__Technical Attachment__

SR/SSD **98-18**

**5-1-98**

1. **INTRODUCTION**

In order to support the National Weather Service (NWS) efforts to improve flash flood watches and warnings and heavy precipitation forecasts, the NOAA/NESDIS Satellite Analysis Branch (SAB) in Camp Springs, MD are currently providing satellite precipitation estimates and outlooks to the NWS Forecast Offices and River Forecast Centers around the country. Satellite-derived precipitation estimates have been provided operationally since 1978 using the Interactive Flash Flood Analyzer (IFFA) and an algorithm developed by Scofield and Oliver (1977) and Scofield (1987). Due to the considerable time needed for image processing, interpretation and computation, the meteorologist using IFFA is able to analyze only one cloud precipitation system at a time, When there are a large number of thunderstorm complexes, an automatic fast rainfall rate estimation technique would be very useful to provide rainfall estimates over the whole country. In response to NWS interest, ARAD scientists have developed an experimental automated satellite derived rainfall rate estimation technique (called the Auto-Estimator). The Auto-Estimator, Vicente (1997), uses the full spatial and temporal resolution of the GOES-8 longwave infrared (IR) window (10.7 um). Instantaneous estimates, average rainfall rates over the past one hour and 3, 6, and 24 hours accumulations are processed for convective systems, tropical and extratropical cyclones in real time both for the continental US and coastal areas. The technique has been tested daily by SAB. In this paper we describe the main points of the Auto Estimator and show some validation work.

The instantaneous and 1-hour average estimates as well as the 3, 6 and 24 hour accumulations are available in real time to the public through the NOAA/NESDIS Flash Flood Home Page located on the Web site http://orbit-net.nesdis.noaa. gov/ora/ht/ff/.

**2.** **TECHNIQUE DESCRIPTION**

The Auto-Estimator initially computes instantaneous rainfall amounts based on a regression relationship between cloud top temperature (10.7 um brightness temperature) and surface rainfall rate. Next it uses model generated relative humidity (RH) and precipitable water (PW) to analyze the environmental moisture and scales the rainfall amounts accordingly. Finally, two additional tests are used to screen nonraining pixels; rain is assumed to be associated with growing clouds exhibiting overshooting domes. Consecutive half-hourly satellite IR images are used to indicate growing and decaying cloud systems. The cloud top temperature gradients are used to identify overshooting tops and sinking (subsiding) areas within the cloud system.

2.1 Rain rate versus cloud top temperature

The precipitation amounts are initialized using a power law fit between instantaneous surface radar derived rainfall estimates and satellite measured 10.7 um IR cloud top brightness temperatures, figure 1. The power law curve is given by

Fig. 1. Mean rainfall rate for each temperature from 195.0 K to 260.0 K computed from 6837 collocated pares of radar derived rainfall rate estimates and IR cloud top temperature(dotted curve). Power law fit between radar derived rainfall estimates and cloud top temperature (solid curve). |

where Rain is the rain rate (mm/hr) and T is the cloud top brightness temperature (K). Both rain and non-rain situations are considered in the regression fit.

This relationship has been computed using well calibrated and navigated pairs of GOES-8 10.7 um IR images and collocated instantaneous radar rainfall estimates from the US operational network of 5.0 and 10.0 cm NWS radar (WSR-57S, WSR-74C, WSR-88D), covering the US central plains and the Gulf of Mexico. Both the radar rainfall estimates and the IR cloud top temperature data have a spatial resolution of 4 km. The data set from observations during the months of March to June 1995 was limited to well defined cloud systems where convective core rainfall could be distinguished from cirrus contamination.

2.2 Moisture correction factor

The moisture correction used in this work is based on the product of the PW and RH (surface to the 500 mb) from the ETA model forecast for the nearest synoptic time. This product is scaled from 0 to 2 (Scofield and Oliver, 1980) and is called the PWRH moisture correction factor. It has the property of decreasing the rainfall rates in very dry environments and increasing rates in very moist ones. The IFFA Technique (Scofield, 1987) uses a similar moisture correction factor. Experience has suggested that if T < 210 K and PWRH > 1.0 the environment is already too wet and the computed rainfall rate should not be multiplied by the PWRH correction factor. On the other hand, if T < 200 K the rainfall rate should be limited to 72.0 mm/h, approximately the extreme rainfall rate found over the US.

