Automatic rainfall recharge model induction by evolutionary computational intelligence

Water Resources Research

2008

1] Uncertainty analysis is starting to be widely acknowledged as an integral part of hydrological modeling. The conventional treatment of uncertainty analysis in hydrologic modeling is to assume a deterministic model structure, and treat its associated parameters as imperfectly known, thereby neglecting the uncertainty associated with the model structure. In this paper, a modeling framework that can explicitly account for the effect of model structure uncertainty has been proposed. The modeling framework is based on initially generating different realizations of the original data set using a non-parametric bootstrap method, and then exploiting the ability of the self-organizing algorithms, namely genetic programming, to evolve their own model structure for each of the resampled data sets. The resulting ensemble of models is then used to quantify the uncertainty associated with the model structure. The performance of the proposed modeling framework is analyzed with regards to its ability in characterizing the evapotranspiration process at the Southwest Sand Storage facility, located near Ft. McMurray, Alberta. Eddy-covariancemeasured actual evapotranspiration is modeled as a function of net radiation, air temperature, ground temperature, relative humidity, and wind speed. Investigating the relation between model complexity, prediction accuracy, and uncertainty, two sets of experiments were carried out by varying the level of mathematical operators that can be used to define the predictand-predictor relationship. While the first set uses just the additive operators, the second set uses both the additive and the multiplicative operators to define the predictand-predictor relationship. The results suggest that increasing the model complexity may lead to better prediction accuracy but at an expense of increasing uncertainty. Compared to the model parameter uncertainty, the relative contribution of model structure uncertainty to the predictive uncertainty of a model is shown to be more important. Furthermore, the study advocates that the search to find the optimal model could be replaced by the quest to unearth possible models for characterizing hydrological processes.Citation: Parasuraman, K., and A. Elshorbagy (2008), Toward improving the reliability of hydrologic prediction: Model structure uncertainty and its quantification using ensemble-based genetic programming framework, Water Resour. Res.,

Section: Genetic Programmingmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Toward improving the reliability of hydrologic prediction: Model structure uncertainty and its quantification using ensemble‐based genetic programming framework

Water Resources Research

2008

“…Moreover, the results indicate that it is diffi cult, if not impossible, to achieve better prediction and less uncertainty simultaneously. Abbott, 1997), rainfall-runoff processes (Whigham and Crapper, 2001;Savic et al, 1999), runoff forecasting (Khu et al, 2001), urban fractured-rock aquifer dynamics (Hong and Rosen, 2002), temperature downscaling (Coulibaly, 2004), the rainfall-recharge process (Hong et al, 2005), soil moisture (Makkeasorn et al, 2006), and evapotranspiration (Parasuraman et al, 2007).…”

mentioning

confidence: 99%

Estimating Saturated Hydraulic Conductivity Using Genetic Programming

Soil Science Soc of Amer J

Cheng

2007

Saturated hydraulic conductivity (Ks) is one of the key parameters in modeling solute and water movement in the vadose zone. Field and laboratory measurement of Ks is time consuming, and hence is not practical for characterizing the large spatial and temporal variability of Ks As an alternative to direct measurements, pedotransfer functions (PTFs), which estimate Ks from readily available soil data, are being widely adopted. This study explores the utility of a promising data‐driven method, namely, genetic programming (GP), to develop PTFs for estimating Ks from sand, silt, and clay contents and bulk density (Db). A data set from the Unsaturated Soil Hydraulic Database (UNSODA) was considered in this study. The performance of the GP models were compared with the neural networks (NNs) model, as it is the most widely adopted method for developing PTFs. The uncertainty of the PTFs was evaluated by combining the GP and the NN models, using the nonparametric bootstrap method. Results from the study indicate that GP appears to be a promising tool for developing PTFs for estimating Ks The better performance of the GP model may be attributed to the ability of GP to optimize both the model structure and its parameters in unison. For the PTFs developed using GP, the uncertainty due to model structure is shown to be more than the uncertainty due to model parameters. Moreover, the results indicate that it is difficult, if not impossible, to achieve better prediction and less uncertainty simultaneously.

“…Hence, compared to other regression techniques, it is not required to choose the model structure a priori. In water-related studies, GP has been applied to model: flow over a flexible bed (Babovic & Abbott, 1997), the rainfall-runoff process (Savic et al, 1999;Whigham & Crapper, 1999), runoff forecasting (Khu et al, 2001), urban fracturedrock aquifer dynamics (Hong & Rosen, 2002), temperature downscaling (Coulibaly, 2004), and the rainfall-recharge process (Hong et al, 2005).…”

Section: Introductionmentioning

confidence: 99%

Modelling the dynamics of the evapotranspiration process using genetic programming

Hydrological Sciences Journal

Carey

2007

125

Evapotranspiration constitutes one of the major components of the hydrological cycle and hence its accurate estimation is of vital importance to assess water availability and requirements. This study explores the utility of genetic programming (GP) to model the evapotranspiration process. An important characteristic of GP is that both the model structure and coefficients are simultaneously optimized. The method is applied in modelling eddy-covariance (EC)-measured latent heat (LE) as a function of net radiation (NR), ground temperature (GT), air temperature (AT), wind speed (WS) and relative humidity (RH). Two case studies having different climatic and topographic conditions are considered. The performance of the GP model is compared with artificial neural network (ANN) models and the traditional Penman-Monteith (PM) method. Results from the study indicate that both the datadriven models, GP and ANNs, performed better than the PM method. However, performance of the GP model is comparable with that of the ANN model. The GP-evolved models are dominated by NR and GT, indicating that these two inputs can represent most of the variance in LE. The results show that the GP-evolved equations are parsimonious and understandable, and are well suited to modelling the dynamics of the evapotranspiration process.Key words eddy-covariance; evapotranspiration; modelling; genetic programming; artificial neural networks; Penman-Monteith method Modélisation de la dynamique du processus évapotranspiratoire par programmation génétique Résumé L'évapotranspiration est l'une des principales composantes du cycle hydrologique et son estimation précise est par conséquent d'une importance vitale pour estimer la disponibilité et les besoins en eau. Cette étude explore l'utilité de la programmation génétique (PG) pour modéliser le processus évapotranspiratoire. Une caractéristique importante de la PG est que la structure du modèle et les coefficients sont optimisés simultanément. La méthode est appliquée à la modélisation de la chaleur latente (CL) mesurée par la méthode des covariances turbulentes (CT) comme une fonction du rayonnement net (RN), de la température du sol (TS), de la température de l'air (TA), de la vitesse du vent (VV) et de l'humidité relative (HR). Deux études de cas sont considérées, qui présentent des conditions climatiques et topographiques différentes. La performance de la modélisation PG est comparée à celles de modèles de type réseaux de neurones artificiels (RNA) et de la méthode traditionnelle de Pennman-Monteith (PM). L'étude montre que les modélisations forcées par les données, PG et RNA, donnent de meilleurs résultats que la méthode PM. Cependant, la performance de la modélisation PG est comparable à celle de la modélisation RNA. Les modélisations PG évoluées sont dominées par RN et TS, ce qui indique que ces deux variables d'entrée peuvent représenter la plupart de la variance de CL. Les résultats montrent que les équations PG évoluées sont parcimonieuses et intelligibles, et sont bien adaptées à la mo...