5:15 PM - 6:30 PM
[ACG36-P09] Comparative analysis of the GSMaP and quantitative precipitation estimation based on surface observation by radars and gauges in heavy rainfall events over Vietnam
Keywords:satellite precipitation observation, radar precipitation observation, Asian monsoon, data validation
1. Introduction
It is necessary to grasp the horizontal distribution of precipitation to forecast fluvial floods, inland flooding, and landslides. The Viet Nam Meteorological and Hydrological Administration (VNMHA) has implemented their automatic observation network of surface weather radars and rain gauges covering the whole country and built the quantitative precipitation estimation (QPE) system from these observations. This might improve the ability to forecast the pluvial disasters in Vietnam, although the performance of this system is still not clear. Moreover, the horizontal distribution of hourly precipitation can be optimized by combining surface observations and satellite data. In the present study, we compared surface and satellite rainfall observations to find out about consensus and inconsistency between them.
2. Data
Version 7 of the GSMaP (Aonashi et al. 2009, Ushio et al. 2009; Kubota et al. 2020) was adopted as a precipitation dataset based on satellite observation. We used four products including MVK, NRT, Gauge, and Gauge_NRT (Mega et al. 2019). Their resolution is horizontally 0.1 degrees and temporarily one hour. The QPE data provided by the VNMHA was used as a precipitation dataset based on surface observations. Their resolution is horizontally one kilometer and temporarily one hour. The QPE data were horizontally binned as fitting to the GSMaP's resolution. We chose 11 rainfall events in 2019 and 2020 for the analysis.
3. Results
3.1 Horizontal distribution
Figure 1 shows one heavy-rainfall case concerning the Tropical Cyclone ETAU on 9 November 2020. Scattering diagrams indicate a high correspondence between the QPE and GSMaP data with correlation coefficients exceeding 0.5 in three snapshots. Especially, both data show a higher agreement over the sea. This result implies that the surface and satellite observations can be complementarily useful in a forecast of a typhoon which often approaches from the eastward ocean in Vietnam, by providing horizontally continuous observations. Whereas, over the land, the GSMaP had heavier precipitation and an unnatural overestimating spot. Such a discrepancy should be overcome in the next step.
3.2 Synthetic analysis
We statistically compared the two datasets in all horizontal points and hours for each case in 2019.
In analyses for hourly precipitation, the GSMaP MVK had heavier precipitation than the QPE data in 8 out of 10 cases in 2019 (Fig. 2). Most of the selected cases were observed over lowlands relatively close to the coast. This characteristic roughly agreed with the previous study (Nodzu et al. 2019). The GSMaP relatively overestimated in light-rain cases and underestimated in heavy-rain cases if we regard the QPE as ground truth data. The GSMaP Gauge (Gauge_NRT) modified by gauge data has a higher similarity with the QPE than the GSMaP MVK (NRT) without modification in the analyses including root mean square difference, biases, and probability of detection (not shown).
The correspondence between the GSMaP and the QPE was assessed by varying accumulation time of precipitation (ATP) from one hour to 12 hours. Hourly or 2-hourly precipitation had a lower correspondence between the two datasets than precipitation with longer ATP. On the other hand, in the longer ATP, normalized RMSD and POD for 6-hourly precipitation had similar values with those for 12-hourly precipitation.
4. Summary
The GSMaP and the QPE precipitation datasets showed a high agreement in their horizontal distribution, especially over the ocean. The GSMaP often had larger values compared to the QPE. This could be a shared feature over lowlands relatively close to the coast observed in the previous studies. It seems that the adjustment algorithm with rain gauge data in the GSMaP successfully works judging from the analyses on the Gauge and Gauge_NRT products. Two common features could be seen in the comparison between the GSMaP and the QPE data: (1) the GSMaP had larger values in light-rain cases and smaller values in heavy-rain cases; (2) the GSMaP and the QPE had a higher correspondence for the precipitation with a longer ATP. Discussion based on the above comparison is helpful in the construction of an optimized and integrated precipitation dataset from the satellite and surface observations.
References
Aonashi, K. et al., 2009, J. Meteor. Soc. Japan, 87A, 119-136.
Kubota, T. et al., 2020, In: Levizzani V. et al. (eds) Satellite Precipitation Measurement. Advances in Global Change Research, vol 67., Springer, Cham.
Mega, T. et al., 2019, IEEE Trans. Geosci. Remote Sens., 57, 1928–1935.
