Procedure

Direct monitoring

Direct monitoring sought to verify the presence or absence of insects, diseases and other factors that can limit or optimize the productivity of the crop.

Block monitoring was carried out by a trained monitor, who moves around the bed and directly observes the presence of pests, diseases and natural enemies in different parts of the bed, as well as the plant strata, with the help of implements such as magnifying glasses, sticks, bags to collect plant material, among others. This monitoring was carried out once a week, checking 100% of the beds on one side only; the following week the opposite side was covered and, thus, in 2 weeks, 100% of the area has been checked. Beds generally have an agronomic length of 32 linear meters, but there are some beds that, due to the topography of the site, are shorter (Tab. 1). Depending on the length of the bed, the product and the monitored object, the points where the work is to be done were defined, usually two fixed points at the beginning and end of the bed, and three additional points distributed along the bed for a minimum total of 5 points. In each monitored point, an area of approximately 1 m² and the three thirds of the plant (upper third, middle third, lower third) was checked.

TABLE 1 Characteristics of sampled beds. 

The presence of any pest or disease was recorded in the monitoring application. At the end of the check (only for some pests and diseases defined by the technical department), the beginning of the bed was marked with ribbons with the color corresponding to the biological target.

To ensure the process, the person in charge was checking 5 positive and 5 negative beds for each monitor and verifying what was reported in each table by the monitor. The quality level of direct monitoring had to be greater than 90%, obtained through the assurance.

Indirect monitoring

Indirect monitoring sought to detect the presence, absence or migration of pests or diseases out of the bed, through different strategies such as:

a) Wet chamber. This monitoring consists of an assembly of flowers of six each of the species and/or varieties (67 in total) in a space saturated with relative humidity, which makes it possible to identify the presence and reproduction of the disease. The reading in a humid chamber is carried out according to the frequency established by the work center, where the presence of the disease is evaluated in all samples, and those affected are eliminated. The results of the trace are recorded in a form;

b) Pheromone and kairomone traps. These traps (50 traps in 32 ha, 1.56 traps ha^-1) release a pheromone or sexual attractant, which attracts the adults of the species (moth, fly, or thrips) to the trap. Diffusers are designed to reproduce a proportion of the insect's own attractive chemical components, but with a higher emission rate. The detection and count of captured insects are carried out weekly to identify migrations or increased pest pressure on the farm;

c) ICA traps. These consist of placing kiosk traps at the four cardinal points of the farm determined by the ICA (4 traps for 32 ha, 0.125 traps ha^-1), used specifically for monitoring thrips and especially Thrips palmi, reading the trap tubes located there. The reading is carried out twice a week and is recorded in the ICA "form for the weekly report of detection of thrips". This information must be registered on the platform https://sisfito.ica.gov. co/

According to the schedule, the vials (sample of captured thrips) are sent to the ICA for analysis.

Monitoring plates

Internal plates

To quantify the presence of pests in the crop, the farms must place plaques at the end of or interspersed in the block, using colors depending on the biological target. The plates should be read weekly by the monitor in charge of recording, and immediately cleaned and prepared for re-reading. These readings must be entered into the Colibri app for indirect monitoring. This application is proprietary software developed by the company Sunshine Bouquet S.A.S.

External plates were placed as follows: plate No. 1 near the entrance of the farm and the following plates every 100 m according to the clockwise movement, until the periphery of the farm is covered (around the blocks). Each plate was located on a 2 m long pole, and this should have a height above ground level of 1.5 m, located at a maximum of 1.5 m from the greenhouses. The plate was identified with the abbreviation EXT and the number corresponding to its distribution. The identification had to be placed perpendicularly on the top of the mast.

The plates were read every 7 d on one side only (external side facing the periphery of the farm), and the number of individuals of each pest trapped per plate was counted. Once the reading of the plate was finished, it was removed and replaced by a clean plate. The information from the reading was recorded in the form (with the average number of individuals per plate/pest/week) and recorded electronically.

Model analysis

Pearson's correlation coefficient and multiple linear regression analysis with equations were used.

The following ranges were used for the correlation coefficient:

TABLE 2 Correlation coefficient ranges (adapted from ^{Hernández-Lalinde et al, 2018}). 

