Since September 2014, NASA's Orbiting Carbon Observatory-2 (OCO-2) satellite has been taking measurements of reflected solar spectra and using them to infer atmospheric carbon dioxide levels. This work provides details of the OCO-2 retrieval algorithm, versions 7 and 8, used to derive the column-averaged dry air mole fraction of atmospheric CO₂ (XCO₂) for the roughly 100 000 cloud-free measurements recorded by OCO-2 each day. The algorithm is based on the Atmospheric Carbon Observations from Space (ACOS) algorithm which has been applied to observations from the Greenhouse Gases Observing SATellite (GOSAT) since 2009, with modifications necessary for OCO-2. Because high accuracy, better than 0.25 %, is required in order to accurately infer carbon sources and sinks from XCO₂, significant errors and regional-scale biases in the measurements must be minimized. We discuss efforts to filter out poor-quality measurements, and correct the remaining good-quality measurements to minimize regional-scale biases. Updates to the radiance calibration and retrieval forward model in version 8 have improved many aspects of the retrieved data products. The version 8 data appear to have reduced regional-scale biases overall, and demonstrate a clear improvement over the version 7 data. In particular, error variance with respect to TCCON was reduced by 20 % over land and 40 % over ocean between versions 7 and 8, and nadir and glint observations over land are now more consistent. While this paper documents the significant improvements in the ACOS algorithm, it will continue to evolve and improve as the CO₂ data record continues to expand.

The version 10 (v10) Atmospheric Carbon Observations from Space (ACOS) Level 2 full-physics (L2FP) retrieval algorithm has been applied to multiyear records of observations from NASA's Orbiting Carbon Observatory 2 and 3 sensors (OCO-2 and OCO-3, respectively) to provide estimates of the carbon dioxide (CO₂) column-averaged dry-air mole fraction (XCO₂). In this study, a number of improvements to the ACOS v10 L2FP algorithm are described. The post-processing quality filtering and bias correction of the XCO₂ estimates against multiple truth proxies are also discussed. The OCO v10 data volumes and XCO₂ estimates from the two sensors for the time period of August 2019 through February 2022 are compared, highlighting differences in spatiotemporal sampling but demonstrating broad agreement between the two sensors where they overlap in time and space. A number of evaluation sources applied to both sensors suggest they are broadly similar in data and error characteristics. Mean OCO-3 differences relative to collocated OCO-2 data are approximately 0.2 and −0.3 ppm for land and ocean observations, respectively. Comparison of XCO₂ estimates to collocated Total Carbon Column Observing Network (TCCON) measurements shows root mean squared errors (RMSEs) of approximately 0.8 and 0.9 ppm for OCO-2 and OCO-3, respectively. An evaluation against XCO₂ fields derived from atmospheric inversion systems that assimilated only near-surface CO₂ observations, i.e., did not assimilate satellite CO₂ measurements, yielded RMSEs of 1.0 and 1.1 ppm for OCO-2 and OCO-3, respectively. Evaluation of uncertainties in XCO₂ over small areas, as well as XCO₂ biases across land–ocean crossings, also indicates similar behavior in the error characteristics of both sensors. Taken together, these results demonstrate a broad consistency of OCO-2 and OCO-3 XCO₂ measurements, suggesting they may be used together for scientific analyses.

