3 Modeling temporal and spatial variability of leaf wetness duration in Brazil

Monday, 12 May 2014
Bellmont BC (Crowne Plaza Portland Downtown Convention Center Hotel)
Clayton Alcarde Alvares, Forestry Science and Research Institute, Piracicaba, Sao Paulo, Brazil; and E. M. D. Mattos, P. Sentelhas, A. C. Miranda, and J. L. Stape

Handout (4.4 MB)

Leaf wetness is basically defined as the presence of free water over plant tissues and its duration is commonly named leaf wetness duration (LWD). LWD is recognized as a very important conditioner of crops and forests diseases, since it is fundamental for the fungal and bacterial infection process, as well as for the sporulation of some of them. Since 1980s, several studies have been made to overcome the problem of LWD measurement through the development of simulation models with different approaches. LWD models can be divided into two broad categories: empirical – based on the relationship between LWD and one or more weather variables; and physical – based on the physical principles of dew formation and evaporation. Among the LWD empirical models, the number of hours with relative humidity (RH) above a specific threshold is the most common and easy to apply, and in this regard several studies have been developed in many parts of the world. As for large scales, e.g. in a country level, the availability of weather data for estimating LWD is very restricted, the use of hours of RH ≥ 90% appears as the only alternative to assess the temporal and spatial LWD variability. Considering that, the objective of this study was to develop monthly LWD models based on the relationship between hours of RH ≥ 90% and average RH for Brazil, and based on these models to characterize the temporal and spatial LWD variability across the country, in order to support crop and forest diseases control strategic plans. The present study used two different relative humidity databases, being one in an hourly basis (RHh) and another in a monthly basis (RHm), both of them provided by the Brazilian National Institute of Meteorology (INMET). We selected 58 automatic weather stations (AWS) distributed across the country. Having the hourly database checked, we programmed a simple counter of hours of RH ≥ 90%, and this number was admitted as the LWD for that day. The total months with data for each AWS were 3156, which was used in the modeling phase. The consisted database was then randomly divided in two data sets. The first, referred to fitting set, with 80% of n ( 2525), was used to adjust non-linear regression equations in order to build the monthly LWD models. Another group, called testing set, with 20% remaining data ( 631), was used to perform the validation of the monthly and annual LWD models. After the validation, monthly LWD maps for the entire country were prepared. For that, the RHm from the 358 conventional weather stations (CWS) were interpolated using geostatistical techniques. Through structural parameters obtained from experimental semivariograms, RHm maps were created with a geographic information system (GIS). Finally, using geoprocessing techniques, the monthly RH and LWD estimated by the non-linear regression models were converted into maps using GIS, processing the independent (LWD) and dependent (RHm) variables as raster layers. Relative humidity and LWD showed sigmoidal relationship. Monthly and annual sigmoidal models showed determination coefficient above 0.84 and were highly significant (p<0.0001). In relation to the validation of the LWD monthly models, based on independent data (test set), the results showed a very good performance for all months, with very high precision, with r between 0.92 and 0.96. Regarding the errors, mean error showed a slight tendency of over estimation during February (0.29 h day-1), April (0.03 h day-1), May (0.31 h day-1), July (0.14 h day-1), August (0.34 h day-1) and November (0.01 h day-1), whereas for the other months such tendency was of underestimation, with mean error ranging from -0.27 to -0.03. The model with the best adjust to the RH and LWD spatial variability in Brazil was the spherical theoretical semivariogram, with coefficient of determination higher than 0.89, and cross-validation above of 0.67. The range of spatial dependence was markedly seasonal, between mid-spring and late summer, which are warmer and wetter months in most parts of Brazil, the semivariograms showed twice the range than cooler and drier months, from April to September. Monthly RH and LWD were spatially described in maps with 1 km, and considering every month, the RH ranges from 38 to 91%. March and April are the wettest months since they have monthly RH below 70% in only few small areas of the Northeastern Brazil. On the other hand, August and September are the driest months, with RH lower than 60% in great part of the country. Average monthly LWD in Brazil ranges from less than 2 h day-1, like in the semi-arid region in the northeast of the country, to more than 16 h day-1, mainly in the north part of the country in the Amazon region. In the Amazon region, where are the states of Amazonas and Pará, the average monthly LWD is higher than 10 h day-1 in all months, as well as observed in the Brazilian coastal region, from the state of Santa Catarina to the south of state of Bahia. Considering the promising performance of the LWD monthly models, they can be used to generate LWD data for disease risk studies.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner