Woods experiment revisited

-does greenhouse effect depend on infrared retention?



In 1909, professor Robert Wood conducted an experiment designed to test whether the retention of longwave infrared inside a glass greenhouse, thought to be the reason form greenhouses becoming warmer than the surrounding terrain under strong sunshine, was in fact the reason for this effect. 

Wood's experiment in its original form consisted of two closed boxes containing thermometers, one with a window of standard silica glass, the other with a window of rock salt. These materials were chosen because standard glass blocks (absorbs) almost all longwave infrared, whilst rock salt largely allows it to pass through. Modern replications tend to use infrared-transparent plastics instead of the hard-to-obtain rock salt, which was the only suitable material in Wood's day.

Robert Wood's results were that after exposure to strong sun, both boxes showed a near-identical temperature rise. He put forward the explanation that longwave infrared has little heating effect, with the main temperature rise being due to convective heat losses (due to air currents created when warm air in the box rises) being blocked by either type of barrier.

This result would appear to have some profound implications for climate change theory. Although the infrared-blocking material was a transparent solid, not carbon dioxide gas, the implications of Wood's experiment are that in proportion to the other direct or indirect heating effects of solar radiation, the contribution of longwave infrared is negligible. If that is the case, then it matters not what material is used to block the outgoing longwave infrared, since no matter how effective a greenhouse gas or solid may be, you can never achieve more heating effect than is available in toto.


Since that time, there have as far as I know been only two properly-controlled replications of Wood's experiment. This in itself is surprising, in view of its importance to a theory on which a global budget running into trillions is based. The two replications were by a Professor Vaughan Pratt of Stanford Unversity, and by Professor Nasif Nahle of Monterrey, New Mexico.

Their respective results are available online  (1)  (2)  so I won't repeat them in detail.

The lowdown is that Pratt claimed Wood's results of 1908 to be in error. His results indicated that longwave infrared did contribute significantly, in fact very significantly, to the warming effect. Temperatures inside his glass-fronted box stabilized at 15-20C higher than the film-covered box, with even more heating achieved when perspex was used as the infrared-blocker.

Meanwhile, Nahle's results reaffirmed those of the original 1908 experiment, that trapped longwave infrared has little or no warming effect, the majority of the 'greenhouse effect' being due to the trapping of convection currents. Nahle hypothesised that dampness in one or other of Pratt's test boxes might have given rise to his published conclusion that Wood's original results were in error.

The anomalous findings come as something of a surprise, considering the relatively simple and repeatable nature of the experiment, and the seemingly meticulous attention to detail by both experimenters. What intrigued me more is that no-one has since tried to resolve these different findings.

I therefore got to thinking about doing a Wood's experiment of my own to see what actually happens, and also investigating the possibility that dampness in either box might bring about inaccuracy.


1;image 2;image 3;image 4;image 5;image 6;image

Both of the previous replications had used relatively large boxes, but I could see no reason why a smaller version should not produce similar results. In fact, Wood did not document all of the parameters of his original experiment, and the sizes of the original boxes are unknown.

For my own Wood's I therefore chose two plastic 'Kitchen Corner microwave oven boxes' as suitable containers. These are of polystyrene or similar plastic, with a transparent base and blue lid. The lid achieves a reasonable, though not perfect, air seal. I painted the interiors of the boxes with several coats of ProDriver matt black paint, and the exteriors with ProDriver silver paint. The exact properties of these coatings are not known, however attempts were made to achieve an equal coating on both boxes, this being the most important parameter in a comparative experiment such as this.

Dimensions of these boxes are 196x142 at inside rim, 175x125 inside base, 60mm depth.

The central part of each lid was cut away in its entirety with a 'hot knife' to leave only the rim, thus providing a mounting frame for the glass or plastic window of each box. A sheet of 1.7mm thickness ordinary silicate glass was epoxied to one lid frame. The other was covered with a layer of 'Snappies catering film' -As far as I know Saran Wrap is not available here, and this is the nearest equivalent available to me. Its properties should be similar.

