Combining Geographical Information System (GIS) and machine learning to monitor and predict vegetation vulnerability: An Empirical Study on Nijhum Dwip, Bangladesh

Abstract

Vegetation loss has become a global concern as it is directly and indirectly harmful to all living beings, specifically to humans. By realizing the dimension of this issue, we have developed an Artificial Neural Network (ANN) based model by combining GIS and machine learning to investigate and predict the vegetation vulnerability with reliable accuracy. This model incorporates a sequence of geostatistical analyses, i.e., Normalized Difference Vegetation Index (NDVI), Hot Spot Analysis (Gi*) and Inverse distance weighted (IDW). 8 drivers, used in this study, were shuffled differently to obtain the highest accuracy possible and investigate their influence on land shift. However, the model was implemented on Nijhum Dwip island to quantify its credibility and evaluate the ecological stability and vulnerability status of this island. According to the findings, the island has undergone significant change between 2001 and 2021. The overall vegetative area has increased in this time as a result of the ecological reforestation projects undertaken after 2001. Then, our developed hybrid model was used to simulate the hot spot map of 2021 to quantify the accuracy of the model. Anyway, the kappa statistics was found more than 0.86 with 88.75% overall correctness, and the same weight values were utilized to predict the hotspot map of 2026 and 2031. The predicted maps showed a gradual increase in the vulnerable zones, which is the outcome of the uncontrolled extracting of natural resources. Eventually, the methodological knowledge of this study can help researchers as well as policymakers to estimate vegetation vulnerability and legislate new policies that support sustainable development, and the quantitative knowledge on Nijhum Dwip can facilitate future planning of this island.