Estimating Leaf Chlorophyll Content of Crops Using Multispectral Remote Sensing and Machine Learning Models

Abstract

Tomatoes (Solanum lycopersicum L.) are the second most cultivated vegetable in the world, contributing significantly to global production and offering a wide range of health advantages and nutrients. Precise and timely measurement of leaf chlorophyll content (LCC) on a small scale spatial level is paramount for monitoring a plant's photosynthetic ability and nutritional health. Hence, this study used LCC to monitor the dynamic changes in tomato crops using various machine learning (ML) models. This study aimed to estimate LCC in tomato crops by leveraging ML models, specifically DTR, RFR, PLSR, and XGB. The models were developed in two phases: Before-Feature-Selection (BFS) and After-Feature-Selection (AFS). In BFS, twenty VIs were used as predictors, while in AFS, the three most relevant VIs for LCC estimation were determined through Pearson's correlation coefficient and the Gini index. Noteworthy findings identified GNDVI, GRVI, and GCI as pivotal VIs, demonstrating a moderate and positive correlation with LCC. The evaluation and assessment of the machine learning models were conducted utilizing leave-one-out cross-validation (LOOCV) and statistical metrics such as R2, MAE, and RMSE. The XGB model revealed the highest R2 of 0.53 and 0.54, RMSE values of 9.54µg/cm2 and 9.36µg/cm2 and MAE values of 6.39µg/cm2 and 6.04µg/cm2 in both phases. In contrast, the PLSR exhibited the lowest R2, with the highest RMSE values of 11.94µg/cm2 and 11.86µg/cm2 and MAE values of 8.79µg/cm2 and 8.55µg/cm2 in both phases. The results revealed that the AFS models demonstrated higher R2 than the BFS models. The study also generated chlorophyll maps using the ML models, capturing the spatial dynamics of leaf chlorophyll within the tomato field. Therefore, this study highlights the potential of multispectral UAVs combined with ML for LCC estimation, offering valuable insights for decision-making and management in small-scale farms.