Understanding heterogeneity in agricultural technology adoption is critical for designing effective rural development policies. This paper employs spatial Hidden Markov Models to identify unobserved farmer types and quantify neighborhood effects in irrigation technology and seed certification adoption, using data from the 2024 Peruvian Agricultural Census (37,492 farms). We compare a baseline HMM without spatial structure against a spatially-augmented specification incorporating k-nearest neighbor weights. The spatial model substantially improves fit (ΔAIC = 36,303), revealing three latent farmer segments: traditional subsistence (43.7%), technological transition (13.3%), and sustainable-oriented (43.0%). Moran’s I tests confirm significant positive spatial autocorrelation (I = 0.199, p < 0.001). The transition segment, despite representing only 13.3% of farms, concentrates 32% of potential water-saving capacity (5.3 M m3/year). Results suggest that spatially-targeted interventions exploiting neighborhood effects could yield substantial water efficiency gains. We validate robustness across alternative neighborhood specifications (k = 4 to 12) and discuss implications for agricultural policy design.