Ecosystem vigilance: leveraging NDVI analysis and Deep Learning for sustainable plant health management in Ziarat and Sherani districts, Balochistan, Pakistan

Abstract

In the Baluchistan province of Pakistan, Ziarat and Sherani districts hold significant economic potential for plant cultivation, yet they face severe environmental challenges, including illegal tree cutting, forest fires, and plant diseases. The first part of our comprehensive study, using advanced technology like Landsat 8–9 Operational Land Imager (OLI) imagery from 2013 and 2022 and the Normalized Difference Vegetation Index (NDVI) has revealed alarming statistics: 99% of the vegetation land cover consists of dead plants, 97% are categorized as unhealthy, and very healthy plants are extinct. Projections indicate that moderately healthy plants will disappear in 3 years in Sherani and 6 years in Ziarat. The second part of our study focuses on early disease detection, especially for exotic tree species like olive as Ziarat and Sherani districts are rich in exotic tree species such as pine nut, juniper, and wild olive. We utilized advanced deep-learning techniques and a dataset comprising 5,334 olive leaf images, including those affected by Aculus Olearius and Olive Peacock Spot diseases, in addition to healthy leaves. Innovative transfer learning models such as Inception V3, Inception Resnet V2, MobileNet, and Convolutional Neural Networks (CNN) have been applied to enhance disease identification accuracy. The results highlight the promise of these technologies in early disease detection, with MobileNet demonstrating exceptional performance by reducing execution time through the strategic use of fewer training epochs, achieving a 99% accuracy rate for binary classification and 97.6% for multiclass classification, along with the highest F1 score of 99.4. These findings underscore the urgent need to preserve plant health, protect vegetation, and safeguard species, highlighting the importance of biodiversity and forest conservation in critical regions. Keywords: Environmental challenges, Vegetation health, Disease detection, Deep learning models, Biodiversity conservation

Plant phenotyping relevance

葉画像から植物病害状態を推定する深層学習手法を複数比較し、精度やF1スコアを評価しており、植物表現型取得・推定が中心的である。NDVIによる地域植生監視部分は対象外だが、病害推定部分が収載基準を満たす。

Code and data availability

The supplied blocks describe NDVI analysis of Landsat imagery and deep-learning classification of 5,334 olive leaf images, but contain no public dataset, image repository, code deposit, or trained model release. The only hosted file ('Classification paper with NDVI.docx') is the article manuscript itself, not a paper-­

No evidence-backed public reproduction asset is currently recorded.