| dc.description.abstract | A new technique is described for estimating daily rainfall by means of visible and infrared geostationary satellite imagery. It is designed for the tropics and warm-season midlatitudes. Because it operates on a grid of points and measures time changes at these points, the technique has been named ?grid history.? It is assumed that at any grid point in some image belonging to a sequence, by means of spectral, textural and evolutionary information, it is possible to classify instantaneous rain rate as nil, light, moderate or heavy. Then the total rainfall over a day is the sum over three classes of the product of frequency and class average rate. The class average rates have been determined by least-squares multivariate linear regression of frequencies on observed rainfalls. The areas treated are South China Sea, India, Arabian Sea, tropical North Atlantic Ocean and Amazonia. Inland India had the lowest (driest) class average rates (coefficients), coastal India the largest (wettest) coefficients. Differences in coefficients were least for the Arabian Sea and Atlantic Ocean. There the class average rates were roughly zero (by definition), 1.5, 6 and 15 mm h?1. For the strongly convective rain regimes treated here, it was found to be important to ?look? at the area at least once per hour. A loss of accuracy in estimates over land apparently was due to unexpectedly large terrain and synoptic effects. Best-circumstance estimates of daily rainfall for an area 100 km on a side should be within a factor of 2 of true rainfall. | |
