3.3. Verification of correction results
This study selects the precipitation process occurring from May 16-17, 2020, for a case study. Affected by an upper-level trough and shear line, Guangdong experienced widespread intense convective weather during the period. The convective activity commenced in northern Guangdong from 08:00 (UTC+8) on May 16 to 02:00 (UTC+8) on May 17. Subsequently, convection in western Guangdong began to spread from the northwest to the southeast, while convection in eastern Guangdong gradually shifted from the southwest to the northeast. From 02:00 (UTC+8) to 10:00 (UTC+8) on May 17, a convective pause period followed the previous energy release. From 10:00 (UTC+8) on May 17 to 04:00 (UTC+8) on May 18, widespread strong convection moved from west to east, resulting in short-term heavy precipitation over a vast area and accompanied by gale-force winds.
The Guangzhou radar data underwent analysis three times, specifically at 19:06 (UTC+8) on May 16, 2020, 21:12 (UTC+8) on May 17, 2020, and 21:54 (UTC+8) on May 17, 2020.
Figure 6 showcases the reflectivity factor during these times. Interestingly, when heavy precipitation occurs locally (
Figure 6a), the convective cells exhibit a high intensity, reaching 50-55 dBZ at their peak. However, these cells are isolated, having a smaller area. As the precipitation transitions into the stratocumulus mixed stage (
Figure 5b,c), the convective system organizes itself, and the strength of the cells maintain. Furthermore, the northeastward movement of convective cells becomes swifter. Consequently, the frontal heavy echo region diminishes, leaving a vast expanse of stratiform precipitation at the rear.
The Z
H comparisons between SDPR and XPAR-D at three typical times (
Figure 7) reveal a typical instance of locally intense rainfall occurring at 19:06 on May 16 (
Figure 7a–d). The short attenuation path results in minor differences in Z
H between XPAR-D and SDPR (
Figure 7a,b). However, attenuation mostly affects the strong-echo area adjacent to the radar center (x in
Figure 7b). Murkily, XPAR-D fails to depict an area where the strong echoes extend beyond 50 dBZ.
At 21:12 (
Figure 7e–h) and 21:54 (
Figure 7i–l) on May 17, we can see typical patterns of stratocumulus and stratiform precipitation. Due to the longer attenuation path and more extensive echo region, the XPAR-D detection of Z
H (
Figure 7f,j) generally exhibits smaller values than the SDPR detection (
Figure 7e,i), particularly for strong echoes that are distant from the radar centers. In situations involving stratocumulus precipitation, the distant, high-intensity echoes detected by the SDPR reach 45-55 dBZ (y in
Figure 7e), while for the XPAR-D, they range from 25-50 dBZ (y in
Figure 7f), which implies that they are 10-20 dBZ lesser than the SDPR. Additionally, the weaker echoes in the same distance detected by the SDPR, the smaller deviation between the two radars (z in
Figure 7j).
The various attenuation correction algorithms exhibit different correction effects depending on the scenario. In the case of stratiform precipitation, both the Z
H-K
DP and MZ
H-K
DP methods yield good correction results for XPAR-D data (
Figure 7k,l), with relatively minor differences.
For instances of local heavy rainfall, despite having intense echoes, the deviations between the XPAR-D and SDPR detections remain small due to the short attenuation path. The XPAR-D echoes near the radar center tend to be weaker (x in
Figure 7b). After the MZ
H-K
DP correction, the strongest echo value of the XPAR-D agrees well with that of the SDPR, and the strong echo area expands slightly (x in
Figure 7d). However, after the Z
H-K
DP correction, the strongest echo of the XPAR-D reaches 55-60 dBZ, which is an overcorrection, and the strong echo region increases (x in
Figure 7c).
Regarding stratocumulus precipitation, the attenuation path of strong echoes is longer, with attenuation ranging from 5-10 dBZ and possibly up to 20 dBZ for the strong echoes that are distant from the radar center (y in
Figure 7f). After the MZ
H-K
DP correction, the strong echo area aligns well with that of the SDPR (y in
Figure 7h), while the Z
H-K
DP method overcorrects the echoes (y in
Figure 7g). In conclusion, overall, the MZ
H-K
DP method outperforms the Z
H-K
DP method.