2.3 Cloud growth rate correction factor and the cloud top temperature gradient

The correction for rate of cloud top growth or decay uses collocated pixels in two consecutive half hour images. A convective system is more active and produces the greatest rainfall rates when the tops are becoming colder and expanding (Woodley at al. 1985, Griffith at al. 1978 and Scofield, 1977). Thus detection of active or decaying portions of thunderstorms can be attempted by searching for those pixels in the IR image that became colder, warmer, or stay at the same temperature during the half-hour interval. This makes the assumption of a stationary convective system, which is somewhat reasonable given the 4 km and 0.5 hr resolution of the data. If the coldest IR pixels in the first image are colder in the second image, the convective system is intensifying and associated with the heaviest precipitation rates. In this case the rainfall rate computed from the regression curve remains unchanged. On the other hand, if the coldest IR pixels in the first image are warmer in the second image or have the same temperature, the convective system is weakening and upward vertical motion has likely ceased. In this case, the rainfall rate is adjusted to zero. The temperature gradient correction searches for the maximums (coldest) and minimums (warmest) cloud tops within a 3 x 3 pixel area centered on every pixel. If there is a maximum, the rainfall rate given by the regression curve remains unchanged. On the other hand, if there is a minimum, or no maximum or minimum, the rainfall rate is set to zero.

2.5 Rainfall rate computation

The Auto-Estimator rainfall rate for each pixel in the IR image is given by the product of the rainfall rate derived from the power law regression curve, times the moisture correction factor, times the growth rate or gradient correction factor. The instantaneous rainfall rate images are calculated for each GOES-8 IR image available 15 and 45 minutes after the hour. The average hourly rainfall rate is computed on a pixel by pixel basis using a weighted average where the median of the three values receives twice the weight, so that for every pixel the hourly rainfall rate is given by

3. **VALIDATION**

In order to assess the impact of each of the correction factors. rainfall rates were computed through four different scenarios of combination: using the regression curve alone; the regression curve plus the PWRH correction" the regression curve plus the PWRH and the gradient correction factor,- and the regression curve plus the PWRH and the growth rate correction factor. Our experience has shown that the PWRH correction factor is fundamental in eliminating excessive rainfall amounts usually present within areas of very cold cloud tops and dry environments. On the other hand, the combined use of the gradient and growth rate correction factor have the tendency of causing severe decrease in the rainfall rate amounts.

A total of 20 hours of hourly rainfall estimates for three different case studies were completed within the southern region of the US for storms systems on October 28, 1996, March 18/19, and June 22, 1997@ These heavy rainfall events (over 100 mm within a 24 hour period), include Mesoscale Convective Systems (MCSS) and squall lines. Estimates area averaged to 12, 48 and 100 km grids, and 3 x 3 degree grids, and verified against gauge-adjusted radar measurements at the same resolution. Several statistical measures, shown in Table 1, are used to compare the GOES estimates at each grid resolution with the rain gaugeadjusted radar measurements. These include the correlation coefficient, the mean rainfall rate, the bias, the root mean square error (RMSE), the standard deviation (SD), the false alarm ratio (FAR), the probability of detection (POD) and the error (ERR). The best statistical values are boldfaced.

A detailed sensitivity analysis have shown that the PWRH correction factor significantly improves the power law regression rainfall rate estimates. The gradient/growth rate corrections on the other hand have shown effective as a masking tool of the non-precipitating areas in the cloud system.

