Ruixi Liu , Qifeng Lu , ? , Min Chen , Lu Mo , Shuiyong Fn

a Key Laboratory of Radiometric Calibration and Validation for Environmental Satellite, China Meteorological Administration, National Satellite Meteorological Center,Beijing, China

b Institute of Urban Meteorology, China Meteorological Administration, Beijing, China

c Key Laboratory for Meteorological Disaster Monitoring and Early Warning and Risk Management of Characteristic Agriculture in Arid Regions, China Meteorological Administration, Yinchuan, China

d Ningxia Meteorological Observatory, Yinchuan, China

Keywords:FY-3C MWHS2 RMAPS-ST Data assimilation Precipitation forecast

A B S T R A C T In order to evaluate the impact of assimilating FY-3C satellite Microwave Humidity Sounder (MWHS2) data on rainfall forecasts in the new-generation Rapid-refresh Multi-scale Analysis and Prediction System—Short Term(RMAPS-ST) operational system, which is developed by the Institute of Urban Meteorology of the China Meteorological Administration, four experiments were carried out in this study: (i) Coldstart (no observations assimilated); (ii) CON (assimilation of conventional observations); (iii) FY3 (assimilation of FY-3C MWHS2 only);and (iv) FY3 + CON (simultaneous assimilation of FY-3C MWHS2 and conventional observations). A precipitation process that took place in central-eastern China during 4—6 June 2019 was selected as a case study. When the authors assimilated the FY-3C MWHS2 data in the RMAPS-ST operational system, data quality control and bias correction were performed so that the O-B (observation minus background) values of the five humidity channels of MWHS2 became closer to a normal distribution, and the data basically satisfied the unbiased assumption. The results showed that, in this case, the predictions of both precipitation location and intensity were improved in the FY3 + CON experiment compared with the other three experiments. Meanwhile, the prediction of atmospheric parameters for the mesoscale field was also improved, and the RMSE of the specific humidity forecast at the 850—400 hPa height was reduced. This study implies that FY-3C MWHS2 data can be successfully assimilated in a regional numerical model and has the potential to improve the forecasting of rainfall.

1. Introduction

Studies have shown that assimilating observational data can improve the initial conditions and thus improve the accuracy of numerical weather prediction (NWP) models ( Zou and Xiao, 2000 ). Meteorological satellites can provide a complete global coverage of data, which can supplement mesoscale information for the initial field of the model,and have thus become an important way to improve the accuracy of numerical weather forecasts ( Gong, 2013 ). The positive effect of assimilating satellite data in global numerical prediction models has been fully reviewed in a study by Eyre et al. (2020) . There has been an increasing number of applications of assimilating satellite data in regional numerical prediction models in recent years. Du et al. (2012) performed an assimilation of FY-3 satellite data into the variational system and found that the FY-3 data have an application potential in predicting the paths of typhoons. Dong et al. (2014) found that an assimilation of FY-3A microwave data improved the forecasting of intensity much better than that of the track by taking typhoon Fungwong as a case study.Ren et al. (2009) indicated that the value of the application of satellite data is affected by clouds. NCEP (National Centers for EnvironmentalPrediction) and ECMWF (European Centre for Medium-Range Weather Forecasts) have realized the assimilation of many kinds of satellite data in their operational 3DVar and 4DVar systems ( Lawrence et al., 2018 ).

It is well known that water vapor provides the main source of weather phenomena, such as rainstorms and typhoons. The polar orbiting meteorological satellite FY-3, independently developed by China,carries a Microwave Humidity Sounder (MWHS), which can detect the vertical distribution of atmospheric humidity globally. The data of FY-3’s MWHS have been assimilated in the ECWMF model and in China’s Global/Regional Assimilation and Prediction System (known as GRAPES), where they have proven to be effective in improving forecasts of rainstorms and typhoons ( Lawrence et al., 2018 ; Zhu and Qin, 2018 ).

In this study, in order to evaluate the impact of assimilating FY-3C MWHS2 data in a numerical weather forecast system on the subsequent rainfall forecast, FY-3C MWHS2 data were assimilated in the RMAPS-ST(Rapid-Refresh Multiscale Analysis and Prediction System—Short Term)operational model developed at the Beijing Institute of Urban Meteorology. A rainfall process that took place in central China during 4—6 June 2019 was taken as a case study. Section 2 introduces the FY-3C MWHS2 data; Section 3 introduces the RMAPS-ST model; the results are presented in Section 4 ; and conclusions are given in Section 5 .

