Global warming arises from an energy imbalance where increased radiative forcing from greenhouse gases traps additional heat in the Earth system. Over recent decades, more than 90% of this heat has gone into the ocean, causing thermal expansion that leads to sea level rise. Additionally, warming of the cryosphere has caused significant amounts of freshwater from melting ice to enter the ocean, representing another major contributor to rising sea levels. It is therefore essential to obtain accurate estimates of the historical changes in ocean heat content and sea level rise, as these are necessary to inform our understanding of future climate scenarios and some of the regional impacts from climate change. Improving these historical estimates, however, requires overcoming uncertainty related to instrumental and spatial sampling biases.

Prior to 2004 estimates of ocean heat content rely primarily on temperature measurements from mechanical and expendable bathythermograph (BT) instruments that were deployed on large scales by naval vessels and ships of opportunity. These BT temperature measurements are subject to well-documented biases, but even the best calibration methods still exhibit residual biases when compared to high-quality temperature datasets. In Chapter I, we use a new approach to reduce biases in historical BT data. Our method consists of an ensemble of artificial neural networks that corrects biases with respect to depth, year, and water temperature in the top 10 meters. A global correction, as well as corrections optimized to specific BT probe types are presented for the top 1800 m. Even with comparable performances at reducing the instrumental biases, distinct patterns emerge across the calibration methods when they are extrapolated to BT data not included in our cross-instrument comparison. Multiple bias corrections should therefore be incorporated into studies of ocean heat content.

Sampling the deep ocean has remained a large technical challenge, leading to a sparse observational record below 2000 m. Due to methodological limitations at handling such sparse temperature data, many historical reconstructions of ocean heat content neglect this large volume of the ocean deeper than 2000 m. In Chapter II, we provide a global reconstruction of historical changes in full-depth ocean heat content based on interpolated subsurface temperature data using an autoregressive artificial neural network, providing estimates of total ocean warming for the period 1946-2019. We find that cooling of the deep ocean and a small heat gain in the upper ocean led to no robust trend in global ocean heat content from 1960-1990, implying a roughly balanced Earth energy budget within -0.16 to 0.06 W m-2 over most of the latter half of the 20th century. However, the past three decades have seen a rapid acceleration in ocean warming, with the entire ocean warming from top to bottom at a rate of 0.63±0.13 W m-2. These results suggest a delayed onset of a positive Earth energy imbalance relative to previous estimates, although large uncertainties remain.