Prediction of tomato plant disease with meteorological condition and Artificial Intelligence

Abstract

Climate change is one of the most challenging problems that humanity has ever faced. Weather plays an important role in the field of agriculture. Precise prediction of crop disease and continuous monitoring are crucial to improving crop yield in agriculture. Processes for disease development and weather parameters have been well understood and used to develop prediction models and decision support systems [1]. Due to variable environmental conditions, an efficient technique is required to facilitate crop cultivation and to help farmers produce and manage. The loss due to disease in plants can be reduced with the proper management step and achieves sustainability and food security. This could help future farmers to improve agriculture. A farmers can be provided with a system of recommendations to assist with data mining in crop growing processes. Plants based on climatic factors and quantities are recommended to implement this approach [2]. The parameters responsible for the occurrence of disease in tomato plants are maximum and minimum temperature, dewpoint, relative humidity, and rainfall. The temperature in the range 22-38°C and relative humidity in the range 55 to 90%  are favourable for fungal diseases in tomato plants  [3]–[7].   The bacterial disease in tomato plants occurs favourably in the temperature range of 24-32°C with a relative humidity of more than 80% [8]. The mosaic virus in tomato plants occurs in the temperature range 21-31°C with a relative humidity of 55-70% [9] and yellow leaf curl virus occurs when the temperature rises above 40°C [10].

While predictive methods develop quickly to more complex statistical and physical models, routine predictions still prefer the use of simple and traditional methods [11]. In this work, time series meteorological data of the Pune region of India is selected from the year 2009 to 2019. The data consists of information of maximum and minimum temperature, humidity, rainfall, dewpoint, wind, etc. The relative humidity is a vital parameter in the evaluation of favourable conditions of disease in tomato plants. Relative humidity [12] is calculated as shown in equation (1). It is the ratio of saturated vapor pressure and actual vapor pressure.

                      Relative Humidity=(E/Es)X100  (1)

Where saturated vapor pressure (E)= 6.11X 10X (7.5 X dewpoint/(237.7 + dewpoint))

And actual vapor pressure (Es)= 6.11X 10X (7.5 X temperature/(237.7 + temperature))

The weather data of 11 years shows the seasonality of the parametric values in the range. Fig 1 shows the seasonal change in the temperature values of the dataset. Fig 2 shows the relative humidity values over the years. As relative humidity is dependant on temperature value, the outlier values in temperature affect the relative humidity value.

Fig.1. Temperature values showing the seasonal change over the years 2009-2019

Fig. 2. Relative humidity values showing the seasonal change over the years 2009-2019

In this work, Support vector regression and Random Forest regression algorithms are used for the prediction of favourable conditions for the occurrence of disease in the tomato plant. The dataset is divided as a training dataset- the testing dataset is in the combination of 70-30%. Tunning of hyperparameters gives better performance of the model and more accurate predictions can be attained [13]. The calibration of the algorithmic parameters is carried out with the GridSearchCV optimization is used for tunning the hyperparameters to fit the model and achieve better performance. The Root Mean Square Error (RMSE) measures the discrepancy of forecasted yield around observations [14]. RMSE is calculated as shown in equation (2).

          (2)

Where N is the number of samples.

Table 1 shows the RMSE of the Support vector regression and random forest regression model with the effect of GridSearchCV optimization. The RMSE values are improved after the optimized hyperparameter values are set.

Table 1. RMSE of the Support Vector Regression and Random Forest Regression

Model

RMSE

RMSE with optimization

Support Vector Regression

1.1978

0.0301

Random Forest Regression

0.0001

0.000003

The RMSE value for Support Vector Regression is improved from 1.1978 to 0.0301 and for Random Forest Regression, the RMSE value is improved from 0.0001 to 0.000003. the comparative result shows that the Random Forest Regression model is performing outstandingly in prediction and can be used further to detect the favourable condition of disease occurrence in tomato plants. This information can be used by farmers to predict the upcoming weather conditions and take the necessary step towards the management of the disease.