Nodzu, M. I. et al., 2019, Prog. Earth Planet Sci. 6:58.
Ushio, T. et al., 2009, J. Meteor. Soc. Japan, 87A, 137-151.
It is necessary to grasp the horizontal distribution of precipitation to forecast fluvial floods, inland flooding, and landslides. The Viet Nam Meteorological and Hydrological Administration (VNMHA) has implemented their automatic observation network of surface weather radars and rain gauges covering the whole country and built the quantitative precipitation estimation (QPE) system from these observations. This might improve the ability to forecast the pluvial disasters in Vietnam, although the performance of this system is still not clear. Moreover, the horizontal distribution of hourly precipitation can be optimized by combining surface observations and satellite data. In the present study, we compared surface and satellite rainfall observations to find out about consensus and inconsistency between them.
2. Data
Version 7 of the GSMaP (Aonashi et al. 2009, Ushio et al. 2009; Kubota et al. 2020) was adopted as a precipitation dataset based on satellite observation. We used four products including MVK, NRT, Gauge, and Gauge_NRT (Mega et al. 2019). Their resolution is horizontally 0.1 degrees and temporarily one hour. The QPE data provided by the VNMHA was used as a precipitation dataset based on surface observations. Their resolution is horizontally one kilometer and temporarily one hour. The QPE data were horizontally binned as fitting to the GSMaP's resolution. We chose 11 rainfall events in 2019 and 2020 for the analysis.
3. Results
3.1 Horizontal distribution
Figure 1 shows one heavy-rainfall case concerning the Tropical Cyclone ETAU on 9 November 2020. Scattering diagrams indicate a high correspondence between the QPE and GSMaP data with correlation coefficients exceeding 0.5 in three snapshots. Especially, both data show a higher agreement over the sea. This result implies that the surface and satellite observations can be complementarily useful in a forecast of a typhoon which often approaches from the eastward ocean in Vietnam, by providing horizontally continuous observations. Whereas, over the land, the GSMaP had heavier precipitation and an unnatural overestimating spot. Such a discrepancy should be overcome in the next step.
3.2 Synthetic analysis
We statistically compared the two datasets in all horizontal points and hours for each case in 2019.
In analyses for hourly precipitation, the GSMaP MVK had heavier precipitation than the QPE data in 8 out of 10 cases in 2019 (Fig. 2). Most of the selected cases were observed over lowlands relatively close to the coast. This characteristic roughly agreed with the previous study (Nodzu et al. 2019). The GSMaP relatively overestimated in light-rain cases and underestimated in heavy-rain cases if we regard the QPE as ground truth data. The GSMaP Gauge (Gauge_NRT) modified by gauge data has a higher similarity with the QPE than the GSMaP MVK (NRT) without modification in the analyses including root mean square difference, biases, and probability of detection (not shown).
The correspondence between the GSMaP and the QPE was assessed by varying accumulation time of precipitation (ATP) from one hour to 12 hours. Hourly or 2-hourly precipitation had a lower correspondence between the two datasets than precipitation with longer ATP. On the other hand, in the longer ATP, normalized RMSD and POD for 6-hourly precipitation had similar values with those for 12-hourly precipitation.
4. Summary
The GSMaP and the QPE precipitation datasets showed a high agreement in their horizontal distribution, especially over the ocean. The GSMaP often had larger values compared to the QPE. This could be a shared feature over lowlands relatively close to the coast observed in the previous studies. It seems that the adjustment algorithm with rain gauge data in the GSMaP successfully works judging from the analyses on the Gauge and Gauge_NRT products. Two common features could be seen in the comparison between the GSMaP and the QPE data: (1) the GSMaP had larger values in light-rain cases and smaller values in heavy-rain cases; (2) the GSMaP and the QPE had a higher correspondence for the precipitation with a longer ATP. Discussion based on the above comparison is helpful in the construction of an optimized and integrated precipitation dataset from the satellite and surface observations.
References
Aonashi, K. et al., 2009, J. Meteor. Soc. Japan, 87A, 119-136.
Kubota, T. et al., 2020, In: Levizzani V. et al. (eds) Satellite Precipitation Measurement. Advances in Global Change Research, vol 67., Springer, Cham.
Mega, T. et al., 2019, IEEE Trans. Geosci. Remote Sens., 57, 1928–1935.
Nodzu, M. I. et al., 2019, Prog. Earth Planet Sci. 6:58.
Ushio, T. et al., 2009, J. Meteor. Soc. Japan, 87A, 137-151.