The paper presents an improved reconstruction of sea surface temperatures (SSTs) for the period 1854–1997, referred to as Extended Reconstruction SST version 2 (ERSST.v2). The new analysis addresses several shortcomings of the previous version (ERSST.v1), particularly in the western tropical Pacific, Atlantic, and Indian Oceans. Key improvements include: 1. **High-Frequency Analysis**: ERSST.v2 uses anomaly modes instead of anomaly increments for high-frequency analysis, which provides better resolution of tropical variations. The method involves fitting data to spatial modes derived from a well-sampled period, and the weights for these modes are determined to minimize squared error. 2. **Sea Ice Information**: The analysis incorporates sea ice concentrations to improve high-latitude SST estimates. A piecewise linear adjustment method is used to convert sea ice concentrations to SSTs, which is found to be more effective than a quadratic adjustment method. 3. **Bias Correction**: The historical bias correction for the period 1939–1941 is adjusted to linearly reduce the correction over this period, addressing issues with abrupt changes in the previous version. 4. **Error Analysis**: The total analysis error is separated into random, sampling, and bias error components. The low-frequency analysis is shown to have the largest error, particularly before 1900 and during the two world wars due to sparse sampling. The high-frequency error is usually the smallest, while the bias error can be significant before 1940. 5. **Results**: ERSST.v2 shows stronger variance in the western tropical Pacific and better representation of the tropical Atlantic and Indian Oceans compared to ERSST.v1. The analysis also includes improved error estimates, with the total standard error being consistent with the root-mean-squared difference (RMSD) between ERSST.v2 and HadISST, another SST analysis. 6. **Conclusions**: ERSST.v2 is an improvement over ERSST.v1 due to its stronger variance in key regions, inclusion of sea ice information, and more accurate error estimates. The analysis is available online at the National Climatic Data Center (NCDC) website, starting from January 1853. The paper also discusses the limitations of the analyses, particularly in regions with sparse data, and emphasizes the importance of using methods that incorporate large-scale spatial modes to efficiently utilize sparse observations.The paper presents an improved reconstruction of sea surface temperatures (SSTs) for the period 1854–1997, referred to as Extended Reconstruction SST version 2 (ERSST.v2). The new analysis addresses several shortcomings of the previous version (ERSST.v1), particularly in the western tropical Pacific, Atlantic, and Indian Oceans. Key improvements include: 1. **High-Frequency Analysis**: ERSST.v2 uses anomaly modes instead of anomaly increments for high-frequency analysis, which provides better resolution of tropical variations. The method involves fitting data to spatial modes derived from a well-sampled period, and the weights for these modes are determined to minimize squared error. 2. **Sea Ice Information**: The analysis incorporates sea ice concentrations to improve high-latitude SST estimates. A piecewise linear adjustment method is used to convert sea ice concentrations to SSTs, which is found to be more effective than a quadratic adjustment method. 3. **Bias Correction**: The historical bias correction for the period 1939–1941 is adjusted to linearly reduce the correction over this period, addressing issues with abrupt changes in the previous version. 4. **Error Analysis**: The total analysis error is separated into random, sampling, and bias error components. The low-frequency analysis is shown to have the largest error, particularly before 1900 and during the two world wars due to sparse sampling. The high-frequency error is usually the smallest, while the bias error can be significant before 1940. 5. **Results**: ERSST.v2 shows stronger variance in the western tropical Pacific and better representation of the tropical Atlantic and Indian Oceans compared to ERSST.v1. The analysis also includes improved error estimates, with the total standard error being consistent with the root-mean-squared difference (RMSD) between ERSST.v2 and HadISST, another SST analysis. 6. **Conclusions**: ERSST.v2 is an improvement over ERSST.v1 due to its stronger variance in key regions, inclusion of sea ice information, and more accurate error estimates. The analysis is available online at the National Climatic Data Center (NCDC) website, starting from January 1853. The paper also discusses the limitations of the analyses, particularly in regions with sparse data, and emphasizes the importance of using methods that incorporate large-scale spatial modes to efficiently utilize sparse observations.