Spatial Temperature Prediction - A Machine Learning and GIS Perspective

Abstract

Abstract Temperature plays a crucial role in influencing the growth and development of crops. Rice, being a significant cereal crop, is highly sensitive to both low and high temperature stresses. Upcoming season temperature forecasts help in site-specific crop management, such as assessing crop growth, pest outbreaks, and suggesting heat-tolerant varieties and prediction models. This paper focuses on the prediction of temperatures using a combination of GIS and machine learning models. The study involved in the estimation of daily maximum temperatures for India, by utilizing the Random Forest algorithm within a machine learning model. The training and testing dataset encompassed Indian Meteorological Department (IMD) grid-based daily temperature data (307 grid points across 365 days) from 2010 to 2019. In contrast, the data for 2020 was reserved as a validation dataset. Since the daily temperature data is available for only 307 grid points and there are more than 700 districts in India, GIS tool has been used for estimating temperatures at unknown points using spatial interpolation method. QGIS software was employed to handle the spatial interpolation of grid-based data using Thiessen polygons. The model's performance was evaluated using the R-squared (R2) values, which ranged between 0.8 and 0.9, indicating a strong and accurate performance of the model. The model performed particularly well when comparing the predicted temperature values for 2020 with the validation dataset, as evidenced by a correlation coefficient (r2) of 0.88. Furthermore, it demonstrates that machine learning algorithms, combined with GIS models, are capable of spatial temperature prediction when forecasting large-scale temperature patterns across India. The ability to continually update the training data with real-time IMD information enhances self-learning modules and results in more accurate temperature predictions enabling the timely generation site-specific crop management advisories.