Land cover Classification using Spatial and Spectral Features from Remotely Sensed Data

In Ethiopia, the agricultural sector relies on traditional methods like the GCES technique for estimating average crop yields. This method, based on crop-cutting experiments, faces significant limitations. It fails to provide timely information on critical aspects such as crop health, growth stages, acreages, and estimated yields for different crop types. Moreover, field surveys, which are often used for data collection, are prone to errors and require considerable time to analyze, leading to delayed and unreliable decision-making. This results in poor yield estimates, neglect of dynamic environmental conditions, and dependence on subjective expert judgments. To address these challenges, we proposed a crop classification and yield estimation algorithm utilizing hyperspectral image data. The study area was carefully selected to include diverse crop types, enabling effective classification and yield estimation. For classification, we employed machine learning algorithms, including Maximum Likelihood, Random Forest, and Support Vector Machines (SVM). Landsat images were acquired from the same study area three different times to monitor crop growth patterns. The proposed method achieved classification accuracies of 93%, 98%, and 97%, respectively. For yield estimation, high-resolution spectral image data were utilized, requiring a separate dataset. The results demonstrated that integrating remote sensing technology into Ethiopia’s agricultural practices significantly improves production efficiency. It enables better crop growth management, land-use monitoring, real-time data analysis, and informed decision making. These advancements highlight the transformative potential of remote sensing and machine learning in modernizing Ethiopia’s agriculture sector and addressing its longstanding challenges.