Combining Statistical, Physical, and Historical Evidence to Improve Historical Sea-Surface Temperature Records
CitationChan, Duo. 2020. Combining Statistical, Physical, and Historical Evidence to Improve Historical Sea-Surface Temperature Records. Doctoral dissertation, Harvard University Graduate School of Arts and Sciences.
AbstractSea surface temperature (SST) is a crucial quantity for interpreting past climate variations, yet reconstructing historical SSTs from ship-based measurements is challenging because of varying biases that are on the order 0.1°C to 1°C. Whereas SST biases are complicated and can arise from a variety of physical and historical reasons, existing corrections are oversimplified and insufficient because of limited metadata. In this thesis, I combine statistical, physical, and historical evidence to perform more refined SST corrections and explore the associated implications for interpreting past climate variations.
I first develop a statistical linear-mixed-effect (LME) model to intercompare nearby measurements and quantify systematic offsets between distinct nations and data-collecting groups. After applying this method to groups of SSTs collected by buckets, I find systematic groupwise offsets (P.05). Removing groupwise offsets leads to SSTs in better correlations with independent station-based coastal air temperatures. Without groupwise corrections, the North Atlantic has been thought to warm more than twice as fast as the North Pacific in the early-20th century. Groupwise corrections increase SST trends over the North Pacific and lead to spatially more homogeneous early-20th century warming. The corrected inter-basin difference in warming rates between the North Atlantic and the North Pacific becomes consistent with the range of CMIP5 simulations. This result indicates an increased contribution of anthropogenic activities (compared with internal variability) to drive climate variations since the 20th century. Corrections in the Pacific mainly come from correcting a drop in Japanese SSTs in the 1930s, which is attributed to truncation errors during digitization, as indicated by a document from the U.S. Air Force. In addition to historical evidence, I also use physical evidence (i.e., diurnal cycles) to understand the origins of groupwise offsets. Investigating the evolution of the relationship between the amplitude of diurnal cycles and groupwise offsets reveals that more than 50% of groupwise offsets after the 1930s arise from incorrect metadata that misclassifies warmly biased engine-room-intake (ERI) SSTs as coming from buckets.
I then apply the LME method to homogenize all ship-based SSTs. I find that the long-standing warm SST anomaly during World War II reflects mainly warm biases. Identified biases arise mostly from unusual wartime measurement practices that involved (1) a rapid increase in the proportion of ERI SSTs measured by naval ships and (2) collecting nighttime bucket SSTs inside ships. Although previous studies hypothesized these data problems, existing corrections are insufficient because they rely on assumed magnitudes of corrections. On the contrary, the LME method quantifies groupwise offsets directly from the data and, therefore, allows for sufficient removal of data heterogeneity. Groupwise-corrected daytime SSTs show no significant World War II anomaly and become consistent with CMIP5 model simulations.
Finally, I explore the implication of groupwise SST corrections on century-long simulations of the North Atlantic hurricane frequency. When forced with SSTs without groupwise bucket corrections, the GFDL-HiRAM (a high-resolution atmospheric model) fails to reproduce hurricane counts that are statistically consistent with observational hurricane reconstructions. The simulated hurricane counts, however, become in line with observational estimates when using groupwise-corrected SSTs. This result demonstrates possible advances that groupwise SST corrections can bring to a broader community in the atmospheric, ocean, and climate sciences.
In sum, this thesis aims at developing more refined SST corrections and improving the interpretation of historical SST variability. It also provides a linear-mixed-effect intercomparison method that can be used to quantify and clean up data heterogeneity for many other climate and non-climate datasets that pool information from heterogeneous sources. Groupwise SST corrections will not be realized and confirmed without combining statistical, physical, and historical evidence. Correcting for groupwise offsets reveals a simpler and smoother evolution of past SSTs and reconciles several data-model discrepancies. Bringing observational estimates into accord with our current knowledge of forcing, climate sensitivity, and internal variability leads to greater confidence in future predictions of global warming made by climate models.
Citable link to this pagehttps://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37368188
- FAS Theses and Dissertations