Separate and Coupled Testing of Meteorological and Hydrological Forecast Models for the Susquehanna River Basin in Pennsylvania

Thomas T. Warner Department of Meteorology and Earth System Science Center, The Pennsylvania State University, University Park, Pennsylvania

Search for other papers by Thomas T. Warner in
Current site
Google Scholar
PubMed
Close
,
David F. Kibler Department of Civil Engineering, The Pennsylvania State University, University Park, Pennsylvania

Search for other papers by David F. Kibler in
Current site
Google Scholar
PubMed
Close
, and
Richard L. Steinhart Department of Civil Engineering, The Pennsylvania State University, University Park, Pennsylvania

Search for other papers by Richard L. Steinhart in
Current site
Google Scholar
PubMed
Close
Restricted access

Abstract

Forecasts of river flow are traditionally performed using observed rainfall as input to a model of the surface hydrology. However, this sometimes allows for only a short forecast lead time, especially for small watersheds. One procedure by which this lead time can be increased involves the use of a meteorological model to produce forecasts of the rainfall. As a test of this approach, the Pennsylvania State University–National Center for Atmospheric Research mesoscale meteorological model was used to produce ten 72-h precipitation forecasts for the Susquehanna River basin (SRB) in New York, Pennsylvania, and Maryland. These forecasts were evaluated in two ways: 1) the observed and predicted hourly rainfall, averaged over the SRB, was compared, and 2) the forecast rainfall was used as input to a river-flow model, and the forecast river discharge was compared with the observed discharge. The ten study periods, from both winter and summer, each included a significant precipitation event over the SRB. In every case, the meteorological model predicted the precipitation-producing event. The average magnitudes of the error in the start time and the end time of the precipitation events were both less than 6 h, and the magnitude of the error in the duration of the precipitation events was ∼ 5 h. This represents approximately 15% of the average event duration of 33 h. The model overpredicts the precipitation volume in the SRB by an average of 61%, and the time of maximum precipitation is in error by ∼ 7 h. Because the area of the SRB represents less than 1% of the area of the forecast domain of the meteorological model, these error statistics for the SRB represent essentially a “point” verification of the model.

Two of the ten study periods were selected for use in the testing of the coupled meteorological-river-flow modeling system. The river-flow model, developed by the U.S. Army Corps of Engineers, is a commonly used system known as HEC-1. In this application, the upper SRB was divided into 25 subareas. For each of the two rainfall events, the observed and model-predicted rainfall in each of the subareas were used separately as input to HEC-1, and discharge forecasts were generated for the Susquehanna River at Sunbury, Pennsylvania. When the parameters used in HEC-1 reflected an average partitioning of rainfall between runoff and infiltration, the discharge forecasts based on observed and predicted rainfall for the two cases both showed a reasonably accurate time of peak discharge at Sunbury. However, the discharge was overpredicted by the forecast rainfall and underpredicted by the observed rainfall. When, in one case, the HEC-1 parameters were modified to reflect the existence of snow on the ground, reasonable peak discharges were predicted using the observed rainfall and the model-predicted rainfall that had been normalized for the bias error in the prediction.

Abstract

Forecasts of river flow are traditionally performed using observed rainfall as input to a model of the surface hydrology. However, this sometimes allows for only a short forecast lead time, especially for small watersheds. One procedure by which this lead time can be increased involves the use of a meteorological model to produce forecasts of the rainfall. As a test of this approach, the Pennsylvania State University–National Center for Atmospheric Research mesoscale meteorological model was used to produce ten 72-h precipitation forecasts for the Susquehanna River basin (SRB) in New York, Pennsylvania, and Maryland. These forecasts were evaluated in two ways: 1) the observed and predicted hourly rainfall, averaged over the SRB, was compared, and 2) the forecast rainfall was used as input to a river-flow model, and the forecast river discharge was compared with the observed discharge. The ten study periods, from both winter and summer, each included a significant precipitation event over the SRB. In every case, the meteorological model predicted the precipitation-producing event. The average magnitudes of the error in the start time and the end time of the precipitation events were both less than 6 h, and the magnitude of the error in the duration of the precipitation events was ∼ 5 h. This represents approximately 15% of the average event duration of 33 h. The model overpredicts the precipitation volume in the SRB by an average of 61%, and the time of maximum precipitation is in error by ∼ 7 h. Because the area of the SRB represents less than 1% of the area of the forecast domain of the meteorological model, these error statistics for the SRB represent essentially a “point” verification of the model.

Two of the ten study periods were selected for use in the testing of the coupled meteorological-river-flow modeling system. The river-flow model, developed by the U.S. Army Corps of Engineers, is a commonly used system known as HEC-1. In this application, the upper SRB was divided into 25 subareas. For each of the two rainfall events, the observed and model-predicted rainfall in each of the subareas were used separately as input to HEC-1, and discharge forecasts were generated for the Susquehanna River at Sunbury, Pennsylvania. When the parameters used in HEC-1 reflected an average partitioning of rainfall between runoff and infiltration, the discharge forecasts based on observed and predicted rainfall for the two cases both showed a reasonably accurate time of peak discharge at Sunbury. However, the discharge was overpredicted by the forecast rainfall and underpredicted by the observed rainfall. When, in one case, the HEC-1 parameters were modified to reflect the existence of snow on the ground, reasonable peak discharges were predicted using the observed rainfall and the model-predicted rainfall that had been normalized for the bias error in the prediction.

Save