Evaluation and Impact Factors of Doppler Wind Lidar during Super Typhoon Lekima (2019)

Doppler wind lidar (DWL) has been shown to obtain fairly accurate wind speeds in normal wind conditions. However, the evaluation of DWL winds under typhoon conditions is less common. This study evaluated the accuracy of wind data measured by two types of DWLs (WindPrint S4000 and WindCube V2), and investigated the impact of factors (e.g., precipitation and humidity) on the DWL-observed wind speed and direction. Data were collected from joint observations in 10 Baoshan, Zhoushan and Taizhou (China) by the Shanghai Typhoon Institute during the passage of Super Typhoon Lekima in 2019. The DWL observations were compared with measured data from balloon-borne radiosonde released at the same location. The results showed that the 1-min average wind speed and direction of WindPrint S4000 were more consistent with the instantaneous observation data of the sounding balloon than those of WindCube V2. The applicability of DWL was poor when the precipitation intensity was larger than 50 mm·h. The DWL wind speed bias significantly increased when the relative 15 humidity exceeded 85%. When the drift distance of the sounding balloon (ldrift) was less than 1 km, the DWL wind speed bias decreased with an increase of ldrift, whereas it increased with an increase of ldrift when the drift exceeded 1.5 km. Within a radius of 700 km, the root mean square of wind speeds between DWL and sounding balloon measurements showed a trend of increasing as the distance from the typhoon center decreased.

The wind measurement principle of a DWL relies on the inversion of the radial velocity by calculating the Doppler 60 frequency shift of laser beams, which can be backscattered by atmospheric aerosol particles and atmospheric molecules . Therefore, the influence of external environmental factors on DWL performance, such as precipitation intensity, air humidity, and underlying surface topography, must be considered in measurement and research. Previous studies have indicated that precipitation could affect both the accuracy and range of DWL measurements. Roadcap et al (2001) pointed out that low humidity would reduce the backscattering of aerosol particles when they detected the wind speed using a CO2 DWL 65 and dropsondes. Träumner et al (2009) compared wind data collected by a scanning 2 μm Doppler lidar and a scanning 35.5 GHz cloud radar. They found that the average wind speed of DWL was larger than that obtained from the cloud radar during rain events. Davis et al (2013) analyzed DWL measurements in convective precipitation. They believed that precipitation caused the downward movement of aerosol particles, which led an increase of vertical velocity. However, the wind speed and precipitation intensity were relatively small in their observations, and no further quantitative analysis about the detection error 70 of the DWL winds with the change of precipitation intensity was reported. Li et al (2020) used a shipborne micro-pulse lidar to measure the aerosol extinction coefficient over the South China Sea. They found that high humidity weakened the echo signal and they then carried out quality control of lidar measurements under high humidity conditions. However, there was no quantitative evaluation towards the impact of air humidity on the DWL winds.
It is worth mentioning that the wind data collected by balloon-borne radiosonde and DWL are not strictly observed at the 75 same position. Sounding balloons drift with the wind during their ascent and collect instantaneous wind speeds at a fixed time interval. This leads to a deviation in wind measurements between the DWL and sounding balloons. However, in both normal showed that the agreement of vertical wind profiles between airborne DWL and dropsonde was generally good. However, the wind speed bias of DWL at different distances away from the typhoon center was not analyzed. Jiangsu Province because of the heavy rainfall caused by the typhoon landfall and transit. During the typhoon, the average accumulated precipitation in Shandong and Zhejiang provinces reached 158 mm and 165 mm, the highest and second highest amounts in their history, respectively; (3) the typhoon's time over land ranked sixth in China's history, because of its long duration, wide range of influence and slow speed. The typhoon stayed for 20 hours in Zhejiang, which was its longest typhoon period. Figure 1

Joint observations
The joint observations were carried out at three sites at the same time, namely Baoshan, Taizhou and Zhoushan, as shown in Fig. 2

Radiosonde
In the typhoon joint observations, sounding balloons were released at the same place as the DWL observations to compare their measurements. The GTS1 digital radiosonde was used in Baoshan and Taizhou observation points. This sonde is a new digital high-altitude detection instrument developed by the Shanghai Changwang Meteorological Science and Technology 145 Company. It can continuously track, locate and measure the trajectory of the sounding balloon through GFE (L) 1 (L-band radar) secondary wind-finding radar (Fig. 2). The wind data were obtained by mathematical model calculations within the specified time interval (1 s) The radar receiver receives the radiosonde code from the radiosonde continuously, and compiles the radiosonde data, namely atmospheric temperature, air pressure, humidity and other meteorological elements. A Vaisala RS41-SG radiosonde was used in Zhoushan station, which was developed by Vaisala Company in Finland. It uses a GPS The types of measuring instruments and the observation start and end time in three sites are summarized in Table 1. The detailed measurement parameters of the DWLs and radiosondes are listed in Table 2. 155

Measurement principle of DWL
A Doppler beam swinging (DBS) 5 beam scanning method was used for wind field observations for both WindCube V2 and WindPrint S4000, as shown in Fig. 3. Compared with the velocity-azimuth display method which uses a complete cone 160 scan to collect dense radial wind speed data (Holleman, 2005), the DBS scanning mode has a shorter scanning time.

