**4. Conclusions**

In this study, we extended the previous study and applied six machine learning algorithms: NB, RPART, NNET, SVM, RF and GBM to evaluate the surface water extraction using a Landsat 8 OLI images in Nepal. Using the previous reference dataset and Landsat scene, six different models were developed using the CARET package in R software. Cross-validation was completed to minimize the overfitting then train the model to predict the surface water and validate the full reference dataset. With three secondary bands: Slope, NDVI and NDWI, the algorithms were evaluated for performance with the addition of extra information. The results were compared, case by case, and the following conclusions were drawn from the test scene and applied machine learning algorithms:


It seems that machine learning methods could be very useful for the accurate automated binary classification of surface water in Nepal. The use of RF with original LS8 data or with the addition of the slope or NDWI with another algorithm can be undertaken. Based on this and previous work [13], it is recommended to segment the study area with and without snow or low and high elevation, then apply RF or GBM for better performance.

For further investigation, this study aimed to evaluate the application of convolutional neural networks or deep learning for better accuracy. In addition, individual original bands and secondary bands with the RF and GBM can be evaluated so that high accuracy can be achieved with minimum bands.

**Author Contributions:** Conceptualization, T.D.A.; formal analysis, T.D.A.; investigation, T.D.A. and A.S.; methodology, T.D.A.; resources, D.H.L.; software, T.D.A.; supervision, D.H.L.; validation, A.S.; visualization, A.S.; writing—original draft, T.D.A.; writing—review and editing, A.S. and D.H.L.

**Funding:** This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. 2018R1A2B6009363).

**Acknowledgments:** The authors are grateful to the U.S. Geological Survey server (http://glovis.usgs.gov) for providing the Landsat data that was used in this manuscript freely. The authors would also like to thank the anonymous reviewers for their constructive comments and improving this manuscript.

**Conflicts of Interest:** The authors declare no conflicts of interest.
