At least 10 different methods to determine exposure for hemispherical photographs were used by scientists in the last two decades, severely hampering comparability among studies. Here, an overview of the applied methods is reported. For the standardization of photographic exposure, a time-consuming reference measurement in the open land towards the unobstructed sky was required so far. The two Histogram Methods proposed here make use of the technical advances of digital cameras which enable users to assess a photograph’s histogram directly at the location of measurement. This avoids errors occurring due to variations in sky lighting happening in the time span between taking the reference measurement and reaching the sample location within the forest. The Histogram Methods speed up and simplify taking hemispherical photographs, and introduce an objectively applicable, standardized approach. We highlight the importance of correct exposure by quantifying the overestimation of gap fraction resulting from auto-exposed photographs under a wide range of canopy openness situations. In our study, gap fraction derived from auto-exposed photographs reached values up to 900% higher than those derived from non-overexposed photographs. By investigating the size of the largest gap per photograph and the number of small gaps (gaps contributing less than 0.1% to gap fraction), we concluded that the overestimation of gap fraction resulted mainly from the overexposure of vegetation surrounding large gaps.
Solar radiation is the ultimate source of energy for life and a major determinant of the physiology, morphology, behavior, and life history of most organisms (
Hemispherical photography is a multi-step process prone to errors at each step (
Considering these errors,
Usually, camera positioning is 1.2 meters above ground level facing exactly vertical. Standardization of the thresholding can be achieved by applying an automated thresholding procedure as recommended by several authors (
Various studies have been published over the past decades on how to standardize exposure in hemispherical photography (
Taking standardized photographs in the field requires a reference measurement taken in the open land by aiming an external exposure meter (angle of view lower than 10°) at the brightest spot in the scene,
Although
In the present study, a literature review was compiled to identify existing methods for the determination of the exposure of hemispherical photographs. By analyzing the grey value histograms of differently exposed photographs we illustrate the influence of the exposure determination method on a photograph’s grey values. Further, we present two protocols by which the correct exposure can be rapidly determined in the field. To highlight the importance of a standardized method to determine exposure in hemispherical photography, we compare gap fraction estimates derived from auto-exposed photographs to those derived from correctly exposed photographs. The errors arising from auto-exposure are quantified for a NIKON® D70s DSLR for a wide range of canopy openness situations found in subtropical ecosystems in southern China.
We analyzed 61 publications appeared between 1991 and 2012 in which hemispherical photography was applied and identified ten different exposure determination methods (
A scene’s dynamic range is defined as the ratio of the maximum light intensity (brightest spot) to the minimum light intensity (darkest spot) in the scene (
When capturing a scene which contains a dynamic range that exceeds that of the camera, a loss of detail occurs either in the lowlight areas, the highlight areas, or both (
The grey value histogram of a photograph can be used very effectively to assess a possible mismatch between a scene’s and a camera’s dynamic range. Only when frequencies of grey values decrease towards both ends of the histogram’s x-axis, the dynamic range of a scene is completely covered. Frequency bins grouped against the extreme left end of the histogram indicate underexposure or blocked up shadows, which are areas of pure black (
Via photographic exposure one can control which section of a scene’s dynamic range is captured in the photograph. Exposure is defined as the total density of light (measured in lux seconds) allowed falling onto the sensor; it is controlled by the settings for shutter speed and lens aperture. Changing these settings while accounting for the sensor’s sensitivity (ISO), the user can control which light levels will be captured, and which light levels will be lost due to saturation of the camera’s dynamic range (
Common camera light meters assume the scene to be photographed as a mid-grey surface which reflects 18% of incident light (
With hemispherical photography, auto-exposure has the effect that the more vegetation covers, thus darkens the scene, the higher is the exposure set by the camera. For dark scenes, this results in photographs in which the forest interior is colorful depicted but the sky and as well brightly illuminated vegetation parts are overexposed and appear completely white (
For exposure, DSLRs offer the possibility to choose between different metering modes to determine a scene’s luminance, for example, spot, center-weighted, and matrix metering in NIKON’s D70s. Spot metering measurements are made from a circular spot (2.3 mm diameter) which are averaged for exposure determination. Bright or dark areas within the spot will give extreme readings (
The first step in processing hemispherical photographs is the photograph’s binarization (
The proposed Histogram Methods require that at each sample location a series of photographs is shot. After each shot, it needs to be assessed whether the photograph is affected by overexposure. If so, exposure needs to be decreased. This can be easily done in steps of 1/2 or 1/3 EV by using the camera’s exposure compensation function.
The following two approaches can be applied to prevent overexposure while maximizing contrast in the photograph:
Take a hemispherical photograph of a chosen site. Assess the photograph’s histogram
Take a hemispherical photograph of a chosen site. Review the photograph using the camera’s “highlight clipping warning” (Nikon) or “highlight alert“ (Canon) playback mode. In this playback mode all overexposed areas are marked as blinking lights which show the spatial distribution of the rightmost frequency bin in the corresponding grey value histogram. Exposure has to be decreased until the warning lights disappear. If no warning lights appear, exposure has to be increased towards one step below that exposure at which the warning lights appear.
