Paper Title
FOREWARNING MODELS FOR PESTS & DISEASES USING CLIMATIC VARIABLE AND DECISION SUPPORT SYSTEM THEREOF

Abstract
Reliable and timely forecasts provide essential support for proper, foresighted, and informed planning, particularly in agriculture, which is inherently vulnerable to various uncertainties. Modern agriculture has become highly input- and cost-intensive. Without the judicious use of fertilizers and plant protection measures, farming is no longer as profitable as it once was. Uncertainties related to weather, production, government policies, and market prices often place farmers under severe financial stress, sometimes leading to tragic consequences such as farmer suicides. In addition, the emergence of new pests and diseases poses an increasing threat to agricultural production. Under these changing circumstances, forecasting various aspects of agriculture has become indispensable. However, despite the strong need for reliable and timely forecasts, the present forecasting system remains inadequate and unorganized in many sectors. A well-tested weather-based model can serve as an effective scientific tool for the advance forewarning of insect pests and crop diseases, enabling farmers to undertake timely plant protection measures. For quantitative data, models were developed to forecast the time of first appearance of pests or diseases, the timing and magnitude of maximum disease severity or pest population, and age-wise, standard meteorological week-wise, or year-wise pest and disease incidence. These models utilized weekly weather data. For each weather variable, two indices were developed: 1. A simple total index representing the cumulative value of the weather parameter over different weeks, and 2. A weighted index, where weights were assigned based on the correlation coefficients between the forecast variable and the weather parameter in the corresponding weeks. The first index reflects the total amount of a weather parameter received by the crop during the period under consideration, while the second accounts for the distribution and relative importance of the parameter across different weeks. Similarly, indices based on the products of two weather variables were computed to study their joint effects. In some cases, previous disease incidence or pest population data (or their indices), as well as the previous year’s final pest population, were also included as independent variables. In general, the models showed good fit for the available datasets, with highly significant coefficients of determination. These models were capable of providing reliable forewarnings at least two weeks in advance. For qualitative data, logistic regression models were developed to forecast the occurrence or non-occurrence of pests and diseases by classifying observations into two categories: occurrence and non-occurrence. To describe pest development patterns during the crop season, non-linear models using time as an independent variable were also developed. These models were further refined by replacing time with suitable functions of both time and weather parameters, thereby capturing fluctuations in pest populations more accurately. For forecasting aphid populations in advance, additional models were developed using weather data lagged by one week. Modelling approaches with the capability to learn from experience are highly useful for solving practical agricultural problems, provided sufficient data are available. Remotely sensed data can also provide valuable information regarding crop area and crop condition. Compared to conventional land-use statistics, remote sensing offers several advantages, including multispectral, synoptic, and repetitive coverage. With the advent of computers, more advanced techniques based on machine learning such as decision tree induction algorithms, genetic algorithms, neural networks, rough sets, etc. have been explored. The developed models were further integrated into web-based forecasting systems for the timely forewarning of pests and diseases.