Qualitative evaluation of well defined and isolated storm systems indicate that the growth rate correction tends to be more efficient than the gradient correction as a rain non-rain discriminator during the early stage of the life cycle of a precipitation system. Once the system has reached a mature stage, the growth and decay rates are obscured by cirrus debris in the satellite data and the gradient correction begins to screen out non-raining parts of the cloud in the areas of low temperature (lower than 213 K) and weak gradients found in anvil cirrus clouds.

versions of the Auto-estimator for one hour estimates and 5 difference grid size for all cases

IR + PWRH +GRADIENT |
IR + PWRH +GROWTH RATE |
|||||||
---|---|---|---|---|---|---|---|---|

Grid size (Km) | 12x12 | 48x48 | 100x100 | 3x3 deg | 12x12 | 48x48 | 100x100 | 3x3 deg |

Sample size | 5188 | 348 | 98 | 20 | 5188 | 348 | 98 | 20 |

Correlation | 0.59 | 0.72 | 0.77 | 0.91 | 0.54 | 0.64 | 0.68 | 0.87 |

Satellite mean (mm) | 2.43 | 2.50 | 2.38 | 2.45 | 2.38 | 2.28 | 2.07 | 2.41 |

Radar mean (mm) | 1.87 | 1.93 | 1.81 | 1.88 | 1.87 | 1.93 | 1.81 | 1.88 |

Bias (mm) | 0.56 | 0.57 | 0.57 | 0.57 | 0.51 | 0.35 | 0.26 | 0.53 |

RMSE (mm) | 3.21 | 2.43 | 1.72 | 0.99 | 3.45 | 2.63 | 1.86 | 1.21 |

Satellite SD (mm) | 3.27 | 3.22 | 2.48 | 1.87 | 3.44 | 3.01 | 2.22 | 2.09 |

Radar Sd (mm) | 3.64 | 3.14 | 2.37 | 1.41 | 3.64 | 3.14 | 2.37 | 1.41 |

FAR | 0.34 | 0.19 | 0.06 | 0.00 | 0.34 | 0.18 | 0.11 | 0.00 |

POD | 0.85 | 0.92 | 0.98 | 1.00 | 0.80 | 0.89 | 0.97 | 1.00 |

ERR | 0.29 | 0.21 | 0.12 | 0.00 | 0.30 | 0.22 | 0.13 | 0.00 |

**5.** **CONCLUSIONS**

This paper presented the description of a new rainfall estimation technique, called the Auto Estimator. The rain rate is derived from a power law regression relationship between the GOES-8 IR 10.7 um brightness temperature and surface radar measurements of rainfall. It is adjusted by model derived PW and RH, and detailed evaluation of changes in cloud top temperatures attributed to cloud growth and cloud structure. The technique runs in real time and is designed for application to flash flood watches/warnings, heavy precipitation forecasting, and initialization of numerical weather prediction and hydrological models. Validation and statistical analysis have been performed on three storm systems by comparing hourly estimates (from the Auto Estimator) with collocated radar-adjusted estimates. Preliminary results on the potential application of the Auto Estimator for flash floods/heavy precipitation prediction are encouraging. Results show that the Auto Estimator has some skill at one hour time resolution and spatial resolutions of 12 km and greater.

*Reprinted from the preprint volume of the 16th Conference on Weather* *Analysis and Forecasting and Symposium on The Research Foci of* *the U. S. Weather
Research Program 1 *1 *- 16 January 1998, Phoenix.* *AZ, by the American Meteorological Society, Boston, MA*-

**6.** **REFERENCES**

Griffith, C.G., W.L. Woodley, P.G. Grube, D.W. Martin, J. Stout, and D.N. Sikdar, 1978: Rain Estimates from Geosynchronous Satellite Imagery: Visible and Infrared Studies. Mon. Wea. Rev, 106, 1153 - 1171.

Scofield, R. A., and V. J. Oliver, 1977: A scheme for estimating convective rainfall from satellite imagery. NOAA Tech. Memo. NESS 86, U.S. Dept. Commerce, Washington, DC, USA, 47 pp--- 1987: The NESDIS operational convective precipitation technique. Mon. Wea.Rev., 115, No. 8, pp 1773-1792.

Vicente, G. A., 1997: Real time rainfall rate estimates derived from the GOES-8/9 satellites for flash flood forecasting, numerical modeling and operational hydrology. 13th Conference on Hydrology, February 2-7, 1997, Long Beach, California, by AMS, Boston, MA, J115-J118.