At this stage I got to thinking that if the glass window blocks infrared, then one issue might be that part of the insolation is blocked by the glass and liberated to the room as convection, therefore this could result in a lower temperature than would result if the only effect of the glass were to block outbound infrared. Therefore, in an attempt to counter any such effects both boxes were 'triple-glazed' with two further layers of catering film, spaced apart by supports of self-adhesive draught excluder foam strip. It was felt that this would be a fairer test, that any thermal disadvantage of the glass front would be reduced, leaving only its 'greenhouse effect' to be comparatively measured. The now triple-fronted and rather fragile lids were put aside in a safe place whilst the rest of the equipment was assembled.

The boxes were mounted on a sheet of approx. 5mm hard polyethylene foam, glued to a hardboard backplate. The same foam was used to construct a thermally insulating surround for the two boxes of approx 20mm width. Again, no exact thermal properties are on-hand for this foam, however it was felt to be sufficient to ensure that both boxes had as near an identical thermal environment as was possible. In fact thermal insulation is not strictly necessary, but will increase the range of the temperature rise in both boxes, and thus give better resolution to results.

For temperature measurement I had considered using a pair of 10k thermistors coupled to a Raspberry Pi or similar microprocessor-based logging system, but in the end decided that the effort involved in writing software for such an arrangement was not worthwhile, manual pen and paper (or netbook  and texteditor) recording involving less effort. As mentioned I wanted to control humidity, and an attractive solution was found in the form of a Chinese no-brand temperature and humidity sensor sold in numerous Ebay stores. The units in question have a digital readout of both humidity and temperature, hardwired to a remote sensor head. These were not expected to provide lab-standard accuracy, but adequate repeatabilty was expected to be possible.

The sensor heads were thought to be a little bulky for this application, so the outer covers were removed to reveal an industry-standard type of bead thermistor and capacitive humidity sensor. After calibration these were directly mounted onto a shaped piece of hard foam, its outside surface covered with aluminium foil to protect them from direct sunshine. This assembly was then mounted into the upper side of the box, the cable being taken out through a hole sealed with contact adhesive.


7;image 8;image 9;image 10;image

Prior to final mounting, the bare temperature sensors were calibrated against each other, and against two other thermometers. This was done by placing the sensors on a kevlar sheet on the top shelf of an Ariston 2085 fan oven. The oven was heated to 60C, and the temperature allowed to stabilise with the circulating fan running but the heat off. The peak temperature was 72C. In the heating phase it was noted that the processors of the digital thermometers could not keep pace with the temperature rise in this powerful oven, and gave readings as much as 10C different due to sampling lag.  In the slower cooling phase the readings did however agree to a satisfactory extent, with thermometer 2's reading  being consistently about 0.6C higher than thermometer 1 across most of the range. The two reference thermometers suggested that thermometer 2 was nearer to correct, thermometer 1 reading a little low. 

Both thermistors were then placed in contact with ice floating on water. This gave consistent readings of just above zero. Interestingly, the 0.6C discrepancy seemed to have disappeared at low temperatures. A test with wet steam was not done, as the units are not rated to withstand 100C.

Overall impression was that the thermometers were sufficiently accurate for our purpose. A minor issue was envisaged in that very rapid heating of the boxes in strong sun might not be accurately recorded due to sampling lag, but the more important final temperature should be measurable to within acceptable limits.

The accuracy of the humidity sensors was then assessed by way of placing them in desiccant, and in saturated water vapour. Results were not as encouraging as for the thermistors, with humidity readouts jittering randomly by 10% or more, for no apparent reason. The conclusion was reached that 10% would probably be the best resolution achievable, and that even so, all humidity readings should be taken as ballpark figures only. This was somewhat disappointing but reflects the fact that these are not lab-grade units. It was decided to accept this mediocre humidity sensor performance, but to note the limitation. 

Test runs

11;image 12;image 13;image 14;image 15;image 16;image 17;image 18;image

August 9th, 2014 saw a nice sunny morning, and at 9am the equipment was set up indoors facing an open window with an unobstructed view of the sun. This was intended only as a preliminary run  to get a feel for how the equipment responds, and hence no detailed results were logged. On the night before the test a packet of self-indicating silica gel had been placed inside each box and the lids sealed with masking tape, to largely eliminate moisture.  At the start of the test the gel indicator was observed to be orange (dry) in both.

