A Semi-automated Method for Analysing Hemispherical Photographs for the Assessment of Woodland Shade

Transcription

1 Biological Conservation 54 (1990) A Semi-automated Method for Analysing Hemispherical Photographs for the Assessment of Woodland Shade Julie Barrie, a* J. N. Greatorex-Davies, a R. J. Parsell b & R. H. Marrsa~ *NERC, Institute of Terrestrial Ecology, ~NERC, Computer Services, Monks Wood Experimental Station, Abbots Ripton, Huntingdon PEI7 2LS, UK (Received 30 October 1989; accepted 30 January 1990) ABSTRACT Hemispherical photography is a useful technique for the assessment and monitoring of the light climate in woodland habitats. Usually the interpretation of the photographs is a slow and tedious task when done manually. A faster semi-automated procedure is described which compares favourably with the manual method. INTRODUCTION Light is one of the major factors influencing the distribution and abundance of woodland species (Warren, 1981, 1985; Hall & Greatorex-Davies, 1989), and it is becoming increasingly clear that woodland reserve management requires routine assessments of the light climate in woodland areas with a high conservation value (Hall & Greatorex-Davies, 1989). There is, therefore, a need for a simple, rapid and inexpensive technique for assessing woodland light climate. Whilst there are many ways of estimating the amount and quality of light per se, ranging from solarimeters to integrating photometers, usually this approach is expensive and requires continuous visits to the site. Even the * Present address: Trent Polytechnic, Burton Street, Nottingham NG1 4BU, UK. To whom all correspondence should be addressed. 327 Biol. Conserv /90/$03" Elsevier Science Publishers Ltd, England. Printed in Great Britain

2 328 Julie Barrie, J. N. Greatorex-Davies, R. J. Parseli, R. H. Marrs cheapest integrating equipment (Woodward & Yaqub, 1979; Mitchell & Woodward, 1987) costs between 70 and 100, and requires monthly or bimonthly visits to collect data. Moreover, the information provided by continuous recording is more detailed than needed for conservation management. An obvious alternative is the use of hemispherical photography, first developed by Evans and Coombe (1959) and Anderson (1964) for estimating amounts of direct and diffuse light received by woodland gaps, and since used as a research tool in conservation studies of woodland Fig. 1. A typical hemispherical photograph taken in a woodland ride ofmoderate shade in southern England; the markers indicate north (with hole) and south.

3 Assessment of woodland shade 329 butterflies and other insects (Warren, 1981, 1985; Hall & Greatorex-Davies, 1989). In conservation work the amount of diffuse light reaching a site has been found to be a useful measure of the light climate, and is highly correlated with the abundance of several insect groups (Hall & Greatorex- Davies, 1989). The main advantage of this technique is that it is quick to use in the field, requires only one site visit per observation, and is relatively cheap. The drawback for its use in extensive conservation assessment and monitoring studies is that the photographs are usually interpreted manually, which is a slow and tedious process. This paper, therefore, describes a semiautomated technique for assessing the amounts of open sky and shade (strictly the amount of open sky obstructed by the canopy) of woodland sites from hemispherical photographs, and compares the results with the manual method. As Anderson's (1964) manual method has been used in the past for this purpose (Warren, 1981, 1985; Hall & Greatorex-Davies, 1989), it is essential that there is good agreement between the two methods. Although computerized assessment of hemispherical photographs has been developed by physiological ecologists (Chadzon & Field, 1987), no attempt has been made to use this approach in applied conservation work. PHOTOGRAPHIC METHODS Hemispherical photographs used in this study were taken with a Minolta MD Fisheye-Rokkor (-X) lens, which had a 180 field of view, fitted on a Minolta SRT101 camera. The camera was mounted on a carriage similar to that described by Evans et al. (1976) and Warren (1981). The carriage was aligned on a north-south axis with direction markers visible in the photograph, and levelled horizontally. Most of the photographs were taken on overcast days, but if there was any direct sunlight present it was blocked out of the picture by a piece of card. Exact details of the method are found in Warren (1981, 1985). An example of a hemispherical photograph taken in a woodland ride is shown in Fig. 1. MANUAL METHOD Anderson (1964) described a method for producing a transparent overlay with 20 concentric annuli, with each annulus containing an equal area of sky. These 20 annuli are then dissected with 50 equally spaced radial lines producing a grid of 1000 segments (Fig. 2(a)). Each segment contributes 0.1% of the total area from a standard overcast sky. In order to calculate the percentage shade or open sky the overlay is placed over the photograph, and the numbers of segments which are clear and obstructed by the tree canopy are counted. An obvious drawback to this technique is that the scale of detail

