INNOVATIVE IDEAS FOR USING THE HYPERESPECTRAL LEVEL 1 DATA OF THE NEXT GEOSTATIONARY MTG-IRS IN NOWCASTING

Transcription

1 INNOVATIVE IDEAS FOR USING THE HYPERESPECTRAL LEVEL 1 DATA OF THE NEXT GEOSTATIONARY MTG-IRS IN NOWCASTING Miguel A. Martinez, 1 Xavier Calbet 2 (1) Agencia Estatal de Meteorología (AEMET), Spain (2) EUMETSAT Abstract The traditional use of hyperespectral data is mainly related to NWP and physical retrieval algorithms. This has been motivated by the polar orbit and the spatial resolution of the current hyperespectral instruments. But in 2020, the next MTG-IRS on geostationary orbit with a spatial resolution of 4 km in the subsatellite point and 30 minutes temporal resolution will open new possibilities on the use of hyperespectral Level 1 data. In order to foster the use of MTG-IRS in nowcasting applications and especially on cloudy pixels (where the NWP and physical retrieval algorithms may not be applied), some innovative ideas to use the MTG-IRS L1B data as an imagery instrument are shown here. INTRODUCTION RGB images from geostationary satellites have been a great aid for the weather forecasters, most notably in the field of nowcasting. The main reason for this resides on the ability of the human eye to quickly comprehend image information, especially when the image is colour coded and the individual has been trained to interpret them. An added advantage is that RGB coded images are also quickly generated from the original image data directly provided by the satellite. They also provide a global coverage of the complete visible disk as seen from the satellite perspective. In the first section, as a starting point, IASI RGB images are used as a proxy for MTG-IRS ones. The way these IASI RGB images are constructed in this section is equivalent to the existing MSG-SEVIRI RGB images. The reason for this is to start off from previous heritage from MSG-SEVIRI such that the user of these RGB composites is already familiar with their appearance. As a first step, the equivalent MSG-SEVIRI channels are chosen from IASI. Then, the IASI RGB images, which will be the equivalent of MSG-SEVIRI RGB composites, are built using the same process for IASI data as for the MSG-SEVIRI RGB images. In order to promote the use of hyperspectral RGBs some IASI RGB images on Polar Regions, where MSG RGBs images cannot be built, are shown. Other examples, such as the airmass IASI RGBs over the Globe and the IASI dust RGB to monitor the volcanic ash in the case of the Chilean Cordon volcano eruption are shown. In the second section, early brainstorm ideas for direct use of IASI L1 data are shown: a) the use of the separated local maxima and local minima envelope lines to represent IASI spectra in order to allow easy comparison between spectra from different spatial or time coordinates. b) one representation of the peaks and valleys as a potential spectral-gram, c) loops using a selection of channels on "CO 2 or WV" absorption bands with channels peaking from the top of the atmosphere gradually down to the surface. In the third section, the search for new RGBs and some early tests on the generation of Principal Component s images with direct physical interpretation are shown. THE STARTING POINT: HYPERSPECTRAL RGBs WITH MSG RGB HERITAGE The key point is the search of the IASI (or MTG-IRS) channels nearest to each one of the MSG IR channels. The search of the nearest IASI (or IRS) channel closest to a MSG IR channel has been made developing one algorithm identifying the synthetic IASI channel that simultaneously meets a high correlation and a low RMSE with synthetic MSG RTTOV BTs on datasets calculated from one ECMWF GRIB file. The IASI channels used to build the RGB images can be seen in the Table inside

2 the Figure 1. More details of the search of these IASI channels closest to each MSG IR channel can be seen in (Martinez, 2010). Two main RGB images, derived from MSG and based on IR or WV channels, are widely used by forecasters operationally. These are the airmass RGB and the dust RGB. The airmass RGB is used in forecasting tasks. The dust RGB is used for several issues like dust monitoring and volcanic and SO2 monitoring. Examples of the use of MSG RGBs can be seen in GRIB ECMWF MSG RTTOV coefficients correlation analysis between synthetic RTTOV IASI or MTG-IRS and MSG-SEVIRI channels IASI RTTOV coefficients IASI 8461 channels MSG Conversion IASI to IRS IRS 1738 channels θ = 0º Search of IASI or MTG-IRS channels with lower mean square error and greatest correlation versus MSG-SEVIRI channels using the ECMWF analysis of 25th May 2009 at 12Z for MSG full disc region and RTTOV-9.3 MTG-IRS θ = 0º Table 1. IASI channels used for airmass and dust RGB instead of MSG channels Figure 1. IASI channels used for IASI airmass and IASI dust RGBs equivalent to the commonly used MSG channels. IASI AIRMASS RGB METOP-A and METOP-B IASI channels, which are chosen such that they are equivalent to MSGSEVIRI channels, can be used to build global coverage airmass RGB images. The IASI RGB images are easily generated once the brightness temperatures for the channels involved in the IASI RGB are read from the hyperspectral files. The computational cost of the RGB image generation is very low and any software package such NWC SAF/PPS or CIMSS ones could be used to generate the hyperspectral RGB images in near real time. th Figure 2: METOP-A and METOP-B IASI global airmass RGB for day 20 February 2013 morning (cyclone Haruna near Madagascar case study) using McIDAS-V Globe display (left) and IDL with transparent PNGs for polar displays (right) Since the IASI zenith angles are always low, hyperspectral RGB images from polar satellites do not show the bluish colours that appear on MSG airmass RGB images in the borders or the disc (caused

