Atlantic surface freshwater fluxes in the HOAPS satellite climatology

by Stephan Bakan, Rainer Hollmann, Jörg Burdanowitz, Julian Kinzel, Christian Klepp, Marc Schröder and Stefan Bühler

Phase transitions among ice, liquid water and water vapour link the global atmospheric water cycle to the energy cycle through latent heat conversion. This mainly involves evaporation (E), precipitation (P), and their difference, the freshwater flux (E-P). Specifically over the ocean, in-situ measurements of E-related parameters remain scarce. In this regard, the Hamburg Ocean Atmosphere Parameters and Fluxes from Satellite data (HOAPS; Andersson et al., 2010; Fennig et al., 2012) provides E-related parameters over the global ice-free ocean from 1987 to 2013. However, its application in climate research and assimilation experiments demands systematic and random uncertainties. To estimate the uncertainty of E in HOAPS, ship and buoy records from the Seewetteramt (DWD) and ICOADS (V 2.5) served as ground reference. To derive the systematic uncertainty, multi-dimensional look-up tables were created using double collocation that allows for assigning instantaneous biases to each HOAPS E-related record as a function of the concurrent atmospheric state. Random uncertainties were derived by excluding the effect of random collocation and in-situ measurement errors by utilizing the novel approach of Multiple Triple Collocation analysis (Kinzel et al., 2016). On climatological scales, largest total uncertainties in E occur in the vicinity of subtropical high pressure systems, over the Arabian Sea, and the western boundary currents (Fig. 1). Locally, these uncertainties exceed 1.5 mm d-1Detailed analyses indicate that the maximum subtropical errors are mainly related to uncertainties in near-surface specific humidity, whereas the errors along the boundary currents are mainly caused by uncertainties in wind speed. In contrast, uncertainty maxima in the Arabian Sea are likely linked to SST retrieval uncertainties.

The lack of in-situ global surface-ocean P data limited the capabilities to validate satellite P retrievals. Since 2010, the Ocean Rainfall And Ice-phase precipitation measurement Network (OceanRAIN; Klepp 2015) collects surface P data from optical disdrometers (ODM470) deployed on various research vessels (RVs). The ODM470 was developed to measure under open-sea conditions with frequently varying high wind speed, phase changes, and sea state. Deployed in the highest ship mast, the ODM470 continuously measures the P particle size distribution (PSD). The PSD is converted into a P rate, which requires knowing the P phase (PP) of the falling particles. The PP has so far been manually derived from 3-hourly observations and RV-measured ancillary data in a time-consuming procedure without uncertainties. Thus, we developed a novel automatic algorithm to derive the PP from available ancillary RV data (Burdanowitz et al., 2016). We applied and validated a logistic regression approach over the Atlantic Ocean based on 4 years of RV Polarstern data using a method by Koistinen and Saltikoff (1998). For RV Polarstern, we find that a combination of air temperature, relative humidity and 99th percentile of the P particle diameter serves best to determine the PP. A rain/snow distinction excluding mixed-phase P reaches the highest accuracy of more than 91%; including mixed-phase P it decreases to 81%. Our novel approach uses two individual PP distributions to determine the PP, which outperforms the commonly used one-distribution approach. The new automatic PP distinction algorithm remarkably speeds up the data post-processing while introducing an objective PP probability at one-minute resolution. Thus, OceanRAIN serves well to estimate the P error in HOAPS, required to estimate the E-P error. However, HOAPS currently overlaps with OceanRAIN in less than 3.5 years. This limitation introduces a bias due to a sparse data sampling density over the tropics (Fig. 2). This issue will be adressed with a prolonged HOAPS-3.3 and OceanRAIN time series until Dec 2015, provided by CM-SAF and UHH in May 2016. The increased OceanRAIN–HOAPS overlap of 5.5 years makes more RVs available that adds to the statistical weight of the analysis. Furthermore, we investigate the spatial-scale difference between OceanRAIN along-track ship data and areal HOAPS satellite data by simulating this point-to-area comparison using subtropical island S-Pol radar data from the trades that previously served well to compare to gridded satellite data (Burdanowitz et al., 2015). First results indicate that the largest representativeness error of the along-track P occurs at low rain-area coverages and short P events along the tracks. Knowing this collocation uncertainty essentially contributes to derive the P-related error in HOAPS.

Combining E- and P-related uncertainty enables to derive  E-P error estimates. As E-P directly imposes changes on the surface ocean salinity budget and hence the density structure of the upper ocean, E-P modifies the large-scale thermohaline circulation. To investigate regions of moisture sources (E-P>0) and sinks (E-P<0), we performed atmospheric water vapour transport (WVT) calculations over the Atlantic Ocean, following Sohn et al. (2004). For this, we derived monthly mean WVT using HOAPS E-P (ocean only) and vertical profiles of ERA-Interim wind speed and humidity (land only). Special focus was put on Isthmus of Panama (that usually loses water vapor to the Pacific), link to climate indices (El Niño, NAO), and signals of (sub-) tropical cyclones by means of anomalous WVT convergence. The observed anomalous westward moisture transport and associated WVT convergence over the Labrador Sea during negative NAO phase (Fig. 3) are of specific interest because they tend to reduce the local deep water convection and hence impact the strength of the global oceanic conveyor belt.

The scientific objective for the 2nd phase is to link the small-scale variability E and P  to the large scale dynamics over the Atlantic Ocean in order to better understand E-P fluxes. In specific, this includes understanding the mean state and inter-annual variability of Atlantic E-P sources and sinks and the associated atmospheric water vapour transports from extratropical storm tracks to tropical rain belts. From the atmospheric side of the freshwater budget, we aim to provide global E-P flux fields as well as high-resolution E and P data from ships to calculate the along-track E-P to the oceanographic community, and link our findings to the larger atmospheric research community. The mostly data-driven project  includes HOAPS satellite and OceanRAIN in-situ observations over the ocean as well as the GECCO ocean synthesis. For comparison, CMIP6 climate model ensembls will be analyzed.

Key tasks comprise:

  1. Analysing mean state and inter-annual variability of P and E areas over the Atlantic Ocean by correlation with dynamical phenomena on multi-annual timescales, such as the North Atlantic Oscillation (NAO) and the El Niño Southern Oscillation (ENSO).

  2. Reassessing to what extent climate models correctly cover the mean state and variability of E and P, given the new HOAPS4 satellite data and OceanRAIN in-situ validation data as well as the latest available GECCO data for the ocean state.

  3. Analysing the sources and sinks as well as spatiotemporal variability of E-P in HOAPS.

  4. Verifying the budget closure between E, P, and water vapour transport (WVT). This includes an error budget assessment of WVT in HOAPS.

  5. Assessing consistency between observed sea surface salinity variability and conclusions from 4.