The work performed during this project is divided into two streams - a statistical stream and an experimental stream. In the statistical stream, I approached cloud reconstruction in two ways. First, I investigated the community-based climate reconstruction method, cRacle, to determine the suitability of the method for reconstructing cloud regimes. This revealed latitudinal impacts on accuracy that may be linked to cloud-data availability at high latitudes, or to a stronger influence of seasonality on cloud impacts at higher latitudes. Further investigations are underway.
Encouraged by the preliminary results, but stepping back the species level specificity of cRacle, I investigated cloud associations with biomes. In collaboration with reserchers at the University of Leeds, I tested the correlation between biomes and cloud in a palaeoclimate model as compared to the modern distributions using machine learning techniques (e.g. Random Forests). This required my training in machine learning, big data and climate model specific data types, which was identified as an area of technical skill for development during this fellowship. Current results suggest a high degree of correlation in the modern observed biome and cloud data, but poor ability of a model trained on this data to predict biomes from cloud in the Pliocene model data. Further work is validating the method of assignment of model data to 'Weather States' and looking more closely at CMIP6 cloud validation studies. We will also analyse the spatial pattern of disagreement in the model output for a possible cause of the mismatch.
To constrain data-model mismatch, and thus its potential causes, I collaborated with Julia Tindall, Alan Haywood and Ulrich Salzmann, investigating the seasonal signal of mismatch between the reconstructions in the Arctic and the model output. We demonstrated a winter signal of mismatch with good agreement in warm season temperatures. This suggests that either our reconstruction methods are bias to warm winter estimates, or that the models are not capturing a feedback that leads to winter warming. Previous studies found that reduced sea-ice increased winter temperatures via cloud - further strengthening the need for the pursuit of a cloud proxy and increased investigation of cloud in palaeoclimate models.
The experimental stream comprises growth experiments growing silver birch in light conditions mimicking sun, cloud and shade. I constructed a novel growth set up, then grew the plants from seed. After six months the seedlings were harvested and their macromorphological characteristics were measured and analysed. Significant differences were found across multiple measured variables (e.g. leaf size, internode distance) between all three conditions. Leaves were cleared, stained and mounted for micromorphological analysis. Acquisition of the images for measurement and analysis is ongoing.
In support of this work, I conducted a project to test whether new methods in computer vision and machine learning could be applied to the leaf microscopic images for micromorphological analysis. We tested the method of preparation for the leaves, how to image them, and the viability of LeafNet for automated character data collection, as well as acquiring training and experience in student supervision that was identified as a training goal of the proposal. The results of the automated segmentation and measurement, with a community-science version, and our manual measurements, will be compared for speed and precision.
During my time at the University of Leeds, I have taken on the role of terrestrial data lead on the PlioMIP3 steering committee. As part of that role, I have contributed to the experimental design of the phase three experiments. I have also taken on the lead of the Neogene Terrestrial Climate database as part of the PAGES working group PlioMioVAR. This aims to improve data-model comparison for both PlioMIP3 and MioMIP2 and contributes to the greater project goal of identifying the causes of Pliocene data-model mismatch. This has allowed me to apply and develop the leadership training that was identified as part of my professional development for this fellowship.