The box internal temperatures started out at 23C on this relatively warm morning, and rose within 15min to around 42C.  During the heating phase the temperatures tracked together nicely. Thermometer 1 was in the glass-fronted box. As expected, Thermometer 2 gave a slightly higher reading, except  the discrepancy was somewhat more than the 0.6C found in calibration, suggesting that some mechanism or other was reducing the amount of energy entering the glass-fronted box. 

The humidity sensors performed less well, initially giving wildly varying readings from near below 10% to over 30%, but settling as the temperature increased to a reading between 10% and 20%. Thus we can conclude that the in-box humidity was relatively low, well below typical room levels of 50-60%, and similar in both boxes.

Unfortunately I had to go out that morning, so the experiment was halted at that point. As it was I ended up having to eat humble pie as I was 15min late to see my client. But, that is the cost of the advancement of science. ;-)

August 22nd: After the initial test I had to wait some weeks for suitable weather. The morning of the 22nd started bright and clear, and thus the experiment was set up at 08:55 BST in an East-facing window. Unfortunately, the moderate wind brought with it intermittent cloud cover, causing the amount of sunshine to vary over intervals of a few minutes. This made it hard to obtain any consistent results. At 30min into the run the lids were swapped to determine if the performance of the boxes themselves matched. Box 2 exhibited the anomalously high readings seen on the 9th. In fact, apart from an interval during initial warming, Box 2 always gave the higher temperature regardless of whether the glass or plastic lid was in place, which seemed to indicate an equipment problem. This and the randomly varying illumination made the results somewhat questionable.

A run was performed with Box 2's silica gel replaced with 5ml of water on a folded paper towel.  Box 2 humidity readout during this run rose to 57% after five minutes, and remained in that region. Condensation was observed forming on the box window. The effect  was a substantial initial cooling of the 'wet' box (presumably due to the energy needed to heat and vaporise the water) but only a small effect on the final temperature relative to a dry box. In spite of the equipment discrepancy noted above, this was conclusive enough to satisfy me that water vapour content is not a major determinant of final temperature.

A mechanical problem noted during these runs was the multiple layers of film becoming attached to each other as a result of vibration or wind, and their being hard to separate without excessively disturbing the apparatus.

August 25th: This day promised to be free of cloud cover, so the apparatus was set-up at 07:55 to make good use of the early morning sun. With full sun available all morning, a comprehensive set of tests was done, starting with a 30min run for each lid position. Two pairs of outliers from the start of the second run, due to the boxes having been open to air, were not included in the graph. These two runs seemed to indicate a measurable increase in heating in the glass-fronted box. The discrepancy between boxes seen on the 22nd was still very apparent, and this made it hard to estimate the magnitude of the effect of the glass. But, it was clear that there was an effect.

A calibration run with both box lids removed and a double layer of plastic film over the entire apparatus confirmed an anomalous difference in final temperature between the two boxes, Box 2 being the warmer, all other items being equal.

An attempt was therefore made to identify the reason for the discrepancy in box performance. It was noticed that there were small air gaps around the box sides, which had had to be allowed to facilitate lid changing. Additional insulation was fitted to cover this area, and any remaining small gaps were sealed with masking tape. This would make lid exchange harder, however it was felt to be necessary. At the same time, the 'triple glazing' on the lids was reduced to two layers, to eliminate the sticking-together issue.

August 26th: A brief run was made with the two-layer lids and additional sealing of side insulation. The time available was too limited for a full run, but results seemed to indicate far better consistency.

August 27th: With good sunshine and more available time, a 60min run was performed with the well-sealed box insulation.  The results this time were more satisfactory, and suggest that the discrepancy in box performance had been reduced, if not totally eliminated. The plastic-fronted box took the lead in the initial heating phase, only to be overtaken by the glass-fronted box once temperatures began to level out - a curiosity also noted by Pratt.  Owing to other duties the results were not logged between 15min and 50min, however the sunshine was noted to be consistent during that time. At end of run, the glass-fronted box showed a 3.33C greater temperature rise, allowing for the sensor calibration results. 

The outside air temperatures were 19.8C on the sunward side of the building, and 12C on the shady side.

A 'wet run' was also performed, its results agreeing  with those of the 22nd, in that humidity has little effect on final temperatures. 

