Spatial prediction of gully erosion using ALOS PALSAR data and ensemble bivariate and data mining models
- Publication Type:
- Journal Article
- Geosciences Journal, 2019, 23 (4), pp. 669 - 686
- Issue Date:
|Arabameri2019_Article_SpatialPredictionOfGullyErosio_GeoSci.pdf||Published Version||23.95 MB|
|Geoscience MS_14Sept2018_Clean version.docx||Accepted Manuscript Version||81.18 kB|
Copyright Clearance Process
- Recently Added
- In Progress
- Closed Access
This item is closed access and not available.
© 2019, The Association of Korean Geoscience Societies and Springer. Remote sensing is recognized as a powerful and efficient tool that provides a comprehensive view of large areas that are difficult to access, and also reduces costs and shortens the timing of projects. The purpose of this study is to introduce effective parameters using remote sensing data and subsequently predict gully erosion using statistical models of Density Area (DA) and Information Value (IV), and data mining based Random Forest (RF) model and their ensemble. The aforementioned models were employed at the Tororud-Najarabad watershed in the northeastern part of Semnan province, Iran. For this purpose, at first using various resources, the map of the distribution of the gullies was prepared with the help of field visits and Google Earth images. In order to analyse the earth’s surface and extraction of topographic parameters, a digital elevation model derived from PALSAR (Phased Array type L-band Synthetic Aperture Radar) radar data with a resolution of 12.5 meters was used. Using literature review, expert opinion and multi-collinearity test, 15 environmental parameters were selected with a resolution of 12.5 meters for the modelling. Results of RF model indicate that parameters of NDVI (normalized difference vegetation index), elevation and land use respectively had the highest effect on the gully erosion. Several techniques such as area under curve (AUC), seed cell area index (SCAI), and Kappa coefficient were used for validation. Results of validation indicated that the combination of bivariate (IV and DA models) with the RF data-mining model has increased their performance. The prediction accuracy of AUC and Kappa values in DA, IV and RF are (0.745, 0.782, and 0.792) and (0.804, 0.852, and 0.860) and these values in ensemble models of DA-RF and IV-RF are (0.845, and 0.911) and (0.872, and 0.951) respectively. Results of SCAI show that ensemble models had a good performance, so that, with increasing of sensitivity, the values of SCAI have decreased. Based on results, determination of gullies and assessing the process of gullying through remote sensing technology in combination with field observations and accurate statistical and computer methods can be a suitable methodology for predicting areas with gully erosion potential.
Please use this identifier to cite or link to this item: