Resulting pesticide cloud rendered as GeoTIFF (.tiff) 

Context in source publication

Context 1

... we will focus on atmospheric dispersion and the induced deposition of pesticide on the ground. Atmospheric dispersion of pesticide is a complex process but can be basically explained as follows (Bozon et al, 2009). Phytopharmaceutical products are usually spread over the plots using a sprayer, which is an agricultural machine allowing to spray pesticide over large areas. It is most of the time towed or suspended from a tractor, as shown in figure 1. Sprayed pesticides then spread in several directions while the sprayer runs through the vineyards. Some of it reaches the vine leaves and grapes, some reaches the ground and run-o on soils and the rest leaves the plot to be transported by the wind. This forms a pesticide cloud, as shown by figure 2, that is prone to dispersion and depositing in the environment (Bozon et al, 2009). Many agro-meteorological studies are focusing on the spray drift process at different scales, and on its impacts on the environment. Near-field studies tend to become very accurate using Computational Fluid Dynamics (CFD) principles, and allow scientists to characterize pesticide emissions to the air during and after the treatments. This is a major asset to model the long-range transport of pesticide, as the results of validated micro-scaled models can be used as input data in the different atmospheric dispersion models available. In our former research, a long-range transport model was developed with taking some aspects of the GIS data model into account and by using open source tools (Bozon et al, 2007). Atmospheric dispersion modeling (ADM) is an essential tool in air quality management because it provides a relationship between source terms locations (i.e where discharges to the air occur) and observed adverse effects on the environment and the neighborhood. Atmospheric dispersion models refers to the mathematical simulation of air pollutants dispersion in the ambient atmosphere. They are intimately related to numerical simulations as most models are performed with computer programs that solve the mathematical equations and algorithms which simulate the pollutant dispersion. As atmospheric dispersion is complex and because air pollution cannot be measured in every place it occurs, models are used to simplify and simulate the dispersion of air pollutants from emission sources, and to to predict the downwind concentrations or depositions on a given area. Despite the fact that many models prove to be efficient on small domains (i.e a few square meters), only a few are adapted and validated for larger areas (i.e a few square kilometers) and simulations of atmospheric spray drift are seldom performed at the watershed scale. The decision of building a GIS- based dispersion model was thus taken, focussing on cartographic projections, scales changes and DEM layers in the original model 's code (Bozon et al, 2009). Accurate wind data-sets are difficult to acquire over large areas and long time-series but required by most CFD models. Some of the eolian processes included in pesticide atmospheric dispersion are also to complex to be solved by numerical simulations at that stage, due to their uncertainty and variability. Furthermore, CFD tools appear to be quite long to use for simulations, as both input data and domain geometry have to be pre-processed through the use of several softwares. Their use necessitate quite expensive hardware and software and most of all involves very long calculation costs. Given that wind data are quite poor for our concern and that a large set of assumptions has to be done to model the whole dispersion process at different spatial and temporal scales, the use of CFD tools to solve our problem was ill-advised. In relation to the necessary accuracy and because calculation costs have to stay low in order to perform fast simulations, an original alternative to the use of CFD is therefore proposed (Bozon et al, 2008). Drift-X model is a probabilistic simplified Gaussian atmospheric dispersion model able to forecast pesticide spray drift after the treatment, from the plot to the watershed scales. The model operates within a domain of a several square kilometers, corresponding to a typical southern French small wine- growing area. Drift-X is based on a reduced-order modeling approach to flow field reconstruction with a small number of measurements, as well as to Gaussian plume transport over realistic topographies and unsteady wind flows (Bozon et al, 2009). The main goal of Drift-X is to provide the mean trajectory of a pesticide cloud after spraying applications, by forecasting the wind field and the pesticide concentrations for a permanent state. The wind flow is calculated according to the parameters and the DEM layer provided by the user. A user- defined quantity of pesticide is then transported and deposited according to the wind flow. Here is an example Drift-X simulation based on the following parameters and