Accurate classification of land cover data can facilitate the intensive use of urban land and provide scientific and reasonable data support for urban development. Rapid changes in land cover due to economic growth are occurring in the megacities of developing countries more and more. A land cover classification method with a high spatiotemporal resolution and low cost is needed to support sustainable urban development for continuous land monitoring. This study discusses better machine learning algorithms for land cover classification in Ho Chi Minh City. We used band combination 764 and band combination 543 of LANDSAT8-OLI image data to classify the land cover in Ho Chi Minh City by combining three machine learning algorithms: Back-Propagation Neural Network, Support Vector Machine, and Random Forest. We divided the land cover into six types and collected 2221 samples, 60% of which were used for training and 40% for validation. Our results show that using the band combination 764 combined with the Random Forest algorithm is the most appropriate, with an overall classification accuracy of 99.41% and a Kappa coefficient of 0.99. Moreover, it shows a more significant advantage regarding city-level land cover details than other classification products.

The spatiotemporal patterns and shifts of net ecosystem productivity (NEP) play a pivotal role in ecological conservation and addressing climate change. For example, by quantifying the NEP information within ecosystems, we can achieve the protection and restoration of natural ecological balance. Monitoring the changes in NEP enables a more profound understanding and prediction of ecosystem alterations caused by global warming, thereby providing a scientific basis for formulating policies aimed at mitigating and adapting to climate change. The accurate prediction of NEP sheds light on the ecosystem’s response to climatic variations and aids in formulating targeted carbon sequestration policies. While traditional ecological process models provide a comprehensive approach to predicting NEP, they often require extensive experimental and empirical data, increasing research costs. In contrast, machine-learning models offer a cost-effective alternative for NEP prediction; however, the delicate balance in algorithm selection and hyperparameter tuning is frequently overlooked. In our quest for the optimal prediction model, we examined a combination of four mainstream machine-learning algorithms with four hyperparameter-optimization techniques. Our analysis identified that the backpropagation neural network combined with Bayesian optimization yielded the best performance, with an R² of 0.68 and an MSE of 1.43. Additionally, deep-learning models showcased promising potential in NEP prediction. Selecting appropriate algorithms and executing precise hyperparameter-optimization strategies are crucial for enhancing the accuracy of NEP predictions. This approach not only improves model performance but also provides us with new tools for a deeper understanding of and response to ecosystem changes induced by climate change.