Sensitivity of Disease Management Decision Aids to Temperature Input Errors Associated with Sampling Interval and Out-of-Canopy Sensor Placement

Pfender, W. F., Gent, D. H., and Mahaffee, W. F. 2012. Sensitivity of disease management decision aids to temperature input errors associated with sampling interval and out-of-canopy sensor placement. Plant Dis. 96:726-736. Many plant disease epidemic models, and the disease management decision aids developed from them, are created based on temperature or other weather conditions measured in or above the crop canopy at intervals of 15 or 30 min. Disease management decision aids, however, commonly are implemented based on hourly weather measurements made from sensors sited at a standard placement of 1.5 m above the ground or are estimated from off-site weather measurements. We investigated temperature measurement errors introduced when sampling interval was increased from 15 to 60 min, and when actual in-canopy conditions were represented by temperature measurements collected by standard-placement sensors (1.5 m above the ground, outside the canopy) in each of three crops (grass seed, grape, and hops) and assessed the impact of these errors on outcomes of decision aids for grass stem rust as well as grape and hops powdery mildews. Decreasing time resolution from 15 to 60 min resulted in statistically significant underestimates of daily maximum temperatures and overestimates of daily minimum temperatures that averaged 0.2 to 0.4°C. Sensor location (in-canopy versus standard-placement) also had a statistically significant effect on measured temperature, and this effect was significantly less in grape or hops than in the grass seed crop. Effects of these temperature errors on performance of disease management decision aids were affected by magnitude of the errors as well as the type of decision aid. The grape and hops powdery mildew decision aids used rule-based indices, and the relatively small (±0.8°C) differences in temperature observed between in-canopy and standard placement sensors in these crops resulted in differences in rule outcomes when actual in-canopy temperatures were near a threshold for declaring that a rule had been met. However, there were only minor differences in the management decision (i.e., fungicide application interval). The decision aid for grass stem rust was a simulation model, for which temperature recording errors associated with location of the weather station resulted in incremental (not threshold) effects on the model of pathogen growth and plant infection probability. Simple algorithms were devised to correct the recorded temperatures or the computed infection probability to produce outcomes similar to those resulting from in-canopy temperature measurements. This study illustrates an example of evaluating (and, if necessary, correcting) temperature measurement errors from weather station sensors not located within the crop canopy, and provides an estimate of uncertainty in temperature measurements associated with location and sampling interval of weather station sensors. Decision aids based on disease epidemic models are being developed increasingly for improving disease management programs (4,5,7,30–32). Models and decision aids vary in complexity, ranging from simple, single-day indices of infection favorability or cumulative favorability indices to complex, season-long simulations (3,10,12,33,35). Most are designed to account for effects of fungicide use on disease development, either quantitatively or based on simple rules. Some also include crop loss components in the decision calculation (13). All of these decision aids depend on disease models that use weather data as inputs, and many were developed using weather data collected at 15or 30-min intervals from sensors placed in the crop canopy. Therefore, the optimum input data for running the disease models are accurate measurements taken in-canopy at a time resolution (interval) of 15 or 30 min. When the decision aids are implemented, however, such weather data may not be available because of the cost and time required to install, maintain, and manage weather-monitoring equipment and data acquisition. Instead, disease models may be run with available weather data, often from nonagricultural or other off-site locations (8). The need for accurate, site-specific weather data in the absence of on-site field measurements has motivated development of various systems to estimate surface weather at relatively high spatial resolution over large geographical areas (1,18,33). These systems typically produce estimates of weather variables at a standard meteorological placement (usually 1.5 m above the ground, over low vegetation such as mowed grass) and a time resolution of 1 h. Accuracy of site-specific weather estimates is particularly challenging in geographical regions such as the western United States that have topographic variation which can greatly complicate spatial interpolation of weather data (2). Importantly, in many regions of the western United States, there are intensively managed perennial or high-value crops for which disease management decision aids are in great demand (34). An important consideration in implementing weather estimation protocols for supporting crop disease management decision aids is the inaccuracy introduced by errors in weather measurements relative to other sources of error. Sources of weather estimation error in on-site measurements include sensor placement (in-canopy versus standard placement) and time resolution (e.g., 15 versus 60 min). In cases where the standard-placement weather data are estimated from off-site measurements, there are additional errors associated with these estimation procedures. We propose that the impact of the weather data errors on a disease management decision aid are affected by sensitivity of the aid to the type, magnitude and frequency of errors in weather monitoring that occur. To investigate how weather input data errors affect disease model performance, we chose three examples of decision aids to Corresponding author: W. F. Pfender, E mail: [email protected] The use of trade, firm, or corporation names in this publication is for the information and convenience of the reader. Such use does not constitute an official endorsement or approval by the United States Department of Agriculture or the Agricultural Research Service of any product or service to the exclusion of others that may be suitable. Accepted for publication 7 December 2011. http://dx.doi.org/10.1094 / PDIS-03-11-0262 This article is in the public domain and not copyrightable. It may be freely reprinted with customary crediting of the source. The American Phytopathological Society, 2012.