Baines, Oliver
(2023)
Quantifying geodiversity–vegetation relationships in the Arctic using remote sensing.
PhD thesis, University of Nottingham.
Abstract
Context
The Earth is changing drastically as a result of human-induced climate change. Understanding contemporary distributions of species, and what drives them, is therefore essential, in order to understand how these may change in the future. Geodiversity – the diversity of the non-climatic abiotic environment – has been shown to be correlated with biodiversity across a range of terrestrial environments. Conservation initiatives underpinned by geodiversity have emerged, suggesting that protecting geofeatures such as landforms, rock types or water bodies could act as a coarse-filter strategy for protecting biodiversity. However, key questions underpinning geodiversity–biodiversity relationships remain unanswered. The Arctic is warming nearly four-times faster than the global average, with plant communities changing rapidly as a result. Incorporating geodiversity into Arctic vegetation research could provide important insights into how and why changes are occurring, the implications of these changes, as well as providing answers to outstanding questions surrounding geodiversity–biodiversity relationships. Using data gathered across a variety of scales, this thesis examines the importance of geodiversity for explaining pan-Arctic changes in vegetation through time, as well as exploring geodiversity–vegetation relationships across very fine scales. In doing so, I explore novel methods for measuring both geodiversity and biodiversity using remote sensing, including the treatment of uncertainty. I hope to provide both evidence and tools to further demonstrate the need to consider environmental changes holistically and improve our understanding of the coupling between plants and their abiotic surroundings.
Study area
Studies constituting the chapters of this thesis were either carried out using data from the Arctic (Papers 1, 3 and 4), or constituted methodological tests which were of relevance to Arctic vegetation (Paper 2). The Arctic is the area lying northward of the treeline, characterised by a short growing season and the presence of permafrost, and its extent was determined using the Circumpolar Arctic Vegetation Map (CAVM). Paper 1 examines the whole of the Arctic, at three spatial extents (pan-Arctic, CAVM subregions, and 500 * 500 km Arctic subsets). Papers 3 and 4 use data gathered from field plots on Qeqertarsuaq (Disko Island), Western Greenland. Paper 2 is based on data gathered while at the Department of Biology, Aarhus University, Denmark.
Methods
I obtained data primarily using remote sensing (or from products derived using remote sensing), and in situ data. Data for Paper 1 were compiled from a variety of freely available sources and included climate, geodiversity, and productivity datasets (see Table 1.1), which were all resampled to 16 * 16 km pixels. Paper 2 used hyperspectral measurements in the visible – shortwave infrared (VIS–SWIR; 350–2500 nm) portion of the electromagnetic spectrum, as well as images taken using a portable leaf scanner. Paper 3 also used hyperspectral measurements, gathered from leaf and soil samples obtained in situ in Qeqertarsuaq between June and August 2019. Biodiversity metrics were calculated from field vegetation plots, and laboratory measurements (nutrient, texture, and pH) made for a subset of soil samples. Soil hyperspectral and laboratory measurements were combined to produce estimates of soil characteristics for the remaining samples which weren’t measured in the lab. Paper 4 used these estimates, alongside trait and vegetation cover measurements made in the field.
Models were built to recreate observed patterns of Arctic Gross Primary Productivity change from 2001–2015 as part of Paper 1, using Boosted Regression Trees to capture non-linear relationships between variables. Models with just climate, and then climate plus geodiversity data, were compared across three spatial extents to test the importance of different geofeatures. Linear regression models were used in Paper 2, to identify the effect of gaps between small leaves (in images) upon proximal leaf spectra, and to test different methods of correcting for these gaps. In Paper 3, Partial Least Squares Regression models were used to relate hyperspectral measurements of soil and leaves to soil characteristics and biodiversity measures, respectively, using a novel nested spatial-leave-one-out cross-validation procedure to account for spatial autocorrelation. Paper 4 then used these data alongside bootstrapped Generalised Additive Models to test whether biodiversity, trait diversity, and vegetative cover measurements can be used to predict underlying soil characteristics, and test whether these relationships are scalable using remote sensing.
Results
Models incorporating geodiversity were found to better capture patterns of Arctic productivity change than those using just climate information, particularly within regional subsets of the Arctic (Paper 1). At very fine scales, models making use of hyperspectral data were able to estimate soil nutrient and pH (and to a lesser extent, texture) status, whilst accounting for spatial autocorrelation. Accounting for uncertainty in soil characteristics was found to be crucial, producing different estimates of soil diversity versus using just a single point prediction for each sample (Paper 3). Soil characteristics were found to vary hugely across sub-metre scales, with some field plots (4 m diameter) showing as large a range in nutrient, pH and texture as found across the entirety of our study site. Plant diversity (taxonomic and phylogenetic) and trait diversity were found to only be weakly related to variation in soil within and between plots, but shrub (and to a smaller degree, cryptogam) cover was strongly, positively related to soil nutrient stocks (Paper 4). The effect of leaf gaps upon proximal hyperspectral measurements could be corrected for using simple empirical models that I developed, performing better than traditional methods such as leaf stacking (Paper 2). Models created using hyperspectral measurements were then able to estimate biodiversity (Paper 3) and separate between plant functional types, whilst patterns of soil nutrient variation (both within and between plots) were detectable using freely available satellite imagery (Paper 4).
Implications and future work
Through the work presented in this thesis, I provide empirical evidence for the role of geodiversity in vegetation change through time, for fine-scale relationships between geodiversity and plants, as well as providing new tools to further assess the relationships between geodiversity and biodiversity. Whilst changes in climate are the driving force behind plant productivity changes, I find that accounting for variation in landforms, soil moisture, topography and rock types significantly improves models, with geodiversity mediating the effect of temperature change upon vegetation. No one geofeature was of universal importance, suggesting that decoupling between warming and productivity change in some areas could be ascribed to a range of locally specific factors. I find strong evidence for variation in topsoil conditions across very fine scales, and that by considering biodiversity–geodiversity (rather than the typical geodiversity–biodiversity) relationships, one can leverage plant characteristics to estimate soil diversity. This could then provide important insights into current conditions in the Arctic, and how they might change in the future under continued warming. Remote sensing was found to be valuable in capturing many of these patterns. By correcting for gaps in hyperspectral measurements and effectively accounting for spatial autocorrelation in model estimates, it is possible to reliably estimate variation in plant and soil characteristics across small distances. This may add new tools to further examine biodiversity–geodiversity relationships at a range of spatial scales, moving beyond coarse-scale predictors. In doing so, geodiversity studies should seek to more effectively model relationships by including uncertainty within both predictor and response variables, to avoid drawing erroneous conclusions about plant–environment relationships.
The work presented here represents a contribution to our understanding of geodiversity and of the Arctic, but much research still needs to be completed. Key questions remain about the importance of geodiversity for biodiversity through time, which could be tested using permanent vegetation plot datasets, such as from the International Tundra Experiment (ITEX). Hyperspectral data could be used to investigate intraspecific differences in leaf characteristics and identify how these might relate to variation in the abiotic environment. Linking this to UAV or satellite derived data could then allow the scale dependence of relationships between plants and their abiotic surroundings. Through work across scales, I have shown new ways to measure the abiotic environment, and its relationships to vegetation. There is still much we don’t know about the world around us, however. Considering the variation in different parts of the Earth system together – as with geodiversity – is of paramount importance to help us learn more.
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
![[img]](/style/images/fileicons/application_pdf.png)