Improving PERSIANN System in High Resolution Precipitation Estimation K' Hsu, S' Mahani, Q' Fan, J'

1
Improving PERSIANN System in High Resolution
Precipitation EstimationK. Hsu, S. Mahani, Q.
Fan, J. Li, X. Gao, and S. SorooshianDepartment
of Hydrology and Water Resources, The University
of Arizona
Cross Section of Cloud Ice Particle Liquid
Water Content from RAMS for 6-Hours
Southwest US SNOTEL Stations
ABSTRACT
Improving PERSIANN Rainfall Estimate by Using
Mesoscale Model Information

• Selecting stations with only snowfall, over the
  Southwest U.S., based on the following
• Average Daily Surface-Air Temperature lt 32
  F
• Daily Snowfall gt 5 Inches

Tb-RR Relationship of Various Cloud Patches
Under the support of the NSF-SAHRA and NASA
EOS-TRMM programs, our efforts have been focused
on high-resolution rainfall estimation and
extending our model estimates from rainfall to
include snowfall depth during the winter seasons.
Regarding the high-resolution rainfall
estimation, our final goal is providing hourly
rainfall estimates at a 4 km x 4 km scale.
Several efforts have been implemented in refining
our developed rain algorithm (PERSIANN,
Precipitation Estimation from Remotely Sensed
Information using Artificial Neural Networks).
In this presentation, several recently developed
rainfall and snow depth estimation algorithms,
over the Southwestern United States, are
introduced.
2000 UTC, 07/8/99
2100 UTC, 07/8/99
2200 UTC, 07/8/99
18.3 14.3 9.3 5.4 3.0 1.5
0.5 0.0
18.3 14.3 9.3 5.4 3.0 1.5
0.5 0.0
18.3 14.3 9.3 5.4 3.0 1.5
0.5 0.0
3.0 2.5 2.0 1.5 1.0 0.5 0.0
Relationship between rainfall and cloud-top
brightness temperature is different for each
cloud class.
In following example, severe rain and flood
occurring on July 08, 1999, demonstrate an
ability to improve rainfall estimates from
PERSIANN through considering mososcale model
information.
Column Ice Content (g/mm3)
Height (km)
Map of All Stations
Map of Snowfall Stations
50 45 40 35 30 25 20 15 10 5 0
44 43 42 41 40 39 38 37 36 35 34 33
44 43 42 41 40 39 38 37 36 35 34 33
4.25 3.50 2.75 2.00 1.25 0.50
35.5 36.0 36.5
37.0
35.5 36.0 36.5
37.0
35.5 36.0 36.5
37.0
6-Hour GOES Satellite Cloud-Top Tb
2300 UTC, 07/8/99
0000 UTC, 07/8/99
0100 UTC, 07/8/99
3.0 2.5 2.0 1.5 1.0 0.5 0.0
18.3 14.3 9.3 5.4 3.0 1.5
0.5 0.0
18.3 14.3 9.3 5.4 3.0 1.5
0.5 0.0
18.3 14.3 9.3 5.4 3.0 1.5
0.5 0.0
Surface Elevation (km)
1600 UTC, 07/8/99
1700 UTC, 07/8/99
1800 UTC, 07/8/99
oK
Latitude (Degrees, North)
Latitude (Degrees, North)
240
38 37 36 35
38 37 36 35
38 37 36 35
Ground-Based Radar Rainfall Rate (mm/h)
230
Column Ice Content (g/mm3)
Height (km)
Latitude (Degrees, North)
Cloud-Top IR (Kelvin)
220
210
115 113 111 109 107
105
115 113 111 109 107
105
Rainfall Generation System (PERSIANN) PERSIANN
Algorithm, input data from geostationary
satellite, high resolution TRMM radar
precipitation, ground-based radar and gauge
rainfall for model training, and output rainfall
rate products (Global Tropical, Pan American, SW
US, and Africa
35.5 36.0 36.5
37.0
35.5 36.0 36.5
37.0
35.5 36.0 36.5
37.0
Longitude (Degrees, West)
Longitude (Degrees, West)
200
Latitude (Degrees, North)
Latitude (Degrees, North)
Latitude (Degrees, North)
117 116 115
114
117 116 115
114
117 116 115
114
2000 UTC, 07/8/99
2100 UTC, 07/8/99
1900 UTC, 07/8/99
oK
200 210 220 230 240
250 260 270 280
290 300
240
38 37 36 35
38 37 36 35
38 37 36 35
Calibration Case (Model Training)
Cloud-Top IR Brightness Temperature (Kelvin)
230
Comparison between observed and estimated
snowfall depths from PERSIANN algorithm (16
SNOTEL observations used for calibration and
training the model) on January 21, 1999.
Extending PERSIANN Ability to Snowfall Depth
Estimation
Latitude (Degrees, North)
Cloud-Top IR (Kelvin)
220
NEXRAD and Estimated Rainfall Rates Comparison
Comparison between ground-based radar rainfall,
and estimated Instantaneous rainfall from
PERSIANN algorithm and cloud type classification
