Evaluation of a multi-model approach to estimate leaf wetness duration: an essential input for disease alert systems

Disease alert systems (DAS) of the AgroClimate platform are intended to facilitate grower decision-making when planning fungicide applications. These DAS provide disease risk estimates that are linked to management recommendations. If disease risk exceeds a threshold, a fungicide application is recommended. Temperature and leaf wetness duration are often required inputs to calculate daily disease risk. While temperature estimation is straightforward, calculating leaf wetness duration lacks standardization, being estimated by predictive models and/or by sensors. For the DAS on the AgroClimate platform, a four-model leaf wetness system and dielectric Campbell 237-L sensors are used. However, the four-model system was never formally assessed. Our objectives were to compare the performance of four leaf wetness models and their two-, three-, and four-model combinations-number of hours with relative humidity equal or greater than 90% (NHRH90), dew point depression (DPD), classification and regression tree (CART), and Penman-Monteith (PM)-to well-calibrated 237-L sensors. The performance of each model was satisfactory and overall comparable among the models used individually or in combinations of three and four models. Three of the two-model combinations did not perform well. PM tended to overestimate wetness, whereas NHRH90 tended to underestimate wetness. CART- and DPD-estimated leaf wetness resulted in satisfactory disease management recommendations. We confirmed the potential for operational use of models or their combinations in our DAS, although PM and NHRH90 should be used carefully. The other models and three- or four-model combinations are viable options for operational use in automated disease alert systems.

Automated summary

The performance of four leaf wetness models and their combinations were compared to well-calibrated sensors. The results showed that each model performed satisfactorily and the CART and DPD-estimated leaf wetness provided satisfactory disease management recommendations. The study confirmed the potential operational use of these models and combinations in automated disease alert systems.