165
The measurement principle of DWL adopts the principles of laser pulse Doppler frequency shift. The lidar emits a laser pulse to the atmosphere and receives a backscattering echo signal from atmosphere. The relationship between the Doppler frequency shift and the lidar radial wind speed is as follows: where vr is the radial wind speed; λ is the laser wavelength, where λ of WindCube V2 is 1543 nm and that of WindPrint S4000 170 is 1550 nm; Δf is the Doppler frequency shift.
In fact, the radial wind was directly measured by DWL. The laser beams were emitted from five directions. Four laser beams were scanned upward in the zenith angle from the direction of east, south, west and north. The middle laser beam pointed vertically to the zenith (Fig. 3). The Doppler frequency shift produced by the laser beam in the scattering of atmospheric aerosol particles was measured, and then the radial wind speed in the direction of laser emission was retrieved. The horizontal 175 wind field and the required data products were obtained by the following methods.
Under the assumption that the horizontal wind field has a linear distribution, the wind speed information of each radial direction was obtained using DWL. The three-dimensional wind speed component at the center of the circle was obtained by taking the radial wind speed in each direction and using the trigonometric function relationship: where u, v and w are wind components in the Cartesian coordinates (x, y, z),  is the zenith angle, which represents the angle between the lidar beam and the vertical direction. The wind speed and direction in the horizontal direction can then be obtained as: where U is the horizontal total wind speed, and β is the horizontal wind direction, rotating clockwise from due north (0°).

