Development of an agrometeorological model integrating leaf wetness duration estimation and weather radar data to assess the risk of head blight infection in wheat A sensitivity and uncertainty analysis R OGER 1, D BUFFET 1, P DETRIXHE 2, A CHANDELIER 2, and M CAVELIER 2 Ministry of the Walloon Region Agricultural Research Centre (1) Unit of Biometry, Data processing and Agrometeorology Rue de Liroux n 9, B-5030 Gembloux, Belgium E-mail: oger@cra.wallonie.be (2) Department of Biological Control and Plant Genetic Resources Rue de Liroux n 4, B-5030 Gembloux, Belgium E-mail: detrixhe@cra.wallonie.be
Plan of the presentation Introduction General objectives of the study Estimation of head blight infection risks Agrometeorological model Interpolation of meteorological data Weather radar data Validation with field observations Sensitivity and uncertainty analysis Conclusions Radar data Meteorological parameters Spatialisation of data
Estimation of head blight infection risks Agricultural factors Previous crop Soil preparation Crop variety Fungicide treatments At parcel level Meteorological factors Leaf Wetness duration Temperature Continuous coverage Risk factors are combined into a GIS
Estimation of leafwetness duration utilizing standard weather stations and radar data Energy budget concept for a leaf or a droplet Rn + LE + H = 0 Rn : net radiation LE : latent heat flux H : convective heat flux Estimation of latent heat flux transfer at the leaf level and for a drop Air temperarure Ambient vapour pressure or relative humidity Incoming short wave radiation Incoming long wave radiation Wind speed
Interpolation of meteorological data Interpolation method : protocol used by the European Crop Growth Monitoring System (CGMS) Use of synoptic and meteorological automatic weather station networks Remark : the available meteorological parameters change from a station to an other continuous observed temperature data are more frequent than radiation or vapour pressure data Interpolation of meteorological data from weather stations ( 10 km x 10 km, 10 minutes period, short wave radiation, long wave radiation, vapour pressure, wind speed, temperature)
Interpolation of precipitations using weather radar data Source Royal meteorological Institute (Belgium) Site: Wideumont (Lat 49.915 - Long 5.505) Band: (S,C,X) Doppler Range : 240 km Number of images /day: +/- 100 Pixel size : +/-560 m2 Step of time between two images: 5 minutes Geometric correction of radar images Filtering of radar images (remove noise) Construction of grid data tables for presence/absence of rain events compatible with the model structure and the meteorological data (1 km x 1km grid; 10 minute period) Stack of temporal series of grid data to produce rain duration maps
Estimation of of head blight infection risk (expressed in hours with a temperature >12 C during the reference period) Simulations for each grid (1km x 1km) were performed each 20 minutes. Hours of cumulated leaf wetness duration for a T > 12 C Reference period : 7 days before and after flowering date
Combination of risk factors into GIS to produce a risk of infection indicator at parcel level Risk of infection map (grids 1 km x 1 km) Attribute information (crop type, history ) is stored in a separate parcel database. IACS = Integrated Administration and Control System used by each member state to distribute subsidies in the frame of the CAP IACS parcel map Risk of infection map (parcel level) Final objectives : produce a risk indicator for control purposes and limit the number of mycotoxins analysis
Localisation and Fusarium sp. infection level for collected samples of winter wheat. The size of the circles is proportional to the percentage of Fusarium sp. determined by microbiological analysis. Percentage of Fusarium sp.
Evaluation of model performances % of Fusarium sp. infection Leaf wetness duration < 10 % ³ 10 % TOTAL < 50 hours 18 6 24 ³ 50 hours 5 14 19 TOTAL 23 20 43 POD = 0.70 Percent correct = 0.74
Quality assessment of rainfall occurrence forecasts (periods of 20 minutes) Rainfall occurrence forecasts Nearest meteorological station POD 0.501 POFD 0.033 PC 0.936 Bias 0.970
Quality assessment of rainfall occurrence forecasts (periods of 20 minutes) Rainfall occurrence forecasts Nearest meteorological station Radar POD 0.501 0.688 POFD 0.033 0.033 PC 0.936 0.949 Bias 0.970 1.191
Sensitivity of head blight infection risk to meteorological parameters Sensitivity of leaf wetness duration and risk of head blight infection to variation of temperature Leaf wetness duration (hours) 125 100 75 50 25 0-1 -0,5 0 0,5 1 Variation of temperature ( C)
Sensitivity of head blight infection risk to meteorological parameters Sensitivity of leaf wetness duration and risk of head blight infection to variation of vapour pressure Leaf wetness duration (hours) 125 100 75 50 25 0-1 -0,5 0 0,5 1 Variation of vapour pressure (mb)
Sensitivity of head blight infection risk to meteorological parameters Sensitivity of leaf wetness duration and risk of head blight infection to variation of solar radiation Leaf wetness duration (hours) 125 100 75 50 25 0-10 -5 0 5 10 Variation of solar radiation (in %)
Sensitivity of head blight infection risk to meteorological parameters Sensitivity of leaf wetness duration and risk of head blight infection to variation of wind speed Leaf wetness duration (hours) 125 100 75 50 25 0-1 -0,5 0 0,5 1 Variation of wind speed (m/s)
Uncertainty of head blight infection risk estimation due to the interpolation of meteorological data Influence of interpolated parameter on leaf wetness duration and risk of head blight infection root mean square of prediction 25,0 Root mean square error 20,0 15,0 10,0 5,0 0,0 Temp of NN station Ea of NN station RH of NN station Wind speed of NN station Radiation of NN station All the parameters of NN station
Uncertainty of head blight infection risk estimation due to the interpolation of meteorological data : confidence interval of POD 1 0,8 ROC associated with lower and upper uncertainty intervals POD 0,6 0,4 0,2 0 25 30 35 40 45 50 55 60 65 70 Risk of infection threshold (leafwetness duration with temperature >12 C)
Conclusions Use of radar data can improve rainfall occurrence interpolation The model used for infection risk estimation is very sensitive to vapour pressure Interpolation of vapour pressure and wind speed are also the major sources of uncertainty for the estimation of infection risk when these parameters are considered separately. Relative prediction root mean square error is close to 10 % Even if the interpolation process seems to give good results, its influence on the probability of detection of head blight infection or contamination risk remains very high This uncertainty must be taken into consideration to interpret maps and results.