3 by the high zenith angles of MSG data in the borders). They therefore show a good RGB performance even in Polar Regions, as is illustrated in Figure 2. Airmass RGB with other hyperspectral instruments A similar process as the one described above for searching IASI channels that are equivalent to SEVIRI channels could be repeated for other hypespectral instruments. In Figure 3, one example using actual AIRS data is shown. This demonstrates the possibility to generalize the RGBs construction to other instruments like CrIS on board NPP-SUOMI. The combined use of RGB images from several hyperspectral instruments could improve monitoring regions not covered by RGB images from geostationary satellites; especially over Polar Regions due to the high revisiting frequency. Airmass RGB AIRS Channel RED AIRS AIRS GREEN AIRS AIRS BLUE AIRS Range g K K K 1.0 th Figure 3: AIRS airmass RGB for day 25 May 2009 afternoon using actual AIRS HDF files (first CWG case study strong convection on Europe) IASI DUST RGB The MSG dust RGB has been widely used for dust and sand storm monitoring. In Figure 4, it can be seen that the IASI dust RGB image is able to detect the dust storm over the Sahara desert and the Canary Islands. The first direct use of an IASI dust RGB could be to help in the subjective selection of dust contaminated pixels for further comparison of the IASI spectra. They could be archived and later used as a collection of spectral probes to train automatic classifications of clear or dusty IASI spectra. MODIS AQUA 2012/03/08 IASI 2012/03/08 10:11:59Z 11:50:55Z Figure 4: (left) MODIS true color RGB in one strong dust storm on the Sahara reaching the Canary Islands, (right) IASI dust RGB for the same case.

4 MSG dust RGBs have also been used for volcanic eruption monitoring (as example in the Iceland volcanoes eruptions). In the same way, IASI dust RGB images can also be used for global monitoring of volcanic ash and SO 2 with a single common instrument, with the added advantage that on Polar Regions the low zenith angles also improves the detection. IASI dust RGB of the eruption from the Chilean Puyehue-Cordon-Caulle Volcano using IASI L1 data from 9 th June :24Z to 10 th June :35Z are shown in Figure 5. The IASI dust RGB image has been created using McIDAS-V on Globe display and IASI L1 HDF-5 files from EUMETSAT UMARF. Volcanic ash and SO 2 Figure 5: IASI dust RGB showing the volcanic ash and SO 2 from the eruption of Chilean Puyehue-Cordon-Caulle Volcano using IASI L1 HDF-5 files from 9 th June :24Z to 10 th June :35 and McIDAS-V globe display. BRAINSTORM IDEAS FOR DIRECT USE OF SPECTRA IN NOWCASTING The comparison of the spectra from several pixels is nearly impossible due to the confusion of the lines due to the high oscillation present in the spectra. To avoid this fact, the first brainstorm idea is to use separately only the wave numbers of the relative maxima ( peaks ) and minima ( valleys ) in the spectra of pixels labelled as clear sea. To obtain theses wave numbers, a set of collected IASI spectra labelled as clear sea pixel have been used to calculate an average BT spectra; then, the wave numbers of the relative maxima ( peaks ) and minima ( valleys ) has been found using an iterative process repeated twice. This representation is adequate for the comparison of the spectra slopes and shapes of several kinds of pixels. In Figure 6, an example of peaks comparison for several kinds of pixels can be seen. The peaks representation allows to clearly distinguish the shape of the spectra and could be used later to develop algorithms based on these features. This technique would avoid the use of differences between only two channels that could be misleading and could produce errors. Figure 6: Comparison of IASI spectra for clear desert pixel (red line), dust desert (green line) and volcanic pixel (black line) using the peaks representation for IASI L1 data from the eruption from the Chilean Puyehue-Cordon-Caulle Volcano case study 9 th June :24Z to 10 th June :35Z.