Evaluation
To evaluate the DWL observation data under typhoon conditions using a sounding balloon, first quality control of the DWL observation data was carried out. According to Goit et al (2020), quality control with 80% data efficiency can significantly reduce the RMS of wind speed observed by DWL. Therefore, referring to the research of Goit et al (2020), this 190 paper took the effective rate of 1 min data as the threshold value, and a data efficiency below 80% was recorded as data missing.
Quality control of the DWL measured data of Baoshan, Zhoushan and Taizhou was carried out, and the 1 min mean horizontal wind speed (U), wind direction, the data missing rate and precipitation intensity (I) with time were plotted, as shown in Fig. 4. From the changes of wind speed in Fig. 4 (a), (c) and (e), it can be seen that the DWL measured U in Baoshan was 200 basically less than 20 m s -1 before 16:00 on August 9. After that, the average wind speed began to increase significantly and reached its maximum value (35 m·s -1 ) at 07:00 on August 10. In terms of wind direction, before 00:00 on August 10, the wind at the measured site was a southeast wind. Then the wind direction clearly changed from southeast to south, which was mainly because the typhoon center of Lekima moved north after landfall. The U of Zhoushan station reached its maximum (30.6 m·s -1 ) about 4 hours before the typhoon made landfall (21:50 on August 9). In terms of wind direction, before 21:50 on August 9, 205 the wind at Zhoushan was an east wind, and this gradually changed to a south wind. During the period from 11:00 on August 9 to 14:00 on August 10 in Taizhou, the data for wind speed and direction were missing, and the maximum wind speed was about 35 m s -1 . The wind direction was north before 02:00 on August 10, and then changed sharply to south.
Inspection of data efficiency ( Fig. 4 (b), (d), (f)) shows that the rate of missing DWL data increased with the increase of height. Generally, the greater I was, the greater the data missing rate became. The missing rate of the DWL measured data 210 below 200 m in Baoshan was less than 20%, and that of data above 800 m was 100%. After 16:00 on August 9, because of the obvious increase of I, the rate of missing DWL data increased significantly. When I was greater than 50 mm h -1 (16:20 on August 9 and 01:00 on August 10), the missing rate of data above 200 m was as high as 100% (Fig. 4b). The data missing rate of the Zhoushan observation point was less than 20% when the height was below 100 m. However, the data missing rate was high when the height was more than 200 m. In some time periods, such as 11:30 on August 10 (I>50 mm·h -1 ), the missing rate 215 of data above 200 m was as high as 100% (Fig. 4d). When I was small or there was no precipitation (before 12:00 on August 9), the data missing rate of the Taizhou observation point was less than 20%. However, when the precipitation intensity was high (I>50 mm·h -1 ), the missing rate of DWL data was more than 80% (Fig. 4f). The above analyses show that the applicability of DWL is poor when the precipitation intensity is greater than 50 mm·h -1 .
From the observation results of DWL at the three observation points, it can be seen that the observation capacity of both 220 WindPrint S4000 and WindCube V2 DWL was reduced under precipitation conditions, and the data missing rate increased significantly. It can be seen from Fig. 4 (b), (d) and (f) that when there was no precipitation or when the precipitation intensity was small, the data missing rate was basically less than 20%. When the precipitation intensity I>50 mm s -1 , the data missing rate above 200 m height began to increase rapidly with height, and the data missing rate was as high as 100%. reason for the missing data was that with the increase of detection height, aerosol particles decreased, resulting in a decrease 225 of the angular scattering probability of lidar-the continuous weakening of the echo signal resulted in data missing. Another reason may be related to precipitation, whereby heavy rainfall accumulating on the window surface led to a decrease of the atmospheric transmission of laser light, which continuously weakened the echo signal. In general, except for the Taizhou observation point, the data missing rate of WindPrint S4000 and WindCube V2 DWL was mostly less than 20%, which means the observation data had a data efficiency greater than 80%.  As can be seen from the figure, below 100 m, the DWL-measured wind speed was significantly greater than that of the sounding balloon. With the increasing height of the sounding balloon, the gap between them became smaller and smaller. When the 260 height reached above 100 m, there was a crossover for wind speed between the DWL and sounding balloon measurements, and the observed results were basically the same. This was because the release of sounding balloon made it accelerate from a static position. The wind speed value observed by sounding balloon from static to complete drift with the wind was basically less than the real wind speed value, causing a deviation in wind speed measurement. As with the results of Zhang et al (2018) and Tang et al (2020), the drift effect of the balloon-borne radiosonde was the main reason for the large wind speed deviation 265 below 100 m. When the height reached about 100 m, the sounding balloon ended the initial acceleration and began to drift completely with the wind, so it was in good agreement with the DWL measurements. Therefore, this paper considered that the measured data of sounding balloons below 100 m were not credible. In addition, the horizontal distance between the sounding balloon at the maximum observation height of DWL (600 m) and the release point was within 12 km. In general, the trends of average wind speed in 30 s, 1 min and tGPS in vertical height were the same.  Owing to the large data missing of WindCube V2 DWL at the Taizhou observation point from 11:00 on August 9 to 14:00 on August 10 (Fig. 4f), it was impossible to compare it with the wind speed observed by sounding balloon. Therefore, Fig. 7 only shows the comparison of the measured horizontal wind speed at different heights between the DWL and the sounding 305  Table 3 that the average wind directions in tGPS of the three observation points were in good agreement with the wind directions measured by sounding balloon: the minimum value of RMS was 2.56°, 6.99° and 12.97°, and all the values of R exceeded 330 0.95. Owing to the different length of each tGPS, it was impossible to compare these in a unified standard. Through comparison, the differences between the RMS and R of mean wind speed and direction in 1 min and tGPS were small. Considered comprehensively, the 1-min average wind speed and direction were adopted for further research in this paper.
By comparing the RMS and R of wind speed measured by WindPrint S4000, WindCube V2 and sounding balloon, the coincidence degree of wind speed with sounding balloon measured by WindPrint S4000 in Baoshan was higher than that 335 measured by WindCube V2 in Zhoushan and Taizhou. This may be because the Baoshan observation point was far from the typhoon landfall center and, therefore, the wind speed and precipitation intensity were small.

