Home
Publications
Asa
2012 16th AAC
Disease
Enhancing Plant Disease Prediction Using Leaf Wetness Forecasting

Enhancing plant disease prediction using leaf wetness forecasting

Simone Kreidl¹, Malcolm McCaskill², Robert Holmes¹, Subhash Sharma³, Oscar Villalta¹ and Debra Partington²

¹ Department of Primary Industries, 621 Burwood Hwy, Knoxfield, Victoria, 3180²Department of Primary Industries, Private Bag 105, Hamilton, Victoria, 3300 Email: malcolm.mccaskill@dpi.vic.gov.au³Department of Primary Industries, 32 Lincoln Square North, Carlton 3053

Abstract

Leaf wetness is a key input for fungal and bacterial disease warning systems in broadacre and horticultural crops. Direct measurements of leaf wetness can lead to delays in treatment after infection occurs, with control options limited to post-infection control agents that are prone to resistance development.

A leaf wetness model developed in the USA was compared with hourly leaf wetness observations in stone-fruit orchards across 11 sites in south-eastern Australia between 2007 and 2011. The model uses hourly temperature, humidity and wind speed as input data. The root mean square error (RMSE) of leaf wetness duration per day ranged from 3.0 to 6.0 hr/d over the various site-year combinations with a mean error of 4.6 hr/d. This was similar to that achieved by developers of the system, who reported a mean absolute error of 4.2 hr/d. The accuracy of leaf wetness prediction was found to be insufficient for its use in predicting brown rot infection periods and as a sole guide for scheduling fungicide at the orchard level. Nevertheless, its current accuracy could be useful to guide the application of pre-infection chemicals when large infection events are forecast.

The model is currently being developed into a real-time forecasting system for brown rot of stone fruit, with input data of hourly temperature, humidity and wind speed provided by the Bureau of Meteorology (BoM). The progress of potential brown rot infections can be predicted for 7 days in advance for over 100 locations around Victoria and will be made available to growers via the internet.

Key Words

Disease forecasting, brown rot, stone fruit

Introduction

The duration of leaf wetness is a key input for disease warning systems because spores of many fungal pathogens that cause diseases such as apple scab and brown rot of stone fruit require free moisture to germinate and infect susceptible tissue (Holmes et al. 2007, Villalta et al., 2002). Leaf wetness can be monitored electronically by automated weather stations fitted with wireless telemetry located within vineyards and orchards. The “Model T” weather station (Western Electronic Design, Loxton SA) is extensively used in south-eastern Australia for monitoring vineyard micro-climatic variables required to identify periods of high risk for Downy mildew. Leaf wetness duration is usually measured with contact surface sensors mounted at canopy level to provide information on how long the canopy has remained wet following a rainfall or dew event. Leaf wetness and temperature data are combined to identify periods conducive for spore infection according to established pathogen infection criteria. A warning is then issued to vineyard or orchard managers to alert them of the need for fungicide intervention.

The limitation of this type of system is that the warning is only available after the predicted infection event, when chemicals with post-infection activity are the only control option. Overuse of post-infection treatments has its limitation because there is a history of pathogens developing resistance to these fungicides. Therefore using weather forecast data to predict periods of high infection risk is more desirable for applying preventive measures and minimising the use of post-infection treatments. The BoM currently uses weather forecast data to issue brown rot warnings for stonefruit growers in Northern Victoria. These are district-wide forecasts, however, utilising expected periods of leaf wetness calculated on the probability of precipitation being >55% (T. Williams, Bureau of Meteorology, September 2011, personal communication) using climate parameters which have been measured out in the open, and not necessarily representative of the drying characteristics within tree canopies used in modern intensive production systems. In addition, the warnings do not indicate the severity of infection events, and sometimes rain does not occur resulting in unnecessary spray applications. The availability of leaf wetness forecasts at the location level, rather than across large geographic regions, would greatly improve the precision of infection risk assessment and provide advanced warning enabling preventive treatments using fungicides less prone to resistance development.

Several methods have been proposed to estimate leaf wetness from standard weather data. Bregaglio et al. (2011) compared 6 methods across 12 sites in the USA and Italy and found an empirical classification system developed in the USA by Kim et al. (2002) to be the most robust. None of these models have been tested under Australian conditions. This paper compares leaf wetness predictions using the model of Kim et al. (2002) with observed data from several fruit-growing locations in south eastern Australia, and its use in a warning system for brown rot of stone fruit.

Methods

Leaf wetness was monitored using surface contact leaf wetness sensors connected to “Model T” weather stations. The sensors recorded wet (yes) or dry (no) conditions at 10-minute intervals in 12 stone fruit orchards in south eastern Australia (Table 1). Hours were counted as “wet” in the observed data if more than half the 10-minute periods were wet. Predicted leaf wetness was estimated from hourly air temperature, relative humidity, and wind speed according to Kim et al. (2002). Since wind speed was not measured in the orchards, data from the nearest automatic weather station of the BoM network were used. These stations were up to 41 km away from the orchard, with a mean of 19 km. Wind speeds were measured at 10 m height, and multiplied by 0.41 to estimate wind speed at 0.3 m, and 0.63 to estimate wind speed at 1.0 m (based on Rosenberg et al. 1983).

