The motivation behind the work lies in a fundamental tension in pesticide fate modelling. Highly detailed models such as PEARL, which is coupled to the hydrological model SWAP, can describe the fate of pesticides in the plant-soil system with great precision, calculating changes in concentrations as affected by various physical and chemical processes. But these models simulate dynamics at a point location, and when the goal is to predict how organisms moving through a landscape are exposed, precision matters less than accuracy and tractability. Simulation runs over many years, typically thirty, with a fine temporal resolution of one day, and calculation time must be reduced to make the model workable. This is particularly the case when assessing effects at higher organisational levels, such as populations, where the precise exposure of individuals is not critical.

The original ALMaSS pesticide model, dubbed ToxImpact and introduced for modelling pesticide effects in skylarks, considered the pesticide amount in only one compartment, a single total environmental amount that decayed following a fixed rate. Refinements introduced temperature-variable decay rates, but the model still tracked only a single amount. The blueprint for the current expansion came from the EFSA Panel on Plant Protection Products and their Residues, which wished for separate but linked vegetation and soil compartments. Further development added amounts of pesticide inside plants and differentiation between plant parts to support the evaluation of pesticide impacts on pollinators. The new model fulfils these feature wishes with a configurable hierarchy of complexity.

At its simplest, the model considers only a single compartment, the 1-compartment model. A more complex version splits the pesticide between the plant canopy and the soil, the 2-compartment model, which also considers rain wash-off from the canopy to the soil. The 3-compartment model adds an additional compartment inside the plant, and it is the amount in this in-plant compartment that is used to calculate the pesticide concentration in pollen and nectar, given simply by the amount of pesticide divided by the green biomass times a pollen or nectar specific partition coefficient. Building on the 3-compartment model, turning on seed coating adds another compartment and enables additional transfer, effectively resulting in a 4-compartment model. In practice, a pesticide map is added for each compartment per pesticide, and the maps can have the same resolution as the landscape or coarser.

Pesticides can be applied in three ways: sprays, granules and seed coating. Sprayed pesticides use the most complicated application model since it must consider drift caused by wind and the division of the pesticide between the plant canopy and the soil. Granular application is assumed not to experience drift and is applied only to the soil compartment map based on the application rate. For seed coating, the pesticide is added to the seed coating compartment map at the time of sowing and from that time on can decay and be transferred to the other compartments. All three application events are managed by adding them to an event queue, which is executed once per day when the weather allows. Each event includes information on the pesticide type, application rate and the landscape element to be treated, typically a field.

The drift model is the most technically demanding part. Most available drift data, such as the studies by Rautmann and Ganzelmeier, investigated the total drift outside a whole field, assuming the field is broad enough that pesticide sprayed at one side leads to negligible drift on the opposite side. But ALMaSS simulates complex field geometries, including very narrow stretches of field, so the researchers needed to know the drift caused by spraying a single square metre to avoid overestimating drift outside those parts. Here the studies on drift from a single nozzle by Stallinga and colleagues proved more applicable, providing ground deposit from a single nozzle one metre upwind and ten metres downwind for two different forward speeds and seventeen different nozzles. These data were fitted to a power function relating the pesticide proportion of the application rate to the distance from the nozzle, and the resulting drift vector is used to diffuse the pesticide in a temporary map to its surrounding cells, especially along the wind direction. In ALMaSS, drift is assumed to happen up to ten metres along the wind direction and one metre upwind and perpendicular to the wind direction, an assumption supported by studies comparing drift measurements with detailed simulations of the spray cone.