5 The separate peaks and valleys lines can be considered as the envelope lines of the spectra. The simultaneous representation of the peaks and valleys provides useful information about the temperature profiles and the amount of absorbers present in the pixel that originated the spectra. Thus, the area between the peaks line and the valleys line could provide an idea of the amount of absorbent present. Since the valleys line has a higher oscillation than the peaks line it is shown also a moving average on the valleys line. The vertical representation of the peaks and valleys in the Figure 7 is one early attempt for one future spectra-gram. Taking into account the geostationary orbit of MTG-IRS, the change of these lines with the time (every 30 minutes) on a fixed location (for example at one airport) could provide the forecasters an indirect and subjective monitoring about the processes that are taking place. Figure 7: Comparison of IASI peaks and valleys for a dust desert pixel on the 9 th June (black line) is the peaks line, (red line) is the valleys line and (green line) is the 9 moving average of the valleys line. To avoid the roller coaster effect when images of all IASI channels are displayed in one loop a solution is to make loops with just the images of the main peaks or valleys on selected spectral region as the CO 2, O 3 or WV absorption bands. This allows creating gently varying image loops which probe deeper and deeper in the atmosphere from very high levels to the surface. One example of usage is the search of low levels jets on satellite images. The images in Figure 8 have been normalized removing the mean and dividing by the standard deviation. The process could be made interactively if the adequate software is developed. It can also be made with the BT relative minima (i.e. valleys ) cm cm cm cm cm cm -1 Figure 8: Probing display of normalized images using main peaks on CO 2 (left) and WV (right) bands for the Chilean Volcano case study 9 th June 2011.

6 SEARCH FOR NEW RGB In this area, only early tests and a few experiments have been done. The early tests shown in this section should be understood as a demonstration and as an example that it is possible to build new RGBs which could be useful for weather forecasting. These could constitute an important application of MTG-IRS L1 data which is far beyond its sole use for NWP assimilation or generation of L2 products. Therefore, the creation of new RGBs for MTG-IRS should be considered a high priority task. There are several possibilities: Search of new RGB: use of synthetic BTs The synthetic MTG-IRS BTs (or synthetic IASI BTs) and NWCSAF precipitable water layers (BL, ML, HL), TPW and Total Ozone have been calculated directly from ECMWF profiles for the same ECMWF GRIB file and are shown in Figure 9. Correlation and RMSE matrices have been calculated for the regressions between every two synthetic IRS (or IASI) BTs and each one of the parameters. After that, a fuzzy-logic scheme has been used to search the regressions with lower RMSE and highest correlation. The results of these regressions and a RGB image built with them are shown in Figure 9. Red ML LPW ( hpa) C1=1717 (λ = ) C2=1539 (λ =4.8751) Regression , 5.25, MTG-IRS ML ( ) Total ozone Green Total Ozone C1=465 (λ = ) C2=457 (λ = ) Regression , , θ = 0º Blue BL LPW (Psfc-850 hpa) C1= 1506 (λ = ) C2= 802 (λ = ) Regression , , BL (P sfc -500) Figure 9: RGB has been built using directly the regressions of two synthetic MTG-IRS channels and some meteorological fields. (top left red) regression of two MTG-IRS synthetic BTs and ML (precipitable water in the layer hpa), (medium left green) same but with Total ozone, (bottom left blue) same but with BL. Synthetic MTG-IRS and BTs from ECMWF analysis 25 th May 2009 at 12Z This algorithm works well with synthetic MTG-IRS BTs, but before applying it to real data a previous bias correction of the BTs would be needed. Search of new RGB: use of spectral features. Example, low levels inversions RGB using Fingers-up feature after Paul Menzel s communication After reception of an from S. Tjemkes (EUMETSAT) about one personal communication from Paul Menzel (CIMSS-Wisconsin) about the existence of features in the spectra to detect inversions on low levels and that one event which had strong low levels inversion on the 17 th January 2013 in Finland, an early test to develop one RGB image for low levels inversions has been made. After downloading the IASI HDF-5 file from UMARF, the McIDAS-V software has been used to analyze the image. Following Paul Menzel s personal communication, IASI pixels that present the finger up signal in the spectra has been searched on the IR window wave numbers. Some absorption lines change from local minima to local maxima in the case of low level inversion spectra; see the paper (Sieglaff, 2009). In Figure 10, this effect can be seen on the pink probe (inside of the low level inversion region) which shows a finger up lines configuration in this part of the spectra and on the cyan probe (outside of the low level inversion area) which shows a finger down lines configuration in the spectra. Afterwards, the difference between two IASI channels, one inside the