The accuracy of the model of Kim et al. (2002) for predicting leaf wetness was examined by comparing the number of hours of leaf wetness predicted per day with observed values using root mean squared error (RMSE) and correlation coefficient estimates using Genstat 13 (VSN international, Hemel Hempstead, UK). In addition, the percentage of actual brown rot infection periods in agreement with predicted infection periods was investigated at two locations (Cobram and Lake Boga) for four years. Brown rot infection periods were estimated by multiplying the hours of wetness by the mean temperature (degree hours °h) during the wet period (Tate and Manktelow 1992). The spore infection process was assumed to stop if the leaf wetness sensor was dry for more than 4 hours between wet events. The severity of infection periods was categorised as marginal (90-120 °h), light (121-150 °h), moderate (151-180 °h) and severe (>181 °h). Infection period thresholds over 120 °h calculated with observed leaf wetness were used to find the percentage of infection periods correctly estimated with predicted leaf wetness.

A brown rot warning system was developed for internet application by combining predicted leaf wetness and the brown rot development algorithm of Tate and Manktelow (1992) with hourly numerical BoM weather forecasts for 7 days in advance and recent past weather data from their automatic weather station network.

Results

Predicted leaf wetness

Correlation coefficients between predicted and observed leaf wetness data sets ranged from 0.695 to 0.015 and RMSE values from 3.0 to 6.0 hr/d for the different locations and years examined (Table 1). In most cases the highest correlation coefficients and the lowest RMSEs were obtained using leaf wetness predicted at 0.3 m. There was a trend for lower RMSEs, indicating less difference measured and predicted values, at orchards closer to BoM weather stations (P = 0.07, data not shown).

Infection period prediction

At both locations, the percentage of correctly predicted infection periods was greatest at 0.3 m wind speed height and lowest at 10 m (Table 2). However, the percentage of falsely predicted infection periods was also greater at 0.3 m due to over-prediction of leaf wetness.

Implementation

Since the leaf wetness predictor at a 0.3 m height was the most reliable for predicting leaf wetness, it was incorporated into a real-time forecasting system for brown rot made available on the internet (www.dpi.vic.gov.au/vro). The forecast is portrayed as an animated Flash graph with background colours graphically representing the brown risk thresholds (Figure 1). Graphical presentation allows users to distinguish between events that only just reach minimal damage thresholds before being killed off by daytime dryness, compared with high risk events when leaves remain wet most of the day, requiring fungicide intervention.

Table 1. Orchard locations, distance from a Bureau of Meteorology automatic weather station (km), and correlation coefficients and RMSE for comparison between observed and predicted leaf wetness data sets.

	Site	BoM	Correlation coefficient R²			Root mean sq error
Year		km	0.3m	1m	10m	0.3m	1m	10m
2010-11	Swan Hill 2	19	0.486	0.368	0.219	4.73	5.66	7.49
	Lake Boga 1	15	0.555	0.358	0.169	4.42	5.93	8.08
	Lake Boga 2	14	0.521	0.365	0.215	4.54	5.02	6.35
	Swan Hill 1	37	0.472	0.326	0.162	4.22	5.34	7.52
	Cobram 1	41	0.373	0.325	0.211	5.09	5.80	8.07
	Cobram 2	33	0.353	0.377	0.244	5.07	5.18	7.50
	Ardmona	7	0.625	0.625	0.362	3.72	3.94	6.92
	Tatura	0	0.594	0.506	0.190	4.33	4.89	7.43
	N E Shepparton	18	0.354	0.267	0.152	5.20	5.65	7.65
	N Shepparton 1	28	0.556	0.455	0.270	5.02	4.90	6.10
2009-10	Swan Hill 2	19	0.597	0.471	0.230	3.86	4.71	5.95
	Lake Boga 1	15	0.608	0.476	0.303	3.97	5.05	6.26
	Lake Boga 2	14	0.608	0.575	0.374	3.27	3.41	4.45
	Swan Hill 1	37	0.519	0.515	0.364	3.55	3.77	4.77
	Cobram 1	41	0.475	0.387	0.211	4.69	5.51	7.07
	Cobram 2	33	0.361	0.358	0.230	5.49	5.97	7.52
	N E Shepparton	18	0.512	0.423	0.237	4.52	5.35	7.14
	N Shepparton 1	28	0.563	0.517	0.380	4.30	4.45	5.50
2008-09	Swan Hill 2	19	0.608	0.499	0.019	3.35	3.79	5.39
	Lake Boga 1	15	0.695	0.408	0.113	2.96	4.44	5.72
	Lake Boga 2	14	0.250	0.184	0.044	4.79	4.01	4.06
	Swan Hill 1	37	0.322	0.237	0.033	4.09	4.09	4.84
	Cobram 1	41	0.524	0.333	0.169	4.48	5.76	6.71
	Cobram 2	33	0.230	0.175	0.038	6.22	6.65	7.18
	N E Shepparton	18	0.182	0.177	0.042	5.65	5.35	5.62
	N Shepparton 1	28	0.164	0.134	0.046	6.60	6.58	6.67
	Shepparton	5	0.399	0.405	0.328	4.26	4.03	4.64
2007-08	Lake Boga 1	15	0.357	0.100	0.015	4.73	5.59	6.15
	Swan Hill 1	37	0.258	0.056	na	4.37	4.60	4.87
	Cobram 1	41	0.351	0.244	0.145	5.31	6.25	6.82
	Shepparton	5	0.256	0.186	0.053	4.24	4.45	4.97
	N Shepparton 2	28	0.269	0.199	0.162	5.49	6.13	6.68
Mean		19	0.439	0.345	0.184	4.58	5.07	6.32

