12:00 PM - 12:15 PM
[ACG36-12] Correction of path integrated attenuation estimates by considering soil moisture effect for the GPM Dual-frequency Precipitation Radar
★Invited Papers
Keywords:precipitation radar, attenuation correction, soil moisture
Spaceborne precipitation radars, the Tropical Rainfall Measuring Mission’s Precipitation Radar (PR; 13.8GHz) and the Global Precipitation Measurement mission’s Dual-frequency Precipitation Radar (DPR), continue measurements for more than 23 years. They can measure not only precipitation echoes but surface echoes, the latter of which are used to estimate path integrated attenuation (PIA) in surface reference technique (SRT; Meneghini et al. 2004; 2021).
Seto and Iguchi (2007) analyzed the measurements by PR, and found that attenuation-free surface backscattering cross section (denoted by σ0e) becomes higher during precipitation over land. This can be considered as a result of an increase in surface soil moisture, then it is called soil moisture effect. SRT may underestimate PIA if it does not consider this effect. In the PR standard algorithm (Iguchi et al. 2009), a constant offset of 0.5 dB is given to the PIA estimates by SRT (denoted by PIASRT) for the precipitation rate retrieval over land.
In this study, the measurements by KuPR (13.6GHz) and KaPR (35.5GHz) radars, which compose the DPR, are analyzed to examine if KuPR and KaPR show soil moisture effect as well as PR. Based on the analysis, a method to correct for PIASRT is developed and is implemented in the DPR standard algorithm (version 06A; Seto et al. 2021). The effect of the PIASRT correction on the precipitation rate estimates is introduced.
The average of measured surface backscattering cross section at no-rain pixels is calculated for each 1 degree by 1 degree grid, month, and angle bin (the value is denoted by σ0NR). At rain pixels, measured surface backscattering cross section (denoted by σ0m) is investigated. The anomaly of σ0m from the σ0NR is denoted by Δσ0m. Figure shows the relation between the average of Δσ0m and the surface precipitation rate estimated by KuPR (denoted by R) at the grid (100-95W, 30-35N) and for angle bin numbers 21-29. R (mm/h) is classified into 9 categories; Category 1 is for R < 0.5, Category k is for 2k-3 < R < 2k-2 (k = 2 … 8), and category 9 is for R > 64. For KuPR, Δσ0m increases with R up to category 4, and is higher than 0 up to category 5, though σ0m is subject to precipitation attenuation. This suggests the existence of soil moisture effect in KuPR. The anomaly of σ0e from σ0NR (denoted by Δσ0e) should be higher than Δσ0m. PIA estimated by Hitschfeld-Bordan method (PIAHB), which is independent of SRT, is used to estimate Δσ0e as Δσ0m + PIAHB. Δσ0m + PIAHB increases with R up to category 5, and is higher than 0 for categories 7 and lower. PIAHB is consider to be underestimated for heavy precipitation. On the other hand, Δσ0m + PIASRT is not clearly dependent on R and is slightly higher than 0. Δσ0m + PIASRT is lower than Δσ0m for categories 5 and lower and is lower than Δσ0m + PIAHB for categories 7 and lower. This reveals the necessity for the correction of PIASRT.
The offset can be regarded as Δσ0m + PIAHB – (Δσ0m + PIASRT). However, as PIAHB is likely to be underestimated for heavy precipitation, the maximum value of Δσ0m + PIAHB among 9 categories is searched. In this case for KuPR, it is found at category 5, then the maximum value replaces Δσ0m + PIAHB for categories 6 and higher (as shown by black solid line in the figure). Also as Δσ0m + PIASRT does not show clear dependence on R, the value is averaged regardless of R and the average value replaces Δσ0m + PIASRT (as shown by purple solid line). So as shown in the figure, the difference between the black solid line and purple solid line becomes offset value. If the value is negative, it is replaced by 0.
In the single-frequency algorithms of DPR, the retrieval processes are iterated twice. In the first loop, no offset is given to PIASRT and R is estimated. In the second loop, the offset is determined by R, then PIASRT is corrected and R is finally estimated. This correction is implemented into the DPR standard algorithms version 06A, and is tested for June 2016. In KuPR (inner swath only), the unconditional average of surface precipitation rate for over land is 2.55 mm/day with a 18.3% increase from 2.16 mm/day in the original product. In KaPR, the average of surface precipitation rate for over land increases by 15.1 % from 1.35 mm/day to 1.55 mm/day. The offset value is not very different between for KuPR and for KaPR. The same change of PIA leads to larger change in R for KuPR than for KaPR, but SRT is less reliable for KuPR and the change of PIASRT does not affect the final estimates of PIA and R. Then, the change ratio of R of KaPR is slightly lower than that for KuPR.