A control experiment was then performed by way of placing the unit behind double glazing, in full view of sun. After a run of 20min the box temperatures matched to within less than a degree. This is shown for comparison purposes as the two short lines on the graph. Notably, the heating effect for both boxes behind double glazing was only marginally less than that for the plastic-fronted box in full sun, at just under 60C peak. 

June 11th, 2015: In view of the questions left by the 2014 test runs, especially as to whether the two boxes were producing equivalent results, it was decided to do a few more. Starting at just after 7am to get the best of the morning sunshine, the apparatus was set up for a repeat of the double-insulated run, this time without desiccant, starting with clingfilm on on box 1, glass on box 2.

Partway into the first run it was noticed that temperatures in box 1 were falling in spite of strengthening sunshine, and a problem was suspected. It turned out that the sealing tape, after nearly a year in storage, had decided to let go, and allowed a substantial air leak. Thus the experiment was dismounted for resealing.  After the repair the boxes were both ventilated to approximately equalise the temperatures, and the run restarted from that point. Once the temperatures had levelled the box lids were quickly swapped and a further 15min run done, to check for any differences in performance. 

The results indicated a very good correlation between the two boxes, suggesting that the sealing issue had been resolved. They again indicated a significantly greater warming of the glass-fronted box, of around 11C out of a rise above sunny-side ambient of  39.5C.

As a final session, the 'secondary double glazing' plastic film was removed from both boxes and the experiment quickly returned to sunshine, so as to allow a continuation without significant cooling. Whilst doing this it was noticed that the outer glass surface was radiating an amount of heat  which was easily detectable by a hand placed in front of it, without touching. No such sensation was felt from the exposed inner plastic film. This suggests that the hot glass, being opaque to far infrared, was indeed radiating significant amounts of heat energy out of the box, and presumably into it as well. It would have been useful to know if the glass surface was warmer or cooler than the box interior, but this was not instrumented.

The findings of this single-fronted run were perhaps the biggest surprise so far, as the boxes now showed virtually zero temperature difference. The glass-lidded box had started-out significantly hotter from the previous run, but within 10min the temperatures had equalised and remained so throughout. As per standard practice a lid swap was done to eliminate other variables, and this too had very little effect, the temperatures remaining within a degree of each other, both showing a temperature rise of 32C from sunny-side ambient, or 41C above ambient air temperature in the shade.  

19;image 20;image 21;image 22;image

June 27th, 2015:To confirm the unexpected 'single-glazing' run of June 11th, a longer run was performed with single windows. This run there was no lid swap, as it was felt preferable to allow the boxes plenty of time to stabilise at a final temperature. The results were similar to the previous single-layer run, and confirm that insulation of the outside of the glass layer is essential in order to see any temperature difference.


  • A glass/plastic double-fronted box shows a greater temperature rise than one with two plastic film layers
  • The difference is removed if the incoming solar radiation has been pre-filtered through glass
  • Single-fronted boxes in direct sunlight do not show this temperature difference
  • The increased warming with the glass/plastic combo may not be immediately apparent
  • Humidity reduces initial rate of warming, but has no significant effect on final temperature

With double-fronted boxes, an inner glass layer gives rise to a significant additional warming over that of a box with two plastic film layers. The difference is however only a fraction of the temperature rise caused by the prevention of convection currents, which mechanism would seem to make-up the bulk of the 'greenhouse effect' in an actual greenhouse. The actual fraction is a little hard to quantify with accuracy as it varies between experiment runs, but  probably lies between 15% and 25% of the total warming effect over shaded air temperature.

With singe-fronted boxes, no such effect is observed, the temperatures being equal.

When double-fronted boxes are placed behind an overall glass filter with an air gap, likewise no such temperature difference is seen.


It was noted in calibration that Sensor 1 read 0.53C lower than the reference thermometer, however in calculating the temperature rises I subtracted 0.53C from Sensor 2's readings instead of adding 0.53C to Sensor 1's readings, as should have been done. The effect on results is not of any significance since relative change is what we are measuring,  therefore I don't intend to repeat the calculations for the sake of this. I merely note it for completeness.


Comparing this with the previous results, both Wood and Nahle both found no measurable increased in warming through the use of an infrared-absorbing glass window. Nahle reported a negative result, in that the plastic film lid resulted in a very slightly warmer box than the glass-faced one. Both appear to have been using single-fronted boxes, and their results of NO significant temperature difference would seem to correlate well with our single-fronted test runs.

