Designing profitable and climate-smart farms using virtual reality

three forest types that are commonly grown on New Zealand farms. The following sections describe the methods used to estimate the impacts of tree planting on the farm profitability and carbon emissions. Virtual reality application The VR application was developed using Unreal Engine (UE, version 4.26.2) and designed to run on an Oculus Quest 2 headset. UE provides a blueprint visual scripting function that enables the development of VR applications with limited coding which in turn accelerates the development phase and reduces costs [28]. The virtual landscape within UE was made using a grid of 4033*4033 vertices, each covering one square meter, which is the recommended land- scape size to optimise headset performance. Therefore, the area considered in the application is covering 1,600 ha centred on the pilot farm. So, the user can navigate through the farm of interest and its surrounding. To improve the performance, the application was developed to be run with the VR headset connected to a high-performance computer. A high-resolution height map was generated in ArcGIS Pro (version 2.5.0) using a one- meter resolution DEM, which was converted from Geotiff to PNG format and imported into UE. The 300 mm aerial imagery was resampled to 500 mm spatial resolution and imported as a PNG file into UE. This 500mm aerial photo was ‘draped’ on top of the 3D terrain to create landscape texture. The map of pasture production, based on the NDVI classification, was also imported (as a PNG file) into UE as an overlay to the landscape texture and displayed with

high transparency. This enables the user to visualise simultaneously the landscape texture and the pasture production categories. At the end of a session, a map representing the spatial foot- print of the planting done by the user can be exported and stored as a jpeg image. Alongside the planting map, screenshots of the virtual environment can also be saved and exported as jpeg to support the user in communicating new planting designs. Carbon emissions and financial modelling Within the VR twin of the farm and using the VR headset and hand controls, the user selects 30m x 30m grid cells where trees can be located. When the user validates a planting site, the whole farm profitability is estimated by Eq (1). Farm profit ¼ Ps ðPp þ CtÞ Eq ð1Þ Where; farm profit (NZD/year); Ps is farm profit before planting (NZD/year); Pp is the profit that was generated by the livestock operations on the selected grid cell prior forestry conver- sion (NZD/year) and Ct is the establishment cost of the planted species (NZD). The overall farm carbon emissions are estimated in the application by Eq (2). Farm emissions ¼ Es ðEp þ StÞ Eq ð2Þ Where farm emissions (kgCO2eq/year); Es is the farm emissions before planting (kgCO2eq/ year); Ep is the emissions generated by the selected grid cell prior to being converted to forest (kgCO2eq/year) and St is the sequestration generated by the planted tree species (kgCO2eq/ year). Pasture production and stock units The profit and the carbon emissions arising from a given area of pasture (respectively Pp and

Ep in Eqs 1 and 2) depends on the stocking rate which varies across grid cells according to the pasture production (i.e. the feed resource). Based on the distribution of the NDVI across the PLOS ONE Designing profitable and climate-smart farms using virtual reality PLOS ONE | https://doi.org/10.1371/journal.pone.0286723 June 2, 2023 5 / 16