Keywords

3.1 Introduction

Aerial photographs have been and continue to be a widely used source of information in the forestry sector (Hall, 2003; Howard, 1991). Currently, remote sensing is used in most countries that conduct large-scale inventories. Many forest inventories use remote sensing images such as aerial photos, orthophotos and satellite images for the classification of land use and land cover at the sampling points in order to select the points to be later detected on the ground or for estimation of the forest area, at least preliminarily. Some inventories also use aerial photos to acquire data on the characters of forest formations (Tomppo et al., 2010).

In the first inventory phase, the Italian national forest inventory INFC uses orthophotos for the preliminary classification of land use and cover to stratify the sample to be used for the field surveys of the two following sampling phases (cf. Chap. 2). The classification carried out during the first inventory phase of the third inventory INFC2015 also allowed a preliminary estimate of the changes relating to the forest use and cover that have occurred since the previous inventory survey, useful for reporting purposes at the end of the first commitment period of the Kyoto Protocol.

After some hints on the characteristics of aerial photos and orthophotos and on the photo interpretation process (Sect. 3.2), the chapter illustrates the land use and cover classification system adopted for INFC (Sect. 3.3) and the tools and procedure used in the first phase of INFC2015 (Sect. 3.4). Finally, the chapter describes the set of controls used to evaluate the quality of the data resulting from the classification of photo interpreters (Sect. 3.5).

3.2 Materials and Methods for Photointerpretation of Land Use and Cover

The quality of digital images (and the use that can be made of them) primarily depends on their resolution, which is divided into spatial or geometric, spectral, radiometric and temporal resolution (Brivio et al., 2006). Spatial resolution is linked to the size of the elementary area on the ground where the electromagnetic energy is detected, that is to the size of the pixels. The spectral resolution indicates the number and width of the spectral bands (wavelength intervals) in which an image is acquired.Footnote 1 Radiometric resolution is given by the minimum difference in electromagnetic energy detectable by the sensors on the photographed surfaces. Finally, temporal resolution indicates the time interval between two successive shots of the same area.

Orthophotos simultaneously provide a photographic and cartographic representation of the territory. They are obtained through geometric correction (orthorectification) and georeferencing of digital aerial photos or previously digitized frames. Orthorectification entails straightening and projection on the horizontal plane of the images, in order to allow a correct representation on the plane of distances, angles and surfaces. Georeferencing allows each point of the territory represented by the orthophoto to be associated with its position in space, referable to a system of geographic or plane coordinates (Gasparini et al., 2014).

The interpretative process of an orthophoto, or any remotely sensed image, is characterised by the presence of two fundamental phases: (a) examination, recognition and, if necessary, measurement of the elements in the image; (b) the formulation of deductive and inductive reasoning, based on the observations made, in order to classify what is represented in the image (Dainelli, 2011). Knowledge of the landscape and its characteristic elements is crucial for effective photointerpretation. The classification is based on the analysis of spatial (localisation, association), spectral (color, tone) and geometric (shape, size) characters. Photointerpreters are recommended to proceed first with a broad observation of the territorial context, based mainly on the analysis of the shape and size of the different elements or polygons and their arrangement in space, defined as structure or pattern. The individual elements are recognizable on the image based on the brightness and intensity of the color, the texture given by micro-changes in the distribution of color tones, and their shapes and sizes. The presence of shadows on an image plays a double role, and they can act as a disturbing element or as a contribution to photointerpretation. This presents an obstacle, especially when large portions of the territory are obscured, but it can also provide important clues for identifying the vertical profile and the height of the elements to be interpreted, facilitating, for example, the distinction between trees and shrubs.

3.3 Classification System

The land use and land cover classification system used for the first INFC phase (Table 3.1) includes three hierarchical levels, the first corresponding to the same level of the CORINE Land Cover classification system (European Commission, 1993) and the following two aimed at further detailing the classes of greatest interest to produce the inventory statistics (Gasparini & Di Cosmo, 2016). The first level is divided into the five main classes of the CORINE Land Cover system (Artificial surfaces, Agricultural areas, Forest and semi-natural areas, Wetlands, Water bodies), from which it differs for the inclusion of Castanea sativa forests for fruit production and pastures in the class Forest and semi-natural areas rather than in Agricultural areas. The second level of INFC classification includes 12 subclasses, of which two (Plantations for timber and wood production and Woodland) are of interest for the subsequent sampling phases, together with the residual class of non-classifiable points. The third level of classification is present only for the subclass Woodland, which is divided into three further subclasses on the basis of coverage thresholds of tree and shrub species consistent with the definitions of Forest and Other wooded land and used since 2000 for the Global Forest Resources Assessment (FRA) (FAO, 2001) and adopted for INFC (cf. Chap. 2).