Precipitation
To reveal the influence of typhoon precipitation on wind field observations of DWL, the SNR and data missing rate under into seven grades: 0-0.01 mmh -1 (sunny), 0.01-3 mmh -1 (light rain), 3-6 mmh -1 (moderate rain), 6-12 mmh -1 (heavy rain), 12-30 mmh -1 (rainstorm), 30-90 mmh -1 (heavy rainstorm), and above 90 mmh -1 (severe rainstorm). The variation of 1-min average SNR with I of the two lidars is shown in Fig. 8. The figure shows that I has different effects on the SNR at different heights, and that the SNR decreases with an increase of I, which is consistent with the findings reported by Tang et al (2020). 350 Inspection of Fig. 8(a) suggests that the SNR of the WindPrint S4000 DWL at the Baoshan observation point did not change with I below 600 m and was between 2-4 dB. The SNR decreased with the increase of I when the height was over 600 m, and the effect was more obvious. Under the conditions of I of 0.013 mmh -1 (light rain) and 36 mmh -1 (moderate rain), the SNR decreased sharply. In Fig. 8(b), the SNR of the WindCube V2 DWL in Zhoushan and Taizhou decreased with the increase of I. During the study period, the precipitation was mostly in the sunny class, followed by light rain. The above analyses show 355 that the SNR of WindPrint S4000 DWL at high levels is easily affected by I, while the SNR of WindCube V2 DWL at all observation heights is greatly affected by I.  Figure 9 shows the variation of the DWL 1-min data missing rate with the SNR at Baoshan, Zhoushan and Taizhou observation points. Except for the Taizhou observation point, the data missing rate decreased with the increase of the SNR, and the higher the ground height was, the greater the data missing rate became. As shown in Fig. 9(a), the data missing rate of the WindPrint S4000 DWL at the Baoshan observation point was less than 40% below 400 m. When the SNR exceeded 1 dB, the data missing rate was less than 20%. However, the data missing rate above 600 m was high-when the SNR was less than 370 4 dB, the data missing rate was as high as 100%. In Fig. 9(b), the data missing rate of WindCube V2 decreased with the increase of the SNR, and the data missing rate of the lower layer was clearly less than that of the higher layer. When the SNR was greater than −10 dB, the data missing rate at each height remained at less than 5%. When the SNR was lower than −15 dB, the data missing rate of each layer began to increase significantly. Inspection of Fig. 9(c) shows that the data missing rate of WindCube V2 DWL at the Taizhou observation point did not change significantly with the SNR. In general, except for the 375 Taizhou observation point, the data missing rate of DWL showed a decreasing trend with the increase of the SNR.

Humidity 385
The water vapor content in the atmosphere has a great influence on the optical properties of aerosols (Beyersdorf et  study the influence of the concentration of water vapor molecules in the air, Fig. 10 shows the variation of DWL wind speed 390 and wind direction bias with RH at Baoshan, Zhoushan and Taizhou. Figure 10(a) shows that the influence of humidity on the DWL-measured wind speed was mainly concentrated in RH>85%. When RH>85%, the wind speed bias increased sharply, and the maximum value exceeded 7 m·s -1 . When RH<85%, the wind speed bias was less than 3 m·s -1 . As for the wind direction, Fig. 10(b) indicates that starting from RH>85%, the DWL-measured wind direction was gradually affected by humidity. When RH was greater than 90%, the DWL wind direction bias increased sharply, and the maximum value exceeded 60°. When RH 395 was less than 85%, the DWL wind direction bias was relatively small, within 10°. This shows that with an RH>85%, the DWLmeasured wind field may have a large deviation, and with an RH<85%, the DWL-measured wind fields are more accurate.

Drift distance of sounding balloon
As mentioned in Section 4.1, the wind speeds observed by the DWL and the sounding balloons were not strictly in situ observations. The sounding balloons drifted with the wind during the ascent and gradually moved away from the release point. 405 Therefore, there was a deviation between the sounding wind speed observations and the DWL observations. To reveal the relationship between the drift distance of sounding balloon (ldrift) and the DWL wind speed and wind direction deviation, Fig. observation point. It demonstrates that the wind speed bias generally decreased at first and then increased with ldrift. Within 1 km, the wind speed bias decreased with the increase of ldrift. In the range of 1-1.5 km, the average wind speed bias was 410 approximately equal to zero. When the drift distance exceeded 1.5 km, the wind speed bias increased sharply, and the maximum value of the average wind speed bias exceeded 1.5 m·s -1 . From Fig. 11(b), the wind direction bias of DWL did not change much with the drift distance of sounding balloon. This shows that the influence of ldrift on the DWL wind speed and wind direction deviation differed, and was mainly concentrated in ldrift>1.5 km. The wind direction deviation was not affected by