The Thermal And Near infrared Sensor for carbon Observation – Fourier Transform Spectrometer (TANSO-FTS) on the Japanese Greenhouse gases Observing SATellite (GOSAT) has been returning data since April 2009. The version 9 (v9) Atmospheric Carbon Observations from Space (ACOS) Level 2 Full Physics (L2FP) retrieval algorithm (Kiel et al., 2019) was used to derive estimates of carbon dioxide (CO₂) dry air mole fraction (XCO₂) from the TANSO-FTS measurements collected over its first 11 years of operation. The bias correction and quality filtering of the L2FP XCO₂ product were evaluated using estimates derived from the Total Carbon Column Observing Network (TCCON) as well as values simulated from a suite of global atmospheric inversion systems (models) which do not assimilate satellite-derived CO₂. In addition, the v9 ACOS GOSAT XCO₂ results were compared with collocated XCO₂ estimates derived from NASA's Orbiting Carbon Observatory-2 (OCO-2), using the version 10 (v10) ACOS L2FP algorithm. These tests indicate that the v9 ACOS GOSAT XCO₂ product has improved throughput, scatter, and bias, when compared to the earlier v7.3 ACOS GOSAT product, which extended through mid 2016. Of the 37 million soundings collected by GOSAT through June 2020, approximately 20 % were selected for processing by the v9 L2FP algorithm after screening for clouds and other artifacts. After post-processing, 5.4 % of the soundings (2×10⁶ out of 37×10⁶) were assigned a “good” XCO₂ quality flag, as compared to 3.9 % in v7.3 (<1 ×10⁶ out of 24×10⁶). After quality filtering and bias correction, the differences in XCO₂ between ACOS GOSAT v9 and both TCCON and models have a scatter (1σ) of approximately 1 ppm for ocean-glint observations and 1 to 1.5 ppm for land observations. Global mean biases against TCCON and models are less than approximately 0.2 ppm. Seasonal mean biases relative to the v10 OCO-2 XCO₂ product are of the order of 0.1 ppm for observations over land. However, for ocean-glint observations, seasonal mean biases relative to OCO-2 range from 0.2 to 0.6 ppm, with substantial variation in time and latitude. The ACOS GOSAT v9 XCO₂ data are available on the NASA Goddard Earth Science Data and Information Services Center (GES-DISC) in both the per-orbit full format (https://doi.org/10.5067/OSGTIL9OV0PN, OCO-2 Science Team et al., 2019 b ) and in the per-day lite format (https://doi.org/10.5067/VWSABTO7ZII4, OCO-2 Science Team et al., 2019 a ). In addition, a new set of monthly super-lite files, containing only the most essential variables for each satellite observation, has been generated to provide entry level users with a light-weight satellite product for initial exploration (CaltechDATA, https://doi.org/10.22002/D1.2178, Eldering, 2021 ). The v9 ACOS Data User's Guide (DUG) describes best-use practices for the GOSAT data (O'Dell et al., 2020). The GOSAT v9 data set should be especially useful for studies of carbon cycle phenomena that span a full decade or more and may serve as a useful complement to the shorter OCO-2 v10 data set, which begins in September 2014.

A novel method for classifying Arctic precipitation using ground based remote sensors is presented. Using differences in the spectral variation of microwave absorption and scattering properties of cloud liquid water and ice, this method can distinguish between different types of snowfall events depending on the presence or absence of condensed liquid water in the clouds that generate the precipitation. The classification reveals two distinct, primary regimes of precipitation over the Greenland Ice Sheet (GIS): one originating from fully glaciated ice clouds and the other from mixed-phase clouds. Five years of co-located, multi-instrument data from the Integrated Characterization of Energy, Clouds, Atmospheric state, and Precipitation at Summit (ICECAPS) are used to examine cloud and meteorological properties and patterns associated with each precipitation regime. The occurrence and accumulation of the precipitation regimes are identified and quantified. Cloud and precipitation observations from additional ICECAPS instruments illustrate distinct characteristics for each regime. Additionally, reanalysis products and back-trajectory analysis show different synoptic-scale forcings associated with each regime. Precipitation over the central GIS exhibits unique microphysical characteristics due to the high surface elevations as well as connections to specific large-scale flow patterns. Snowfall originating from the ice clouds is coupled to deep, frontal cloud systems advecting up and over the southeast Greenland coast to the central GIS. These events appear to be associated with individual storm systems generated by low pressure over Baffin Bay and Greenland lee cyclogenesis. Snowfall originating from mixed-phase clouds is shallower and has characteristics typical of supercooled cloud liquid water layers, and slowly propagates from the south and southwest of Greenland along a quiescent flow above the GIS.

