A field test of forest canopy structure measurements with the CanopyCapture smartphone application

CanopyCapture

CanopyCapture uses tonal contrast to interpret pixels as “vegetation” or “sky”. No user manual or technical specifications have been made available by the developer (Patel, 2018), so the thresholding algorithm cannot be reported here. CanopyCapture generates a false colour photo of canopy structure, rendering pixels interpreted as sky in red (Fig. 1). Although percentage “canopy cover” is reported on the screen, what CanopyCapture actually appears to estimate is the complement of percentage gap fraction (100 – gap fraction). Jennings, Brown & Sheil (1999) point out that canopy cover (the percentage of the forest floor covered by the vertical projection of tree crowns and other vegetation) is often confused in the literature with canopy closure i.e. the percentage of the sky obscured by vegetation when viewed from a singular point, like a camera lens. Moreover, standard smartphone cameras produce rectilinear images, which like any other projection that is not perfectly equisolid, will distort the relative areas of gaps in the canopy as a function of zenith angle (Fleck, 1995). Gap fractions must therefore be corrected for area distortion in order to calculate canopy openness, or its complement i.e. canopy closure (Frazer, Trofymow & Lertzman, 1997). As CanopyCapture does not require any input about lens projections or field-of-view, the figure it reports must be the complement of uncorrected gap fraction (100 – % gap fraction). CanopyCapture output is therefore reported below as “gap fraction”. Photos taken with CanopyCapture can only be saved as screen shots, so the “canopy cover” figure must be noted down before taking the next measurement.

Comparison with LAI-2200C canopy analyzer

Gap fraction estimates by CanopyCapture were compared with gap fraction values computed by the LAI-2200C Canopy Analyzer, in two evergreen temperate forests in New Zealand.

An old-growth podocarp/broadleaved forest was sampled in the Lake Okataina Scenic Reserve (38.089°S, 176.424°E). The podocarps Dacrydium cupressinum and Dacrycarpus dacrydioides up to 45 m tall emerge from a canopy dominated by Beilschmiedia tawa (Lauraceae) with fewer Knightia excelsa (Proteaceae), Laurelia novae-zelandiae (Atherospermataceae) and Litsea calicaris (Lauraceae). The treeferns Dicksonia squarrosa (Dicksoniaceae) and Cyathea smithii (Cyatheaceae) are common in the understorey, and sometimes form dense thickets in treefall gaps (Lusk & Laughlin, 2017). The monocotyledonous liane Ripogonum scandens (Ripogonaceae) forms dense tangles in places. Although some stands were selectively logged during the mid-20th century, most forest in the reserve approximates an old-growth structure, including treefall gaps in varied stages of regeneration. Nomenclature follows Allan (2002–2021). Research at Lake Okataina Scenic Reserve was authorized by the Department of Conservation (66760-RES).

A southern beech (Nothofagus) stand was sampled on a poorly-drained site near Mamaku on the Patetere Plateau (38.018°S, 176.067°E). Nothofagus truncata up to 22 m tall formed most of the canopy, accompanied by fewer and slightly smaller N. menziesii. An approximately normal distribution of N. truncata diameters between 15 and 70 cm, and the presence of occasional larger cut stumps in an advanced state of decay, were consistent with establishment of an even-aged Nothofagus stand after logging of the previous forest. Weinmannia racemosa (Cunoniaceae) and Ixerba brexioides (Strasburgeriaceae) were common in the subcanopy and understorey. There were only a few small canopy gaps, resulting from standing deaths of N. truncata trees. Access to this forest was provided by Hancock Forest Management (OTPP_91459) and Department of Conservation (97727-RES).

The LAI-2200C quantifies the attenuation of diffuse sky radiation in five angular bands within a quasi-hemispherical field of view, using five sensors arranged in concentric rings. One of two wands was mounted on a tripod and used to take measurements 1.5 m above each sampling point. The other wand was placed in a large clearing that offered an unobstructed view of the sky over the 148° field of view of the instrument, and programmed to take readings at 30 s intervals. Combining data from the two wands enables calculation of diffuse non-intercepted irradiance beneath a forest canopy as a proportion of simultaneous irradiance outside the forest. Computation of gap fraction (G) is based on averaging the logarithms of diffuse light transmittance in the five angular bands, weighted by their relative areas:

$$G=e^{\left[\sum _{i=1}^{5}ln{\displaystyle \frac{B_i}{A_i}{W}_i}\right]}$$

where the subscript i refers to the five concentric optical sensor rings (i = 1…5), B_i is diffuse irradiance reaching the understorey wand from each of the five zenith angle arcs, A_i is diffuse irradiance reaching the clearing wand from each of the same arcs, and W_i is the weighting of each of the five zenith angular bands according to their relative areas.

CanopyCapture installed on a Samsung Galaxy A21s smartphone was used to estimate gap fraction at the same sampling points. The 13 MP front camera used by CanopyCapture has a 45° field of view, meaning it samples a much smaller area of the forest canopy than the LAI-2200C. After each measurement with the LAI-2200C, the phone was placed on the same tripod and CanopyCapture measurements taken with and without a clip-on 37 mm wide-angle HD lens (SourceTon, Guangdong, China) fitted to the phone camera. This lens increased the camera’s field of view to 65°. Measurements were made under overcast conditions, between 11:00 and 15:00 h on summer days (3rd January at Okataina, 9th February at Mamaku).

