Estimation of Mesoscale Atmospheric Motion Vectors and Its Application at Kma/nimr

Atmospheric motion vectors (AMVs) derived with high resolution satellite data are useful for analyzing mesoscale motion such as convective clouds and ageostrophic flow. National Institute of Meteorological Research at Korea Meteorological Administration (KMA/NIMR) has developed a mesoscale AMV algorithm. In the mesoscale AMV algorithm, visible channel data with 1 km resolution from the geostationary satellite are used and the search area is determined dynamically to detect the mesoscale winds effectively. The optimal conditions of target selection and grid interval for mesoscale AMVs have been investigated through sensitivity tests. The quality of the estimated AMVs is leveled as quality indicator (QI) by comparing with model predicted winds. However the QI considers mostly consistency and homogeneity between the AMVs and model winds suitable for synoptic scale motions. Therefore, mesoscale AMVs with large gradient in space and time tend to be treated as vectors of low quality. In this regard, KMA/NIMR investigates the optimal threshold of QI for mesoscale winds and additional methodology as the expected error (EE).. Mesoscale AMVs using high-resolution visible channels can derive relatively lots of vectors in low levels. Low-level winds are involved in various situations including low-level temperature inversions and their heights should be assigned carefully. In order to re-evaluate height assign approaches of mesoscale AMVs, the heights of AMVs are compared with the heights of clouds from CALIPSO satellite observations. TARGET SELECTION Target size The target size is one of most important parameters to estimate mesoscale AMVs. While target sizes from 32 km X 32 km to 128 km X 128 km were investigated for synoptic AMVs, target sizes smaller than 32 km X 32 km have been tested for mesoscale AMVs (Figure 1). As a result, the optimal condition is determined as a target size of 24 km X 24 km. When the target size is smaller than this optimal size, the number of vectors decreases but vector-RMSEs increases. Although the speed bias at this optimal condition is slightly larger than at 20 km X 20 km, the number of vectors is the largest and vector-RMSEs is the smallest. The flow of cyclonic circulations could be depicted comparatively effectively. Figure 1: Result from sensitivity tests of target sizes for mesoscale AMVs (QI ≥ 0.5). Target sizes are investigated from 32 km X 32 km to 12 km X 12 km for 2315 UTC 23rd May 2010. Color of AMV indicates the individual vector’s speed. Target location method The regular method specifies the locations of targets with equidistant grid intervals while the optimal method specifies it to have maximum contrasts based on the local standard deviations. The optimal method can derive more meaningful vectors with high accuracy than the regular method. Target Size (km) 96 80 64 48 32 24 Number of Collocated Vectors Regular 1057 1609 2404 3594 5117 4934 Optimal 1579 2244 3160 4740 5896 6155 Vector-RMSE (ms-1) Regular 5.51 5.43 5.52 5.71 5.94 6.41 Optimal 5.56 5.51 5.40 5.61 5.82 6.34 Speed-BIAS (ms-1) Regular 1.20 1.42 1.35 1.41 1.53 1.75 Optimal 1.07 1.23 1.28 1.43 1.46 1.60 Table 1: Result from sensitivity tests of target location method for mesoscale AMVs (QI ≥ 0.5) for Feb. 2010. QUALITY CONTROL (QC) For a quality control (QC), mesoscale AMV algorithm uses a Quality Indicator (QI, Holmlund 1998) and filters vectors with a QI below 0.85. This condition is same as QC method for synoptic AMVs. The current QI is calculated from 5 consistency tests, which is for a speed, a direction, a vector, spatial, a forecast. In order to detect mesoscale AMVs, a new QC method, a expected error (EE, Le Marshall et al 2004) is tried in NIMR’s algorithm and is calculated from 5 QI tests and additional 4 tests (wind speed, assigned height, simulated wind shear and temperature gradient). Log(AMV-sonde+1) = a0+a1x1+a2x2+...a9x9 EE=exp(a0+a1x1+a2x2+...a9x9)-1 As a result of sensitivity tests, new QC method (QI≥0.5 and EE≤4) is more effective than the current method (QI ≥0.85). The more vectors with high quality and high accuracy are extracted from the new QC method. Figure 2: Comparison of AMV speed and AMV – Radiosonde speed according to thresolds of quality control indices for Feb. 2010. HEIGHT ASSIGNMNET (HA) The height of AMV is assigned to cloud top pressure estimated from the coldest 15% pixels in a target. However, the low-level clouds tend to move according to the speed of cloud base. In order to correct this problem, the inversion layer height correction (IHC) and cloud base correction (CBC) methods were tested. MS CBC (Le Marshall, 1996) and EU CBC (EUMETSAT, 2009) methods reassign the vector allocated below 650 hPa to the height of cloud base. T ebbt base T T σ ⋅ + = 2 The equation of EU CBC is similar to MS CBC but it contains empirical variables instead of 2 . When both CBC methods are applied, the number of vectors with high accuracy increases and vectorRMSE decreases. Because the number of vectors with negative speed-biases is reduced, mean speed-biases increase. EU CBC method is effective to acquire more vectors. Figure 3: Comparison of AMV wind speed and radisosonde wind speed according low level correction method for Feb. 2010. Height assignment validation with CALIPSO The assigned height is validated with CALIPSO cloud top pressure data specified according to cloud characteristics. The vectors over single layered clouds are only used. CALIPSO data is collocated with AMVs within 1hour and 100 km in a horizontal distance. In figure 4, the heights of AMVs are shown differently with heights of clouds from CALIPSO observations. Figure 4: (a) Mesoscale AMVs collocated with CALIPSO cloud top pressure at 0515UTC 24th May 2010. The colors are pressure height. (b) is vertical cross sections, blue dots indicate AMVs height. AMVs at the transition stratocumulus show good agreements in low level. AMVs heights at the transparent overcast clouds in low level are assigned slightly higher than CALIPSO’s. In middle level, the number of AMVs derived is small and the comparison result show poor agreements. Most of AMVs are assigned slightly lower than CALIPSO cloud heights in high level. AMV heights over transparent clouds show lower accuracy than opaque clouds. AMVs are assigned using only IR channel, which detects poorly transparent clouds. Figure 5: Comparison of AMV height and CALIPSO cloud top pressure classified by cloud characteristics vertical feature mask in low level. OPTIMAL CONDITIONS FOR MESOSCALE AMVS Comparison of the optimal conditions for mesoscale AMVs and synoptic scale AMVs are shown in table 2. Figure 6 is AMVs derived by the optimal conditions in table 2. Mesoscale (HRV) Synoptic (VIS) NWP model Unified Model Resolution of scene analysis (km) 1 X 1 4 X 4 Target tracking Moving search area using NWP winds Time interval between satellite images 15-minute Target size (km) 24 X 24 96 X 96 Grid size (km) 24 X 24 48 X 48 How to decide the location of target Optimal method Regular method Height assignment EBBT Pixel selection method in height assignment Coldest pixels (15%) Low level correction Inversion height correction, EUMETSAT cloud base correction Inversion height Correction Quality Control QI ≥ 0.5 and EE ≤ 4 QI ≥ 0.85 Table 2: Comparison of the optimal condition of mesoscale AMV and synoptic AMV at KMA/NIMR. Figure 6: Comparison of mesoscale AMVs (left) and synoptic scale AMVs (right) for tropical cyclone OMAIS, 2315 UTC 23th March 2010. The color of AMVs indicates the individual vector’s height and only 50% of all vectors are displayed.