FY-4 AGIIRS atmospheric temperature profile data assimilation method

Abstract

The present application relates to a kind of FY-4A GIIRS atmospheric temperature profile data assimilation method, the method comprises the following steps: (1) the quality control method of FY-4A GIIRS atmospheric temperature profile ATP data is established, and the ATP data after quality control is obtained;(2) the statistical method of observation error of FY-4A GIIRS atmospheric temperature profile ATP data is established, and the root mean square error of FY-4A GIIRS ATP and ERA5 data is obtained;(3) the preprocessing method of assimilating FY-4A GIIRS atmospheric temperature profile ATP data is established, and the ATP data after preprocessing is obtained;(4) the cyclic assimilation and numerical prediction of FY-4A GIIRS atmospheric temperature profile ATP data.The present application is low in cost, high in efficiency, and innovatively proposes to use the observation error estimated based on reanalysis data in three-dimensional variational assimilation, and the effective assimilation time frequency of inversion temperature profile data is determined by sensitivity test.

Description

Technical Field

[0001] This invention relates to the field of weather forecasting technology, and in particular to a method for assimilating FY-4A GIIRS atmospheric temperature profile data. Background Technology

[0002] In satellite data assimilation studies, considerable research has been conducted on infrared hyperspectral data carried on polar-orbiting satellites. McNally et al. (2006) assimilated clear-sky AIRS (Atmospheric Infrared Sounder) radiation data in the ECMWF (European Centre for Medium-range Weather Forecasts). Following this, many scholars applied the assimilation of AIRS radiation data to typhoon forecasting with beneficial results (Reale et al., Zhou et al., 2010). Prunet et al. (1998) studied the application potential of simulated IASI (Infrared Atmospheric Sounding Interferometer) in numerical weather prediction. Hilton et al. (2009a) confirmed the assimilation performance of IASI in the UK Met Office numerical model through a series of OSSEs. Subsequently, Hilton et al. (2009b) assimilated IASI in the ECMWF operational system. Numerous scholars have conducted a series of assimilation studies on CrIS (Cross-track Infrared Sounder) radiometric data (Gambacorta and Barnet, 2012; Liang and Weng, 2014; Smith et al., 2015; Eresmaa et al., 2017; Li and Zou, 2017; Li et al., 2019).

[0003] The GIIRS, carried aboard my country's FY-4A geostationary satellite, is the world's first infrared hyperspectral instrument mounted on a geostationary satellite. It uses Michelson interferometry to observe infrared radiation in different spectral bands to obtain the vertical distribution of atmospheric temperature, humidity, and ozone (Yang et al., 2017). The National Meteorological Satellite Center has publicly released GIIRS primary regional observation data since January 24, 2019, and secondary inversion products, including atmospheric temperature profiles (ATP), since November 27, 2018. Due to the relatively short data release period, there are currently few studies on the assimilation of GIIRS data. Yinet et al. (2020) used the GRAPES global 4DVAR system to study the bias characteristics and bias correction of GIIRS. The results showed that after quality control and cloud detection processing of the brightness temperature in the long-wavelength temperature channels, the average bias of most channels in uncontaminated channels was within ±2 K. Subsequently, Yin et al. (2021) directly assimilated GIIRS radiation data in the GRPAES global 4DVAR assimilation system. The assimilation experiment was able to describe the temporal and spatial structure of typhoon processes in more detail and improve precipitation forecasts in coastal areas during typhoon landfall.

[0004] However, the above-mentioned direct assimilation of infrared radiation data requires the construction of complex observation operators (radiative transfer models), channel selection, bias correction, cloud detection, and quality control, which places high demands on computing power and assimilation systems, and is difficult to implement and has higher costs. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a low-cost and high-efficiency method for assimilating FY-4A GIIRS atmospheric temperature profile data.

[0006] To address the above problems, the present invention provides an assimilation method for FY-4A GIIRS atmospheric temperature profile data, comprising the following steps:

[0007] (1) Establish a quality control method for FY-4A GIIRS atmospheric temperature profile ATP data:

[0008] ① Each point on each vertical profile in ATP contains its own quality control label, which is divided into the following four types: perfect, marked as perfect, QF = 0; good, marked as good, QF = 1; bad, marked as bad, QF = 2; unusable, marked as do not use, QF = 3; ATP data with a quality control label greater than 0 are removed, and data with a quality control label of perfect are selected.

