In the field of agriculture, optimizing crop yield while conserving resources is a significant challenge, given the variability in environmental conditions and crop responses. Current methodologies for improving crop growth often lack the precision required to accommodate the diverse factors influencing crop yield, such as soil nutrient content, water intake, and weed levels. To address this, we propose an innovative multimodal predictive framework that integrates various modalities, such as crop images, soil nutrient content (pH, organic carbon, cation exchange capacity (CEC), nitrogen (N), phosphorus (P), potassium (K), calcium (Ca), magnesium (Mg), and soil moisture content), water intake, and weed levels. Our methodology employs the transformation of these modalities into frequency, Z-transform, S transform, entropy, and convolutional features, which are then classified using an efficient graph convolutional neural network (GCNN). This novel approach enables the prediction of optimal vesicular–arbuscular mycorrhizal (VAM) fungi dosages to enhance maize crop growth parameters including shoot and root biomass, root length, plant height, leaf mass and area, cob and grain yield, number of grains per cob, and silage yield. When tested on soil samples from Nagpur and its surrounding areas, our framework significantly improved various growth parameters including a 10.4% increase in shoot and root biomass, 4.9% increase in root length and plant height, 2.5% increase in leaf mass and area, and a remarkable 14.5% increase in cob and grain yield compared to crops grown using traditional methods. The implications of this work are profound, highlighting the potential of our proposed model in revolutionizing agricultural practices by providing a data-driven, precise, and efficient method for enhancing crop yields and resource utilization in agriculture.