Abstract The research aims to explore the application of the deep residual network-spatio-temporal fusion model in the estimation of net primary productivity. The deep residual network-spatiotemporal fusion model is an advanced deep learning model specifically designed to process and fuse remote sensing images from different time points to generate synthetic images with higher spatiotemporal resolution. The study first collected remote sensing data at different time points within the research area, and then processed these data using the deep residual network-spatio-temporal fusion model to reconstruct images with higher resolution. Then, by analyzing the features in these high-resolution images, the model estimates the net primary productivity of the study area. The research results indicate that the deep residual network model combined with the Carnegie-Ames-Stanford model was used to estimate the net primary productivity of vegetation. The mean value of the proposed deep residual network model in estimating net primary productivity of vegetation is 455 gC/m 2 /month, with a standard deviation of 25 gC/m 2 /month, and the coefficient of determination R 2 reaches as high as 0.98, which is significantly better than other algorithms. Compared with the mean value of net primary productivity of vegetation measured on the ground in the study area during the same period of 448 gC/m 2 /month, the deviation is only 7 gC/m 2 /month, with a deviation rate of 1.56%. This further verifies the accuracy of the model’s estimation results. The deep residual network-spatio-temporal fusion model can enhance the accuracy of net primary productivity estimation, providing a new tool for ecological research and environmental monitoring.
Yuan Sun (Tue,) studied this question.