PhD: Predicting water stress of Amazon forest trees from space

University of Leeds

Leeds, UK 🇬🇧

Year: 2023

Primary Supervisor: Dr David Galbraith <>

Institution: University of Leeds (School of Geography)

Academic Supervisors: Dr Amanda Lenzi <> (University of Edinburgh, School of Mathematics), Dr Marcos AdamiProf Beatriz Schwantes MarimonProf Ben Hur Marimon JuniorProfessor Emanuel Gloor (University of Leeds, School of Geography), Professor Patrick Meir <> (University of Edinburgh, School of Geosciences )

Research Themes: AmazonEarth ObservationForestwater

Project Partners: Pixxel and Planet

Research Keywords: ForestsWater

Supervisors: Prof. David Galbraith (University of Leeds), Prof. Emanuel Gloor (University of Leeds), Dr. Amanda Lenzi (University of Edinburgh), Prof. Patrick Meir (University of Edinburgh), Prof. Beatriz Schwantes Marimon (UNEMAT), Prof. Ben Hur Marimon Junior (UNEMAT), Dr. Marcos Adami (INPE)

Proposed Industrial Partners (contacted): Pixxel and Planet.

Background and Motivation. The Amazon’s forests are the most biodiverse on the planet, absorb 5-10% of global CO2 emissions1 and sustain rainfall over continental scales2. However, the invaluable ecosystem and climate services these forests provide are currently under severe threat from deforestation and changing climate. This threat is particularly severe in southern Amazonia (SA), where temperature increases and dry season rainfall reductions3 have been particularly pronounced and where physiological and atmospheric measurements suggest that forests are already experiencing significant water stress, enhancing mortality risk. To better understand plant physiological thresholds that result in drought-induced tree mortality in this critical region, the project team is leading the installation of a new 1-ha drought experiment in the southern limits of the Amazon biome in the Brazilian state of Mato Grosso4.  At this site, a large number of measurements will be made on forest canopy trees to evaluate plant water stress status (e.g. leaf water potential, embolism status) and its impacts on plant function (e.g. sap flow, growth).  The experiment also presents a unique opportunity to assess how the spectral properties of leaves change as plant water stress intensifies and to relate these to metrics derived from high-resolution satellite images that can be used to evaluate plant water status over larger scales and predict regions close to experiencing extensive tree mortality5.

Aims, Objectives & Methodological Approach.  The key aim of this PhD is to develop a methodological basis for predicting water stress of trees in the southern Amazon and predicting regions that are more likely to experience drought-induced mortality. This aim is achieved through three objectives: 1) evaluation of the relationship between leaf spectral reflectances measured in the field and field measures of plant water status, 2) evaluation of the relationship between in situ leaf reflectance signatures under drought and satellite-derived indices from multispectral sensors (e.g. red-edge CI from Sentinel 2) and hyperspectral sensors (e.g. DESIS, Pixel and Planet) and 3) prediction of plant water status over the broader southern Amazon using deep/machine learning models trained on data from the new drought experiment site, additional sites in the southern Amazon and from a long-standing drought experiment site in the north-eastern Amazon coordinated by co-supervisor Patrick Meir. Attainment of these objectives is made possible through the unique combination of: 1) repeated sampling of plant water stress in both the experimentally droughted and a nearby control plot, 2) repeated leaf spectroscopy measurements of canopy trees with a high-resolution field spectroradiometer (400-2500 nm) that will allow determination of shifts in leaf reflectance spectra under drought, 3) harnessing of a novel generation of high resolution hyperspectral satellites (e.g. – DESIS: 30 m spatial resolution, 3.3 nm spectral resolution, Pixel: 300 bands, 5 m spatial resolution, daily imagery; Planet: broad spectral range (400-2500 nm), 5 nm spectral resolution) and 4) powerful deep learning (e.g. convolutional neural networks, Resnet) and machine learning approaches (e.g. random forest) that will be applied to multispectral and hyperspectral imagery across the broader southern Amazon to determine water stress over larger scales and provide new insights into the climatic threat faced by these forests.

Supervision. The diverse supervisory team has a sustained track record in studying the impacts of climate change in Amazonia and is highly multidisciplinary in nature, with expertise ranging from plant ecophysiology6-7 to machine learning approaches8 and its application to satellite imagery9.  The proposed industrial partners (Pixel and Planet) would provide access to high-resolution hyperspectral imagery at no cost to the project.  The student will be based at the School of Geography, University of Leeds within the Ecology and Global Change cluster, a world-leading research group with a strong focus on tropical forests. Although the PhD studentship will focus largely on relating the satellite imagery to field observations, there would be opportunities to contribute to field data collection in the southern Amazon and to also to spend time in co-supervisor Adami’s research group in the Brazilian Space Institute (INPE).

References. 1.Brienen et al. 2015, Nature 519:344-48. 2. Spracklen et al. 2012, Nature 489:282-285. 3. Haghtalab et al. 2020. Theoretical and Applied Climatology 140:411-427. 4. Lethal Psi: Characterising critical embolism thresholds for Amazon tree survival.  4. Doughty et al. 2020.  Biotropica 53:581-595.  5.  Signori-Muller et al. 2021, Nature Communications 12:2310.  6. Rowland et al. 2015. Nature 528-119-122.  7. Lenzi et al. 2021. arXiv 2107:14346. 8.  Wang et al. 2019. Remote Sensing of the Environment 221:474-88.





You ad could be here!