[0009] ② Data with temperatures below -100 °C and above 60 °C are removed based on physical constraints, and data outside the observation range (30 °E-180 °E, 0 °E-60 °E) are removed based on geographical constraints;

[0010] ③Based on cloud detection data, remove ATP data containing cloud data to obtain quality-controlled ATP data;

[0011] (2) Establishing a statistical method for observational errors in FY-4A GIIRS atmospheric temperature profile ATP data:

[0012] i. Using a 2-hour time window, the ATP data within one hour before and after the nearest ERA5 data point are unified to the grid points of the ERA5 data using an interpolation method;

[0013] ii. Using the quality-controlled ATP data, calculate the root mean square error of the FY-4A GIIRS ATP and ERA5 data according to the following formula:

[0014]

[0015] In the formula: N This represents the total number of grid points. M i For the first i ATP data from each network point; O i For the first i ERA5 data at grid points;

[0016] (3) Establish a preprocessing method for assimilating FY-4A GIIRS atmospheric temperature profile ATP data:

[0017] Based on the background field at the assimilation time, the quality-controlled ATP data is first unified onto the background field grid points, and then the data located outside the analysis area and outside the analysis time in the quality-controlled ATP data are removed to obtain the processed data;

[0018] b. Convert the processed data into the LITTLE_R format used by the assimilation system to obtain the preprocessed ATP data;

[0019] (4) Cyclic assimilation and numerical forecasting of FY-4A GIIRS atmospheric temperature profile ATP data:

[0020] A uses the root mean square error as the observation error of the ATP data in the assimilation system, and uses a three-dimensional variational assimilation system to assimilate the preprocessed ATP data.

[0021] In the assimilation procedure of step A, B designs time windows of 15 minutes, 30 minutes, 1 hour and 2 hours to cycle and assimilate ATP data, tests the sensitivity of different time windows when assimilating ATP data, and determines the impact of different time assimilation windows on analysis and forecasting.

[0022] Compared with the prior art, the present invention has the following advantages:

[0023] 1. To achieve efficient and high-quality assimilation of GIIRS inverted temperature profile data, this invention innovatively proposes the use of observation errors estimated based on reanalysis data in three-dimensional variational assimilation, and determines the effective assimilation time frequency of the inverted temperature profile data through sensitivity experiments.

[0024] 2. This invention was validated through a nationwide precipitation event numerical forecasting experiment. The results show that using this invention to assimilate the GIIRS ATP cycle yields an analytical field with smaller errors compared to the background field and ATP observation data. Compared to the control experiment, the temperature in the assimilation experiment's forecast field was effectively adjusted, with temperatures in certain regions and at specific altitudes closer to the ERA5 reanalysis data. From the spatial distribution of the predicted 24-hour cumulative precipitation, the assimilation experiment effectively suppressed spurious precipitation events observed in the control experiment and adjusted the morphology and magnitude of heavy precipitation centers.

[0025] 3. Compared with existing direct assimilation of GIIRS radiation data, this invention adopts indirect assimilation of ATP data to improve the initial field of the model. Compared with traditional assimilation and inversion data, this invention achieves effective assimilation of GIIRS inversion temperature profile data (ATP) from my country's geostationary satellite FY-4A in a three-dimensional variational assimilation system based on the observation error estimated by the inversion data. This improves the accuracy of the initial field of the numerical model and enhances the level of numerical weather prediction in my country.

[0026] 4. The temperature profile obtained by assimilating GIIRS inversion in this invention can easily and quickly improve the thermal environment in the initial field, requiring relatively low computational costs. For some operational forecasting systems that do not yet have the capability to directly assimilate satellite data, assimilating GIIRS inversion products from FY-4A is a simple and efficient method. Attached Figure Description

[0027] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0028] Figure 1 This is the assimilation experiment design diagram for GIIRS ATP of the present invention within a 12-hour assimilation duration at different cycle assimilation time windows.

[0029] Figure 2The RMSE of ATP in the vertical direction in June (a), July (b) and August (c) of 2020, according to an embodiment of the present invention.