technique, at 0045 UTC, on July, 15, 1999.
Using cloud mask can improve rainfall estimates.
210
200
PERSIANN Medium Spatial and Temporal Global
Gridded Coverage
117 116 115
114
117 116 115
114
117 116 115
114
Longitude (Degrees, West)
Longitude (Degrees, West)
Longitude (Degrees, West)
Hourly Snowfall Depth
Daily Snowfall Depth
Introduction Snow is the main source of water
supply in Southwester United States, that is why
an algorithm was developed for estimating
snowfall depth over this region. In this study,
hourly and daily snowfall depth were retrieved
from multisource variables, including satellite
infrared cloud top imageries, Tb, surface-air
temperatures, and topographic data (elevations).
Daily snowfall observations from 32 SNOTEL
stations, over the SW U.S., during January 1999,
are selected for model calibration and
validation.
Estimation
3.0 2.5 2.0 1.5 1.0 0.5 0.0
25 20 15 10 5 0
GOES
CORR 0.94 RMSE 0.15
CORR 0.93 RMSE 1.01
6-Hour Ground-based Radar Rates
High Temporal, Low Spatial
1700 UTC, 07/8/99
1600 UTC, 07/8/99
1800 UTC, 07/8/99
38 37 36 35
38 37 36 35
38 37 36 35
ANN Estimated Hourly Snowfall (Inches)
ANN Estimated Daily Snowfall (Inches)
TRMM
40
Training
GOES IR Image
Radar Rainfall Map
20
Latitude (Degrees, North)
Radar Rainfall Rate (mm/h)
10
IR Comparable but low temporal Radar High
Spatial, low temporal, Narrow Swath
Pan American
2
0.0 0.5 1.0 1.5
2.0 2.5 3.0
0 5 10 15
20 25
0
Observed Hourly Snowfall (Inches)
Observed Daily Snowfall (Inches)
NEXRAD Gauges
SW U.S
117 116 115
114
117 116 115
114
117 116 115
114
Africa
2000 UTC, 07/8/99
2100 UTC, 07/8/99
1900 UTC, 07/8/99
38 37 36 35
38 37 36 35
38 37 36 35
40
Verification Test
Radar High Spatial Temporal Mountain
Blockage Gauges Spotty Coverage
Global-tropical
20
Latitude (Degrees, North)
Comparison between observed and estimated
snowfall depths from PERSIANN algorithm in
verification test (the other 16 stations) on
January 21, 1999.
Radar Rainfall Rate(mm/h)
10
Methodology
200
220
240
260
280
300
0
5
10
15
20
25
Rainfall Rate mm/hr
Cloud-Top Tb (K)
2
0
PERSIANN Rainfall Map
Cloud Mask Rainfall Map
117 116 115
114
117 116 115
114
117 116 115
114
Identifying Cloud Types and Mask Classification
and Identification of Cloud types according to
the Distribution, of GOES Infrared Cloud-Top
Brightness Temperature, Over the Southwest United
States, at 0045 UTC, on July, 14, 1999.
Hourly Observed Snowfall Depth (for Calibration)
Longitude (Degrees, West)
Longitude (Degrees, West)
Longitude (Degrees, West)
Daily Snowfall Depth
Hourly Snowfall Depth
INPUT Hourly - Pixel Cloud Top Tb - Mean Tb of
Window - Std. Dev. of Tb in Window -
Surface-Air Temperature - Surface Elevation
6-Hour PERSIANN Estimated Rainfall Rates
25 20 15 10 5 0
3.0 2.5 2.0 1.5 1.0 0.5 0.0
OUTPUT Hourly/Daily SNOWFALL
CORR 0.86 RMSE 1.53
CORR 0.79 RMSE 0.28
1600 UTC, 07/8/99
1700 UTC, 07/8/99
1800 UTC, 07/8/99
38 37 36 35
38 37 36 35
38 37 36 35
ANN Model (PERSIANN)
40
ANN Estimated Hourly Snowfall (Inches)
ANN Estimated Daily Snowfall (Inches)
20
Latitude (Degrees, North)
Estimated Rainfall Rate (mm/h)
10
300
2
0
5
10
15
20
25
0
280
Rainfall Rate mm/hr
117 116 115
114
117 116 115
114
117 116 115
114
0.0 0.5 1.0 1.5
2.0 2.5 3.0
0 5 10 15
20 25
2000 UTC, 07/8/99
2100 UTC, 07/8/99
1900 UTC, 07/8/99
260
38 37 36 35
38 37 36 35
38 37 36 35
Observed Hourly Snowfall (Inches)
Observed Daily Snowfall (Inches)
Ground-based radar rainfall, estimated rainfall
from PERSIANN model and cloud mask technique, at
0045 UTC, on July, 14, 1999.
Cloud-Top Brightness Temperature (K)
40
Estimating Hourly Surface-Air Temperatures
240
20
Latitude (Degrees, North)
Estimated Rainfall Rate(mm/h)
10
Discussion and Future Work
Hourly surface-air temperature, Ta , was
estimated from observed daily maximum and
minimum temperatures Thour(t) Tmax (Tmax
Tmin) . Sin(a.t b) b p / 2 12 hours
(depends on geo-location), and a -p / 24.
220
PERSIANN Rainfall Map
Cloud Mask Rainfall Map
Radar Rainfall Map
2
0

