An effective BiLSTM-CNN model for predicting large-scale temporal-spatial dynamics of normalized difference...

Vegetation, though central to terrestrial ecosystems, remains highly vulnerable to fluctuations in climate and human-induced activities. Such combined influence on vegetation health dynamics necessitates the application of robust remote sensing-based vegetation indices, such as the Normalized Difference Vegetation Index (NDVI). The latter enables the detection of subtle structural and functional changes, offering an early warning of plant stress before visible symptoms appear. It is therefore crucial to predict vegetation activities by modelling NDVI that is able to detect and attribute the climate change impacts on vegetation growth through its temporal and spatial variations. For this reason, in this paper we introduce an advanced combined deep learning method (bidirectional long short-term memory and convolutional neural network model, BiLSTM-CNN) for temporal-spatial modelling of NDVI informed by meteorological and soil moisture data. BiLSTM-CNN is a composite progressive processing model that can investigate potential trends of vegetation alterations that may be abrupt and barely obvious, localized and extensive, happening over short or long-time scales. Our proposed BiLSTM-CNN forecasting method has been evaluated and compared with state-of-the-art techniques, and the experimental results have shown clearly that our proposed method is competitive with the existing relevant NDVI deep learning predicting models.