Using multiple linear regression in R, we obtained an expression of the form

where:

% Pathogen = % incidence of the pathogen tested;

X = response variable "Average evaporation/week";

Y = response variable "Average temperature/week";

Z = response variable "Average relative humidity/ week";

W = response variable "Average precipitation/week";

β ₀ = Intercept; β ₁ , β ₂ , β ₃ , β ₄ = slopes corresponding to variables X, Y, Z, and W, respectively.

Utilizing the lm function in the R software for the statistical analysis of the database, we performed the respective multiple linear regression to see the behavior of the response variables with regard to the predictor variables. The result of this function provides an equation that allows determining the prognosis or future behavior of each of the pathogens (Tab. 3).

Example equation for prediction of mites

TABLE 3 Resulting multiple linear regression in the statistical software R. Example is given to predict percentage of mites. 

FIGURE 1 Correlation coefficients between each response variable vs. predictor variable, expressed in percent. 

Predictor equations obtained by means of the R statistical software

For a reliable statistical analysis, a relevant database is obtained to predict the behavior of the different pests and diseases that occur during the development of the rose crop.

This analysis used the multiple linear regression method, which allows us to take more than one independent variable to see the behavior of the response variable; in this case, the independent variables are the climatic conditions, and the response variables are each of the pests and diseases that occur in the crop.

A database was used to record the climatic conditions and the presence of pests and diseases in the Betania farm.

By means of the statistical software R and the statistical method of multiple linear regression, the following equations are proposed as a contribution of this research with the objective of predicting the increase of diseases and pests in rose crops.

These results were similar to the equation proposed (Eq. 9) in the exponential increasing model of ^{González (2013)}, where the variables were the age of the crop in days of planting, the maximum, minimum and average temperature and the maximum, minimum and average relative humidity 7 d before each evaluation. This corroborates that there is a directly proportional relationship between pests and diseases with the behavior of climato-logical variables.

FIGURE 2 Incidence frequency values for each variable and its expression in different ranges. 

where

Tmax = Maximum temperature;

Tmed = Medium temperature;

Tmin = Minimum temperature;

Rhmax = Maximum relative humidity;

Rhmed = Average relative air humidity;

Rhmin = Minimum relative air humidity.

In the different predictor equations, verifications were made by replacing the predictor variables with values already observed to predict the percentage of incidence of each of the pests and diseases and to compare with the values reported.

The monitoring obtained very close values, with differences of below 5% for the downy mildew, botrytis, mites and thrips, and below 10% for powdery mildew.

Table 4 shows the values of the predictor variables provided by the Canal Clima platform (^{CANAL CLIMA S.A.S., 2023}), Table 5 shows the incidence percentages calculated with the predictor equations, and Table 6 shows the incidence percentages reported in the monitoring. These results show that the objective has been achieved, namely, to predict the incidence of a pest and disease based on the values of the predictor variables with an acceptable error percentage of +/- 5%.

The week number corresponds to the week of the year in which the evaluation was carried out.

TABLE 4 Values of the predictor variables provided by the Canal Clima platform. 

TABLE 5 Incidence percentages calculated with the predictor equations. 

TABLE 6 Incidence percentages reported in monitoring. 

Although the correlations between pathogens and predictor variables are moderate or weak, the latter have a significant influence when predicting the percentage of incidence of each pathogen by replacing observed values of the predictor variables in the equations obtained.

Correlation tests are useful to identify the trend of the response variables as a function of each of the predictor variables, which does not necessarily mean that when two variables correlate with each other, it is due to a cause and effect relationship; these variables may be independently related to a third variable.

While this analysis obtained satisfactory predictions, the amount of information in the database should be increased in order to make a more accurate prediction, considering that the data used in this research were limited.

Agriculture has improved due to the use of intelligent agricultural machines for crops and livestock control, intelligent monitoring drones for pest control, as well as to observation sites that record agro-climatic variables to collect large amounts of data at the lowest costs. This has led to more information-oriented, potentially more productive, and efficient agricultural processes.

The above results indicate the need to deepen understanding of the incidence of diseases in the face of the climate changes that are manifesting themselves by constantly feeding the database and thus obtaining better predictions. Further studies to address the subject could be done using different methodologies, such as the application of Machine Learning and Deep Learning, thus creating an autonomous learning algorithm that can make it even more efficient in its predictions.

From the practical point of view, it is important to correct some aspects, to undertake improvements and to include new elements of interest for the solution to the problem addressed.