Data were collected in Xishuangbanna Tropical Botanical Garden (XTBG - WGS 84: 21°55’39.36” N 101°15’51.84” E) which is located in Xishuangbanna, Yunnan, China. The wide range of canopy openness situations found in the botanical garden provided us with ideal conditions to assess the characteristics of gap fraction estimates derived from hemispherical photographs. Photographs were taken at 97 locations in the botanical garden aiming at covering a wide range of canopy openness situations.
A Nikon D70s DSLR equipped with a Sigma Circular Fisheye 4.5mm 1:2.8 lens with a field of view of 180° was used. The camera was mounted on a tripod at 1.2m height to characterize the canopy without the interfering presence of understory vegetation (
The basic camera settings mode “P” (Programmed Auto), ISO = 400, and matrix metering were used. At each location an auto-exposed photograph was taken. Subsequently EV was reduced in steps of 1/2 EV until the photograph’s grey value histogram and the “highlight clipping warning” function of the camera indicated that no overexposed pixels were left in the photograph (Histogram Methods a and b). At each EV level, one photograph was taken. Photographs were stored in JPEG format (3008 × 2000 pixels resolution), since no difference in grey values between TIFF and JPEG format was found (
All processing and analysis was done using R (
An automated global thresholding was applied to avoid variations in threshold setting by manual interpretation of photographs and because it speeded up the processing time (
For each location gap fraction values were derived from: (1) the auto-exposed photograph, and (2) the non-overexposed photograph, resulting from the application of the Histogram Methods. Besides comparing the paired gap fraction values, as well the number of gaps and the largest gap index (LGI) derived from the auto-exposed and the non-overexposed photograph were compared. We defined the LGI as the proportion of the photograph covered by the largest gap.
Photographs were post stratified into stratum “open” (gap fraction > 30%, N=16), “medium” (gap fraction 15-30%, N=8), and “dense” (gap fraction < 15%, N=73). As well figures were analyzed over all strata (gap fraction 0-50%, N=97). Pairs of images were tested with the paired Wilcoxon signed rank test for statistically significant differences; all p-values given refer to this test at the 0.05 alpha level.
Visual examination of photographs taken within forests revealed that auto-exposed photographs appeared brighter and contained a wider range of green and brown tones than non-overexposed photographs.
On average, photographs had to be underexposed by -3.3 EVs to avoid overexposure. We observed that more underexposure was required under dense canopy conditions than under open canopy conditions (
Gap fraction estimates derived from non-overexposed photographs ranged from 0.32 to 45.59%. Estimates based on auto-exposed photographs covered a slightly different range from 3.2 to 49.97% (
In all three strata, medians of gap fraction (
In non-overexposed photographs, the mean LGI was 5.95% while in auto-exposed photographs it was 7.79% (
In stratum “open”, medians did not differ significantly between auto-exposed and non-overexposed photographs (p=0.16), but significant differences were observed for medians in the strata “medium” (p <0.05) and “dense” (p<0.001). While the total difference between mean values was largest in stratum “medium”, relative differences were larger in stratum “dense”. Here, the LGI of auto-exposed photographs was on average 226% larger than the LGI of non-overexposed photographs.
In all strata, auto-exposed photographs tended to have higher mean numbers of gaps (
Over all strata, 99.52% of differences in mean number of gaps could be allocated to changes in the number of small gaps with a size <2.000 pixel (or <0.1% gap fraction). In
Our study made very clear that auto-exposed hemispherical photographs cannot be reliably interpreted. Other studies support these findings (
In forests, errors arising from auto-exposure occurred due to overexposure of vegetation. Compared to non-overexposed photographs, auto-exposed photographs contained more and larger gaps, and therefore, tended to overestimate gap fraction. By visual examination of auto-exposed photographs, it was observed that vegetation bordering canopy gaps was more illuminated than vegetation farther away from gaps. Larger gaps also tended to embrace fine structures, for example, in-growing branches and leaves that were brightly illuminated. In auto-exposed photographs most of these bright components were overexposed and appeared as solid white areas (
Auto-exposure resulted in increasing overestimation of gap fraction for decreasing canopy openness. Hence, relative distances in gap fraction among sample locations were not preserved by auto-exposure. Our study has also shown that underexposure with a fixed EV value, as done by,
The wide availability of DSLR cameras with histogram display modes solves the problem of standardizing exposure for hemispherical photographs. In the reviewed literature we found only two other publications that describe a method similar to the Histogram Methods:
Due to problems related to auto-exposure, we strongly advise to apply either the method described by
Our thanks are due to the Advisory Group on International Agricultural Research (BEAF) at the German Agency for International Cooperation (GIZ) and the German Ministry for Economic Cooperation (BMZ) for funding the research project MMC (“Making the Mekong Connected”, Project No. 08.7860.3-001.00) within which this study had been carried out. We are also grateful to all members of the MMC-project for their support. Especially, we thank Rhett D. Harrison for facilitating the data collection in Xishuangbanna Tropical Botanical Garden (XTBG). In addition, this study was supported by the CGIAR Research Program 6 on Forests, Trees and Agroforestry.
