AN ADVANCED NRCS-CN MODEL FOR RUNOFF ESTIMATION USING GIS AND REMOTE SENSING

Abstract

Selecting the suitable method to estimate rainfall generated surface runoff in a specific watershed has always been a challenge for hydrologists. The Natural Resources Conversion Service-Curve Number (NRCS-CN) is one of the broadly accepted methods for runoff estimation since its inception for the reason that it can be applied easily on gauged and ungauged type of watersheds. Though the method has been applied around the globe for modeling hydrological processes, and even to the fields, it was originally not developed for. In many cases, the existing NRCS-CN method gives unsatisfactory performance due to lack of sound procedure for soil moisture accounting (SMA) in the model. In this study, an advanced NRCS-CN model was developed by incorporating soil moisture accounting (SMA) in the existing NRCS-CN model to pass up the problem of abrupt jump in runoff estimation with retaining the inherited simplicity. The performance of the proposed model was compared with the existing models of Ajmal et al. (2015), Woodward et al. (2003) along with the existing NRCS-CN model using a large data set of USDA-ARS data base and one Indian watershed. A total of 12022 event rainfall-runoff data from 48 USA watersheds and 9 events of Kalu watershed were used in this work. The goodness-of-fit of the models was compared using root mean square error (RMSE), Nash and Sutcliffe efficiency (NSE) and percent bias (PBIAS) along with Rank Grading System (RGS). The proposed model performs much better than the models of Ajmal et al., Woodward et al. and the existing NRCS-CN model in most of the applications. The mean values of RMSE and NSE were found to be 5.00, 5.39, 5.60, 6.40 and 66.30%, 60.80%, 55.56%, 46.65% respectively, for models M1-M4. Based on PBIAS, the proposed model was found to be ‘very good’ in 39 watersheds, ‘good’ in 1 watershed, ‘fair’ in 6 watershed and unsatisfactory in 2 watersheds. The Ajmal et al. model was found ‘very good’ in 28 watersheds, ‘good’ in 6 watersheds, ‘fair’ in 8 watersheds and unsatisfactory in 6 watersheds. The performance of Woodward model was found ‘very good’ in 22 watersheds, ‘good’ in 3 watersheds, ‘fair’ in 7 watersheds and unsatisfactory in 16 watersheds. The performance of the existing NRCS- CN model was found to be ‘very good’ in 16 watersheds, ‘good’ in 4 watersheds, ‘fair’ in 7 watersheds and unsatisfactory in 21 iv watersheds. However, the existing NRCS-CN model was found to overestimate the computed runoff when it was coupled with GIS and remote sensing. Based on Rank Grading System (RGS), the proposed advanced NRCS-CN model scores highest mark (183; Rank I) followed by Ajmal et al. model with 151( Rank II), Woodward et al. model with 100 (Rank III) and the existing NRCS-CN model with 48 mark (Rank IV) out of the maximum 192. The value of initial abstraction coefficient (λ) was found to vary from 0.001 to 0.26 with a mean of 0.06 during application to 12022 events from 48 watersheds. A similar value of λ was also suggested by Hawkins et al. (2001) who reported that λ = 0.05 is gives better fit to data and is more proper for use in runoff calculations. This suggests that λ cannot always be taken as a fixed amount of total rainfall.