ABRFC Precipitation Processing
Click here for a more in-depth discussion on precipitation measurements and their relationship to hydrologic modeling.
Most RFCs use the Multisensor Precipitation Estimator (MPE) for generating gridded precipitation estimates. The ABRFC employs a different software program to produce the gridded, multi-sensor precipitation estimates mosaicked over our forecast area. Both use the same basic input: the hourly digital precipitation product (HDP) computed by each radar within the ABRFC's area, and hourly rain gage reports (usually at automated reporting sites). Both also use Hydrologic Rainfall Analysis Project (HRAP) grid, a 4km x 4km grid.
The ABRFC has written its own software for precipitation processing, beginning with P1 in 1996 leading up to its current version P3. The basic methodology follows: Every hour, a mosaic of all the HDPs is created by combining them in one product that covers the entire ABRFC basin. Where radars overlap, distance-weighted average values are used. (Note, the maximum overlapping value can be used to provide better estimates in certain weather conditions.) Also each hour, a collection of all reporting rainfall amounts from gage sites is created. A irregular triangulated grid field is created using the locations of the gage sites. The radar mosaic is overlayed on this triangulated grid and a 'bias' field is created based on the difference between the radar value and the gage value. Where there is no gage site, a 'bias' is computed using the triangular grid and the distance from the nearest gage sites. The resultant bias field is then used to create the final precipitation product. P1 was heavily based on a similar method of daily precipitation processing created by the Tulsa District Corps of Engineers.
Hour-by-Hour Quality Control
There are quite a few sources for error inherent in the process of estimating precipitation. All ABRFC-generated precipitation estimates undergo extensive quality control to identify and correct all of the following issues.
- Radar as a source of error.
- Hail Contamination - Radar-based precipitation measurements are based on the relationship between "reflectivity" of raindrops and the rainfall rate. Wet hail stones within a storm clouds relfect much more energy back to the radar than an equivalent amount of all-liquid precipitation, which results in overestimation.
- Beam Blockage - Mountains, forests, towers, etc., very near the radar can block the radar beams from adequately sampling the atmosphere. This can result in large areas of underestimated rainfall.
- Anomalous Propagation (AP) - Under certain atmospheric conditions, the radar beams actually bend back toward the ground, and reflect off of buildings, hills, etc. This "ground clutter" appears as rain where none fell.
- Non-Precipitation Echoes - Radar beams occasionally reflect off of items in the air that are not producing rain at the surface. Examples of this include birds, bats, virga (rain that evaporates before it reaches the surface) and chaff (reflective materials used by the military to confuse/counter enemy radar, and tested occasionally on domestic military bases).
- Bright Banding - Bright banding occurs due to the higher reflectivities associated with snow that is melting as it is falling aloft. Because ice scatters more radar radiation compared to liquid water, snow shows a lower reflectivity on radar. When the snow is melting however, a film of water forms on the outside of the snowflake. These snowflakes can be fairly large, and show up on radars as extremely heavy rain. A radar beam normally samples a higher elevation as it moves away from the radar site. Because the melting of the snowflakes occurs within a specific elevation range aloft, there will be a higher reflectivity as the radar beam moves through this layer. This can produce a circular or arcing band of higher reflectivity around the radar site on the reflectivity display.
- Range Degradation - As the radar beams go farther out, they sample higher parts of the storm. Storms with low cloud tops are frequently under-sampled when they are farther away from the radar. This is particularly common with winter weather events.
- Improper Z-R Relationships - Convective storms, tropical storms, and winter storms all require different relfectivity-to-rainfall (Z-R) relationships. An incorrectly set Z-R relationship can seriously impact the rainfall estimates.
- Algorithm Errors - Various other problems can increase the error in the precipitation estimates. For example, radar echoes from light snow often are too weak to be treated as rainfall by the radar programming. This results in a "no-rain" estimate, even though precipitation is reaching the ground.
Rain Gages as sources of error. The "ground truth" isn't always true...
- Sampling Errors - Like any instument, automated rain gages are subject to errors. In heavy rain events, the rain can fall too quickly for tipping-bucket rain gages to keep up. When this happens, some rain can spill over of the side, resulting in inaccurately low "ground truth".
- Fully-Clogged Gages - Bird nests, leaves, spiders, etc, can partially or completely stop the rain from reaching the recording mechanism in a rain gage. Completely clogged gages appear as 0.00" estimates in areas of heavy rainfall.
- Partially-Clogged Gages - Although the 24-hour total may be accurate, ground truth from clogged gages is not temporally accurate. For example, 1.00" of rain in 1 hour may record over 5 hours as 0.85", 0.11", 0.02", 0.01", 0.01".
- Frozen / Melting Precipitation - This can be thought of as a combination of fully- and partially-clogged rain gages. When the frozen precipitation falls it is not recorded. Days or weeks later, when it melts, the rain gage reports rain.
Software as a source of error.
- Both precipitation processing methodologies work best in areas with many observations. When there are fewer rain gages, the bias estimates are less accurate. Additionally, unusual shapes (e.g. small circles, triangles, large polygons) can appear in the final rainfall estimate.
Quality Control of the Hydrologic Day
Supplementing data from automated rain gages, the National Weather Service's Cooperative Observer (COOP) Network reports official 24-hour rainfall totals every day. There are many instances where manual observations are more accurate than automated ones, such as intense rainfall or frozen precipitation. ABRFC HAS forecasters use this collection of high-quality, manual observations as ground truth to perform a daily precipitation "post-analysis". If the ground truth shows that more rain fell than the estimates indicate, the HAS forecaster increases the rainfall. Similarly, if the ground truth shows that the precipitation estimates are too high, the HAS forecaster decreases the estimates. The HAS forecaster uses radar, satellite and surface observations to ensure the rainfall corrections are added to the correct hours.
The 24-hour post-analysis also allows the HAS forecaster to identify systemic problems over the past hydrologic day. Areas of excess AP, bright bands, and other radar phenomena, which may not have been noticeable in individual hours, are frequently easier to identify when looking at the entire 24-hour period. Similarly, rain gages which reported values which were too low tend to appear as erroneous minima on the 24-hour product.