A new algorithm is proposed for the identification of three-layer feedforward artificial neural networks. The algorithm, entitled LLSSIM, partitions the weight space into two major groups: the input- hidden and hidden -output weights. The input- hidden weights are trained using a multi -start SIMPLEX algorithm and the hidden -output weights are identified using a conditional linear- least- square estimation approach. Architectural design is accomplished by progressive addition of nodes to the hidden layer. The LLSSIM approach provides globally superior weight estimates with fewer function evaluations than the conventional back propagation (BPA) and adaptive back propagation (ABPA) strategies. Monte -carlo testing on the XOR problem, two function approximation problems, and a rainfall- runoff modeling problem show LLSSIM to be more effective, efficient and stable than BPA and ABPA.

This report presents a new multiple objective optimization algorithm that is capable of solving for the entire Pareto set in one single optimization run. The multi-objective complex evolution (MOCOM-UA) procedure is based on the following three concepts: (1) population, (2) rank-based selection, and (3) competitive evolution. In the MOCOM-UA algorithm, a population of candidate solutions is evolved in the feasible space to search for the Pareto set. Ranking of the population is accomplished through Pareto ranking, where all points are successively placed on different Pareto fronts. Competitive evolution consists of selecting subsets of points (including all worst points in the population) based on their ranks and moving the worst points toward the Pareto set using the newly developed multi-objective simplex (MOSIM) procedure. Test analysis on the MOCOM-UA algorithm is accomplished on mathematical problems of increasing complexity and based on a bi-criterion measure of performance. The two performance criteria are: (1) efficiency, as measured by the ability of the algorithm to converge quickly, and (2) effectiveness, as measured by the ability of the algorithm to locate the Pareto set. Comparison of the MOCOM-UA algorithm against three multi-objective genetic algorithms (MOGAs) favors the former. In a realistic application, the MOCOM-UA algorithm is used to calibrate the Soil Moisture Accounting model of the National Weather Service River Forecasting Systems (NWSRFS-SMA). Multi-objective calibration of this model is accomplished using two bi-criterion objective functions, namely the Daily Root Mean Square-Heteroscedastic Maximum Likelihood Estimator (DRMS-HMLE) and rising limb /falling limb (RISE/FALL) objective functions. These two multi-objective calibrations provide some interesting insights into the influence of different objectives in the location of final parameter values, as well as limitations in the structure of the NWSRFS-SMA model.

Runoff generation has been shown to be very sensitive to precipitation input. With the use of precipitation estimates from weather radar, errors are introduced from both the transformation from reflectivity to precipitation rate and the spatial and temporal aggregation of the radar product. Currently, a significant degree of uncertainty exists in the accuracy of radar-based precipitation estimates. When uncalibrated or poorly calibrated radar products were used as input to a rainfall-runoff model, the resulting predicted runoff varied severely from the runoff generated using well-calibrated radar products. Another source of uncertainty, errors in the precipitation system structure due to aggregation in time and space, has also been shown to affect runoff generation. This study focuses on separating the primary runoff- generating mechanisms -- infiltration excess and saturation excess -- to assess their responses to variable precipitation inputs individually. For the case of saturation excess runoff, there was minimal sensitivity due to temporal aggregation of the precipitation; however, there was considerable sensitivity to spatial aggregation. For the case of infiltration excess runoff, temporal and spatial aggregation of the precipitation significantly reduced the amount of runoff produced. The magnitudes of these runoff reductions varied between storms and showed a high degree of dependence on storm characteristics, particularly the maximum precipitation intensity.

Infrared (IR) imagery collected by geostationary satellites provides useful information about the dirunal evolution of cloud systems. These IR images can be analyzed to indicate the location of clouds as well as the pattern of cloud top temperatures (Tbs). During the past several decades, a number of different approaches for estimation of rainfall rate (RR) from Tb have been explored and concluded that the Tb-RR relationship is (1) highly nonlinear, and (2) seasonally and regionally dependent. Therefore, to properly model the relationship, the model must be able to: (1) detect and identify a non-linear mapping of the Tb-RR relationship; (2) Incorporate information about various cloud properties extracted from IR image; (3) Use feedback obtained from RR observations to adaptively adjust to seasonal and regional variations; and (4) Effectively and efficiently process large amounts of satellite image data in real -time. In this study, a kind of artificial neural network (ANN), called Modified Counter Propagation Network (MCPN), that incorporates these features, has been developed. The model was calibrated using the data around the Japanese Islands provided by the Global Precipitation Climatology Project (GPCP) First Algorithm Intercomparison Project (AIP-I). Validation results over the Japanese Islands and Florida peninsula show that by providing limited ground-truth observation, the MCPN model is effective in monthly and hourly rainfall estimation. Comparison of results from MCPN model and GOES Precipitation Index (GPI) approach is also provided in the study.