Coastal delta agricultural systems are facing severe threats from saltwater intrusion (SWI), which further impacts global food security. However, how to accurately and timely assess the vulnerability of coastal agriculture to SWI at a regional scale remains challenging. Most existing methods rely on either manually-designed multi-index systems or the identification of severely salinized areas, resulting in high uncertainties and more granular crop-scale evaluations. Meanwhile, they remain inadequate for early-risk warning, as initial crop canopy obscures the underlying soil spectral signatures critical for early SWI detection. Critically, two unresolved issues persist regarding the effects of crop cover on SWI: the objective quantification of early-stage SWI impacts at the field scale under crop cover conditions, and the variation along with precise measurement of differential SWI responses across crop types. Taking the Po Delta in Italy as an example-a well-known hotspot affected by SWI but with unknown quantification, we proposed a crop damage probability assessment framework fully driven by machine learning and satellite imageries. Based on crop classification, a Convolutional Neural Network (CNN) was employed to quantify the probability of SWI effects. Furthermore, this method enables crop damage probability mapping during early-stage SWI upon each satellite revisit. This study presents the first application of CNN to generate crop-specific, monthly mappings of SWI impacts on coastal agriculture, addressing a critical gap in early and objective damage quantification. The results indicated that the CNN-based algorithm enabled quantitative risk assessment of crop damage from saltwater intrusion on a monthly scale. During the saltwater intrusion period (May-September), the CNN-based model achieved an identification accuracy of 82.4% to 92.5% in damaged agricultural areas based on known episodes of salinization, and the quantitative risk maps of crop damage can be used to reveal agricultural hotspots threatened by saltwater intrusion. Analysis of crop-specific damage probability showed that the highest risk occurs in June and July, followed by a gradual decline, with alfalfa in the Po River Delta experiencing the greatest risk. The proposed framework avoided subjective judgment while delivering crop-scale quantification of potential SWI effects on agricultural systems, serving as a preliminary tool for early detection of vulnerable areas and supporting timely management and mitigation strategies.
Xue et al. (Tue,) studied this question.