The averaged Z
H values across various azimuths are utilized for quantitative comparisons (
Figure 8). Results show that after correction, the Z
H curves of different precipitation scenarios almost match the original Z
H curve in the initial 1/2-2/3 segments of the attenuation path, revealing that the attenuation of the XPAR-D echoes is minimal during this time. Hence, it becomes difficult to diminish the deviations between the SDPR and the XPAR-D through attenuation correction. In the latter 1/2-1/3 segments of the attenuation path, the attenuation of the XPAR-D echoes increases with distance, and the differences between the corrected Z
H curve and the original Z
H curve gradually become more prominent.
Concerning stratiform precipitation (
Figure 8a) and local heavy precipitation (
Figure 8c), the disparities between the two correction methods are minor. Both methods exhibit commendable performance for the correction of stratiform precipitation, while the correction outcome is fair for local heavy precipitation. Concerning stratocumulus precipitation (
Figure 8b), the deviations between the two correction methods intensify with distance. Significantly, the Z
H curve corrected by the MZ
H-K
DP method bears a closer resemblance to the SDPR Z
H curve, while the Z
H-K
DP method correction results show prominent discrepancies.
To conclude, the amplitude of attenuation is dependent on the path length and echo intensity. Longer paths and stronger echoes lead to potential overcorrection of the ZH-KDP method and accentuate the benefits of the MZH-KDP method.
Furthermore, by inspecting the Z
H values of all the times of three precipitation types during May 16-17, 2020, we quantitatively analyze the attenuation characteristics of different precipitation types and the performance of the correction algorithms. The statistical indicators employed are correlation coefficient (R), root mean square error (RMSE), normalized absolute error (NAE), and normalized relative error (NRE). Comparing the Z
H values between XPAR-D and SDPR (
Figure 9a,d,g), we discover that when the Z
H exceeds 40 dBZ, the XPAR-D detection values are lower than those of the SDPR for all three precipitation types, indicating notable attenuation for echoes beyond 40 dBZ. Nonetheless, for Z
H values below 40 dBZ, distinct precipitation types display different attenuation characteristics. Specifically, the XPAR-D detections are relatively inferior for the stratocumulus precipitation (
Figure 9d), the difference between the two radars is insignificant for the stratiform precipitation (
Figure 9g), and the XPAR-D detections are better for the local heavy precipitation (
Figure 9a).
The evaluation of the correction algorithm performance reveals that both MZH-KDP and ZH-KDP methods provide only minor correction of echoes below 40 dBZ, and the former method performs somewhat better. For echoes exceeding 40 dBZ, the attenuation correction algorithms perform notably, particularly for stratocumulus precipitation. Moreover, they exhibit some capability for attenuation correction of stratiform precipitation and local heavy precipitation.
In terms of locally heavy rainfall, we observe that the NRE values exceed 0, while positive deviations are present in XPAR-D echoes. Interestingly, these deviations are further amplified after attenuation correction. However, due to the efficient correction of strong echoes, we note significant improvement in R, RMSE, and NAE values post-correction. These outcomes indicate improved overall data quality upon correction, with the additional observation that the MZH-KDP method outperforms the ZH-KDP method.
For stratocumulus precipitation, the NRE values fall below 0, and XPAR-D echoes reflect distinct negative deviations. Notably, ZH-KDP method correction shows a decrease in negative deviations, albeit mainly attributable to the overcorrection of strong echoes. Thus, this does not imply that ZH-KDP method correction is superior. Statistical indicators of R, RMSE, and NAE reveal that the MZH-KDP method still performs better.
Regarding stratiform precipitation, we note overall negative deviation as evidenced by NRE values falling below 0. We can reduce these values to nearly 0 with effective correction by both algorithms. However, various statistical indicators confirm that the MZH-KDP method exhibits better correction performance than the ZH-KDP method.