Table 3.1 Classification scheme adopted for the INFC photointerpretation of land use and cover / Schema di classificazione adottato per la fotointerpretazione dell’uso e copertura del suolo nell’INFC

The classification system for the INFC2015 photointerpretation has remained unchanged compared to the second Italian forest inventory INFC2005, to allow the comparison between the results of the two photointerpretations and highlight the significant changes to and from the subclasses of inventory interest. The only difference concerns the introduction, for the Agricultural areas class, of the new subclass of Fruit plantations, which includes orchards, vineyards and olive groves.

For a detailed description of the photointerpretation classes and instructions for their identification on the orthophotos, refer to the Photointerpretation Manual (Gasparini et al., 2014).

3.4 Classification Tools and Procedure

The classification of land use and land cover for the more than 301,000 inventory points constituting the INFC first phase sample was performed through photointerpretation of digital orthophotos on a dedicated WebGIS platform called GeoInfo, developed by the AlmavivA company with the collaboration of Telespazio (Gasparini et al., 2020). The platform allows operators to view orthophotos (Fig. 3.1) and other useful information layers, such as administrative limits, roads, toponymy, hydrographic network, altitude bands, state nature reserves and areas covered by fires, as well as to acquire the classifications of the photo interpreters and store the data in a national central archive.

Fig. 3.1
figure 1

Viewing of orthophotos on the WebGIS platform GeoInfo / Visualizzazione di ortofoto sulla piattaforma WebGIS GeoInfo

The INFC2015 photointerpretation was conducted on digital orthophotos in color and infrared-false color (RGB + nIR) with a resolution of 50 cm derived from the 2010–2012 AGEA coverage. To solve doubtful cases, the photointerpreters could also consult the digital color orthophotos of equal resolution derived from the AGEA 2007–2009 coverage, and orthophotos with a resolution of 1 m of the 1999–2005 coverage. The latter, used by photointerpreters in the first phase of the second inventory INFC2005, were used to evaluate any significant changes in use and land cover that occurred subsequently, through the visual comparison of the images referring to the two different periods.

The first phase of INFC2015 was conducted by about 50 photointerpreters, partly from the State Forestry Corps and partly from the Forest Services of the regions with special statutes, and the autonomous provinces, suitably instructed and trained through a specific training course.

The INFC classification procedure consists of assigning each sample point to the class and subclass of land use and cover of the polygon in which the point falls, after checking whether or not the minimum dimensional thresholds are exceeded. The polygon represents a homogeneous area for land use and cover, having an area greater than 0.5 ha and a width greater than 20 m. Limited to polygons characterised by a cover of trees or shrubs, the photointerpreter verifies also whether or not the minimum coverage thresholds are exceeded (if tree species, 10% for Forest and 5% for Sparse forest categories, and if shrubs species 10% for Shrubs category) in accordance with the definition adopted for the inventory domain (cf. Chap. 2). An additional 40% coverage threshold, relating to the herbaceous component, is used to distinguish the subclass Grasslands, pastures and uncultivated areas from that of Open areas with little or no vegetation. Sample points falling in smaller polygons, with an area between 500 and 5000 m2 or, if elongated, with a width between 3 and 20 m, are assigned to the land use and cover class of the nearest polygon that respects the minimum area and width thresholds indicated above, 0.5 ha and 20 m, respectively. In these cases, the presence of an ‘included polygon’ and its land use and land cover are also recorded.

Fig. 3.2
figure 2

Analysis window, grid and measuring tools on the WebGIS GeoInfo / Intorno di analisi, griglia e strumenti di misura nel WebGIS GeoInfo

The WebGIS GeoInfo automatically manages the sequence of operations for photointerpretation, proposing to the operators the points to be classified and allowing verification of the minimum thresholds through the display of an analysis window (cf. Chap. 2), a grid and tools to measure distances and surfaces (Fig. 3.2) (Gasparini et al., 2021). The analysis window, consisting of nine contiguous squares of 50 m side for a total area of 22,500 m2 and centred on the sample point, allows for visual estimation of the extension and width of the homogeneous polygons identified. The grid of points spaced 10 m apart, on the other hand, enables quick and objective evaluation of the crown cover degree by counting the points that intercept trees or shrubs crowns.

