Spatio-temporal variation and impacting factors of NPP from 2001 to 2020 in Sanjiangyuan region, China: A deep neural network-based quantitative estimation approach

Abstract

The ecosystem in the Sanjiangyuan region of China is known to all for its simplicity and fragility. Studying the changes in net primary productivity (NPP) of vegetation and its influencing factors will help protect and improve this fragile environment. However, there are missing data in some regions in NPP data collectors, which affect the data availability. Therefore, this article proposes a method for estimating NPP using only a small amount of basic data by constructing a deep neural network (DNN) model suitable for NPP estimation. This method requires only a small amount of basic data to obtain high-precision, complete NPP data. We conducted a precision verification of the NPP estimates in the Sanjiangyuan region and studied the temporal and spatial variation of NPP in the Sanjiangyuan region, the relations between NPP and meteorological factors, and the contributions of climate change and human activities to NPP changes. The results showed that the estimation of NPP was reliable, with coefficient of determination (R2) ranging from 0.893 to 0.933 and root mean square error (RMSE) ranging from 24.904 to 34.550. From 2001 to 2020, the NPP in the study area showed a significant increasing trend (1.32 g C/m2·year−1, P < 0.05). The partial correlation coefficients between climate and NPP were all positive, with the maximum partial correlation coefficient between temperature and NPP (0.308), followed by radiation (0.113) and precipitation (0.067). The contributions of climate change and human activities to NPP changes were 1.298 and − 0.051 g C/m2·year−1, respectively. Climate change was the main driving factor for vegetation regeneration, while human factors were the main cause of vegetation reduction in the study area.