferreira et al. (2019)

Funding:  São Paulo Research Foundation (FAPESP)