Employing sentinel-2 time-series and noisy data quality control enhance crop classification in arid environments: A comparison of machine learning and deep learning methods

Abstract

Accurate and timely mapping of agricultural products is a crucial component in management and decision-making for promoting food security and sustainable development. The intricacy of differentiating diverse croplands due to the existence of small and winding agricultural fragments contributes to the complexity of crop classification in arid environments. In this study, we employed a novel hybrid approach, integrating time-series analysis, noisy data quality control, and different machine learning and deep learning models to classify croplands of complex multi-crop systems in central Iran. The classification was based on time series spectral bands of Sentinel-2 images and indices of crop growth phenology, providing valuable insights into the growth cycles of different crops in the region. Additionally, a neural network-based method was used to assess and enhance the quality of training data before modeling. For crop classification, we used four machine learning and deep learning methods including Extreme Gradient Boosting (XGBoost), Random Forest (RF), Support Vector Machine (SVM), and Temporal Convolutional Neural Network (TCNN), and compared their results before and after quality control measures. The results indicated that after quality control, the overall accuracy and Kappa coefficient of the analyses were considerably improved. RF and TCNN methods demonstrated superior prediction and modeling performance compared to XGBoost and SVM models. The overall accuracy of the four methods increased from 91 %, 87 %, 83 %, and 91 % before quality control to 96 %, 94 %, 89 %, and 95 % after quality control, respectively. The results of this study highlight the effectiveness of employing time-series data and quality control procedures to enhance crop classification in complex agricultural systems. By improving the precision and accuracy of agricultural classifications our findings can contribute to optimizing resource management, food security, and sustainable development goals.