Nowcasting of Thunderstorms from GOES Infrared and Visible Imagery

In this paper, we describe our progress in identifying and tracking storms at multiple scales from satellite infrared (11-micron Band 4) and visible (Band 1) channels. Storms are identified by clustering the pixels in the input images using spatial-contiguity-enhanced K-means clustering. Identified clusters are then processed morphologically to yield self-consistent storms. Identified storms (at all the scales) are tracked using a hybrid scheme that minimizes mean absolute error between frames of the input sequence of images and then smoothed temporally using Kalman filtering. This yields a grid of motion vectors at each pixel in the spatial domain. The motion vector estimated from the sequence is used to nowcast the images. Comparison of the nowcasts with the observed values at the corresponding time gives a measure of skill of the nowcast. Statistical properties are extracted for each cluster. The extracted properties are used as inputs to an automated decision tree training algorithm to identify regions of overshooting tops. Results and measures of skill are demonstrated on a sequence of images from Oct. 12-13, 2001. ∗Corresponding author address: [email protected], also affiliated with NOAA/OAR/National Severe Storms Laboratory. 1. Short-term forecast methods There are two broad techniques to ”track” storms from remotely sensed imagery. One method is to identify the centroids of spatially-contiguous pixels above a particular threshold and to match these centroids across time. A principed approach to such association can use linear programming (Dixon 1994), although most implementations use heuristics based on proximity and size of the storms in question. Change in position and trends of storm properties are then extrapolated. A second technique is to use rectangular sub-grids and find the maximum correlation within a search radius (Rinehart and Garvey 1978; Tuttle and Gall 1999). A modification of this technique is to pre-filter the data so as to track only the larger scales (Wolfson et al. 1999; Lakshmanan 2000). It is also possible to use sub-grids ranging in size from that of the entire image to small 16km x 16km grids and to compute motion estimates at each of these scales. Smoothness criteria can be used to constrain these estimates at different scales. Identifying, matching and extrapolating storm core locations is suitable for small scale storms. The large scale features and cross-correlation technique is suitable for longer forecasts, but with loss of detailed motion estimates. An assumption here is that the storms are of the scale of the sub-grid, not larger. The multiscale estimation is suitable also for large scale forecasts, but with less precise detailed motion estimates. When used for advection, all the correlation techniques rely on reverse projection, so there needs to be