CJEES
IMPACT FACTOR
0.9
CITESCORE
2.3

« Back

ARTICLE IN Volume 15, 2020 - Number 2

COMPARISON OF RUSLE AND SUPERVISED CLASSIFICATION ALGORITHMS FOR IDENTIFYING EROSION-PRONE AREAS IN A MOUNTAINOUS RURAL LANDSCAPE



Kwanele PHINZI1*, Njoya Silas NGETAR2, Osadolor EBHUOMA3 & Szilárd SZABÓ4
1Doctoral School of Earth Sciences, Department of Physical Geography and Geoinformatics, University of Debrecen, Egyetem tér 1., Debrecen H-4032, Hungary, e-mail: phinzi.kwanele@science.unideb.hu
2School of Agricultural, Earth and Environmental Sciences, University of KwaZulu-Natal, Howard College Campus, Durban 4041, South Africa, e-mail: njoya@ukzn.ac.za
3School of Agricultural, Earth and Environmental Sciences, University of KwaZulu-Natal, Westville Campus, Durban 4000, South Africa, e-mail: osadolorebhuoma@gmail.com
4Department of Physical Geography and Geoinformatics, University of Debrecen, Egyetem tér 1., Debrecen H-4032, Hungary, e-mail: Szabo.szilard@science.unideb.hu


Full text Downloads: 3044

Abstract

The identification of erosion prone areas with reasonably high accuracy is a prerequisite for formulating relevant soil conservation measures especially in rural areas where there is much reliance on subsistence agriculture. The aim of this paper was to compare and exploit the complementary advantage of fusing three independent methods including the Revised Universal Soil Loss Equation (RUSLE) and two supervised image classification algorithms: Random Forest (RF) and Maximum Likelihood (ML). All analyses were conducted using a GIS proprietary software, ArcGIS. The results indicated that RF was the best with the highest overall accuracy (OA), producer’s accuracy (PA), and user’s accuracy (UA) of 87%, 78%, and 95%, respectively. RUSLE poorly performed relative to other methods, scoring the lowest PA (34%) and OA (66%), but slightly outperformed ML in terms of UA. From the user’s perspective, the performance of individual methods was satisfactory with each method achieving an UA of greater than 90% although ML and RUSLE were not satisfactory from the producer’s perspective, recording respective PAs of 56% and 34%. When the results from individual methods were fused, the accuracy increased above 90% across all accuracy indices, which is far above the 85% acceptable level for planning and management purposes.

Keywords:

  • RUSLE
  • supervised
  • classification
  • random
  • forest
  • maximum
  • likelihood
  • soil
  • erosion

How to cite

COMPARISON OF RUSLE AND SUPERVISED CLASSIFICATION ALGORITHMS FOR IDENTIFYING EROSION-PRONE AREAS IN A MOUNTAINOUS RURAL LANDSCAPE, Carpathian Journal of Earth and Environmental Sciences, August 2020, Vol. 15, No. 2, p. 405 – 413; Doi:10.26471/cjees/2020/015/140

[Google Scholar]

©2006—2024 Published by Carpathian Association of Environment and Earth Sciences
For the best experience, please use Google Chrome or Firefox browsers.