WHEAT YIELD MODELLING FOR PRE-HARVEST FORECASTING

Abstract

Wheat is one of the most important crops and is used in a variety of food products. It is also a widely grown crop not only in India but in many nations. Prediction of wheat yield at the farm, district or regional levels and at other larger scales has been the subject of several investigations in the past. Earlier studies related crop yield with agrometeorological variables, sometimes even at weekly intervals. Subsequently, there was the emergence of models which extensively used satellite products and several vegetation indices were used either to develop yield models only based on the vegetation indices, or involving vegetation indices and agrometeorological variables. There were also attempts to use physiological parameters in crop yield modeling. The use of a large number of physiological parameters often limits the use of such models in Indian conditions. Similarly, crop water stress index-based relationships have emerged in the literature. These relationships utilize canopy temperature at fully stressed and non-stressed conditions and are reported to work well for other crops. Similarly, various crop models are also available in the literature. Recently, it is suggested to use a combination of methods to improve crop yield prediction. Thus, in this work, farm experiments are done in different phases. Phase 1 experiments are conducted in the agricultural farm of IIT Roorkee in nine plots. In some of these plots, the dose of fertilizer was varied along with the type of fertilizer. This led to wide variation in leaf area and plant height. These experiments were done in 2018-19 and the data of these farm-based experiments are utilized to evaluate the efficacy of two crop models, i.e. InfoCrop and DSSAT. Chapter four provides the comparative evaluation of these two popular crop models. The type of agreement between simulated and observed crop yield indicated to explore the potential of other models. Subsequently, in Chapter 5, a yield model using only two physiological variables, i.e. Leaf Area Index, and plant height is used to explore the development of a new model. Such type of modeling work is not yet reported in the literature and because of the involvement of only two variables, it offers much-needed simplicity in the model development and its use in wheat yield forecasting. Wheat yield is forecasted at 60, 90, and 120 Days After Sowing (DAS). Very good simulations of the wheat yield are achieved using these types of models. In Chapter 6, another type of crop yield modeling based on the Crop Water Stress Index is attempted. To assess the potential of CWSI in crop yield modeling, phase two experiments were planned. These experiments were done in year 2018-19 and 2019-2020. Wheat was grown in ten plots. Some of these plots were well irrigated so that there was no stress due to moisture limitation. In some of the plots, stress due to soil moisture was imposed. Canopy temperature and the air temperature were recorded in each of these plots. The data of canopy and air temperature was used to develop CWSI. Subsequently, the yield was related to CWSI. In the range of data collected. Yield correlated with CWSI very well. Thus, if such relationships are available in any region or farm and one has the idea of how CWSI will evolve till crop maturity, such models can also be used for forecasting wheat yield at different stages of crop growth. Last in the category was the development of the models at the district level. Considering the simplicity of models, the models were developed based on the monthly values of agrometeorological parameters. In the process of model development, only one type of Vegetation Index, i.e. Normalised Difference Vegetation Index, is utilized. In the literature, there are wide views regarding the use of appropriate values of NDVI. Some studies prefer to use integrated values of NDVI at the stage of crop forecast while there are studies that suggest using the values of NDVI prevailing at the forecast stage. A model, known as the INSEY model and described in this work, uses an index based on the rate of NDVI development during a particular period of crop growth near maturity. In Chapter 7, various models are evaluated and it has been observed that it is possible to forecast crop yield at 60, 90, and 120 DAS. It is also seen that once the period of calibration increases, the forecasts tend to improve. To study pre-harvest forecasting, three districts lying in three different agro-climatic zones of northern India are considered. The use of vegetation index (NDVI) along with meteorological variables is found to work well. In this approach, only monthly values of temperature and precipitation are used. Towards this, minimum and maximum temperature along with other defined threshold temperatures are used. The crop yield prediction has been a widely studied topic and it has not been feasible to address all the prevailing modeling approaches in this. Rather the intent has been to develop simple models requiring a lesser number of model parameters and fewer data. It is expected that some of the approaches explored here will add to the subject of wheat yield forecasting in a meaningful manner.