Breeding crops tolerant to abiotic constraints is more efficient if key traits that are closely related with yield are available for screening. This study was undertaken to identify key traits that are closely related to cassava yield under drought stress and also identify stable high yielding cassava genotypes under varying environments. A participatory rural appraisal involving 120 farmers from three districts in the northern region of Ghana was carried out to identify farmers’ perception on drought in cassava cultivation, production constraints, mitigation strategies and preferences for improved cassava genotypes. To identify ideal genotypes that meet farmers’ preferences, 150 cassava genotypes from local and exotic sources were assembled and assessed for diversity using morphological traits and simple sequence repeat markers. The genotypes were also evaluated in an alpha lattice design for growth and yield traits. Thirty six simple sequence repeat markers were used to assess molecular diversity of 89 selected genotypes. Subsequently 20 genotypes were selected and evaluated under irrigation and no irrigation to assess genetic variability in abscisic acid content, carbon isotope ratio, stomatal conductance and leaf temperature. Root yield was assessed at six, eight, ten and twelve months after planting to study the relationship between early bulking and drought tolerance. Stability of genotypes for physiological and yield traits were also assessed using the additive main effect and multiplicative interaction (AMMI) and GGE biplot analyses. The participatory rural appraisal indicated lack of credit as the most important constraint facing cassava cultivation in the region. Drought was the second most important constraint and the intensity was observed to be increasing. Majority of the farmers also preferred early maturing cassava varieties that are high yielding with good plant type and marketability. Factorial analysis of the morphological traits grouped the genotypes based on their origin with few exceptions. Principal component analysis further identified plant height, branching habit, colour of stem exterior, distance between leaf scars, colour of end branches, root yield, harvest index and number of roots per plant as the traits contributing most of the variability in the groups. High level of heterozygosity was revealed by the  simple sequence repeat markers. Genetic diversity and the number of alleles were high for the genotypes reflecting their diverse sources. Heterozygosity levels for the markers ranged between 0.03 and 0.79. Gene diversity also ranged between 0.03 and 0.81. Cluster analysis based on the neighbour- joining analysis and simple matching coefficient grouped the genotypes into seven distinct clusters irrespective of source. Genetic variability was established for abscisic acid content which was higher under stress than irrigation. ABA content was negatively correlated with root yield, harvest index and above ground biomass yield meaning it can be used as indirect selection criteria against unproductive genotypes. Narrow genetic variation was observed for carbon isotope ratio which was higher under irrigation than no irrigation. Carbon isotope ratio was positively correlated with above ground biomass yield but not root yield. Stomatal conductance and leaf temperature were significantly different among genotypes and environments but genotype x environment interaction was not significant. Stomatal conductance was positively correlated with root yield. Broad sense heritability estimates were low for stomatal conductance, above ground biomass yield, root number and stem diameter but high for plant height (82.8%), height at branching (78.32%), dry matter content (77.7%), harvest index (71.7%), root girth (75.03%), root length (77.3%), root length/girth ratio (84.6%) and mean root weight (70.58%). AMMI analysis of plant height, severity of cassava mosaic disease, root yield, root length/girth ratio, above ground biomass yield and harvest index indicated stronger effect of environment than genotypes for all traits except CMD. Based on AMMI selections and the GGE biplot analysis, three genotypes, MM96/1751, UCC2001/449 and 00/0203 were identified as high yielding and stable across envelopments. Six genotypes (UCC2001/449, 96/1708, MM96/1751, 00/0203, 96/409 and I91934) had significantly higher root yield than the best local farmer preferred variety.