4 330 Julie Barrie, J. N. Greatorex-Davies, R. J. Parsell, R. H. Marrs Fig. 2. The overlays used to calculate percentage cover of shade and open sky; (a) overlay used in the manual method with 1000 segments of equal area; (b) overlay used in the semiautomated method with 20 annuli of equal area. on the photograph is often smaller than the grid scale. Anderson (1964) suggested that segments should be classified into five categories, viz., clear, <33%, 33-66%, 66-90%, >90% shaded, corresponding in the final calculation to 100, 75, 50, 25 and 0% open sky. Experience suggests that scoring should be done annulus by annulus rather than by category. In theory, therefore, each analysis requires I000 decisions to be made as well as a final calculation and count, although in practice photographs with large degrees of either open sky or shade can have entire annuli counted in one operation. Where the canopies are moderately dense, the number of decisions that must be made increases, and with it the potential for error. Anderson (1964) suggested that an assessment took between 10 and 15 min, but to maintain reasonable accuracy on a sustained basis a total of 20 assessments per day is realistic. SEMI-AUTOMATED METHOD An overlay of 20 concentric annuli (each containing an equal area of sky) was constructed using Anderson's (1964) method, but here the radial lines were omitted (Fig. 2(b)). Both the overlay and the hemispherical photograph were then digitized using a digitizing camera on to an International Imaging Systems I2S model 75 Image Analysis System (IA). A four-step procedure was then followed. Step 1. The digital values of the overlay were adjusted by the IA system so that each annulus had a unique value (1 for the outer annulus to 20

5 Assessment ~f woodland shade 331 for the inner circle). This adjusted digital overlay was then stored for reference against each of the digitized hemispherical photographs. Step 2. A threshold level was determined manually on the digitized image of the photograph on the I2S screen which separated the image into tree canopy vegetation (black) and open sky (white). This step is the only subjective part of the entire operation. Step 3. The IA system then compared the threshold image prepared in step 2 with the stored image prepared in step 1. Each annulus in the stored image was considered in turn. Each pixel within the annulus was compared with its corresponding pixel on the threshold image of the photograph and allocated a score of 1 if below the threshold and 2 if above it. The total number of pixels scoring 1 and 2 in each annulus was then counted and output to a storage file. Step 4. The data in the storage file were then processed via a computer program which compensated for the differing numbers of pixels in each of the 20 annuli. The percentages of open sky and shade were then determined for the entire photograph. Although the values for open sky and shade were determined independently their sum was 100%. Calibration of this technique using overlays of known shade (produced with Indian ink) gave an extremely good fit (Fig. 3(a)), with a regression coefficient of almost 1.0 suggesting that the technique is extremely accurate. The main potential source of error in this technique is the manual determination of the threshold. It is essential that this is done manually for each photograph to compensate for differences in photographic exposure or development. This method allows at least 100 photographs to be analysed each day. COMPARISON BETWEEN THE TWO METHODS The percentage shade was estimated using both methods for 78 photographs taken in woodland rides and glades in southern England. The linear relationship between the two methods was exceptionally good (Fig. 3(b)) and the regression coefficient was almost 1.0. Six independent repeat estimates of percentage shade were made for three photographs of varying canopy cover, i.e. low shade--open canopy, medium shade---canopy beginning to close, and high shade---canopy obviously closing. Mean estimates + standard errors for the three categories were %, % and % with the manual method and %, % and % with the semi-automated method. The range around any individual mean was always < = + 5% of the mean value.