2. FY ‐3C MWHS2 data

The FY-3C satellite was launched on 23 September 2013. The MWHS2 instrument onboard FY-3C has the ability to synchronously detect the vertical distribution of humidity and temperature. A full list of the channels and their frequencies is given in Table 1 . In summary,there are five humidity-sounding channels around the 183.31-GHz water vapor band, and two window channels at frequencies close to 89 GHz and 150 GHz. In addition, MWHS2 has eight new sounding channels around the 118-GHz oxygen band, which had not been included in a space-borne instrument before, and thus provide an interesting source of new information. These channels and the 183.31-GHz channels work jointly to detect the atmosphere, obtaining more precise vertical distributions of temperature and humidity, and providing rich information for numerical prediction and climate research.

Table 1 FY-3C MWHS2 characteristics.

The peak height of the 183.31-GHz band of the MWHS2 weight function is from 400 hPa to 800 hPa. Compared with NOAA’s AMSUB, 183.31 ± 1.8 GHz and 183.31 ± 4.5 GHz detection channels have been added in FY-3C’s MWHS2. The weight peaks of these two channels are 500 hPa and 700 hPa, which improves the detection ability of vertical distribution information on atmospheric humidity below 400 hPa( Guo et al., 2014 ). In this study, the data of all five humidity channels were used for assimilation experiments.

3. RMAPS ‐ST system

The RMAPS-ST operational system was used in this study, which was developed by the Institute of Urban Meteorology, China Meteorological Administration. The system is based on the Weather Research and Forecasting (WRF) model and the Weather Research and Forecasting model Data Assimilation (WRFDA) system ( Barker et al., 2012 ). It is a system of regional rapidly updated multi-scale data assimilation and short-term prediction (0—12 h). It employs 9-km and 3-km one-way nested domains covering the whole of China and northern China, respectively ( Zhong et al., 2020 ). It has 649 ×500 grid points and 51 vertical levels with a model top of 50 hPa. The ECMWF global medium-range forecast fields (at 0.25°resolution) were used as the initial and boundary conditions of the model. For further details on RMAPS-ST, refer to Zhong et al. (2020) .

RTTOV 11.3 was selected as the radiative transfer model for satellite data assimilation in this study. RTTOV was developed by ECMWF in the early 1990s. Saunders et al. (2013) gave a detailed description of RTROV. It was originally used to simulate the radiance of TOVS satellites. Up to now, it has been able to simulate the infrared and microwave radiance of various satellites, and has been continuously developed and improved.

4. Analysis of the impact of assimilating FY ‐3C MWHS2 data on rainfall forecasts

4.1. Case selection

In this study, a rainfall process that took place in central-eastern China during 4—6 June 2019 was selected for conducting the experiments. Influenced by a high trough, low-level vortex, and low-level jet,heavy rain fell in the provinces of Sichuan, Shannxi, southwestern Shandong, most parts of Henan and Anhui, northern Jiangsu, and northwestern and eastern Hubei.

4.2. Experimental design

In order to study the effect of assimilating MWHS2 data on the precipitation forecast, we carried out four experiments with the RMAPSST operational model. The data from five humidity channels of FY-3C MWHS2 were mainly assimilated in the 9-km outer nest of the model.The four experiments are summarized in Table 2 , in which ‘conventional data’ refers to surface, sounding, and airplane data from Global Telecommunication System and GPS-ZTD (Zenith Total Delay) data.

Table 2 Experiment details.

Our study domain covers the whole of China according to the passing territory of FY-3C, and the MWHS2 data at 1200 UTC on 4 June 2019 cover the study domain. Therefore, this paper gives the analysis and forecast results at 1200 UTC 4 June 2019 as the start time.

4.3. Preprocessing of FY -3C MWHS2 data

4.3.1.