The derived drift vectors reveal striking variation between nozzles. At a distance of one metre from the nozzle, the fraction of applied pesticide varies by roughly one order of magnitude, and at ten metres it varies almost by two orders of magnitude. The amount at the spraying point itself is the amount that has not drifted, calculated as one hundred percent of the applied amount minus the sum of the upwind drift and three times the downwind drift, which also accounts for the amount going perpendicular to the wind direction. This assures that the intended amount of pesticide is spread in the landscape. Once drift is handled, the last step of spraying is distributing the pesticide between the compartments. For the 2- and 3-compartment models, it is shared between the plant canopy and soil using Beer’s Law to calculate the canopy cover, where the fraction of the surface covered by the crop is given by one minus the exponential of minus the extinction coefficient times the leaf area index, with a default extinction coefficient of 0.6.

Once applied, the pesticides are assumed to undergo first-order decay every day, with a temperature-dependent half-life given by the half-life at twenty degrees Celsius multiplied by the exponential of a fixed coefficient times the difference between twenty and the average temperature on the given day. The half-life can vary for different pesticides and compartments and is specified in a configuration file, with a default value of ten days. The remaining amount of pesticide in each cell is checked against a user-defined threshold for infinitesimally small values and set to zero, and a flag prevents running the computationally heavy loop over all cells when there is nothing to decay. The amount of pesticide in the plant compartments can also decrease or be completely removed due to farm management events like harvesting or ploughing, and the in-plant amount decreases when green biomass transforms to dead biomass, ensuring that only the living part of the plant, which is able to transfer pesticide into pollen and nectar, is considered.

Transfer between compartments follows well-established physical relationships. In the 2-compartment model, the only mechanism is rain wash-off, which depends on the daily gross precipitation and is calculated using the leaf area index and surface cover to determine intercepted precipitation, with an empirical coefficient set to 0.25 millimetres per day for agricultural crops. The wash-off factor itself depends on the water solubility of the pesticide, with a default solubility of 10000 milligrams per litre unless another value is given. In the 3-compartment model, part of the pesticide amount is absorbed from the plant canopy and the soil into the plant, with transfer depending on the green biomass of the plant under the assumption that a large plant absorbs more than a small one. When seed coating is used, two additional mechanisms transfer pesticide from the seed coating to the soil and into the plant. The order of transfers is plant canopy to inside the plant, soil to inside the plant, and seed coating to inside the plant and to the soil, with default transfer rates of ten percent.

Because the pesticide code is very computationally intensive with regard to both CPU time and memory consumption, the researchers built in optimisation options. The resolution of the pesticide maps can be decreased to, for example, a four or sixteen square metre grid instead of the one square metre resolution of the landscape, and the loop over the cells in the pesticide map can be run in parallel on several CPU cores for both the decay and transfer methods, which are among the most time-consuming parts of the code, since the cells are independent of each other. Example simulations confirm the model behaves as described: sprayed pesticide on winter rape quickly transfers from canopy to soil due to rain wash-off while the plant absorbs pesticide from both soil and canopy, and the framework keeps track of several pesticides at the same time. The authors caution that the model is not a one-to-one replica of reality and would not be computationally feasible if it were; it merely aims to cover the main exposure routes that would impact the numerous organisms simulated in ALMaSS. One important missing mechanism is runoff, the removal of pesticides from soil by water flow, which would require implementing a surface water model that has not yet been explored. For the more complex 3- and 4-compartment models, detailed calibration data such as residue measurements in soil, plant, pollen and nectar several times after application would preferably be used; if such data are not available, worst-case estimates would have to be applied.

Subject of Research: Environmental fate modelling of pesticides at landscape scale

Article Title: Modelling dynamic pesticide amounts in multiple environmental compartments at landscape scales in ALMaSS

Article References: Poulsen, T., Duan, X., & Topping, C. (2023). Modelling dynamic pesticide amounts in multiple environmental compartments at landscape scales in ALMaSS. Food and Ecological Systems Modelling Journal, 4, Article e107849. https://doi.org/10.3897/fmj.4.107849

Image Credits: AI Generated

DOI: 10.3897/fmj.4.107849

Keywords: pesticide modelling, ALMaSS, landscape simulation, environmental fate, drift, pollinators, risk assessment, ecotoxicology, Modelling, dynamic, pesticide, amounts