Long‐running eddy covariance flux towers provide insights into how the terrestrial carbon cycle operates over multiple timescales. Here, we evaluated variation in net ecosystem exchange (NEE) of carbon dioxide (CO₂) across the Chequamegon Ecosystem‐Atmosphere Study AmeriFlux core site cluster in the upper Great Lakes region of the USA from 1997 to 2020. The tower network included two mature hardwood forests with differing management regimes (US‐WCr and US‐Syv), two fen wetlands with varying levels of canopy sheltering and vegetation (US‐Los and US‐ALQ), and a very tall (400 m) landscape‐level tower (US‐PFa). Together, they provided over 70 site‐years of observations. The 19‐tower Chequamegon Heterogenous Ecosystem Energy‐balance Study Enabled by a High‐density Extensive Array of Detectors 2019 campaign centered around US‐PFa provided additional information on the spatial variation of NEE. Decadal variability was present in all long‐term sites, but cross‐site coherence in interannual NEE in the earlier part of the record became weaker with time as non‐climatic factors such as local disturbances likely dominated flux time series. Average decadal NEE at the tall tower transitioned from carbon source to sink to near neutral over 24 years. Respiration had a greater effect than photosynthesis on driving variations in NEE at all sites. Declining snowfall offset potential increases in assimilation from warmer springs, as less‐insulated soils delayed start of spring green‐up. Higher CO₂ increased maximum net assimilation parameters but not total gross primary productivity. Stand‐scale sites were larger net sinks than the landscape tower. Clustered, long‐term carbon flux observations provide value for understanding the diverse links between carbon and climate and the challenges of upscaling these responses across space.

The presence of 3D cloud radiative effects in OCO-2 retrievals is demonstrated from an analysis of 2014–2019 OCO-2 XCO2 raw retrievals, bias-corrected XCO2bc data, ground-based Total Carbon Column Observation Network (TCCON) XCO2, and Moderate Resolution Imaging Spectroradiometer (MODIS) cloud and radiance fields. In approximate terms, 40 % (quality flag – QF = 0, land or ocean) and 73 % (QF = 1, land or ocean) of the observations are within 4 km of clouds. 3D radiative transfer calculations indicate that 3D cloud radiative perturbations at this cloud distance, for an isolated low-altitude cloud, are larger in absolute value than those due to a 1 ppm increase in CO₂. OCO-2 measurements are therefore susceptible to 3D cloud effects. Four 3D cloud metrics, based upon MODIS radiance and cloud fields as well as stand-alone OCO-2 measurements, relate XCO2bc–TCCON averages to 3D cloud effects. This analysis indicates that the operational bias correction has a nonzero residual 3D cloud bias for both QF = 0 and QF = 1 data. XCO2bc–TCCON averages at small cloud distances differ from those at large cloud distances by −0.4 and −2.2 ppm for the QF = 0 and QF = 1 data over the ocean. Mitigation of 3D cloud biases with a table lookup technique, which utilizes the nearest cloud distance (Distkm) and spatial radiance heterogeneity (CSNoiseRatio) 3D metrics, reduces QF = 1 ocean and land XCO2bc–TCCON averages from −1 ppm to near ±0.2 ppm. The ocean QF = 1 XCO2bc–TCCON averages can be reduced to the 0.5 ppm level if 60 % (70 %) of the QF = 1 data points are utilized by applying Distkm (CSNoiseRatio) metrics in a data screening process. Over land the QF = 1 XCO2bc–TCCON averages are reduced to the 0.5 (0.8) ppm level if 65 % (63 %) of the data points are utilized by applying Diastkm (CSNoiseRatio) data screening. The addition of more terms to the linear regression equations used in the current bias correction processing without data screening, however, did not introduce an appreciable improvement in the standard deviations of the XCO2bc–TCCON statistics.