For details on classification procedures and any specific cases, refer to the Photointerpretation Manual (Gasparini et al., 2014).

3.5 Quality Controls of the Photointerpretation

Land cover classification by visual interpretation of aerial photos always involves a certain degree of subjectivity, and the experience of the photointerpreters plays an important role in determining the quality of the result (Strand et al., 2002). Subjectivity should be limited as much as possible in order to obtain comparable classifications. The implementation of a quality control procedure during the photointerpretation activity and at its conclusion is fundamental, both to evaluate the uniformity of judgment by the photointerpreters and the reproducibility of the classification, and to evaluate the accuracy of the classification made.

The data quality assurance (QA) procedures implemented for the first INFC phase aimed at checking non-sampling errors due to measurement errors (during the verification of polygons minimum extension and width thresholds and of tree and shrub cover), or to an incorrect understanding of the rules of interpretation of the images by the photointerpreters. The control procedure is based on the identification of quality objectives (MQOs, Measurement Quality Objectives) and quality limits (DQLs, Data Quality Limits), respectively corresponding to incorrect classifications or tolerable measurement errors and the relative maximum permissible thresholds (Gasparini et al., 2009). The MQOs related to tolerable incorrect classifications for the QA of INFC2015 are reported in Table 3.2. In regard to the minimum surface and width thresholds, measurement errors of 200 m2 and 2 m, respectively, for the polygons and of 50 m2 and 1 m, respectively, for the included polygons were considered tolerable. For the crown coverage, errors up to 2.8% and 5.5% were considered tolerable for tree or shrub coverage and herbaceous coverage, respectively, values corresponding to one point and two points of the grid used (cf. Chap. 4). Table 3.3 shows the DQLs for the different land use and cover classes, established according to the importance of the individual classes and subclasses and the relative difficulty of recognition on orthophotos. The greater the importance of a class or subclass, the higher its DQL and the lower the percentage threshold of admissible errors; on the contrary, the greater the classification difficulty of a class or subclass, the lower its DQL and the higher the percentage threshold of admissible errors.

Table 3.2 Admitted misclassifications of the land use and cover in the photointerpretation during INFC2015 first phase / Tolleranze di classificazione dell’uso e copertura del suolo nella fotointerpretazione della prima fase INFC2015
Table 3.3 Data Quality Limits (DQLs) for the photointerpretation in the INFC2015 first phase / Limiti di Qualità (DQLs) per la fotointerpretazione della prima fase INFC2015

The QA activity during photointerpretation was carried out through the reclassification of a certain number of randomly selected points by a group of expert operators of the CREA Research Centre for Forestry and Wood, who took on the role of reference operators. The CREA operators also used the GeoInfo platform for the classification of the points. However, the results of the classification, which had been previously performed by the photointerpreters, were not made available to them. Later, the reference operator compared his own classification with that of the photointerpreter and, if there were discrepancies, assessed whether they were admissible based on measurement or classification errors deemed tolerable according to the MQOs. A specially implemented IT platform, accessible from the intranet of CREA Research Centre for Forestry and Wood, was used by internal operators in charge of periodic and final checks to record the identification code of the points concerned and the results of the checks. In total, checks were performed on 9766 points during the photointerpretation, equal to 3.2% of the points of the INFC sample.

In addition to the on-going checks described above, final checks for approval were carried out on a randomly selected subsample of the inventory points per region. The final checks covered 2% of the inventory points in each region. Subsequently, for three regions which had not reached the DQLs for the subclasses of greatest interest, a partial revision of the classification was conducted and a new control was carried out on a further 2% of the sampling points. The results of the final checks for Woodland (Table 3.4) show that the classification discrepancy between photointerpreters and reference operators affects 2.1% of the points at national level, and percentages were always lower than the maximum threshold of 5% in all regions.

Table 3.4 Results of the final checks for approval for the subclass Woodland (Forest formations, Sparse forest formations and Temporary unstocked areas), by region; positive check means concordant classification between control operators and photointerpreters / Risultati dei controlli finali per la sottoclasse delle Aree boscate (Formazioni forestali, Formazioni forestali rade e Aree temporaneamente prive di soprassuolo), per regione; collaudo positivo in caso di classificazione concorde tra fotointerpreti e operatori di controllo

During the classification activity, a further blind check was conducted on a random subsample of inventory points, which were assigned simultaneously to two photo interpreters from the same region without their knowledge. The results of the blind check will be used for subsequent analyses aimed at any modification of the classification system.