Epidemic potential of Maravalia cryptostegiae, a classical biological control agent of rubber vine, in Northeastern Brazil

Rubber vines (Cryptostegia madagascariensis and C. grandiflora) are highly invasive weeds in the semi-arid landscapes of many tropical regions. The rust fungus Maravalia cryptostegiae has provided a spetatular and sustainable control of invasive C. grandiflora populations in Australia, after its scientific and quarantine-authorized release in the context of a classical biological control program. The Caatinga biome of Northeastern Brazil is under a devastating invasion by C. madagascariensis. This species is also susceptible to M. cryptostegia but its epidemic potential towards C. madagascariensis in Brazil remains unknown. We developed a weather-driven epidemiological model integrating experimentally derived relationships for infection as a function of temperature and wetness duration and for temperature-dependent latent development. Three infection-response models were fitted to 20 experimental temperature x wetness combinations from the literature and compared using information criteria, grouped cross-validation and a parametric bootstrap. The selected model, in which thermal breadth increases with wetness duration, was best supported (R2 = 0.931; {Delta}AICc [≥] 3.3 relative to the alternatives), and prolonged wetness significantly broadened the thermal range favorable for infection ({gamma} = 0.30 oC h-1; bootstrap P = 0.002). The infection and latent-development models were coupled with 25 years (2001-2025) of hourly NASA POWER data for 65 grid cells representing rubber-vine-infested locations in Northeastern Brazil, with wetness defined as hours with relative humidity equal or greater than 90%. Infection potential was concentrated in March-June (73% of the annual total, peaking in April), coinciding with the regional rainy season, and was highest in the interior of northern and central Ceara. Predicted potential varied more than a hundredfold among grid cells and was lowest along parts of the Ceara coast and the lower Jaguaribe valley, which include some of the largest infestations; the magnitude of these contrasts depended on the humidity threshold used to define wetness. Applied with identical rules to 13 historical release sites in Queensland, Australia, the model reproduced the documented wet-season timing of rust activity, and predicted potential was on average higher at sites with reported high rust impact, although the difference was not statistically significant. These results indicate strong seasonal and geographic structuring of weather-driven epidemic potential and provide testable hypotheses for selecting release periods and locations for M. cryptostegiae in Brazil.