Spatial temperature prediction—a machine learning and GIS perspective

Temperature and humidity have a significant impact on the growth and development of crops. Rice, an important grain crop, is very sensitive to temperature and heat stress. It helps in crop management in specific areas like monitoring temperature for the upcoming season, measuring crop phenology, pest outbreaks, suggesting heat-tolerant varieties and prediction models. This article focuses on the use of a spatial machine learning model to predict temperatures in India. The study employs a geographic random forest (GRF) algorithm to predict daily maximum and minimum temperatures. The training and evaluation data consist of grid-based daily temperature records from the India Meteorological Department (IMD) spanning 2009 to 2018, covering 307 grid points over 365 days. The 2020 IMD data is reserved as the validation dataset. Python was used to download the grid data and merge the ten years of data into a single CSV file. QGIS software trimmed the outer layers and extracted the Indian boundary map for further analysis. The R statistical environment estimated missing values in the grid data and generated spatial temperature prediction maps using the GRF model. The model's performance was measured with R-squared (R 2 ) values between 0.8 and 0.9, indicating high accuracy. When comparing the 2020 temperature forecast with actual data, the model achieved R 2 values between 0.6 and 0.88. Rice phenology was estimated in terms of cumulative growing degree days (GDD °C days) in four major rice-growing areas. The performance was assessed using the D-Index, with values ranging from 0.91 to 0.95, indicating high accuracy due to their proximity to 1. The study also shows that spatial machine learning algorithms can effectively predict large-scale temperatures across India. The ability to continuously update training data with real-time IMD data enhances self-learning and provides more accurate temperature predictions, enabling timely crop management recommendations.