Unlike traditional geostationary estimation, passive microwave estimation using DMSP SSM/I and TRMM TMI is much more direct and accurate. This is because passive microwave rain rates are based on channels that sense precipitation in clouds and do not rely on cloud top temperature. However, the problem with passive microwave rates is that they are extremely sparse in space and time. It is therefore unlikely that a user needing a current estimate of rain over a particular region could obtain a timely passive microwave rain rate.

The solution is to tune the abundant geostationary IR passes using the sparse but accurate passive microwave rain rates. This is done by co- locating in space and time newly arriving microwave and geostationary data sets for 15-degree latitude/longitude boxes. Using these data within each box, an algorithm periodically derives a new lookup table between infrared brightness temperature and rain rate based on the co-located satellite measurements. An important aspect of this procedure is the derivation of a zero-rain rate temperature. At temperatures greater than this threshold, no precipitation is estimated. At lower temperatures, estimates are derived, which are inversely proportional to temperature. Different relationships and zero-rain rate temperatures are derived for each box. Thus, rainfall estimates will be based on relationships which vary from place to place, optimizing accuracy overall.

There are a number of relatively minor adjustments applied to this technique to improve the estimates. Cloud tops that are warming (measured over an hour's time) are assumed to be decaying. Rain rates are adjusted downward accordingly. Also, there are a number of multi-spectral tests applied over GOES data to test for thin cirrus. If thin cirrus is detected, no rain rates are derived. One difficulty of the technique is that rainfall is sometimes estimated poorly in regions with high mountains. Thus, a procedure checks for orographic effects in the corresponding NOGAPS data and adjusts the totals accordingly.

2. The technique is regionally tuned, meaning that estimates will be valid within local rainfall regimes.

3. The output is given both in terms instantaneous rain rate and accumulations over different time scales. The instantaneous rates are useful over oceans because precipitation in the atmosphere can interfere with line-of-sight measurements and communications. The accumulations over land give information about soil moisture and potential flooding.

4. The potential impact of precipitation-bearing storms headed for land (tropical cyclones and frontal systems) can be inferred by using this product.

5. Preliminary results suggest that assimilation of this product into numerical models can greatly improve forecasts.

6. The technique should have longevity since many more passive microwave instruments are planned for space. In addition, new multi-spectral satellites will allow improvements worldwide that are now just in effect for GOES.

2. In some areas cold, but thin cirrus can be misidentified as precipitation on geostationary infrared images. Often this effect appears on the edges of actual precipitation cells as a "fringe" effect. Thus, precipitation may be correctly identified by the technique, but the precipitating area is too large.

3. The technique is most effective for tropical convection. Precipitation estimates may suffer in midlatitude frontal systems consisting of stratiform clouds, especially over land.

4. The technique should be used with caution over midlatitude land areas in the colder seasons of the year, especially in particularly cold areas. It should not be used when the surface precipitation type is snow.

5. The technique depends on the amount of current passive microwave data that is available. When relatively few passive microwave instruments are orbiting, the resulting match-ups with the infrared can be based on data that are more than twenty-four hours old, resulting in degraded estimates.

6. The technique usually cannot account for orographic precipitation effects that occur near the surface of the earth, well beneath cloud tops. Windward areas can be analyzed as too dry, and leeward areas can be analyzed as too wet. There are partial corrections in effect to account for this problem.

The IR image (Fig. 2), taken three hours later, shows a deep cloud band over Western United States. The advantage of IR images is that they cover an area frequently. Cold cloud tops show regions where precipitation is likely, though it is not possible to determine how heavy precipitation is based on IR images alone.

The geostationary rainfall technique (Fig. 3) shows precipitation rate over the country based on microwave (SSM/I + TRMM) passes (e.g., Fig. 1) and a geostationary pass (Fig. 2). It fills in the huge areas missing from Fig. 1, especially over land and over coastlines, but it lacks the accuracy that Fig. 1 has, especially over water. Also, it tends to put precipitation in places not favored by orographic lift. In Fig. 1, significant precipitation is actually falling in a north/south line over the Sierra Nevada mountains in Northern California (mostly blues with some red). This is caused by westerly flow causing enhanced precipitation over the Sierra Nevada. But the geostationary rain rate shows the bulk of the precipitation displaced into Western Nevada, a desert region with little precipitation. This incorrect result arises because a high band of non- raining cirrus (Fig. 2) over the region is analyzed as a region of significant precipitation.

Fig. 4 shows a six- hour accumulation of precipitation (3-9 Z). This represents the sum of the past six hours worth of estimates, ending with the latest estimate, at 9 Z (Fig. 3). Notice that Fig. 4 shows areas of precipitation that occurred in the recent past but that may not still be occurring in the present as shown in Fig. 3. Thus, caution must be taken to not use accumulated precipitation to estimate precipitation in the present.

Fig. 5 shows the accumulation of 24-hours of precipitation before the 9 Z end time shown in Fig. 3. Notice that the area covered by precipitation is greater because precipitation affects a greater area over 24-hours than over six hours. Notice for example that Texas shows no rain at 9 Z (Fig. 3) and none for a six-hour prior accumulation (Fig. 4). However, for a 24- hour prior accumulation, Texas shows precipitation. This is because the accumulation (Fig. 5) shows precipitation that fell early in the twenty- four cycle but that later stopped.

Fig. 6 shows the accumulated precipitation from microwave sources over the same twenty-four hour period shown in Fig. 5. The precipitation for the sparse microwave passes are extrapolated in time in order to produce this product. Notice that it shows mesoscale detail over precipitation, as opposed to the smoothing apparent in the geostationary accumulation (Fig. 5).

Also, the area of precipitation is smaller than in the geostationary technique. Look for example over Texas. The geostationary technique overestimates precipitation because of thin cirrus at the fringes of precipitation that is often misinterpreted by the algorithm as precipitation.