Remote sensing, GIS and GPS: Geospatial techniques for detecting TOF in Italian traditional Agroforestry systems

Agroforestry, being grounded in traditional land use practices, has developed as an autonomous science to increase productivity and profitability for the farmers, while ensuring the land use sustainability. Agroforestry systems are widespread in many countries, supporting the coexistence of tree, crop and livestock components. Such complex ecological systems offer a wide range of economic, social and environmental benefits, occurring over a range of spatial and temporal scales. The integrated use of GIS, Remote Sensing and GPS technologies is particularly suited for assessing, mapping and quantifying the intrinsic spatial complexity of these systems. One of the major geospatial issues in Agroforestry is detecting, mapping and estimating the forest component of the systems: scattered trees or linear forest formations located either inside the field or along the field boundaries, also known as Trees Outside Forest (TOF) (FAO 1998, 2001). Data on TOF are scarce and the information available is fragmented at regional and national levels (Schnell et al., 2015). Beckschäfer et al. (2017) give an overview of inventory approaches suitable for the science-based assessment of TOF, specifically on agricultural lands. However, up to now there are no guidelines for TOF inventory in agroforestry systems. Aims Traditional tree-based agriculture systems involving different multipurpose trees such as chestnuts (Castanea spp.), oaks (Quercus spp.), and olive (Olea europa), (Eichhorn et al., 2006) are common in Italy and other Mediterranean countries. We investigated the integration of geospatial techniques for TOF inventory in traditional silvoarable systems located in Umbria region (central Italy), where oaks tree hedgerows (THRs) coexist with herbaceous crops. Methods We tested a procedure for the GIS inventory of THRs, through the semiautomatic photo interpretation of high-resolution multispectral Sentinel-2 satellite images and NDVI. Results were compared with GPS field measurements of THRs as control points to assess the ground truth. We also compared THRs picked up by remote sensing products with different spatial resolution (Google Digital Globe, Sentinel-2 and Landsat 8) using the same combination of spectral bands. Results and conclusion The THRs length detected, corresponding to the 14% of the total perimeter of the cultivated fields, fits accurately with the GPS field survey. The THRs' crowns cover the 3% of the total cultivated area, with an incidence of 67 m of linear tree rows for each hectare of cultivated land. We also observed that the THRs' spatial distribution improves the connection between forested patches in the study area, enhancing landscape connectivity. Further development is needed in order to include diverse landscape patterns: the high-resolution Sentinel-2 imagery appear especially suitable for the detection of most TOFs at landscape level.