6 332 Julie Barrie, J. N. Greatorex-Davies, R. J. Parsell, R. H. Marrs (a) 8O 20 io 4"0 6"0 8"0,6o Known shade (%) I00 8O m "~ eo 0 ~ o ~ 4b e~ sb ~6o 128 estimate of shade (%) Fig. 3. Relationship between (a) estimated cover from the semi-automated I2S method and known shade standards; and (b) estimated cover from the manual and semi-automated I2S method. Linear regressions: (a) y = 0"998x (r 2 = ff9994, p < ); (b) y = l'01x - 0"09 (r = 0"977, p < ). DISCUSSION The use of hemispherical photography is a rapid and relatively inexpensive technique for habitat assessment and routine monitoring of woodlands with a high conservation interest. Hitherto, this approach has been confined to

7 Assessment of woodland shade 333 scientific studies, because the manual method for analysing the photographs is slow, and extremely tedious for operators. The semi-automated method reported here will allow hemispherical photography to be used on a much greater scale, because the photographs can be analysed much more quickly. The data presented here showed that the semi-automated method was extremely accurate against known shade standards, and there was a very good correlation with the manual method. The standard errors and ranges of the repeated estimates on individual pictures suggest that the techniques should be able to detect changes in canop~ cover of between 5 and 10%. This level of accuracy is probably good enough for practical conservation work. A major advantage of the semi-automated method is its speed, as it can process at least five times as many photographs per day. Moreover, in extensive studies with many pictures it would be possible to achieve a much higher standard of quality control with the semi-automated method, by including repeated assessments of standard pictures, and replication of randomly chosen pictures in any run. It is difficult to compare costs of each method as the semi-automated method has reduced staff costs (with or without overheads) which must be offset against the increased costs of buying or hiring time on the digitizing equipment. We have found that the costs of the manual methods are % cheaper if done by casual or junior staff (Assistant Scientific Officer grade, differences in costs depends on the overheads included), but if experienced technicians (Scientific Officer grade or above) are used the semiautomated method is at least 20% cheaper. Our costings included the full economic hourly costs of running the I2S system, an advanced machine designed for remote sensing work. Costs could be considerably reduced if cheaper digitizing equipment were used, or perhaps if there were central facilities serving a wide range of conservation groups. ACKNOWLEDGEMENTS We would like to thank Mr J. Moiler for help with computer programming. REFERENCES Anderson, M. C. (1964). Studies of the woodland light climate, I. The photographic computation of light conditions. J. Ecol., 52, Chadzon, R. L. & Field, C. B. (1987). Photographic estimation of photosynthetically active radiation: evaluation of a computerized technique. Oecologia, 73,

8 334 Julie Barrie, J. N. Greatorex-Davies, R. J. Parsell, R. H. Marrs Evans, G. C. & Coombe, D. E. (1959). Hemispherical and woodland canopy photography and the light climate. J. Ecol., 47, Evans, G. C., Freeman, P. & Rackham, O. (1976). Developments in hemispherical photography. In Light as an Ecological Factor, ed. G. C. Evans, R. Bainbridge & O. Rackham. Blackwells Scientific Publications, Oxford, pp Hall, M. L. & Greatorex-Davies, J. N. (1989). Management Guidelines for the Conservation of Invertebrates, Especially Butterflies, in Plantation Woodland (Natural Environment Research Council Report to Nature Conservancy Council HF3/08/12). Institute of Terrestrial Ecology Huntington. Mitchell, P. L. & Woodward, F. I. (1987). Instrument for measuring temperature, photosynthetically active radiation and daylength, and its use in the measurement of daylength and temperature in coppice. J. Appl. Ecol., 24, Warren, M. S. (1981). The ecology of the wood white butterfly Leptidea sinapsis (Lepidoptera, Pieridae). PhD thesis, University of Cambridge. Warren, M. S. (1985). The influence of shade on butterfly numbers in woodland rides, with special reference to the wood white Leptidea sinapsis. Biol. Conserv., 33, Woodward, F. I. & Yaqub, M. (1979). Integrator and sensors for measuring photosynthetically active radiation and temperature in the field. J. AppL Ecol., 16,