Potential benefits of variable rate nitrogen topdressing strategy coupled with zoning technique: A case study in a town-scale rice production system

Integrating remote sensing (RS)-based variable rate nitrogen (N) recommendation (VRNR) algorithms and management zones (MZs) may improve the accuracy and efficiency of site-specific N management. However, its potential benefits for application in commercial rice production systems can hardly be assessed, since it requires to intervene in common agricultural practices and causes certain economic and environmental consequences. Through a machine learning approach, this study aims to comprehensively evaluate the economic and environmental benefits of applying different N management strategies to a town-scale rice production system. Rice field experiments were conducted during 2017–2021 to establish the sufficiency index-based N recommendation algorithm (SIA). The rice fields at Diaoyu town were classified into 12949 rice N management grids. The historical (2020) and in-season (2021) Sentinel-2 images during rice growing seasons were clustered and overlapped to derive RS-based MZs. Five different in-season N topdressing strategies were used to calculate N topdressing rates for each N management grid in the 2021 rice growing season, including: [i] 150 kg N ha−1 of uniform N topdressing rate (Nlocal), [ii] 120 kg N ha−1 of uniform N topdressing rate (Nredu), [iii] SIA-based N topdressing rate (Nrec1), [iv] SIA-based N topdressing rate combined with MZs (Nrec2), [v] average of the recommended N rates of Nrec2 in each MZ (Nrec3). Previously developed random forest models were used to simulate the final yield and field reactive N losses of each grid. Simulation results showed that the Nlocal obtained the highest yield level, while simply reducing 20% N topdressing rate (Nredu) would decrease the net profit (NP). Compared with Nlocal, the Nrec1, Nrec2 and Nrec3 saved N topdressing amounts by 17.9%−22.5%, and reduced N damage cost (NDC) by 6.0%−7.8%; meanwhile, NP, N recovery efficiency (NRE), and partial factor productivity of N fertilizer (PFPN) were increased by 2.3%−3.0%, 4.9%−6.0%, and 10.6%−14.3%, respectively. Comparably, the performance of Nrec2 was the optimum among the three VRNR approaches. Overall, through the synthetical evaluation of different N topdressing scenarios, this study demonstrated that combining VRNR algorithm and RS-derived MZs may help regional rice production systems achieve higher N utilization efficiency and economic benefits, along with lower environmental risks.