• Cloud-type classification and mask approach, and
  cloud physics information from atmospheric
  mesoscale models (RAMS) are used in the
  improvement of rain estimates. Our final goal is
  to retrieve high spatial and temporal resolution
  (up to hourly-based for 4km x 4km pixel) of
  precipitation estimates.

200
117 116 115
114
117 116 115
114
117 116 115
114
Longitude (Degrees, West)
Longitude (Degrees, West)
Longitude (Degrees, West)
Cloud Classification Map of 4 x 4 boxes for
each cloud patch peak
6-Hour RAMS Estimated Rain using Cloud
Microphysics, with 4 Hours Delay
2000 UTC, 07/8/99
2100 UTC, 07/8/99
2200 UTC, 07/8/99
Tmax
Tmax
38 37 36 35
38 37 36 35
38 37 36 35
Tmin
Tmin
time
40

• Remotely sensed IR cloud-top and surface-air
  information are useful to estimate snowfall
  depth. Classification of observed, and estimated
  daily precipitation into snowfall depth and
  rainfall rate is necessary, if both snow and rain
  occurred.

20
Latitude (Degrees, North)
Cloud Mask Rainfall vs Observed Radar Rainfall
PERSIANN Rainfall vs Observed Radar Rainfall
Estimated Rainfall Rate (mm/h)
10
Estimating Hourly Snowfall Observations
2
0
117 116 115
114
117 116 115
114
117 116 115
114
Daily snowfall observations were distributed to
hourly snowfall with respect to cloud-top
brightness temperature, Tb , distribution,
0000 UTC, 07/8/99
0100 UTC, 07/8/99
2300 UTC, 07/8/99
38 37 36 35
38 37 36 35
38 37 36 35
Acknowledgement Financial support provided from
mainly NSF-SAHRA, NASA EOS, TRMM, Hydis, and
Raytheon projects.
40
20
Latitude (Degrees, North)
Estimated Rainfall Rate(mm/h)
10
2
where SNh and SNd are daily and hourly observed
snowfall, respectively, and hi is the hour
that, Ta lt 32 ?F and Cloud-Top Tb ? 245 K .
0
117 116 115
114
117 116 115
114
117 116 115
114
Longitude (Degrees, West)
Longitude (Degrees, West)
Longitude (Degrees, West)