Cloud

and

precipitation

detection

We assimilated satellite data under the clear-sky condition in this study. Therefore, the cloud detection and identification of MWHS2 was carried out first. The Scattering Index (SI) was employed to identify and eliminate the cloud pollution in the WRFDA-3DVar system. SI is defined as the difference in observed brightness temperature between the 89-GHz and 150-GHz window channels i.e., SI = Tb(89) - Tb(150). In this study, we chose the SI thresholds of 3.0, 4.0, and 5.0 respectively, and found SI = 4.0 to be the most appropriate.

As for precipitation detection, the NWP field was regarded as the first-guess field, and then the cloud liquid water path (CLWP) was calculated. If CLWP exceeded the threshold value of 0.2 mm, it was determined as rainfall, and the corresponding satellite data were directly removed in the assimilation process. In Fig. 1 , the white bright area indicates clouds observed by the FY-2G satellite, and the blue dots are the clouds identified in the step of cloud and precipitation detection.It can be seen that most of the MWHS2 observations with cloud in the coverage are effectively eliminated.

Fig. 1. Cloud and precipitation detection results (blue dots) and limb detection results (red dots) at 1200 UTC 4 June 2019.

4.3.2.

Quality

control

procedures

Extreme value detection: Observations with a brightness temperature beyond the range of 150—350 K were removed; brightness temperature deviation within - 20 K to 20 K was retained; and all observations with a residual of greater than three times the standard deviation of the corresponding channel were removed.

Surface type detection: Data on sea ice, snow, and all mixed surfaces were removed.

Limb detection: Data with a scanning angle of less than 8°or greater than 90°were removed. The red dots in Fig. 1 indicate the observations identified by limb detection. As shown, the data at the orbit edge will be removed in the assimilation.

4.3.3.

Bias

correction

Bias correction is mainly used to eliminate systematic bias and to meet the unbiasedness requirement of the variational assimilation technique. There are two kinds of methods to correct bias in satellite radiance: offline-BC ( Harris and Kelly, 2001 ) and Variational Bias Correction(VarBC; Auligne et al., 2007 ). VarBC was selected in our study. Fig. 2 is a histogram showing the distributions of O-B (observation minus background) for MWHS2 channels 11—15 before and after bias correction.The O-B distribution was obviously not a normal distribution before bias correction. After bias correction, the distributions approximated a Gaussian distribution. The variational bias correction effectively eliminated the systematic bias and helped the five MWHS2 channels meet the unbiasedness requirement, especially for channels 12—15.

There were 3645 FY-3C MWHS2 observational pixels in the study domain, and 1634, 1653, 1693, 1679, and 1611 pixels were reserved at channels 11—15 respectively after cloud detection, quality control, and bias correction for assimilation.

4.4. Analysis of the effect of FY -3C MWHS2 data assimilation on the rainfall forecast

4.4.1.

Effect

of

FY

-3C

MWHS2

data

assimilation

on

the

initial

field

of

the model

With the assimilation of five humidity channels from FY-3C MWHS2 in the RMAPS-ST operational forecast system, the humidity fields in the initial conditions differed remarkably from those in the experiments without this assimilation. In contrast, the differences in the temperature and wind fields in the initial conditions were negligible (figures omitted). Hence, we mainly discuss the influence on the humidity field here.Fig. 3 shows the humidity increment of different layers at the levels of 850 hPa, 700 hPa, 600 hPa, 500 hPa, and 400 hPa, before and after FY-3C MWHS2 data assimilation between the CON and FY3C + CON experiment. After assimilation of the FY-3C MWHS2 satellite data, the humidity changed significantly at all five levels, and the increment between 500 and 850 hPa was largest. The areas with large increments were northern China and Guizhou Province, corresponding to the cloudy area of satellite observations ( Fig. 3 (f)). However, the humidity decreased in the clear-sky region (red circle), which corresponded to the clear-sky area of the satellite image. So, the assimilation of FY-3C MWHS2 data increased the water vapor in the cloudy area and decreased the humidity in the clear-sky region.

Fig. 2. Histograms of distributions of O-B for MWHS2 channels 11—15 before(left) and after (right) bias correction: (a) channel 11; (b) channel 12; (c) channel 13; (d) channel 14; (e) channel 15.

4.4.2.

Influence

of

FY

-3C

MWHS2

data

assimilation

on

the

precipitation forecast

We conducted a 24-h precipitation forecast experiment and compared and analyzed the precipitation results simulated in the four forecast experiments at 3 h, 6 h, 12 h, 18 h, 21 h, and 24 h.