Seto and Iguchi (2007) analyzed the measurements by PR, and found that attenuation-free surface backscattering cross section (denoted by σ0e) becomes higher during precipitation over land. This can be considered as a result of an increase in surface soil moisture, then it is called soil moisture effect. SRT may underestimate PIA if it does not consider this effect. In the PR standard algorithm (Iguchi et al. 2009), a constant offset of 0.5 dB is given to the PIA estimates by SRT (denoted by PIASRT) for the precipitation rate retrieval over land.
In this study, the measurements by KuPR (13.6GHz) and KaPR (35.5GHz) radars, which compose the DPR, are analyzed to examine if KuPR and KaPR show soil moisture effect as well as PR. Based on the analysis, a method to correct for PIASRT is developed and is implemented in the DPR standard algorithm (version 06A; Seto et al. 2021). The effect of the PIASRT correction on the precipitation rate estimates is introduced.
The average of measured surface backscattering cross section at no-rain pixels is calculated for each 1 degree by 1 degree grid, month, and angle bin (the value is denoted by σ0NR). At rain pixels, measured surface backscattering cross section (denoted by σ0m) is investigated. The anomaly of σ0m from the σ0NR is denoted by Δσ0m. Figure shows the relation between the average of Δσ0m and the surface precipitation rate estimated by KuPR (denoted by R) at the grid (100-95W, 30-35N) and for angle bin numbers 21-29. R (mm/h) is classified into 9 categories; Category 1 is for R < 0.5, Category k is for 2k-3 < R < 2k-2 (k = 2 … 8), and category 9 is for R > 64. For KuPR, Δσ0m increases with R up to category 4, and is higher than 0 up to category 5, though σ0m is subject to precipitation attenuation. This suggests the existence of soil moisture effect in KuPR. The anomaly of σ0e from σ0NR (denoted by Δσ0e) should be higher than Δσ0m. PIA estimated by Hitschfeld-Bordan method (PIAHB), which is independent of SRT, is used to estimate Δσ0e as Δσ0m + PIAHB. Δσ0m + PIAHB increases with R up to category 5, and is higher than 0 for categories 7 and lower. PIAHB is consider to be underestimated for heavy precipitation. On the other hand, Δσ0m + PIASRT is not clearly dependent on R and is slightly higher than 0. Δσ0m + PIASRT is lower than Δσ0m for categories 5 and lower and is lower than Δσ0m + PIAHB for categories 7 and lower. This reveals the necessity for the correction of PIASRT.
The offset can be regarded as Δσ0m + PIAHB – (Δσ0m + PIASRT). However, as PIAHB is likely to be underestimated for heavy precipitation, the maximum value of Δσ0m + PIAHB among 9 categories is searched. In this case for KuPR, it is found at category 5, then the maximum value replaces Δσ0m + PIAHB for categories 6 and higher (as shown by black solid line in the figure). Also as Δσ0m + PIASRT does not show clear dependence on R, the value is averaged regardless of R and the average value replaces Δσ0m + PIASRT (as shown by purple solid line). So as shown in the figure, the difference between the black solid line and purple solid line becomes offset value. If the value is negative, it is replaced by 0.
In the single-frequency algorithms of DPR, the retrieval processes are iterated twice. In the first loop, no offset is given to PIASRT and R is estimated. In the second loop, the offset is determined by R, then PIASRT is corrected and R is finally estimated. This correction is implemented into the DPR standard algorithms version 06A, and is tested for June 2016. In KuPR (inner swath only), the unconditional average of surface precipitation rate for over land is 2.55 mm/day with a 18.3% increase from 2.16 mm/day in the original product. In KaPR, the average of surface precipitation rate for over land increases by 15.1 % from 1.35 mm/day to 1.55 mm/day. The offset value is not very different between for KuPR and for KaPR. The same change of PIA leads to larger change in R for KuPR than for KaPR, but SRT is less reliable for KuPR and the change of PIASRT does not affect the final estimates of PIA and R. Then, the change ratio of R of KaPR is slightly lower than that for KuPR.