[0030] Figure 3 The above figures represent the Fair Threat Scores (ETS) for the 6-hour cumulative precipitation forecasts from 12:00 on August 1, 2020 to 12:00 on August 3, 2020, in this embodiment of the invention. (a) is a 1 mm threshold, (b) is a 5 mm threshold, (c) is a 15 mm threshold, and (d) is a 30 mm threshold. The vertical axis represents the ETS score, and the horizontal axis represents the forecast duration.

[0031] Figure 4 This is a forecast performance diagram of the 6-hour cumulative precipitation at different thresholds for each experiment in this embodiment of the invention, from 12:00 to 18:00 on August 2, 2020 (24-hour forecast). (a) 1 mm, (b) 5 mm, (c) 15 mm, (d) 30 mm. Detailed Implementation

[0032] An assimilation method for FY-4A GIIRS atmospheric temperature profile data includes the following steps:

[0033] (1) Establish a quality control method for FY-4A GIIRS atmospheric temperature profile ATP data:

[0034] ① Each point on each vertical profile in ATP contains its own quality control label, which is divided into the following four types: perfect, marked as perfect, QF = 0; good, marked as good, QF = 1; bad, marked as bad, QF = 2; unusable, marked as do not use, QF = 3; ATP data with a quality control label greater than 0 are removed, and data with a quality control label of perfect are selected.

[0035] ② Data with temperatures below -100 °C and above 60 °C are removed based on physical constraints, and data outside the observation range (30 °E-180 °E, 0 °E-60 °E) are removed based on geographical constraints;

[0036] ③ Based on cloud detection data, remove ATP data containing cloud data to obtain quality-controlled ATP data.

[0037] (2) Establishing a statistical method for observational errors in FY-4A GIIRS atmospheric temperature profile ATP data:

[0038] i. Using a 2-hour time window, the ATP data within one hour before and after the nearest ERA5 data point are unified to the grid points of the ERA5 data using an interpolation method;

[0039] ii. Using the quality-controlled ATP data, calculate the root mean square error of the FY-4A GIIRS ATP and ERA5 data according to the following formula:

[0040]

[0041] In the formula: N This represents the total number of grid points. M i For the first i ATP data from each network point; O i For the first i ERA5 data at grid points.

[0042] (3) Establish a preprocessing method for assimilating FY-4A GIIRS atmospheric temperature profile ATP data:

[0043] Based on the background field at the assimilation time, the quality-controlled ATP data are first unified onto the background field grid points, and then the data located outside the analysis area and outside the analysis time in the quality-controlled ATP data are removed to obtain the processed data.

[0044] b converts the processed data into the LITTLE_R format used by the assimilation system to obtain the preprocessed ATP data.

[0045] (4) Cyclic assimilation and numerical forecasting of FY-4A GIIRS atmospheric temperature profile ATP data:

[0046] A uses the root mean square error as the observation error of the ATP data in the assimilation system and uses the three-dimensional variational assimilation system to assimilate the preprocessed ATP data.

[0047] In step A's assimilation procedure, time windows of 15 minutes, 30 minutes, 1 hour, and 2 hours were designed to cycle and assimilate ATP data. The sensitivity of different time windows when assimilating ATP data was tested, and the impact of different time assimilation windows on analysis and forecasting was determined. Example

[0048] This study evaluated assimilation effects using case studies from precipitation events occurring in Northeast, North, and South my country during the summer of 2020. Five experiments were designed, including a control trial (CTL) that did not assimilate any observations, and four ATP cycling assimilation trials with different time windows. Figure 1The experimental design diagram for the cyclic assimilation assay is presented. The CTL was initiated at 18:00 on July 31, 2020, with a spin-up time of 6 hours, and a forecast period of 12+48 hours from 00:00 on August 1. To determine the effect of different time windows on assimilation, ATP data from 00:00 to 12:00 were divided into assimilation windows of 15 minutes (C15min), 30 minutes (C30min), 1 hour (C1hour), and 2 hours (C2hour). The C15min assay began its first analysis at 00:00, analyzing data from 00:00 to 00:15, cycling sequentially until 11:45. The C30min assay began its first analysis at 00:15, analyzing data from 00:00 to 00:30, cycling sequentially until 11:45. The C1-hour experiment began its first analysis at 00:30, analyzing data from 00:00 to 01:00, cycling sequentially until 11:30. The C2-hour experiment began its first analysis at 01:00, analyzing data from 00:00 to 02:00, cycling sequentially until 11:00. After the final analysis, all assimilation experiments were projected to 12:00 on August 3rd. The projected results were generated by assimilating ATP data from 00:00 to 12:00 using different assimilation time windows.