Fig. 4 shows the spatial distributions of the 3-h cumulative precipitation forecasted from 1200 UTC to 1500 UTC 4 June 2019 in the four experiments and from the rain gauge data, separately.

The actual precipitation indicated that there were two high-value centers, located in Sichuan and Shaanxi provinces ( Fig. 4 (e)). However, the precipitation in central Sichuan Province predicted by the four groups of experiments was obviously smaller (blue circle in Fig. 4 ), while all four experiments captured the precipitation from northern Sichuan to southern Shaanxi (red circle in Fig. 4 ). For the precipitation area in Shaanxi Province, compared with the actual observations ( Fig. 4 (e)),the intensities of precipitation predicted by the Coldstart ( Fig. 4 (a))and FY3 ( Fig. 4 (c)) experiments were overestimated, and the precipitation area predicted by the CON experiment was smaller ( Fig. 4 (b)). After the assimilation of conventional and MWHS2 observations together(FY3C + CON experiment, Fig. 4 (d)), the precipitation area was more consistent with the actual situation, and the high-value area in southeastern Shaanxi was also captured.

The reasons for failing to predict the precipitation area in central Sichuan Province, even after assimilation of satellite data, is that the MWHS2 orbit did not cover the precipitation area at this time ( Fig. 4 (f)).In addition, assimilating MWHS2 only had a small effect on the adjustment of the humidity field and thus the forecasted precipitation( Fig. 4 (c)). However, if MWHS2 and conventional observations were assimilated simultaneously, the model dynamic field was changed when ingesting wind information from conventional observations, and in turn,the period of influence on hygrometer data assimilation was prolonged.As a result, it enhanced the assimilation effect of the MWHS2 humidity data. Therefore, after the assimilation of both conventional observations and the MWHS2 data (FY3 + CON), the effect on the 3-h precipitation forecast was better than that of assimilating only MWHS2 data (FY3).

Fig. 5 (a—d) shows the 3-h cumulative precipitation predicted from 0900 UTC to 1200 UTC 5 June 2019 in the Coldstart, CON, and FY3C + CON experiments and actual observations, respectively.

As shown in the rain gauge precipitation map ( Fig. 5 (d)), precipitation was below 10 mm in Shanxi Province. However, in the Coldstart( Fig. 5 (a)) and CON ( Fig. 5 (b)) experiments, the precipitation in Shanxi Province was more than observed, even exceeding 30 mm in the south of the province. Meanwhile, the precipitation magnitude was closer to the actual value in the FY3 + CON experiment ( Fig. 5 (c)).

For the maximum center of precipitation in Henan Province, the precipitation range predicted by the Coldstart ( Fig. 5 (a)) and CON( Fig. 5 (b)) experiments was relatively small, while the precipitation predicted by assimilating MWHS2 ( Fig. 5 (c)) was slightly weaker than the actual situation, but the improvement was very obvious compared to the previous two experiments.

In Hubei Province, the spatial pattern of the precipitation distribution in the Coldstart ( Fig. 5 (a)) experiment was considerably different from the actual situation, and the precipitation predicted by CON was more than the actual situation ( Fig. 5 (b)). After assimilating MWHS2( Fig. 5 (c)), the distribution pattern of precipitation was more consistent with the actual observation and with similar intensity, which was better than the Coldstart and CON experiments.

Fig. 3. Distribution of humidity increment at (a) 850 hPa, (b) 700 hPa, (c) 600 hPa, (d) 500 hPa, and (e) 400 hPa, as well as (f) in the satellite image. The white area in (f) indicates clouds observed by the FY-2G satellite, while red dots show the clear-sky area.

Fig. 4. Spatial distribution of 3-h cumulative precipitation from 1200 UTC to 1500 UTC 4 June 2019 in the four experiments and in the rain gauge data: (a)Coldstart experiment; (b) CON experiment; (c) FY3 experiment; (d) FY3C + CON experiment; (e) rain gauge observation; (f) orbital coverage of FY-3C MWHS2 and cloud detection.

Fig. 5. Spatial distribution of 3-h cumulative precipitation from 0900 UTC to 1200 UTC 5 June 2019 in the forecast experiments and the rain gauge data: (a)Coldstart experiment; (b) CON experiment; (c) FY3C + CON experiment; (d) observation.