displayed as common vector and raster formats in figure 3 and 4. – The domain for calculation is 8km 2 . – The cartographic projection is extended Lambert 2 (EPSG:27572) – The number of points for the output grid is 900. – The used input DEM layer is SRTM 90m resolution. – The source plot is 1 ha with 33 rows to treat. – The sprayer treats 3 rows at the same time at the average speed of 1 m/s. – The spraying nozzle output velocity is 7 m/s with an output ow of 0.001 kg/s. Two wind points are used to calculate the flow field (N 60 - 5 m/s wind , N 30 -4 m/s wind.) The reduced order modeling approach used for the modeling made the programing aspects easier. The equations composing the model have been transcribed in Fortran language, including a local spray drift model, the wind flow calculation aand a travel-time based transport model as routines. The choice of Fortran was made because it is one of the languages that is best suited to compute complex mathematical expressions ((Bozon et al, 2009). The fastness of the Fortran compiler and the mesh free approach allows us to compute the solution in only a few seconds depending on the size of the domain and on the elevation data resolution. Indeed, the DEM values are extracted for the domain and then sent to Drift-X, which computes the solution using the x, y, z triplets as base topography. The results are then written to output files which contain point-based information for the whole domain. The program has not been transformed into a independent GIS class yet as the integrated approach would suggest, but is used as a standalone and fast executable program. Both input and output datasets are communicating with Quantum GIS. Despite the fact that a more integrated GIS oriented ADM class will greatly enhance the coupling, there is a major advantage in the resulting coupling which is that the Fortran program stay independent of the GIS software, which will make any modification in the model easier as we will only need to re-compile the Fortran program. Using Quantum GIS which is based on a robust C ++ API that presents plenty of spatial algorithms and native GIS functions, we could easily integrate the Fortran executable (Bozon et al, 2008). QGIS has been designed according to an extensible plugin architecture. This allows new features and user- oriented functions to be easily added to the application and that's why QGIS offers advanced programming possibilities. Plugins can be created using C ++ or the related Python bindings, which allow a simpler programming environment for developing specic plugins that directly interact with the C ++ source code. The Drift-X plugin was thus developed using the QGIS Python bindings, (Bozon et al, 2008). and its interface is presented in figure 5. Once the Drift-X model was integrated into QGIS, we could use it directly into a geo-referenced framework and run many simulations simply, according to various wind parameters and DEM resolution layers. Other analysis tools provided by QGIS could also been used directly to work on the resulting pesticide clouds layers, and to intersect them with other relevant geodata to tend to risk assessment (i.e intersecting the simulated pesticide cloud with population density layers for example). Although it gave satisfying results, the scientists and the drift-X users soon needed a more generic and user-friendly platform, as it was intended to be used also by non GIS experts. Indeed, the use of the QGIS plugin requires the basic knowledge of the software, and the plugin is also subject to instability if some major changes occur in the QGIS API. The decision to adapt Drift-X to the Web was thus taken, as it appeared to be the simplest way to provide a stable, cross-platform and easy-to-use simulation platform. As the development work done into QGIS was based on the chaining of various processing tasks, the idea of using WPS to reproduce the procedure on the server-side was taken rather early (Fenoy et al, 2009). This was also the best way to proceed to such processing in a OGC standardized way. The use of ZOO1.0 for adapting Drift-X to WPS is presented in the next section, and the whole WPS chaining is then detailed. ZOO Kernel is the heart of ZOO 1.0. It is a server-side C Kernel which makes it possible to create, manage and chain WPS 1.0.0 compliant Web Services, by loading dynamic libraries and handling them on-demand. Thus, it can easily connect to geospatial libraries and scientific models, but also with the common cartographic engines and spatial databases. ZOO Kernel is written in C language, but Web Services can be programmed in C, Python, Java, Fortran, PHP and JavaScript. This multi-language support is convenient for developers and allows above all to use existing code to create new Web Services. Open source GIS libraries or specific code can so be ported server-side with very little modifications (Fenoy et al, 2009). Using ZOO was thus very convenient for our concern, as it supports Fortran code, but also includes server-side ...