Hailstone: An automatic sizing procedure T. Montefmale/ C. Rafanelli* & P. Ferrari* "C.N.R. - Instituto di Fisica delvatmosfera. P. le L.

Transcription

1 Hailstone: An automatic sizing procedure T. Montefmale/ C. Rafanelli* & P. Ferrari* "C.N.R. - Instituto di Fisica delvatmosfera. P. le L. Sturzo, MzcAe/e - MacA, g, ^0^0 -,9am Mzc^e/e a/a Abstract Hailstones are usually sized measuring the dents produced on a panel hit by hail. At this time the impressed panels are processed manually by operators. Then the minor axis of each dent is assigned to an equivalent diameter class. This is a low and tedious work also comporting undesirable effects as reading random errors and classification of hailstone diameter in a few classes only. The goals of this paper are the change of error type from random to systematic one, the data quality improvement analyzing not only the minimum dent diameter but perimeter, area, maximum & Feret diameters and shape factor also. Finally the automatic upgrade of hail data base and the lowering of processing costs. The better data quality allows to overcome the classification technique sizing the single hailstone directly This is very useful in many statistical elaborations as those based on maximum likelihood method. Key words : Atmospheric Precipitation, Hail, Image processing. 1 Introduction Since 1973 a dense hailpad network was installed by the S. Michele Experimental Station in the Trento province, Italy. The aim was to gather consistent informations: even on the effects of hail on the ground, well known having heavy impact on the agricultural economy, as on climatology and microphysic of the hail phenomena. Usually the hail is detected in field by hailpads. Those are instruments that record the kinetic energy of the falling hailstones as a permanent indentation of their surface. Various kinds of hailpads are available, Huggins et a/.*, Long et al*. Those of Trento network are Schleusener type, Schleusener & Jennings \ a polystyrene pad of 15x15 cm covered by a 170 urn aluminum foil % 1. The impressed panel, fig. 2, is removed from his support and then his surface rollered by typographic ink, to evidence the boundaries of dents, so that the measurement of their axes can be made easy. A calibration method based on a experimental curve, Montefmale et al*, permits to estimate the equivalent diameters of hailstones and obtain the hail size spectra. Actually the data reduction is done manually by trained operators with a long and expensive work obviously affected by randomized errors.

2 38 Computer Techniques in Environmental Studies To overcome those unfavorable aspects a digitalized reading system is proposed, his prospective advantages are summarized in table I. A reading from digitalized image allows not only the use of min and max diameters but the area and perimeter also, for a best estimate of equivalent diameter. Moreover the all pad examination area permits a more representative sample of stones' number than a partial pad scanning. To have access to the single diameter instead of a classified distribution of radii is meanfull and gives a better Figure 1 : Hailgauges infield statistical elaboration of spectral parameters using Maximum Likelihood techniques. As a matter of fact the M.L. estimates are consistent and asymptotically efficient, i.e. as the size of the sample is increased the estimates converges stochastically to the desired population values as a limit and the variance of estimates approaches the minimum. Two sources of errors takes place in dents reading: the first due to the surface inking that can make the measure of axes different from real size, not improved by digitalization. The second caused by the differences between various readers and by their psychophysic state at the moment as well. Here the digitalization transforms a random error into a systematic one At last must be consid- bigure2:. ered that in planning an

3 Computer Techniques in Environmental Studies 39 hailpad network, the increase of instrument density, always desirable, comports much higher management cost due mainly to the reduction work. The digitalization gives the possibility to building an archive of image and a data base of all dents measurements. This is useful in successive equivalent hailstones diameters data reductions once that a new more approximated calibration method will be made. Table I. Prospective advantages of digitalized reading. Item measured parameters hailpad area examinated diameter classification errors velocity of processing net density (pads * km"^) Traditional min & max diameter classes from 2.5 to 4.5 mm random due to the reader about 8 pads *h * linked to reading time also Digital part weighed rapresentative min & max diameter, area, perimeter, orientation, localization, numbering all pad surface no classes systematic 20 pads * h"* or more, hardware depending. no linked to 2 Measurements The procedure starts with the pad image acquisition using an Epson GT 800 flat bed color scanner connected to a 486 DX2 66 MHz. 16 Mb IBM like computer. The setup image is: 200 dpi resolution, 256 gray tones. The picture is recorded as BMP standard. Every hailpad is lied on the same place of bed to avoid random errors linked to the localization on the scanner. For each side 1 cm Gray level width is cut off to reject border faults. Then a ranking of five filter (median) is 3 times iterated Figure 3 : Intensity of gray scale of a pad to

4 40 Computer Techniques in Environmental Studies smooth the boundaries of dents and reduce the image noise. This last value is the best compromise between the smoothing effect and the preservation of geometrical dimensions. In fact the mean ratio of the length of filtered and no filtered major axes is and the same ratio for the minor axes is ± Before measurement it needs to recognize the dents; the technique used is based on pixel intensity of gray scale. If a pixel falls within a defined range it becomes part of dent. The low limit value of the interval was assumed to be the max relative minimum of the gray level histogram of each scanned pad (valley technique) Phillips*. The figure 3 shows a sample of pad gray scale where the black (0 value, 64%) is not drawn. So that a threshold covering the intensity range from the value given by the valley technique to 255 was applied. An object is defined as a group of selected pixels contiguous at least for a side, the outer pixels draw the boundaries of dent, Rafanelli et alf\ Inside the boundaries all the gray tone is setted to 255, as in figure 4 right. Once defined the dents, various parameter was measured: major and minor axis length, major axis slope, major and minor axis endpomts, Feret diameter, perimeter, area and shape r factor. This, ' ( SF - renmeter *** ^% ) is a measure of how nearly circular the dent is; major axis slope is the angle from an horizontal line, other parameter dimensions are in pixels, keeping in mind that at our scanning setup one millimeter corresponds at about eight pixels. Major axis length is the distance between the two farthest pixels, minor axis length is the longest line that can be drawn perpendicular to the major axis inside the dent, the endpoints are the coordinates of the extremities of axes to recognize the mark on the pad. (a) (b) *f; fo/yzo/? q/\?/%%// f a/)

5 Computer Techniques in Environmental Studies 41 The Feret diameter is the diameter of a fictitious circular dent that has the same area. The slope gives an indication of prevalent wind direction at ground. All those measurements are recorded in a work sheet and then analyzed to gain infbrmations on hailstorm. A filter is applied to the dents, those having an area minor than 64 pixels are rejected because it is incorrect to reduce them to hail. Usually the shape factor ranges from about 1 and 0.3 but values lower than 0.85 indicate two cases: the first is a poor dent due to irregularities or to rebounds of falling hailstones. The second given by a superimposition of two or more stones. Figure 4 shows a portion of pad where are quite listed many of the possible dent shapes that can occur, marks were produced by a mean-high intensity hail ranging from 4 to 8 mm. The dents with an area major of 64 and a shape factor higher than 0.85 are considered as hailstones having as diameter, that estimated from calibration curve. The other are examined again to tray and split the objects. A partially satisfactory recursive method is a combination of erosion and dilatation binaryfilterin this way a part of overlapped dents are splitted mainly those with low mixed area, fig 5. The other complexes cases actually are manually solved and a more specific method is in progress. Finally the data base is upgraded and to have an easy handling archive the image are stored compressed. 3 Concluding remarks The automatic sizing procedure shown, solves easy the major parts of problem that take place in hail data reduction with hailpads. Actually there are two problems not completely settled: thefirst,as said below^ concerns the dents overlapping. The second one occurs when more dents ovenmpose themselves and simulate a bigger one. In this case the shape factor gives an erroneous indication. The rule followed is a statistical check of diameter spectra so that when a extreme value is found, the ambiguity is solved by the operator

6 42 Computer Techniques in Environmental Studies Acknowledgments: The authors wish to thank Dr. R. Leonardi for his helpful suggestion on image processing. This work is fully supported by CNR - IASM contract. Bibliography 1 Huggins A., Crow E.L. & Long A.B., 1980, Errors in hailpad data reduction. Journal Appi Meteorol. Vol. 19, N 6. 2 Long A.B., Matson R.J. & Crow E.L., 1980, The hailpad: Materials, Data reduction and Calibration. Journal Appl Meteorol., Vol 19, N Schleusener R.A. & Jennings P.C., 1960, An energy method for relative estimates of hail intensity. Bull. Am. Meteorol Soc. Vol. 41, N 7. 4Montefmale T, Ferrari P., Rafanelli C & Paoletto P.,1983,Misure dell'attivita grandinigena in provincia di Trento, periodo ,Part I: Calibrazione dei rilevatori di grandine al suolo. Experience e Ricerche, Vol. XII, Staz. Sper. Agr. For, S. Michele a/a (Trento). 5 Phillips D., 1993, Image Processing, Part 9: Histogram-Based Image Segmentation. "C" (AcrA Jowr/za/,Vol. 11, N 2. 6 Rafanelli C, Montefmale T. & Ferrari P., 1991, Hail pad data reduction by computer, Environmental Software, Vol. 6, N 4.