Area estimation of Trees Outside Forest: searching for effective sampling strategies on the basis of aerial photo imagery

Trees outside forest (TOF) are important elements in many landscapes worldwide, they in fact have specific ecological, social and economic values. For example they play a key role for the production of several ecosystem services such as: carbon sequestration, fuel production, agriculture protection against salt and wind, air and water pollutant reduction, nitrogen fixation, biodiversity conservation, fodder and browse supply and aesthetic landscape values. TOF are considered as distinct items in many National Forest Inventories and on the basis of the FAO they are defined as all single discrete forest trees or small groups of forest trees isolated within rural and urbanized areas and not classified as forest or small woodlots. For our purposes and consistently with the classification of the Italian National Forest Inventory, they have an area >500 m2 and <5000 m2. Inventorying TOF means the estimation of population parameters such as total volume or total number of trees. An efficient solution for inventorying TOF is the application of a sampling scheme on the basis of aerial orthophotos. The purpose of this contribution is to test several sampling schemes in order to identify the most effective strategies in terms of precision and costs. To this purpose, a full mapping census of TOF was performed in Regione Molise obtaining the list and the positions of all the 82424 TOF in the Region. Because TOF abundance and total area cover by TOF were parameters known from the census without no field measurement, the precision of some two-phase sampling strategies was checked for the estimation of these two parameters. In the first phase a tessellation stratified sampling (TSS) was considered. Firstly, the study area was partitioned into 4639 hexagons of size 100 ha. Then a point was randomly placed within each hexagon and TOF were selected in accordance with: (a) point sampling (a TOF was selected if at least one of the first-phase points fell within it); (b) centroid-based plot sampling (a TOF was selected if its centroid fallen in at least one of the plots of pre-fixed radius centred in the first-phase points); (c) TOF-based plot sampling (a TOF was selected if it intersected at least one of the plots of pre-fixed radius centred in the first-phase points); (d) line-intersect sampling (a TOF was selected if it intersected at least one of the transect of pre-fixed direction and length centred in the first-phase points). The variance of the Monte-Carlo integration estimator of area and abundance was theoretically derived for each of the four sampling schemes, together with the expected number of selected TOF and the bias of the conservative variance estimator usually adopted under TSS. A second-phase of sampling was also considered. A sample of hexagons was selected in accordance with the one-per-stratum stratified sampling. Firstly, the population of hexagons was partitioned into spatial strata containing approximately the same number of hexagons, and then an hexagon is randomly selected within each stratum. Then sampling schemes (a),(b),(c) and (d) were performed only in the selected hexagons. The variance of the two-phase estimator of area and abundance was theoretically derived for each of the four sampling scheme and second-phase sampling fractions of 50%, 25%, 10%, 5%, 1% together with the expected number of selected TOFs and the bias of the two-phase variance estimator. The achieved results allow to check the precision of the first-phase sampling schemes wrt sampling effort quantified by the expected number of selected TOF and the fall of precision involved by the second phase of sampling.