Abstract Himalayas are ecologically fragile, and unplanned exploitation of natural resources is severely affecting water flow regimes in the mountainous watersheds. Therefore, it is imperative to quantify the effect of different environmental and morphological factors on flow behavior in the micro-watersheds to efficiently plan and execute water management practices in a sustainable manner. In this study, three hilly micro-watersheds were gauged in Uttaranchal State of India to assess the impact of morphological characteristics and land uses on surface runoff, base flow and total flow. Artificial intelligence (AI) models based on the multivariate adaptive regression splines (MARS) technique were employed to predict surface runoff, base flow, and total flow as affected by rainfall and morphological features of the micro-watersheds. Daily rainfall, runoff, base flow, and total flow data recorded from July 1, 2001 to June 30, 2003 in the three watersheds, were used to develop and validate MARS models. The average correlation coefficients between the observed and predicted runoff, base flow, and total flow for the unseen test datasets were 0.573, 0.884, and 0.881, respectively. The corresponding average deviations were −0.113, −0.02, and −0.04 mm, and the average absolute deviations were 0.171, 0.187, and 0.267 mm, respectively. Thus, the analysis revealed that base flows and total flows, as predicted by MARS, were in close agreement with the observed values while the surface runoff predictions were reasonable at best. MARS analysis determined that 5-day antecedent precipitation index (API5), rainfall, day of the year, runoff estimated by using curve number method, and watershed area are the most important variables for simulating runoff in hilly watersheds. Soil cover and watershed geometry parameters also affected runoff generation which are indirectly covered in the estimation of runoff by curve number method. In order to explore applicability of MARS models on ungauged watersheds, data from two watersheds were used to develop MARS models, and tested on the third watershed. The observed and predicted values of flows were found to be in a reasonably good agreement. The correlation coefficients for the unseen test datasets varied from 0.391 to 0.648 for surface runoff, 0.736 to 0.879 for base flow, and 0.789 to 0.886 for total flow. The prediction for surface runoff can improve further if more data on surface flow events are available. Therefore, it is concluded that MARS models have the potential to simulate runoff in hilly areas and can be applied satisfactorily to ungauged watersheds under identical agro-climatic situations.
[1]
Hiromitsu Saegusa,et al.
Runoff analysis in humid forest catchment with artificial neural network
,
2000
.
[2]
Chun-Chieh Yang,et al.
A multivariate adaptive regression splines model for simulation of pesticide transport in soils
,
2003
.
[3]
Sholom M. Weiss,et al.
Computer Systems That Learn
,
1990
.
[4]
J. Arnold,et al.
SWAT2000: current capabilities and research opportunities in applied watershed modelling
,
2005
.
[5]
A. Tokar,et al.
Rainfall-Runoff Modeling Using Artificial Neural Networks
,
1999
.
[6]
Jerome H. Friedman.
Multivariate adaptive regression splines (with discussion)
,
1991
.
[7]
Peggy A. Johnson,et al.
Stream hydrological and ecological responses to climate change assessed with an artificial neural network
,
1996
.
[8]
Ajith Abraham,et al.
MARS: Still an Alien Planet in Soft Computing?
,
2001,
International Conference on Computational Science.
[9]
M. Castellano-Méndez,et al.
Modelling of the monthly and daily behaviour of the runoff of the Xallas river using Box-Jenkins and neural networks methods
,
2004
.
[10]
Will Dwinnell.
Exploring MARS: an alternative to neural networks
,
2000
.
[11]
Ashu Jain,et al.
Comparative Analysis of Event-Based Rainfall-Runoff Modeling Techniques—Deterministic, Statistical, and Artificial Neural Networks
,
2003
.
[12]
J. Friedman.
Multivariate adaptive regression splines
,
1990
.
[13]
A. Sharma,et al.
Problems and prospects of natural resource management in Indian Himalayas - a base paper.
,
1999
.
[14]
Jason Smith,et al.
Neural-Network Models of Rainfall-Runoff Process
,
1995
.
[15]
Peter S. Sephton,et al.
Forecasting recessions: can we do better on MARS?
,
2001
.
[16]
Chun-Chieh Yang,et al.
APPLICATION OF MULTIVARIATE ADAPTIVE REGRESSION SPLINES (MARS) TO SIMULATE SOIL TEMPERATURE
,
2004
.