Table 2. Number of actual brown rot infection periods (> 120 °h) in agreement with similar infection periods estimated using predicted leaf wetness at Cobram and Lake Boga 1, 2007-2011

Site	Year	Infection periods >120	No. of events in agreement			Events predicted where none was recorded
Site	Year	Infection periods >120	0.3m	1.0m	10m	0.3m	1.0m	10m
Lake Boga 1	2010-11	47	29 (61.7%)	22 (46.8%)	10 (21.3%)	10	5	1
	2009-10	23	11 (47.8%)	5 (21.7%)	1 (4.3%)	1	0	0
	2008-09	15	12 (80.0%)	8 (53.3%)	1 (6.7%)	3	2	0
	2007-08	17	7 (41.2%)	5 (29.4%)	0 (0.0%)	1	0	0
Cobram 1	2010-11	93	40 (43.0%)	30 (32.3%)	6 (6.5%)	5	4	0
	2009-10	49	22 (44.9%)	14 (28.6%)	0 (0.0%)	7	3	0
	2008-09	39	14 (35.9%)	8 (20.5%)	0 (0.0%)	0	0	0
	2007-08	37	8 (21.6%)	4 (10.8%)	2 (5.4%)	1	0	0

Discussion and Conclusions

The leaf wetness model had an average RSME of 4.6 hours at the 0.3 m wind speed, which was similar to the mean absolute error of 4.2 hours reported by the developers of the system (Kim et al. 2002). The greatest proportion of infection events was predicted from the 0.3 m wind speeds. However this only predicted between 22 and 80% of infection events. Wind speed estimates from the 1.0 and 10 m heights detected fewer infection events, but there were also fewer events that were falsely predicted. Wind speed is clearly a sensitive parameter in the model, and better methods of estimating wind speeds in orchards from standard weather measurements should lead to improved leaf wetness prediction.

Figure 1. Flash graph of brown rot index from the internet forecasting system

The model accuracy is, at this stage, insufficient to be the sole guide to a fungicide program. The system could be used to apply preventative fungicides prior to forecast large infection events, combined with leaf wetness measurements from orchard-based weather stations to detect smaller or unforecast events. This should lead to lower overall treatment costs because pre-infection control chemicals are cheaper than those applied post-infection.

Similar systems that combine forecast and measured leaf wetness could be built for fungal diseases of other horticultural and broad-acre crops once suitable models that link leaf wetness duration to disease severity have been developed.

References

Bregaglio S, Donatelli M, Confalonieri R, Acutis M, Orlandini S (2011) Multi metric evaluation of leaf wetness models for large-area application of plant disease models. Agricultural and Forest Meteorology 151, 1163– 1172.

Holmes R, Villalta O, Kreidl S, Partington D, Hodson A, Atkins TA (2007) Weather-based model

implemented in HortPlus MetWatch with potential to forecast brown rot infection risk in stonefruit. Acta Horticulturae. 803:19-27.

Kim KS, Taylor SE, Gleason ML, Koehler KJ (2002) Model to enhance site-specific estimation of leaf wetness duration. Plant Disease 86, 179-185.

Rosenberg NJ, Blad BL, Verna SB (1983) Microclimate: The Biological Environment. 2^nd Ed. John Wiley & Sons, New York.

Tate KG, Manktelow DW (1992). Field evaluation of a disease management system for peach rot in Hawkes Bay. Proc. 45^th N.Z. Plant Protection Conference, 129-135.

Villalta ON, Washington WS, Kita N, Bardon D (2002) The use of weather and ascospore data for the forecasting of apple and pear scab in Victoria. Australasian Plant Pathology 31: 205-215.

...

Search Site

The Regional Institute
Help/Feedback
phone: 61 2 4369 6006
email: admin@regional.org.au
Sitemap