Augmenting agroecosystem models with remote sensing data and machine learning increases overall estimates of nitrate-nitrogen leaching

Abstract

Process-based agroecosystem models are powerful tools to assess performance of managed landscapes, but their ability to accurately represent reality is limited by the types of input data they can use. Ensuring these models can represent cropping field heterogeneity and environmental impact is important, especially given the growing interest in using agroecosystem models to quantify ecosystem services from best management practices and land use change. We posited that augmenting process-based agroecosystem models with additional field-specific information such as topography, hydrologic processes, or independent indicators of yield could help limit simulation artifacts that obscure mechanisms driving observed variations. To test this, we augmented the agroecosystem model APSIM with field-specific topography and satellite imagery in a simulation framework we call Foresite. We used Foresite to optimize APSIM yield predictions to match those created from a machine learning model built on remotely sensed indicators of hydrology and plant productivity. Using these improved subfield yield predictions to guide APSIM optimization, total NO3-N loss estimates increased by 39% in maize and 20% in soybeans when summed across all years. In addition, we found a disproportionate total amount of leaching in the lowest yielding field areas vs the highest yielding areas in maize (42% vs 15%) and a similar effect in soybeans (31% vs 20%). Overall, we found that augmenting process-based models with now-common subfield remotely sensed data significantly increased values of predicted nutrient loss from fields, indicating opportunities to improve field-scale agroecosystem simulations, particularly if used to calculate nutrient credits in ecosystem service markets.

Interviews with farmers from the US corn belt highlight opportunity for improved decision support systems and continued structural barriers to farmland diversification

“…The majority of land is the region is used for agriculture, predominantly corn and soybeans, as well as secondary products like ethanol, eggs, beef, chicken, and pork (Green et al, 2018). While highly efficient and economically important (Riccetto et al, 2020), these annual cropping systems also leave the region's land bare for six months of the year and therefore prone to erosion (Thaler et al, 2021), nutrient loss (McLellan et al, 2015;Nowatzke et al, 2022), and greenhouse gas emissions (Griffis et al, 2013;Lawrence et al, 2021). The limited number of crops raised leaves farmers prone to economic vulnerability as individual commodity markets fluctuate (Wright, 2011), a fact well-known to farmers (Roesch-McNally et al, 2018).…”

Field‐scale analysis of miscanthus production indicates climate change may increase the opportunity for water quality improvement in a key Iowa watershed

“…Coarse scale studies suggest that the inclusion of miscanthus could reduce N loss within small agricultural watersheds and HRUs (Hydrologic Response Unit; Housh et al, 2015; Rajib et al, 2016; Teshager et al, 2016) and across the entire Mississippi River Basin but only if policies couple economic and N loss targets (Ferin et al, 2021). Cropping system processes like N leaching vary significantly at small spatial scales (Basso et al, 2019; Nowatzke et al, 2022). Targeted changes to land management, like perennial integration, could lead to disproportionate benefits (i.e., improvements that exceed unity with the proportion of area changed; Asbjornsen et al, 2014; Brandes et al, 2018) therefore signifying the importance of running simulations at the field scale.…”