Prediction and mapping of land degradation in the Batanghari watershed, Sumatra, Indonesia: utilizing multi-source geospatial data and machine learning modeling techniques

• Published in: Modeling earth systems and environment 2023-11, Vol.9 (4), p.4383-4404
• Main Authors: Yulianto, Fajar, Raharjo, Puguh Dwi, Pramono, Irfan Budi, Setiawan, Muhammad Anggri, Chulafak, Galdita Aruba, Nugroho, Gatot, Sakti, Anjar Dimara, Nugroho, Sapto, Budhiman, Syarif
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Artificial intelligence; Bioclimatology; Chemistry and Earth Sciences; Computer Science; Decision trees; Degradation; Earth and Environmental Science; Earth Sciences; Earth System Sciences; Ecosystem management; Ecosystems; Environment; Environmental degradation; Environmental impact; Environmental management; Land cover; Land degradation; Learning algorithms; Machine learning; Mapping; Math. Appl. in Environmental Science; Mathematical Applications in the Physical Sciences; Model accuracy; Modelling; Monitoring systems; Original Article; Physics; Regression analysis; Rehabilitation; Restoration; Spatial data; Statistics for Engineering; Support vector machines; Tropical climate; Tropical environment; Tropical environments; Watersheds
• ISSN: 2363-6203; 2363-6211
• DOI: 10.1007/s40808-023-01761-y

In the present study, the Geospatial Artificial Intelligence (Geo-AI) is proposed to overcome challenges and phenomena related to land degradation identification in the field by integrating multi-source geospatial data and machine learning modeling techniques. The study area is located in the Batanghari watershed, Sumatra, Indonesia, which is a tropical environment. The existence of environmental problems in the study area can be identified based on land degradation. This study’s novelty model is that it is the first to integrate the six main variables of multi-source geospatial data—topographical, biophysical, bioclimatic, geo-environmental, global human modification, and accessibility—in predicting and mapping the potential of land degradation in Indonesia’s tropical environment. Support Vector Machine (SVM), Minimum Distance (MD), Classification and Regression Trees (CART), Gradient Tree Boost (GTB), Naïve Bayes (NB), Random Forest (RF) are machine learning modeling algorithms used to predic and map land degradation in the study area. The prediction results from these modeling algorithms can be compared and evaluated to get the most optimal performance and accuracy. During the modeling phase, 70% of the reference data were divided into training and 30% into validation. The performance of the model was compared using three accuracy assessment processes (producer accuracy, user accuracy, and overall accuracy). The overall accuracy of the results of the comparison and evaluation of machine learning modeling on the SVM, MD, CART, GTB, NB, and RF algorithms in the study area are 52.8, 34.5, 81.2, 85.8, 36.3, and 86.2%, respectively. Therefore, the study concluded that the RF, CART, and GTB are machine learning modeling algorithms that were proposed to be applied and to produce a land degradation map in the study area. The results of this study can be used by policymakers in making decisions related to environmental management, ecosystem rehabilitation, and restoration. Technically, this can also be applied to other watersheds with similar characteristics. In addition, the results of this study can also be used to develop an intelligent watershed monitoring system to effectively monitor land cover change.