Position relative to the typhoon center
To show the influence of the position relative to the typhoon center on the DWL-measured wind speed, the DWL data were divided into two groups according to the observation height, h<200 m and h>200 m. The RMS of each group of data and the measured wind speed of the sounding balloon at the same time and at the same height were calculated. There were five sounding datasets from the Baoshan observation point, nine from Zhoushan, and three from the Taizhou observation point. as the distance from the typhoon center decreased. The RMS of the Baoshan and Taizhou points was relatively small, and the maximum value did not exceed 2 m·s -1 . Considering Fig. 4 (a), (b), (e), (f), this may be due to the fact that the Baoshan and Taizhou observation points were far from the typhoon center at the corresponding time, and the wind speed and precipitation 430 intensity were both small. The RMS of the observation point in Zhoushan increased significantly, and the maximum value was 5.89 m·s -1 when h<200 m (the corresponding time was 02:22 on August 10, 179 km away from the center of the typhoon, with a corresponding wind speed of 25 m·s -1 ).When h>200 m, the maximum RMS value was 5.78 m·s -1 (the corresponding time was 11:50 on August 10, 225 km from the center of the typhoon, and the corresponding wind speed was 16 m·s -1 ). The Zhoushan observation point was relatively closer to the typhoon center (within 250 km). In addition, Fig. 4

(c) and (d) 435
demonstrate that at the corresponding time, the wind speed and precipitation intensity of the Zhoushan observation point were much larger than those of the Baoshan and Taizhou observation points. This showed that when the distance from the typhoon center exceeded 250 km, the DWL was less affected, and the RMS was basically less than 2 m·s -1 . Within a radius of 250 km, the RMS increased significantly as the distance decreased and the RMS reached a maximum of 5.89 m·s -1 .

Conclusions
This study was based on the joint observations of Super Typhoon Lekima in 2019 by the Shanghai Typhoon Institute of 445 China Meteorological Administration in Baoshan, Zhoushan and Taizhou. The performance of the DWL-measured wind field under typhoon conditions was evaluated, and the impact of precipitation intensity, relative humidity, drift distance of sounding balloons and the position relative to the typhoon center on the DWL wind measurements was studied. In this study, we selected the DWL-measured 30-s, 1-min, and tGPS average wind speeds and directions to evaluate the coincidence degree with the measured instantaneous wind speed of sounding balloons, and we analyzed the impact factors that caused the deviation of the 450 DWL wind field measurement. The research height was 0-600 m.
Through comparing the horizontal wind speed between the DWL and sounding balloon measurements, when the height was below 100 m, the wind speed value observed by the sounding balloon was basically lower than the true wind speed value because of the acceleration when the sounding balloon was released. This resulted in a significant difference of the measured wind speed between the DWL and sounding balloon. Therefore, this paper suggests that the measured wind speed data of 455 sounding balloons below 100 m is not reliable. Comparing the deviation statistics of the average wind speed of the DWL, the 1-min average wind speed of WindPrint S4000 was in the best agreement with the measured instantaneous wind speed of sounding balloons. In the range of 100-600 m, its R was 0.98 and its RMS was 1.29 m s -1 . In comparison, the tGPS average wind speed of WindCube V2 was in the best agreement with the instantaneous wind speed of sounding balloons. In the range of 100-290 m, its R was 0.82 and 0.90, and its RMS was 2.19 m s -1 and 1.52 m·s -1 in Zhoushan and Taizhou, respectively. The 460 DWL-measured horizontal wind direction was in good agreement with the sounding balloon measurement-R exceeded 0.7, and the RMS was less than 15°.
By analyzing the influence of precipitation intensity on the DWL-derived SNR, data missing rate and wind speed deviation, it was seen that the SNR decreased with the increase of I. Generally, DWL had poor applicability when I>50 mm·h -1 . The precipitation intensity had a large impact on the SNR of WindPrint S4000 above 600 m and at all observation heights 465 of WindCube V2. The rate of missing DWL data increased with an increase of I. Moreover, the data missing rate of the lower layer, which was not affected much by I and remained within 20%, was significantly smaller than that of the higher layer.
There was an obvious correlation between the DWL wind speed bias and the RH. When the RH was less than 85%, the DWL wind speed bias was small (less than 3 m·s -1 ). When the RH exceeded 85%, the DWL wind speed bias increased sharply, and the maximum value exceeded 7 m·s -1 . The influence of RH on DWL-measured wind direction was mainly concentrated 470 around RH>90%, and the maximum wind direction bias exceeded 60°. In addition, the DWL wind speed bias generally decreased with the increase of ldrift within 1 km. When it exceeded 1.5 km, the wind speed bias increased sharply with the average value exceeded 1.5 m·s -1 . In comparison, the DWL wind direction observation was not strongly affected by ldrift.
Within a radius of 700 km, the RMS of wind speeds between DWL and sounding balloon measurements showed a trend of increasing as the distance from the typhoon center decreased. Beyond 250 km from the typhoon center, the RMS was less 475 than 2 m·s -1 . Within a radius of 250 km, the RMS increased significantly with the decrease of distance, and the maximum RMS reached 5.89 m·s -1 .