[0049] This invention utilizes the three-dimensional compressible non-hydrostatic mesoscale numerical weather prediction model WRF4.2.1 (Skamarock et al., 2019) for numerical weather prediction. The model has a horizontal resolution of 9 km and 50 eta layers in the vertical direction. Initial fields and boundary conditions are provided by the Global Forecast System (GFS) dataset, with a time interval of 3 hours and a resolution of 0.25° × 0.25°. The physical parameterization schemes mainly used are the Thompson microphysical parameterization scheme (Thompsone et al., 2008), the Dudhia shortwave radiation scheme (Dudhia, 1989), the RRTM longwave radiation scheme (Mlawer et al., 1997), the ACM2 boundary layer scheme (Pleim, 2007), and the Noah land surface model (Ek et al., 2003). To better analyze convection processes, the KF cumulus parameterization scheme (Kain, 2004) was used.

[0050] The assimilation data of this invention are ATP data with QF = 0 under clear sky conditions within the model domain from 00:00 to 12:00 on August 1, 2020. The assimilation data is the data after processing in step (3). Figure 2 Step (2) is given to calculate the observation error of ATP. Figure 2The observational errors indicate that the RMSE of ATP between 700 hPa and 100 hPa is greater than 1 K and less than 2 K, and the errors are relatively stable compared to other altitudes. This section proposes to assimilate only ATP data between 700 hPa and 100 hPa. The observational errors in the assimilation procedure use the RMSE of the month in which the assimilation occurred.

[0051] We selected the observed temperature (T) minus the background field temperature (OMB) and the observed temperature minus the analytical field temperature (OMA) at different altitudes during the last analysis time of the C2hour experiment (11:00 AM on August 1st). Near the 500 hPa and 200 hPa altitudes, the deviation of the temperature in the analytical field compared to ATP observations was significantly reduced. Near the 500 hPa altitude, the background field showed a large positive deviation in Xinjiang, northern Inner Mongolia, and northeastern Tibetan Plateau, with positive deviations exceeding 3 K at some locations, while the analytical field significantly reduced this positive deviation. Overall, near the 500 hPa altitude, the average deviation of the background field was 0.1934, the root mean square error was 0.8881, and the standard deviation was 0.8668, while the average deviation of the analytical field was 0.0164, the root mean square error was 0.3926, and the standard deviation was 0.3923. It can be seen that the deviation, root mean square error, and standard deviation between the analytical results and ATP observations are significantly reduced, proving that the analytical results are significantly closer to the ATP observations. The analysis results also showed a reduction in root mean square error and standard deviation near the 200 hPa height.

[0052] The data were selected based on the temperature at 850 hPa of ERA5 at 12:00 on August 1, 2020, and the difference between the control experiment and each assimilation experiment at 850 hPa and ERA5. In the northwestern region (regions G and E) on the northern edge of the Tibetan Plateau, the 850 hPa altitude is close to the ground due to topographical influences, and there are large areas of desert and Gobi. Therefore, the temperature simulated by ERA5 in this region is relatively high. Compared to ERA5 data, the high temperatures simulated by the control experiment in this region exhibit a cold bias (~3 K), while the ATP assimilation experiments at different time windows can reproduce the high temperature distribution at this location well, with smaller deviations from ERA5 data. In East China (region C), ERA5 data shows a warm center in this region, and all experiments exhibit significant cold biases compared to ERA5. The cold bias of the assimilation experiments in this region is greater than that of the control experiment. Comparing the effects of different assimilation time windows, the C15min experiment has the largest cold bias in East China, followed by the C30min, C1hour, and C2hour experiments.