Measurements were taken at 30 points in each forest, randomly spaced 11 to 15 m apart on transects run through the forest. This spacing meant that successive sampling points were unlikely to fall beneath the crown of the same canopy tree, as most crown diameters in New Zealand forests are ≤10 m (Veblen & Stewart, 1982; Walcroft et al., 2005). All transects were located on flat ground or on slopes of <5°.

Do results depend on sky condition?

CanopyCapture installed on a Samsung Galaxy A21s smartphone was used to compare canopy closure values obtained under blue and overcast skies. This aspect of the work was carried out in Jubilee Bush (37.775°S, 175.292°E), a 5-ha urban remnant of evergreen temperate forest in Hamilton, New Zealand. The present forest canopy in Jubilee Bush is dominated by Dacrycarpus dacrydioides that mostly established after 19th-century logging; and Beilschmiedia tawa, some of which are undoubtedly older (Whaley, Clarkson & Smale, 1997). There are also scattered larger D. dacrydioides (some exceeding 1 m diameter) that predate logging. The other common trees are Laurelia novae-zelandiae, Alectryon excelsus (Sapindaceae), and Melicytus ramiflorus (Violaceae). Recent tree deaths have left a few small canopy gaps. Research in Jubilee Bush was authorized by Hamilton City Council.

Sampling points were randomly spaced between 11 and 15 m apart along two transects run parallel to the longest axis of the reserve. At each of 27 sampling points, three flat-topped 25 cm plastic stakes were driven into the soil, to provide a level sampling platform 15–20 cm above the forest floor. CanopyCapture measurements were taken from these sampling platforms under overcast and blue skies between 09:30 and 15:00 h in August (winter) 2021.

Do results depend on make and model of phone?

CanopyCapture installed on three different smartphones was used to estimate canopy closure in a wooded area of the University of Waikato Hamilton campus (37.785°S, 175.316°E). These consisted of a Samsung Galaxy A21s (13 MP front camera with 45° field of view), a Samsung Galaxy S20 (10 MP front camera, 45° field of view), and a Sony H4113 (8 MP front camera, field of view 90°). Fifteen sampling points were spaced at random intervals along a 100-m long transect run along the longest axis of the wooded area. Measurements were made under an overcast sky. The eight different tree species present in the canopy included two deciduous species that were leafless at the time of sampling (August 2021), resulting in wide spatial variation in canopy gap fraction.

Statistical analysis

The relationships of CanopyCapture output with results from the LAI-2200C were analysed by standardized major axis (SMA), using the smatr R package (Warton et al., 2012). SMA (rather than ordinary least squares regression) was appropriate here as the scaling of bivariate relationships tells us how sensitive CanopyCapture is to variation in leaf area index and understorey light availability. Variables were log-transformed before some analyses, to improve model fits and normalize distributions of residuals. SMA was also used to compare CanopyCapture measurements carried out under blue and overcast skies.

CanopyCapture measurements made on different smartphones could not be compared by parametric ANOVA, as the distribution of the residuals could not be normalised by any of the attempted transformations. Consequently the non-parametric alternative of Friedman’s test (Friedman, 1937) was used to compare means, by way of the agricolae R package (Mendiburu, 2015). Variances of CanopyCapture values obtained using the three phones were compared with Levene’s test for homogeneity of variance, using the car R package (Fox et al., 2012).

Comparison with LAI-2200C canopy analyzer

CanopyCapture output was significantly correlated with gap fraction computed by the canopy analyzer in the old-growth podocarp/broadleaf forest at Okataina (R² = 0.39), and use of the wide-angle adapter lifted this figure to 0.56 (Fig. 2A). Use of the wide-angle adapter affected neither the slope nor the elevation of the standardized major axis (SMA) of these relationships (Fig. 2A). Slopes of ~3.0 show that CanopyCapture has low sensitivity to spatial variation in canopy gap fraction (Fig. 2A): the SMA equations indicate that a doubling of gap fraction as measured by CanopyCapture equated to an eight- to nine-fold increase in gap fraction as measured by the LAI-2200C at Okataina.

CanopyCapture output was not significantly correlated with LAI-2200C output in the even-aged Nothofagus stand at Mamaku (Fig. 2B), where there was less spatial variation in canopy structure (Table 1). Although use of the wide-angle adapter raised the R² value by >50%, the relationship remained non-significant (Fig. 2B).

Despite the relatively low sensitivity of CanopyCapture to spatial variation in canopy structure, it did nevertheless detect significant differences in average gap fraction between the two forests—both with and without the wide-angle adapter (Table 1). CanopyCapture did, however, find proportionally smaller means differences than the LAI-2200C (Table 1).

Do results depend on sky condition?

CanopyCapture measurements under blue skies scaled as approximately 1.195x–17.0, where x = gap fraction estimated under an overcast sky (Fig. 3); 95% confidence intervals of the slope did not include unity (1.003–1.423), providing conclusive evidence of a departure from isometry. Under dense canopies, sky condition made little or no difference to gap fraction estimates; under more broken canopies, gap fraction values recorded under blue skies tended to be slightly higher than those recorded in overcast conditions.

Do results depend on make and model of phone?

Friedman’s test indicated that mean gap fraction estimates differed significantly between phones (F = 10.55, P = 0.0004), and post-hoc pairwise comparisons showed the mean value recorded on the Samsung A21s was significantly higher than those obtained with the other two phones (Fig. 4). Levene’s test indicated homogeneity of variances across the three phones (F (2, 42) = 0.17, P = 0.85), despite the narrower range of values obtained with the Sony H4113 phone (Fig. 5).