Abstract:
This report closely examines Version 6 of the NASA Soil Moisture Active Passive (SMAP) Level 4 Surface and Root Zone Soil Moisture (L4_SM) product, which was first released on 8 December 2021. The assessment includes comparisons of L4_SM soil moisture estimates with in situ measurements from SMAP core validation sites and sparse networks. Also provided is a quasi-global evaluation of the product’s anomaly correlation skill relative to the previous version and a model-only version, based on independent satellite radar soil moisture retrievals and an Instrumental Variable approach. The assessment further provides a global evaluation of the internal diagnostics from the ensemble-based data assimilation system that is used to generate the L4_SM product, including observation-minus-forecast (O-F) brightness temperature (Tb) residuals and soil moisture analysis increments. The core validation site comparisons, the assessment of the anomaly correlation skill using independent radar soil moisture retrievals, and the statistics of the assimilation diagnostics are considered primary validation methodologies for the L4_SM product. Comparisons against in situ measurements from regional-scale sparse networks are considered a secondary validation methodology because such in situ measurements are subject to upscaling errors from their native point-scale to the grid-cell scale of the data product. The validation period is April 2015 to March 2021.

The precipitation forcing data of earlier L4_SM versions was derived primarily through temporal and spatial downscaling of the NOAA Climate Prediction Center Unified (CPCU) gauge product using background data from the Goddard Earth Observing System (GEOS) Forward Processing (FP) weather analysis. Validation of the earlier L4_SM versions, however, revealed serious deficiencies in the CPCU product. To address the shortcomings in the CPCU product, Version 6 of the L4_SM algorithm primarily uses satellite-gauge and satellite-only products provided by the NASA Global Precipitation Measurement (GPM) mission. First, the reference precipitation climatology for the Version 6 L4_SM algorithm is based on the NASA Integrated Multi-satellitE Retrievals for GPM (IMERG) Final (Version 06B) product. Where the IMERG climatology is not available (in much of the high latitudes north of 60ºN), the climatology of the Global Precipitation Climatology Project (GPCP) v2.3 product is used as reference. Second, outside of North America, the precipitation forcing in the Version 6 L4_SM algorithm is corrected to match the daily totals from the IMERG (Version 06B) product. The IMERG-Final product, which is informed by satellite observations and monthly totals from precipitation gauges, was used during L4_SM reprocessing through 29 June 2021. Owing to the ~3.5-month latency of the IMERG-Final product, the satellite-only IMERG-Late product, which is available with ~14-hour latency, is used from 30 June 2021 to present. This switch from IMERG-Final to IMERG-Late inputs is reflected in a change in the L4_SM Science Version ID from Vv6032 to Vv6030. In North America, daily precipitation corrections in the Version 6 L4_SM algorithm are based on CPCU data, as in all previous L4_SM versions. Consequently, the mean latency of the Version 6 L4_SM product is still driven by that of the CPCU product and remains at ~2.5 days.

An analysis of the time-average surface and root zone soil moisture shows that the global pattern of arid and humid regions is well captured by the Version 6 L4_SM estimates. Compared to Version 5, the Version 6 surface and root-zone soil moisture is wetter in much of South America and Australia and drier in most of Africa, owing primarily to the revised precipitation climatology. These changes are also reflected in the surface turbulent fluxes and the surface and soil temperatures. Because of these climatological differences, the Version 5 and Version 6 products should not be combined into a single dataset for use in applications outside of the continental United States.

Results from the core validation site comparisons indicate that the Version 6 L4_SM product meets its accuracy requirement, which is formulated in terms of the root-mean square (RMS) error after removal of the long-term mean error, i.e., ubRMSE≤0.04 m3 m-3, where the error is vs. the unknown true soil moisture. Computed directly against core site in situ measurements at the 9 km scale, the average unbiased RMS difference (ubRMSD) of the 3-hourly Version 6 L4_SM data is 0.040 m3 m 3 for surface soil moisture and 0.027 m3 m-3 for root zone soil moisture. When factoring in the measurement error of the in situ data, the L4_SM product clearly meets the 0.04 m3 m-3 ubRMSE requirement.

The ubRMSD values of the Version 6 soil moisture are essentially unchanged from those of Version 5. There is a small increase in the correlation and anomaly correlation skill of the Version 6 surface soil moisture. On the other hand, the mean differences of the Version 6 surface and root-zone soil moisture from the corresponding in situ measurements are slightly larger in Version 6 than in Version 5. Since most of the in situ measurement sites are in North America, where both product versions use CPCU data for the daily precipitation corrections, it is not surprising that the skill of the Version 5 and 6 products is very similar.

The L4_SM estimates are an improvement compared to estimates from a model-only Open Loop (OL6000) simulation, which demonstrates the beneficial impact of the SMAP Tb data. Overall, L4_SM surface and root zone soil moisture estimates are more skillful than OL6000 estimates, with statistically significant improvements for surface soil moisture R and anomaly R values (based on 95% confidence intervals). The correlation and anomaly correlation skill differences between the L4_SM product and the Open Loop are slightly smaller in Version 6 than in Version 5. This is because the Open Loop baseline skill is somewhat larger in Version 6 than in Version 5, which leaves less room for improvement associated with the assimilation of the SMAP Tb observations.

Results from comparisons of the L4_SM product to in situ measurements from more than 400 sparse network sites corroborate the core validation site results.

The evaluation of the anomaly correlation skill based on the independent radar soil moisture retrievals reveals that the improvements in the Version 6 surface soil moisture (relative to Version 5) are concentrated in South America, Africa, Australia, and parts of East Asia. In these regions, the Version 5 system used either the uncorrected (simulated) GEOS FP precipitation (throughout Africa) or CPCU-corrected precipitation, albeit with the CPCU data relying on sparse or faulty regional gauge networks (including parts of South America, central Australia, and Myanmar). Using IMERG data for the daily precipitation corrections in the Version 6 algorithm considerably improves the surface soil moisture anomaly correlation in these regions.

The instantaneous soil moisture analysis increments lie within a reasonable range and result in spatially smooth soil moisture analyses. The long-term mean soil moisture analysis increments make up only a small fraction of the water budget. The O-F Tb residuals exhibit only a small regional bias (less than 3 K) between the (rescaled) SMAP Tb observations and the L4_SM model forecast, which indicates that the assimilation system is reasonably unbiased. The globally averaged time series standard deviation of the O-F Tb residuals is 5.1 K, which represents a reduction of ~0.4 K from that of the Version 5 product. This considerable improvement in the Tb simulation skill is concentrated in the same regions where the anomaly correlation skill is most improved based on the independent satellite radar soil moisture retrievals (that is, in parts of South America and Africa, central Australia, and Myanmar). Regionally, the time series standard deviation of the normalized O-F Tb residuals deviates considerably from unity, which indicates that regionally the L4_SM assimilation algorithm over- or underestimates the total (model and observation) error present in the system, although the consistency is somewhat improved in Version 6 compared to Version 5. The globally averaged time series standard deviation is 3.3 K for the observation-minus-analysis Tb residuals, reflecting the impact of the SMAP observations on the L4_SM system.

In summary, Version 6 of the L4_SM product is sufficiently mature and of adequate quality for distribution to, and use by, the larger science and application communities.