There is one active project within this sub-area:

Estimating Precipitation Over the Southwestern U.S. at 6hr/12km Resolution From Remotely Sensed Data
X. Gao, H. Gupta, Y. Hong, K. Hsu, S. Mahani, S. Sorooshian (UA-HWR)

Obtaining reliable quality of precipitation observations is important to the hydrologic simulation and weather forecasting. Many areas in the mountains of the southwest U.S. lack sufficient and effective means of measuring precipitation, and traditional observations from gauges provide limited and unreliable data. This research attempts to provide and improve precipitation observations over various seasons and scales using multiple sources of observations, including gauges, radar, and multiple-satellites. Our research goals are: 1) to develop techniques that will enable our estimation of rainfall at high spatial-temporal resolutions (6-hourly, 12 km x 12 km), and 2) to improve precipitation estimates and rain/snow classification during the winter seasons.

Providing high quality regional precipitation observations at 6-hour 12 km x 12 km is highly relevant to many SAHRA research activities. This project is in support of water balance study and the model simulation and validation of the SAHRA modeling group studies at both regional and basin scales. The results from this research will provide precipitation input for Thrust Areas 1, 2, and 3 and connect in close coordination with climate modeling in TA4.

Activities and Results

Three research activities are highlighted in this reporting period, and are listed below:

1. Cloud classification and rainfall estimation - A rainfall estimation algorithm based on a cloud classification scheme was developed. Other than using local pixel-based approaches, this cloud classification algorithm identifies various cloud types according to the cloud features extracted from cloud textures at several temperature threshold levels. Rainfall distributions of various cloud types are trained and assigned from the radar rainfall measurements. This procedure separated rainfall estimation from satellite infrared imagery into three stages including cloud segmentation, cloud classification from selected features, and rainfall estimation. The variable threshold approach was applied to separate infrared cloud images into various cloud patches. The fixed threshold used in the constant threshold approach was not able to separate cloud patches in detail. The variable threshold approach, however, gives significant improvement over the constant threshold approach in the segmentation of individual cloud patches. Features consist of threshold temperatures in various cloud heights, cloud texture, and shapes. Initial experimental results show that the rainfall estimates from the cloud classification algorithm provide a tool for fine scale rainfall estimation at a 4 km resolution. The current plan is to implement, test, and include this rain algorithm during the monsoon season rainfall period in the coming reporting year.



2. Winter season snow estimation - Ground/air temperature (every 3 hours from Eta model) and surface elevation (DEM) were used for distinguishing snowfall from rainfall and estimating snowfall depth from Snow Water Equivalent (SWE) where the temperature was less than 2°C. The investigation was applied to the daily snowfall estimates at 0.25 x 0.25 lat/long resolution over the southwest U.S. The model was trained by TRMM satellite-TMI and estimated PERSIANN-SWE was validated using daily SNOTEL-SWE from Natural Resources Conservation Service (NRCS) data source. Daily snow estimates using PERSIANN precipitation and surface temperature from Eta model were produced. Investigation of SWE in one test period, December 2001, showed high correlations between SNOTEL observation and PERSIANN SWE estimates.



3. PERSIANN product evaluation - The PERSIANN global-product was compared with TRMM-satellite (3B43) and GPCP global products at a 1°x1° resolution on a monthly and daily basis. The comparison was applied globally from 40°S to 40°N for a 22-month period from March 2000 to December 2001. Over landmass, PERSIANN estimates are observed to be higher estimates than those estimates of GPCP and TRMM 3B43, during the high rainfall period; however, all three estimates are similar over the oceans. Further evaluation of rain estimates over the southwest is continuing and will be reported in the next reporting period. Evaluation of PERSIANN estimates in higher spatial-temporal scale using local gauge and radar measurements are ongoing and will be reported in the next period.

Plans for the Next Reporting Period

Many regions over the southwest US are not well covered by rain gauges and radar. Observation of precipitation is very difficult and limited. This research will be intensified. Our observations and understanding of precipitation distribution in various seasons and scales from multiple sources of information are providing reliable rainfall/snow estimates over the southwest U.S. In the past three years, we have developed techniques integrating satellites and ground-based measurements to improve our ability of estimating precipitation. In years 4 and 5 we plan to conduct research and data service as follows:

· Develop an algorithm providing high-resolution rainfall estimation at 6-hour, 12 km resolution
Develop/improve rainfall algorithm (cloud classification approach) and select effective satellite cloud features in the classification of cloud type and rain distribution estimation. Currently, a limited selection of satellite cloud features was selected in the cloud type classification and rainfall estimation. We plan to evaluate additional static and dynamic features of the rain cloud image to quantify rain cloud types and rainfall intensity.

· Generate rainfall estimates at 6-hour, 12 km x12 km resolution
Current PERSIANN estimates provide 6-hour, 25 km rainfall. Down scaling of PERSIANN rainfall to the 12 km resolution will be useful to hydrologic applications at the basin scale. Rainfall generated from the cloud classification map is at an hourly 4 km satellite infrared pixel resolution. However, at a 4 km level, the rainfall estimates from satellite imagery do not provide good pixel-to-pixel correspondence, resulting in pixel displacement errors and necessitating geo-location adjustment. Two adjustments are considered: 1) the up-scaling of hourly 4 km to 6-hour 12 km rainfall, which will reduce the pixel miss-location error, and 2) applying adjustment displacement error of images at the pixel level. Six-hour 12 km x 12 km rain product will be available for areas covering the regional (southwest U.S.) to basin (Colorado river, San Pedro, and Rio Grande basins) scales. This precipitation data set is useful to those SAHRA research activities that use precipitation forcing to drive hydrological models.

· Validate rainfall products at basin scale level using gauge and radar measurement
A validation program will be carried out with the selection of experimental sites over the Colorado and Rio Grande river basins. Several sites at a 12 km x 12 km scale with a large density of gauge networks and radar coverage will be selected. Rainfall products covering various spatial-temporal scales at 4 km, 12 km, 24 km and 3-hour, 6-hour, 12-hour and daily accumulation will be evaluated. Bias of rain estimates from the satellite-based algorithms will be quantified. Bias removal of satellite and radar estimates at designed spatial and temporal will be quantified and applied to those regions in which gauge and radar coverage is limited.

· Integrate multiple sources of information in rainfall estimation
Possible observation errors from gauges, radar, and satellite measurements will be quantified. A merged product consisting of satellite, gauge, and radar rainfall measurements will be generated. This product will consider bias of satellite estimates and uncertainty of gauge measurement with respect to the density of gauges, and radar bias corrections from gauge measurement.

· Estimate snow water equivalent in the wintertime
Daily PERSIANN-SWE will be compared to daily-observed SNOTEL-SWE in a different approach from the method that was applied in the 3rd year. Specifically, hourly-based surface/air temperature from the RUC data source will be used for partitioning snowfall from rainfall by applying a 2°C temperature threshold (snow level). Different Indices will be estimated for snow, rain, or a mixture consisting of snow and rain in relation to the ratio of the precipitation event length and time of the event that temperature is less than the threshold for a day. These indices will be used to select a day as a snow, rain, or mixed day. Daily SWE estimates will be validated using SNOTEL-SWE. A time series investigation will be carried out for all storms occurring in 2.5 months (from February 05, 2002) over different 1° x °1 regions in the southwest U.S.