Upscaling Tree Demography to Heterogenous Landscapes Using Models and Remote Sensing

Tree demography is foundational to ecology and conservation, from mass tree die-offs to forest recovery. Plot-level studies of tree demography, including field measurements of tagged individuals, have been fundamental for developing ecological theory and forest management strategies. However, the limited spatial extent of field plots impedes generalizing plot-level models for spatial predictions across heterogeneous landscapes. Novel high-spatial resolution remote sensing imagery has opened the possibility for measuring tree demographic rates with continuous spatial coverage at landscape to regional extents. Remote sensing derived measurements could address pressing research questions, including disentangling causes of high variation in natural regeneration across secondary forest landscapes. Despite the promise of high-spatial resolution imagery for ecology, applying these data to ecological questions will require novel modeling approaches that can account for large amounts of spatial data that often include hierarchical structure. In this thesis, I apply high-resolution remote sensing to upscale tree demography at landscape scales, and provide guidelines for ecologists seeking to parametrize spatially explicit models for neighbor interactions by combining field data, high-resolution remote sensing, and Bayesian quantitative methods. Chapter 1 demonstrates how high-spatial resolution remote sensing can help improve predictions of tree recruitment at the landscape scale. This chapter is the first step towards new support tools that inform restoration projects about where and which species will regenerate naturally in agricultural landscapes. Chapter 2 addresses how to optimize neighbor interaction models using the Hamiltonian Monte Carlo algorithm. I demonstrate how ragged matrices could solve data storage inefficiencies associated with the neighbor interaction models' pairwise structure. I also provide code for a model parametrization that solves a sampling pathology associated with high correlation in hierarchical structures and an overview of metrics to assess when this hierarchical structure pathology is present. Chapter 3 explores the influence of biophysical and anthropogenic drivers on tree mortality in agricultural landscapes using high-resolution remote sensing data. The results suggest that accessibility and land management are core factors that could be managed to prevent the mortality of agricultural trees. Educational initiatives and new policies that address anthropogenic factors could be the answer to reduce agricultural tree loss. Overall, this thesis brings together Bayesian statistical methods with novel high-resolution remote sensing to overcome the spatial limitation of field measurements and produce spatial predictions and inference on drivers of tree demography across heterogeneous landscapes.

Genome assemblies for two Neotropical trees: Jacaranda copaia and Handroanthus guayacan

The lack of genomic resources for tropical canopy trees is impeding several research avenues in tropical forest biology. We present genome assemblies for two Neotropical hardwood species, Jacaranda copaia and Handroanthus (formerly Tabebuia) guayacan, that are model systems for research on tropical tree demography and flowering phenology. For each species, we combined Illumina short-read data with in vitro proximity-ligation (Chicago) libraries to generate an assembly. For J. copaia, we obtained 104X physical coverage and produced an assembly with N50/N90 scaffold lengths of 1.020 Mbp/0.277 Mbp. For H. guayacan, we obtained 129X coverage and produced an assembly with N50/N90 scaffold lengths of 0.795 Mbp/0.165 Mbp. J. copaia and H. guayacan assemblies contained 95.8% and 87.9% of benchmarking orthologs, although they constituted only 77.1% and 66.7% of the estimated genome sizes of 799 Mbp and 512 Mbp, respectively. These differences were potentially due to high repetitive sequence content (> 59.31% and 45.59%) and high heterozygosity (0.5% and 0.8%) in each species. Finally, we compared each new assembly to a previously sequenced genome for H. impetiginosus using whole-genome alignment. This analysis indicated extensive gene duplication in H. impetiginosus since its divergence from H. guayacan.

Baumartenwechsel in den Walliser Waldföhrenwäldern verstärkt nach grossen Störungen

Vegetation shift in Scots pine forests in the Valais accelerated by large disturbances In the past dozen years, several studies have concluded a vegetation shift from Scots pine to oak (pubescent and sessile) forests in the low elevated zones of the Valais. It is, however, not fully clear in which way such a vegetation shift actually occurs and on which processes such a shift would be based. Two studies, one on the tree demography in the intact Pfynwald and the other on the tree regeneration on the large Leuk forest fire patch, serve to discuss different aspects of the shift from Scots pine to oak. The forest stands of Pfynwald consist of 67% Scots pines and 14% oaks. Regenerating trees are 2–3.5 times more frequent in small gaps than under canopy. In gaps of the Upper Pfynwald, seedlings and saplings of Scots pine are three times more abundant than oaks, while both species regenerate in similar quantities under canopy. In the Lower Pfynwald, young oaks – especially seedlings – are more frequent than Scots pines. A different process is going on at the lower part in the Leuk forest fire patch where Scots pines prevailed before the burn of 2003. While Scots pines regenerate exclusively close to the edge of the intact forest, oaks not only resprout from trunk but also profit from unlimited spreading of their seeds by the Eurasian jay. Regeneration from seeds are hence observed in the whole studied area, independent of the proximity of seed trees. After the large fire disturbance, a mixed forests with a high share of oaks is establishing, which translates to a rapid vegetation shift. The two trajectories are discussed in the light of climate change.

Tree Demography Plots

The pantropical network of large tree demography plots coordinated by the Smithsonian’s Center for Tropical Forest Science has now gone global, as part of the Smithsonian Institution Global Earth Observatories. Some four million tropical trees, representing about 10,000 species, are now tagged, provisionally identified and periodically recensused. Some 3,000 species are captured in the six plots within Malesia. These include species rarely collected and many that are now endangered. Easy location of trees for periodic examination for fertile material and detailed ecological data, together with seasoned in-country research teams, provide unique opportunities for research collaboration.

The importance of tree demography and root water uptake for modelling the carbon and water cycles of Amazonia

Amazonian forest plays a crucial role in regulating the carbon and water cycles in the global climate system. However, the representation of biogeochemical fluxes and forest structure in dynamic global vegetation models (DGVMs) remains challenging. This situation has considerable implications for modelling the state and dynamics of Amazonian forest. To address these limitations, we present an adaptation of the ORCHIDEE-CAN DGVM, a second-generation DGVM that explicitly models tree demography and canopy structure with an allometry-based carbon allocation scheme and accounts for hydraulic architecture in the soil-stem-leaf continuum. We use two versions of this DGVM: the first one (CAN) includes a new parameterization for Amazonian forest; the second one (CAN-RS) additionally includes a mechanistic root water uptake module, which models the hydraulic resistance of the water transfer from soil pores to roots. We compared the results with the simulation output of the big-leaf standard version of the ORCHIDEE DGVM (TRUNK) and with observations of turbulent energy and CO2 fluxes at flux tower locations, of carbon stocks and stand density at inventory plots and observation-based models of photosynthesis (GPP) and evapotranspiration (LE) across the Amazon basin. CAN-RS reproduced observed carbon and water fluxes and carbon stocks as well as TRUNK across Amazonia, both at local and at regional scales. In CAN-RS, water uptake by tree roots in the deepest soil layers during the dry season significantly improved the modelling of GPP and LE seasonal cycles, especially over the Guianan and Brazilian Shields. These results imply that explicit coupling of the water and carbon cycles improves the representation of biogeochemical cycles in Amazonia and their spatial variability. Representing the variation in the ecological functioning of Amazonia should be the next step to improve the performance and predictive ability of new generation DGVMs.