(A) Auto-exposed photograph taken with aperture F6.7 and shutter speed 1/125s. Gap fraction: 8.83%; number of gaps: 5848; LGI: 2.19%. (B) Non-overexposed photograph (underexposed by -3.5 EVs) taken at the same location. Aperture F11 and shutter speed 1/500s. Gap fraction: 1.78%; no. of gaps: 1177; LGI: 1.23%. Resolution of circular image area: 1 998 029 pixels.
(A) Grey value histogram of an auto-exposed photograph. The bright end of the dynamic range is cut off due to overexposure. This is indicated by the peak at the right end of the histogram. (B) Grey value histogram of a non-overexposed photograph. The dark end of the dynamic range is cut off.
Scatterplot of gap fraction [%] as measured in non-overexposed photographs against the exposure value required to avoid overexposure. Photographs taken under dense canopy conditions require more steps of underexposure than photographs taken under open canopy conditions. Grey diamonds: per stratum.
Box-plots of gapm non-overexposed and auto-exposed photographs. Box: first, second, and third quartile; whiskers: 1.5 x interquartile range; grey diamonds: mean gap fraction.
Relative difference of gap fraction estimates derived from auto-exposed and non-overexposed photographs. Grey diamonds: mean difference per stratum.
Subsets of hemispherical photographs taken at the same location. (A) Auto-exposed photograph: the vegetation at the border of gaps and branches and leaves growing into gaps were overexposed and disappeared. (B) Non-overexposed photograph (underexposed by -3.5 EVs): all relevant information was retained.
Exposure determination methods found in 61 publications using hemispherical photography.
Exposure determination method | Number of publications | Found in |
---|---|---|
Auto-exposure | 13 | |
Bracketing (auto-exposure, -1, -2 EVs or auto-exposure, +2, -2EVs) and selection of “best” photograph | 5 | |
Underexposed by -2 EVs to reference within forest | 1 |
|
Underexposed by -1 EV to reference within forest | 1 |
|
Underexposed by -0.7 EVs to reference within forest | 1 |
|
Overexposed by 2-3 EVsto reference in open land | 11 | |
Overexposed by 1-2 EVs to reference in open land | 3 | |
Overexposed by 1 EV to reference in open land | 2 | |
Same exposure as reference in open land | 3 | |
Posterior correction of photographs using image manipulation software | 1 |
|
No statement | 20 |
Mean (standard deviation in parenthesis) and median of gap fraction estimates [%] derived from non-overexposed and auto-exposed photographs. And mean and median of relative differences of gap fraction estimates [%] between non-overexposed and auto-exposed photographs.
Stratum | non-overexposed | auto-exposed | relative difference | |||
---|---|---|---|---|---|---|
Mean | Median | Mean | Median | Mean | Median | |
all | 10.96(13.19) | 4.6 | 17.16(11.34) | 12.83 | 56.32 | 152.61 |
open | 37.87(4.76) | 37.63 | 39.62(5.09) | 40.15 | 4.64 | 2.35 |
medium | 20.65(1.8) | 20.93 | 25.29(4.02) | 25.07 | 22.44 | 22.37 |
dense | 4.02(2.6) | 3.62 | 11.35(3.25) | 11.27 | 181.96 | 208.01 |
Mean (standard deviation in parenthesis) and median of Largest Gap Index (LGI) estimates.
Stratum | non-overexposed | auto-exposed | ||
---|---|---|---|---|
Mean | Median | Mean | Median | |
all | 5.95(10.06) | 1.41 | 7.72(10.06) | 3.52 |
open | 24.54(11.04) | 25.71 | 25.68(11.17) | 27.22 |
medium | 10.54(6.70) | 9.30 | 13.99(7.08) | 15.68 |
dense | 1.37(1.78) | 0.62 | 3.10(2.53) | 2.19 |
Mean and median number of gaps within non-overexposed and auto-exposed photographs. Mean and median number of gaps <2000 pixel (<0.1% gap fraction) are in round parenthesis and gap fraction represented by gaps <2000 pixel are in square brackets.
Stratum | non-overexposed | auto-exposed | ||
---|---|---|---|---|
Mean | Median | Mean | Median | |
all | 2735(2729)[2.31%] | 2682(2678)[1.90%] | 3994(3982)[4.58%] | 3778(3767)[4.38%] |
open | 1975(1958)[2.91%] | 1680(1658)[2.54%] | 2435(2417)[3.19%] | 2322(2292)[2.91%] |
medium | 3283(3268)[4.33%] | 3212(3208)[3.94%] | 3487(3740)[4.59%] | 2438(2425)[4.04%] |
dense | 2842(2838)[1.95%] | 2794(2793)[1.66%] | 4392(4381)[4.89%] | 4627(4614)[4.85%] |