In conclusion, further assimilation of FY-3C MWHS2 data in the RMAPS-ST operational model in addition to assimilation of conventional data had a positive effect on precipitation prediction at 1200 UTC 4 June 2019, and improved the prediction of precipitation intensity and the falling area.

4.4.3.

Influence

of

FY

-3C

MWHS2

data

assimilation

on

predicting

the environmental

fields

The predicted 6-h profiles of the humidity field, temperature field,height field,

u

-wind field, and

v

-wind field were compared with sounding profiles, and their corresponding biases and root-mean-square error(RMSE) in each layer were calculated to further evaluate the influence of FY-3C MWHS2 data assimilation on predicting the environmental fields( Fig. 6 ). The bias of the humidity profile was reduced in the FY3 + CON experiment compared to the other experiments. After assimilation of MWHS2 data, the RMSE of three fields including humidity above 500 hPa, the temperature between 700 and 400 hPa, and the wind field in the lower level, were reduced in the forecast. Assimilation of FY-3C MWHS2 data improved the prediction of the environmental fields. The accuracy of humidity was improved after assimilating the MWHS2 data alone, and was even more pronounced after assimilating both MWHS2 and the conventional data simultaneously, as was the accuracy of the precipitation forecast. The main reason for the improvements is that the assimilation of the conventional wind data changes the dynamic field of the model, extends the duration of the influence of hygrometer data,and enhances the assimilation effect of humidity data. The improvement in the 6-h forecasts for the temperature field and wind fields was not obvious.

4.4.4.

Precipitation

forecast

score

For the precipitation predictions, the Critical Success Index (CSI) was employed to evaluate the forecast skill. The CSI indicates the forecast accuracy and reflects the degree of accuracy for an effective precipitation prediction; the ideal value is 1. The observed rainfall data from national rain gauges were selected for verification.

Fig. 7 shows the CSI scores of the precipitation forecast every 6 h for precipitation magnitudes

＞

0.1 mm,

＞

1.0 mm,

＞

5.0 mm,

＞

10.0 mm,

＞

25.0 mm, and

＞

50 mm. The red bars represent the CSIs after assimilation of MWHS2, and the blue bars are the CSIs before assimilation. The CSI scores of 10.0—25.0 mm were significantly improved in all 24 hours, and the prediction scores in the period of 12—18 hours were improved on the whole. In general, for moderate rainfall, the CSI score was enhanced significantly in the 6-h forecast.

Fig. 6. The bias and RMSE profile of the 6-h predicted (a) humidity field, (b) temperature field, (c) height field, (d) u -wind field, and (e) v -wind field.

Fig. 7. CSI scores of the precipitation forecast every 6 h for precipitation magnitudes ＞ 0.1 mm, ＞ 1.0 mm, ＞ 5.0 mm, ＞ 10.0 mm, ＞ 25.0 mm, and ＞ 50 mm.

5. Conclusions

In this study, FY-3 MWHS2 data in five humidity channels were successfully assimilated in the operational model RMAPS-ST. Through a rainfall case study in the Sichuan—Henan area of China on 4 June 2019,four experiments were carried out: (i) Coldstart (no observations assimilated); (ii) CON (assimilation of conventional observations); (iii) FY3(assimilation of FY-3C MWHS2 only); and (iv) FY3 + CON (simultaneous assimilation of FY-3C MWHS2 and conventional observations). The results indicated that:

(1) After assimilating the MWHS2 data, there was a positive humidity increment in the cloudy area and negative humidity increment in the clear-sky area, which improved the information and distribution of the humidity field.

(2) After assimilating MWHS2 data, the distribution pattern of the predicted precipitation area and intensity were closer to the actual observation. Meanwhile, the distributions of forecasted environmental fields improved, leading to a positive effect on the precipitation forecast.

This research demonstrates the potential for improving rainfall forecasts by assimilating FY-3C MWHS2 data in operational models. More operational evaluation experiments are needed to further improve the assimilation algorithm in the future.

Funding

Supported by the Second Tibetan Plateau Scientific Expedition and Research (STEP) program (grant no. 2019QZKK0105) and the National Key Research and Development Program of China ( 2018YFC1506603 ).