Gross primary productivity (GPP) measures the carbon exchange of any terrestrial ecological system with the atmosphere; it serves as a crucial factor in the overall carbon cycle of the planet and contributes as a significant factor in climate mitigation. However, existing process-based models for predicting this carbon flux have several caveats and substantial uncertainties. To address this challenge, we explore the potential of deep learning techniques, which offer innovative solutions for complex and nonlinear problems in forest ecology. Our study represents an elementary effort to compare different deep learning models for predicting daily GPP. The foundation of our approach lies in a deep learning network that combines long short-term memory (LSTM) and gated recurrent unit (GRU) networks. This integrated model considers a time series of climatic and meteorological parameters derived from the European reanalysis data and leaf area index (LAI) biophysical satellite parameters as input variables. Our study focuses specifically on the Western Ghats region in India, chosen as the study site due to its ecological significance. Through extensive analysis, we demonstrate that our hybrid LSTM-GRU model surpasses individual LSTM and GRU models in terms of performance with a coefficient of determination (R2) score of 0.87 and a root mean squared error (RMSE) was found to be 4.81 g C m −2 d −1 • Our results are useful for obtaining precise GPP estimations, and models can be used for predicting the same.