Abstract: Carbon emissions from the planting industry are a significant source of greenhouse gas emissions. The accurate prediction and effective management of these emissions are crucial for mitigating climate change and promoting sustainable agricultural development. Conventional prediction models exhibit a limited capability to capture the complex nonlinear interactions inherent in carbon emission systems of the planting industry, and their insufficient robustness often leads to overfitting. In this study, we used the planting industry in Heilongjiang Province as a case study to explore how to optimize existing methods for predicting carbon emissions from the planting industry. First, the IPCC carbon emission method is applied to comprehensively account for three major sources of carbon emissions: carbon emissions from agricultural land use, CH₄ emissions from rice field, and N₂O emissions from agricultural land. Carbon emissions from planting activities in Heilongjiang Province from 2001 to 2022 were systematically calculated. Based on this, a long short-term memory (LSTM) network model was developed, incorporating three key dimensions: social and economic drivers, production scale effects, and technical energy consumption intensity. To enhance the predictive performance of the model, an improved whale optimization algorithm (IWOA) was introduced to optimize four hyperparameters of the LSTM model: number of hidden units, learning rate, batch size, and training epochs. Then, the IWOA-LSTM model was used to predict future carbon emissions from the planting industry in Heilongjiang Province from 2023 to 2027 under both baseline and low-carbon scenarios. The results were showed as below. 1) Carbon emissions from the planting industry in Heilongjiang Province showed a trend of “rapid growth followed by a fluctuating decline”, reaching a peak of 20.45 million t in 2015. The main sources of carbon emissions included CH₄ emissions from rice field, N₂O emissions from agricultural land, and carbon emissions resulting from fertilizer production and application; their average proportions in the total annual emissions were 41.42%, 38.26%, and 11.65%, respectively. 2) Compared with the unoptimized LSTM model, the IWOA-LSTM model demonstrated significant improvements in both the prediction accuracy and stability. It achieved a mean absolute error of 55.82×10⁴ t, root mean square error of 61.74×10⁴ t, and mean absolute percentage error of 2.83%, all of which were superior to those of the LSTM model (114.41×10⁴ t, 124.72×10⁴ t, and 5.78%). In this study, we demonstrated that the IWOA-LSTM model could effectively predict carbon emissions from the planting industry, thereby providing a scientific basis for the formulation of carbon reduction policies for the planting industry in Heilongjiang Province. 3) The prediction results of the IWOA-LSTM model showed that carbon emissions from the planting industry in Heilongjiang Province could be effectively suppressed by controlling the crop planting area, improving fertilizer application efficiency, and reducing diesel consumption per unit area of agricultural machinery. Based on the aforementioned conclusions, the following recommendations for emission reduction are proposed: optimizing land-use structure and controlling crop planting area, increasing the application and innovation of green agricultural technologies, promoting rural economic development, increasing farmers’ income, and strengthening policy support and incentive mechanisms. Through the above measures, the sustainable development of agriculture in Heilongjiang Province can be further achieved.