Document Type : Research Paper
Authors
1
Cotton Research Institute of Iran, Agricultural Research, Education and Extension Organization, (AREEO) Gorgan, Iran
2
Headquarters, Department of Water Engineering, Faculty of Water and Soil Engineering, Gorgan University of Agricultural Sciences and Natural Resources, Gorgan, Iran
3
PhD student in Water Science and Engineering - Irrigation and Drainage, Gorgan University of Agricultural Sciences and Natural Resources, Gorgan
4
Agricultural Research, Education and Extension Organization (AREEO), Gorgan, Iran
5
Postdoctoral Researcher, Department of Water Engineering, Faculty of Water and Soil Engineering, Gorgan University of Agricultural Sciences and Natural Resources, Gorgan
6
Associate Professor, Department of Desert Areas Management, Faculty of Rangeland and Watershed Management, Gorgan University of Agricultural Sciences and Natural Resources, Gorgan, Iran
7
Cotton Research Institute of Iran (CRII), Agricultural Research, Education and Extension Organization (AREEO), Gorgan, Iran
10.22092/ijcr.2026.372160.1253
Abstract
Background and Objective: Accurate monitoring of crop cultivation patterns and crop separation at a regional scale is considered to be one of the fundamental requirements of sustainable water resources management and efficient agricultural planning. This study aims to evaluate and compare the performance of the C5 decision tree classification model in crop identification, with a particular focus on cotton, using Landsat 8 and 9 satellite imagery over the agricultural lands of Gorgan, Aliabad, and Aqqala counties in Golestan Province. The tree-based model used in this research was implemented stepwise, effectively reducing the greedy nature of the decision tree algorithm and identifying the most influential variables within the model.
Materials and Methods: For this purpose, three time periods of August 6, August 31, and September 23 were selected to determine the most appropriate phonological stage of the growing season to achieve maximum cotton separation accuracy in multispectral data. The classification process was performed using the C5 algorithm as a rule-based model with automatic feature selection capability. A set of vegetation and moisture indices including NDVI, EVI, GNDVI, SAVI, NDWI and NMDI along with visible, near-infrared (NIR) and short-wave infrared (SWIR) spectral bands were used as input datasets.
Results: The comparative results of the error matrix showed that the date of September 23 provided the highest overall crop classification accuracy with a value of 88.10%; while the dates of August 6 and August 31 recorded accuracies of 83.6% and 82.3%, respectively. Also, the cotton crop was identified more successfully on the date of September 23 with an accuracy of 86.8% compared to other dates. The coincidence of this date with the full maturity stage of the cotton plant, which is accompanied by a decrease in chlorophyll, opening of bolls, a change in canopy structure and an increase in reflectance in the red and SWIR bands, creates a distinctive spectral signature of cotton compared to other crops.
Conclusion: These features have increased spectral separability and consequently improved the performance of the classification model at this time point. Accordingly, choosing the 23rd of September as the reference date and using the C5 model at this phonological stage can significantly improve the accuracy of identifying and mapping cotton fields in agricultural monitoring studies.
Keywords
Main Subjects
)