[0053] Analysis of the spatial distribution of 24-hour cumulative precipitation forecasts from various experiments revealed that the main advantage of assimilating ATP data under clear-sky conditions lies in suppressing false precipitation forecasts in the CTL experiment, and it also shows some improvement in identifying the centers of heavy precipitation in certain regions. To further evaluate the impact of assimilating ATP data on precipitation forecasting, Figure 3 The ETS scores for 6-hourly cumulative precipitation forecasts from 12:00 on August 1st to 12:00 on August 3rd, 2020, are presented for each experiment. The CTL experiment showed higher ETS scores than the assimilation experiment within the first 18 hours of the forecast at the 1 mm and 5 mm thresholds. As the forecast duration increased, the ETS scores of the assimilation experiments gradually surpassed those of the CTL experiment. At the 15 mm and 30 mm thresholds representing heavy rain and torrential rain, the assimilation experiments showed higher ETS scores than the CTL experiment, with the highest scores for the C15-minute and C30-minute experiments occurring after the 24th hour of the forecast. In calculating the ETS score, the penalty for false positives is smaller than for false negatives; therefore, the CTL experiment can achieve higher ETS scores than the assimilation experiment at smaller thresholds (1 mm and 5 mm) even with large areas of false precipitation.

[0054] Figure 4 The forecast performance plots for 6-hour cumulative precipitation from 12:00 to 18:00 on August 2, 2020 (24 hours after the forecast) are presented. At the 1 mm and 5 mm thresholds, the scoring techniques for 6-hour cumulative precipitation from each experiment are quite similar, with all experiments located above the diagonal line with a slope of 1, indicating a wet bias in precipitation forecasts for that threshold. At the 15 mm and 30 mm thresholds, it is clear that the CTL experiment has the lowest hit rate and success rate, while assimilating ATP data yields a higher hit rate and success rate.

[0055] In summary, analysis of the spatial distribution and scoring techniques of the cumulative precipitation forecasts from various experiments shows that the assimilation experiment using this invention effectively suppressed spurious precipitation events observed in the CTL experiment in terms of precipitation range and intensity. Furthermore, under the heavy precipitation threshold, the assimilation experiment achieved better precipitation scoring technique. Among all assimilation experiments, the C15min experiment demonstrated the highest scoring technique. This invention indicates that using a 15-minute assimilation time window provides better temporal representativeness and smaller observation errors, thus resulting in better assimilation forecast performance.

Claims

1. A method for assimilating FY-4A GIIRS atmospheric temperature profile data, comprising the following steps: (1) Establish a quality control method for FY-4A GIIRS atmospheric temperature profile ATP data: ① Each point on each vertical profile in ATP contains its own quality control label, which is divided into the following four types: perfect, marked as perfect, QF = 0; good, marked as good, QF = 1; bad, marked as bad, QF = 2; unusable, marked as do not use, QF = 3; ATP data with a quality control label greater than 0 are removed, and data with a quality control label of perfect are selected. ② Data with temperatures below -100 °C and above 60 °C are removed based on physical constraints, and data outside the observation range (30 °E-180 °E, 0 °E-60 °E) are removed based on geographical constraints; ③Based on cloud detection data, remove ATP data containing cloud data to obtain quality-controlled ATP data; (2) Establishing a statistical method for observational errors in FY-4A GIIRS atmospheric temperature profile ATP data: i. Using a 2-hour time window, the ATP data within one hour before and after the nearest ERA5 data point are unified to the grid points of the ERA5 data using an interpolation method; ii. Using the quality-controlled ATP data, calculate the root mean square error of the FY-4A GIIRS ATP and ERA5 data according to the following formula: In the formula: N This represents the total number of grid points. M i For the first i ATP data from each site; O i For the first i ERA5 data at grid points; (3) Establish a preprocessing method for assimilating FY-4A GIIRS atmospheric temperature profile ATP data: Based on the background field at the assimilation time, the quality-controlled ATP data is first unified onto the background field grid points, and then the data located outside the analysis area and outside the analysis time in the quality-controlled ATP data are removed to obtain the processed data; b. Convert the processed data into the LITTLE_R format used by the assimilation system to obtain the preprocessed ATP data; (4) Cyclic assimilation and numerical forecasting of FY-4A GIIRS atmospheric temperature profile ATP data: A uses the root mean square error as the observation error of the ATP data in the assimilation system, and uses a three-dimensional variational assimilation system to assimilate the preprocessed ATP data. In the assimilation procedure of step A, B designs time windows of 15 minutes, 30 minutes, 1 hour and 2 hours to cycle and assimilate ATP data, tests the sensitivity of different time windows when assimilating ATP data, and determines the impact of different time assimilation windows on analysis and forecasting.

Images