Chatterjee, Dwaipayan ORCID: 0000-0001-8757-9884 (2024). Characterization of Cloud Variability through Novel Satellite-Based Observations. PhD thesis, Universität zu Köln.
PDF (Dwaipayan Chatterjee, PhD Thesis, Institute fo Geophysics and Meteorology, University of Cologne)
Thesis_Chatterjee_publication.pdf - Accepted Version Bereitstellung unter der CC-Lizenz: Creative Commons Attribution. Download (77MB) |
Abstract
Cloud formation and evolution across different spatiotemporal scales present some of the most complex mechanisms in Earth's atmosphere. These complexities lead to various cloud types displaying diverse characteristics regarding their water content and impact on the radiative budget. Our limited capacity to represent the governing physical laws, likely because of too many unknown parameters and insufficient high-resolution observations, hampers the semi-empirical sub-grid representation of clouds in the numerical models. Consequently, clouds emerge as substantial sources of uncertainty in climate sensitivity experiments, diminishing our confidence and constraining our capacity to predict the future trajectory of our climate. Along with this, especially low-level clouds strongly modulate the surface solar irradiance (SSI), which impacts solar photovoltaic production. The relevance of this impact has increased with time as our electric grids have been increasingly infused with renewable energy, thus becoming more dependent on their intermittent nature. This causes sudden stress on microgrids, which causes them to struggle to balance the resulting variations around the standard frequency. The current state-of-the-art Geostationary Satellites (GS), like Meteosat Third Generation (MTG), offer spatial resolution at 1km, yet this still falls short of resolving the fine-scale nature of many cloud objects, which extend to hundreds of meters or finer. However, by treating the aggregation of these individual cloud objects spanning up hundreds of kilometers as spatial distributions, the resolution likely becomes adequate to characterize their variability. Research increasingly supports the notion that these cloud fields (CF) exhibit distinct cloud radiative effects and form under different large-scale conditions. Consequently, advanced-generation climate models must consider their variability. This work develops two neural networks (defined together as the framework) based on self-supervision principles to taxonomize the entire daytime visible mesoscale CF spectra from satellite images. Their individual pretext tasks simplify the entire cloud spectrum by focusing on the essential features of the CFs and thus reduce their high dimensional intricate distribution. We use cloud optical depth images retrieved from GS measurements as an input in both of our networks. Exploiting these networks and GS observations over central Europe (land) and north Atlantic trades (NAT, ocean), we demonstrate how different regimes can be distinguished as well as investigate their organizational interconnections. Our investigation includes the spring to autumn period over land and the dry season over NAT which both feature convective activity. The framework maps the spatial distribution of CFs based on their optical depth characteristics, facilitating the identification of distinct classes over both ocean and land. Their physical uniqueness is validated independently using supplementary satellite-retrieved products and reanalysis data. This enables the identification of low-level CFs, which are subsequently further distinguished to capture their difficult-to-separate diverse convective stages. Capturing different regimes of the low cloud family allows us to better characterize their particular contribution to SSI temporal fluctuations using ground-based pyranometer measurements. The exploration of organizational interconnections among low-level shallow clouds takes place within the downstream trades, where a framework systematically arranges them based on their organizational similarities. This method enables the identification of weak break points in the cloud organization continuum, which are interpreted as regime transitions. A novel metric has been introduced to measure the relative changes in CF organizational status during their transition. Further, a simplified representation of the complete CF spectrum, along with its partitioning, facilitates a comparison between state-of-the-art human labels and machine labels, thus assessing their positions in the continuum. Despite these advancements, shortcomings persist, particularly with lower consensus labels being misclassified, leading to an uncertain allocation within the continuum. Our analysis indicates that the results depend on the spatial scale of the CFs. Assessing the generalization potential of the framework by ingesting the continuous temporal data over Jülich, Germany, gives us the flexibility to explore the ground-based measurements from the Jülich Observatory for Cloud Evolution (JOYCE). We track the temporal evolution of the CF and emphasize the reasoning within the latent space. Our study illustrates how intrinsic cloud properties influence changes in neural embeddings, where we validate the results with ground-based radar and pyranometer observations. The developed framework opens up the possibility of using artificial intelligence framework-driven metrics for model evaluation against observational data.
Item Type: | Thesis (PhD thesis) | ||||||||
Translated abstract: |
|
||||||||
Creators: |
|
||||||||
URN: | urn:nbn:de:hbz:38-742158 | ||||||||
Date: | 2024 | ||||||||
Language: | English | ||||||||
Faculty: | Faculty of Mathematics and Natural Sciences | ||||||||
Divisions: | Faculty of Mathematics and Natural Sciences > Department of Geosciences > Institute for Geophysics and Meteorology | ||||||||
Subjects: | Data processing Computer science Natural sciences and mathematics Earth sciences |
||||||||
Uncontrolled Keywords: |
|
||||||||
Date of oral exam: | 12 September 2024 | ||||||||
Referee: |
|
||||||||
Funders: | Federal Ministry for the Environment, Nature Conservation, Nuclear Safety, and Consumer Protection, Federal Ministry for Digital and Transport, Federal Ministry of Education and Research | ||||||||
Projects: | KISTE, Deutsche Forschungsgemeinschaft, Warmworld | ||||||||
Refereed: | Yes | ||||||||
URI: | http://kups.ub.uni-koeln.de/id/eprint/74215 |
Downloads
Downloads per month over past year
Export
Actions (login required)
View Item |