Implication of weather factors on mango (mangifera indica L.) phenology and prediction of yield using geospatial techniques

Abstract

The study evaluated mango phenology through three meteorological indices, Growing Degree Days (GDD), Photothermal Units (PTU), and the Standardized Precipitation Index (SPI). GDD and PTU illustrated cumulative heat requirements, highlighting varietal differences in growth rates and maturation. Banganpalli required higher GDD and PTU, indicating higher heat requirement for maturation, while Totapuri and Sindhooram matured under lower heat accumulation. SPI showed negative values, signifying dry conditions during the mango growth cycle. The study utilized remote sensing indices, like Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), Enhanched Vegetation Index (EVI), Land Surface Temperature (LST) and Soil Moisture Index (SMI), to monitor crop health and environmental conditions. NDVI and EVI effectively assessed vegetation health and canopy vigor, with NDVI peaking during crucial growth stages like pea-size fruit and maturity. LST highlighted temperature impacts during fruit maturity and harvest, higher LST delayed phenophase durations during all the phenophases. SMI was particularly useful in identifying moisture-sensitive stages, such as flowering. For yield prediction, correlation analysis and stepwise regression were initially performed to identify key predictors for each phenological stage. In order to achieve better model and prediction accuracy six machine learning models, Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), Least Absolute Shrinkage and Selection Operator (LASSO), Ridge and Partial Least Square Regression (PLSR) were tested with diverse input feature sets (weather variables, agrometeorological indices and remote sensing indices). Among these PLSR achieved the highest R² of 0.93 during leaf bud development (Phase I), while XGBoost performed best during days to marble stage (Phase V, R² = 0.75) and days to maturity stage (Phase VI, R² = 0.83). Ridge regression showed consistent performance across phases, with R² values 0.89 (using only remote sensing indices) and 0.81 (using all input features) during Phase I, in Phase VI, Ridge achieved an R² value of 0.71. The results demonstrated the feasibility of predicting yield using weather variables, agrometeorological indices and remote sensing indices as input data in machine learning models at key growth stages of mango. The findings of this study can empower farmers by providing actionable insights into optimizing mango cultivation based on weather and crop health data. By using remote sensing and machine learning-based yield prediction models, farmers can anticipate key growth stages, adjust irrigation and nutrient management and mitigate risks from erratic weather. This approach can enable timely decision-making, enhance productivity and maximize income of farmers.