Evaluation of Version 7 products from GPM/DPR

Major improvements in the GPM/DPR Level-2 Version 07 (V07) product include the introduction of a three-dimensional rain/no rain discrimination method in the PRE module. In the CSF module, flagHail, an index for the existence of hail, was introduced, as well as flagGraupelHail, a flag for hail and hail as an experimental product. In the SLV module, the DSD parameter ε was changed from a fixed vertical value to a variable value, and the effect of soil moisture over land was introduced in the estimation of the normalized surface scattering cross section (σ0). The multiple scattering index (MSindex) and non-uniformity index (NUBFindex) in the Trigger module (TRG) are newly released as standard products. Here, as an evaluation of DPR V07, we evaluated this product by comparing with V06 mainly in terms of global precipitation, and also compared the long-term trend for about 20 years from TRMM/PR using the single-frequency product. Next, we compared the solid precipitation products including newly introduced in V07: MSIndex, flagGraupelHail, flagHail, and flagHeavyIcePrecip. Furthermore, since it became clear in V07 that the surface precipitation intensity has an incidence angle dependence (this became apparent after the scan pattern of KaPR was changed to the same as KuPR in May 2018), we performed an analysis on the incidence angle dependence.

First, the comparison of the global precipitation was performed by comparing between V06 and V07 mainly in the range of global precipitation (latitude, 35S-35N and 60S-60N). The Full Swath (FS) product is the result of a change in the scan pattern of the KaPR in May 2018. In addition, the single-frequency algorithm was also revised, so long-term data from the TRMM era were also compared. Large difference didn’t appear between V06 and V07 for DPR product, whereas Ku's V07 has an 8% decrease over ocean and an 18% increase over land. This can be explained by the introduction of the effect of soil moisture content on σ0 from V07. The FS underestimates the DPR in general, and this is an issue to be addressed for V08.

The (large) solid precipitation products in GPM/DPR are flagHail and flagHeavyIcePrecip. The flagGraupelHail product has been added as an experimental product. In addition, MSIndex, an index of multiple scattering, has been added since V07. Since multiple scattering is more pronounced in the presence of hail, MSIndex can also be an indicator of large solid precipitation. We evaluated the characteristics of flagHeavyIcePrecip (flagHIP), flagHail (flagH), flagGraupelHail (flagGH), and MSIndex. From a preliminary analysis, the overlap between these parameters was found to be small: the frequency of MSIndex is small compared to the other products. From this perspective, the inclusion of hail (flagGH and flagHIP) and large snowflakes (flagHIP) may explain the small overlap. In fact, a Contoured Frequency Altitude Diagram (CFAD) was created for each of them, with flagHIP showing a prominent feature in the Zm (Ku) of the observed radar reflectivity factor. The Zm of many profiles peaked at around 27 dBZ at around -10°C, which is smaller than that of hail and hailstorms. In these profiles, Dm and Nw of DPR showed an increase in Dm and a decrease in Nw from around -15°C to -10°C altitude. This can be explained by an increase in size due to the mechanical association of dendritic crystals (Hobbs, 1973). That is, these could be due to large snowflake particles. Therefore, in order to classify flagHIPs as large snowflake particles, we subdivided to the flagHIP2 with a threshold value of approximately 30 dBZ. As a result, it was found that CFAD showed a large difference between the appearance of large snowflake particles and the rest of the time. The introduction of flagHIP2 reduced the overlap between flagGH and flagHIP, but the overlap between the two was approximately less than 50%, indicating that there is still room for improvement or subdivision. 45% of the MSIndex overlapped with flagGH or flagHIP.

To evaluate the incidence angle dependence of the DPR product, we evaluated the behavior of the DPR algorithm in terms of PIA: PIAobs that is estimated from observations of KuPR and KaPR, respectively, PIAhyb that is calculated consistent with ΔPIA in DPR, and the final value, PIAfinal. The relationship between PIAhyb(Ku) and PIAhyb(Ka) for this value is a straight line of 1:6 to be consistent with ΔPIA. The relationship between PIAobs(Ku) and PIAobs(Ka) ranged from 1:3.3 to 1:9. The change from PIAobs to PIAfinal shows that PIAfinal(Ku) is decreased when PIAobs(Ku) is overestimated from the 1:6 line, and PIAfinal(Ka) is decreased when PIAobs(Ka) is overestimated, and PIAfina(Ka) is increased in cases where the observed PIAobs(Ku) and PIAobs(Ka) are large. The distribution of ε values at the binClutterFreeBottom shows that ε is smaller for relatively large PIA(Ka). However, large ε appeared not only where PIA(Ka) was relatively large, but also in the center of the distribution. More detailed analysis is needed for the latter. Next, we examined the incident angle dependence of PIAobs (Ku) and PIAobs (Ka) using PIA (ζ) estimated from ζ of KuPR as an indicator, and found that there was almost no incident angle dependence in any PIA (ζ) in PIAobs (Ku), but the incident angle dependence of PIAobs (Ka) became more pronounced as PIA (ζ) became larger. Possible causes for this could include contamination of precipitation echoes near the ground surface.