Plants exhibit highly plastic stomatal distributions on the leaf epidermis driven by environmental factors. However, methods to model this complex spatial organization as a global topological system remain limited, as most existing tools focus on segmentation and counting. This study proposes a graph-based classification framework to characterize these distributions. We use stomatal centroids to construct networks and employ Life-Like Network Automata (LLNA) to extract dynamical signatures from them. This approach simulates network emergent behaviours to derive feature vectors based on optimized transition rules. Compared against traditional network metrics and some graph-based algorithms, our results indicate that our proposed method can effectively classify environmental stressors across the three studied species. In these restricted data scenarios, the evaluated models achieved accuracies from 90.00 ($\pm $5.65)% to 95.47 ($\pm $4.61)%, outperforming the chosen traditional and graph-embedding techniques. We present a pattern recognition framework capable of distinguishing environmental conditions based on the stomatal topological distribution.