7 finger up wavelengths and another one outside of the finger up it could be used to build a test for the presence of low level inversions. The pink region in Figure 10 (right) is a region showing low level inversions and in the McIDAS-V scatter plot (Figure 10, far right) it is clearly seen as a separate cloud of pixels (marked with a pink line) cm cm-1 Figure 10: (pink line and probe) IASI L1 spectra in the case of one pixel inside low-level inversion (light blue line and th probe) same but outside low-level inversion. 17 January 2013 at 18:01:47 Z Then, two of the fingers up features have been combined to build the RGB image of Figure 11. This RGB image is only an early test and demonstrates the potential to build new IASI RGB images. Difference BTs in finger up 1: ( BT@ BT@801.0) scaled in [-1,5] Difference BTs in finger up 2: ( BT@ BT@815.5) scaled in [-1, 2.5] Inverse of mean of BTs: -(BT@ BT@801.0)/2 scaled in range [-313, -233] Figure 11: The difference between two IASI channels, one inside and other outside of the finger-up line feature, are used to create the mask of the pixels inside the low-level inversion region. The red component and green component th uses two different finger-up lines. The third is an IR image with inverse brightness to complete the RGB. 17 January 2013 at 18:01:47 Z Search of new RGB: use of Principal Components (PC) Due to the high bandwidth that would be needed to disseminate lossless MTG-IRS L1 data via EUMETCast with all channels and at full temporal and spatial resolution, the baseline solution is to disseminate only 300 pieces of spectral information for every pixel at the full temporal and spatial resolution. The pieces of information at the moment of writing this paper are the dissemination of 300 Principal Components (PC) of the spectra. The performance of the reconstructed MTG-IRS spectra derived from the 300 PCs is an ongoing task for EUMETSAT. Based on these facts, it is of interest to investigate the possibilities of a direct use of PCs to obtain RGB images. Several early attempts have been made. Due to the fact that IDL6.2 does not allow to get PCs from the full IASI spectra, they have been calculated using only chopped parts of the IASI spectra. Two of these early tests are shown in Figure 12. In Figure 12 (left), the PCs have been calculated for the wave numbers [770, 1270 cm-1] from one dataset that contains a low number of IASI spectra collected

8 selecting pixels manually from an IASI dust RGBs and labelling them as clear land, clear sea, dust or volcanic. The PC images from one actual IASI L1 dataset are then calculated using the PC coefficients from the dataset with collected and labelled IASI spectra. This test is a proof of concept to calculate PC coefficients from subjective labelled spectra. In Figure 12 (right) the PCs have been calculated from one actual IASI image (low level inversions on Finland case) in the range [780, 840 cm -1 ] and the PC image can be identified subjectively. The second PC (Figure 12 right; in title appear 1 because it is 0-based the number on PCA) is highly correlated with the pixels that present the finger-up feature in and 801 cm -1 in Figure 11. Figure 12: (left) 8 th PC of one IASI orbit from Chilean Volcano case study 9 th June 2011 in range [770, 1270 cm -1 ] after apply PC coefficient from a dataset with clear desert, clear sea dust, volcanic ash and clouds probes, (right) 2 nd PC in the IR window [780, 840 cm -1 ] of the IASI L1 dataset in a low-levels inversion case study (Finland 17 th January 2013). CONCLUSIONS The RGB images generated from low earth orbit hyperspectral instruments like AIRS, IASI and CrIS can be used now operationally in nowcasting, especially on Polar Regions. The use of hyperspectral RGB images could be started following the heritage of MSG-SEVIRI RGB images. New RGB images from hyperspectral L1 data could be developed and used operationally with future hyperspectral geostationary satellite instruments like MTG-IRS. The early tests with PCs from chopped spectral domain demonstrate that it is possible to get RGB images using PCs with direct physical interpretation. EUMETSAT should investigate the possibilities to disseminate not only the regular principal components describing the information content of the spectrum or bands of MTG-IRS as a whole, but also dedicated principal components optimised to describe the information within a limited spectral domain (grouped by spectroscopy or physical applications) to satisfy the needs of a wider number of applications than NWP, like specialised RGBs and for chemistry applications. These early tests should be understood as an example against the preconceived idea that MTG-IRS L1 is only for NWP or generation of L2 produdcts. Due to the low number of IASI spectra and synthetic MTG-IRS spectra used in this paper, further studies should be made for each of these early ideas with higher number of IASI spectra and cases. One Web page where these activities could be collected should be created. Since the main purposes is for nowcasting activities the first option should be the NWC SAF web server ACKNOWLEDGEMENTS This study was carried out in the EUMETSAT MIST and NWC SAF activities framework. REFERENCES Martínez M.A., X. Calbet, Prieto J., Tjemkes S. (2010). Use of synthetic RGB images in training. Proceedings of the 2010 EUMETSAT Meteorological Satellite Conference, Cordoba, Spain. Sieglaff J.M., Schmit T.J., Menzel W.P., Ackerman S.A. (2009). Inferring Convective Weather Characteristics with Geostationary High Spectral Resolution IR Window Measurements: A Look into the Future. J. of Atmospheric and Oceanic Technology Vol 26, Pages