Pratt, on the other hand, reported a far larger increased heating effect in the glass-fronted box, both windows being single layer. Since he does not quote his starting temperatures it is hard to arrive at a figure for temperature rise, but if we assume 25C as his starting point and his claim of 65C with plastic film and 80C with perspex, then his temperature rises would have been 40C and 57C respectively, a ratio of 1.425.   He likewise claimed a 'difference of 15 to 20C' between glass and plastic film. These results far exceed any effect which we found. It is possible that he was operating in conditions of stronger sun, but even so it seems doubtful if that could account for the difference.

Pratt also performed tests with double-layer windows, but seems to have drawn the conclusion that these exhibited less difference than between single layers. Which, is the reverse of our findings here.

Our double-fronted boxes indicated a significant additional 'greenhouse effect' from the the inside layer being glass, of up to 11C in strong sun conditions. Which was still significantly less than Pratt's findings, but nevertheless is a positive result, and one which could not be obtained at all from single-layer windows.

Pratt also noted that the increased warming of the glass-fronted box only became apparent after perhaps ten or twenty minutes of solar heating, with the plastic-fronted box's temperature leading in the initial stages. I also noted this rather surprising effect, which tends to suggest that too short an experimental run may give misleading results. Half an hour would seem a sensible minimum.

Nahle hypothesised that water vapour in one or other of Pratt's test boxes may have given rise to the large difference in heating  Therefore, one of the objectives was to establish whether or not water vapour has any significant effect on final temperature. The answer would appear to be no, it does not.

It would be interesting to see if prof. Pratt has any advice or comments on the nature of his setup that might explain the differences.


It is said that in science there are few genuinely easy experiments. Whilst sounding deceptively easy to perform, in fact it is not easy at all to get consistent results from a Wood's Experiment. In spite of best efforts at making them identical, the two boxes exhibited temperature differentials even when the same lid types were used. Eliminating these sources of error proved nontrivial. Issues were also experienced with variations in insolation to to sun angle and cloud formation, and with varying wind chill. These made consistent experiment runs far more difficult than might have been expected. The results that were achieved do nevertheless satisfy the experiment's objectives, which were to gain some understanding of the effects and magnitude of infrared retention on solar heating.

It was felt that the use of lid-swapping proved effective in providing a control experiment, to identify unwanted influences.

The temperature rises achieved were better than hoped for, and are quite surprising for a Northern climate like Scotland, being only slightly lower than those obtained in much sunnier regions of the USA.

The results obtained by Wood and Nahle seem to correlate with our findings of no effect when boxes are single-fronted. Professor Pratt's results seem to correlate to some extent with our double-fronted results, except that his boxes were seemingly not double-fronted and therefore should not have shown this effect. In any event, the magnitude of his results far exceeded that of anything we found. These differences remain unexplained.

The main point of interest here is that double glazing yields a result, when single glazing does not. Since filtering the infrared part out of the incoming radiation also eliminates the effect, this suggests that the additional heating of the glass-fronted box may be due to solar infrared warming the glass layer directly, rather than its action in trapping 'back radiation' from the box interior. When inbound infrared is eliminated, both boxes appear to warm equally from the visible component of the sunshine.

As to why no such effect is seen with a solo glass layer, this may be since the warming of the exposed glass gives rise to convection currents on the outside of the box which rob it of any additional heat. The additional plastic layer prevents these losses by inhibiting convection from the outside glass surface. It could reasonably be assumed that the hot glass will then re-radiate some infrared both outward and inward equally, and will also warm the air inside the box by convection.

This aspect might warrant some more experimental work, perhaps tests on an instrumented glass window with no 'black body' box behind it, to determine how much heating of the glass itself arises from absorbtion of solar infrared.  I'm thinking of something along the lines of glass window, insulated on both sides with gapped plastic film, a black-painted metal tank containing iced water behind the glass so as to keep the amount of rear 'back radiation' constant as the glass warms, along with circulated cool air in the cavity between this tank and the inside plastic film layer. That would keep the other effects  constant and thus leave only the temperature change due to interception of inbound solar IR.

The cancellation of the effect when a glass pre-filter is in the light path would seem to indicate that inbound longwave infrared plays an important part in the additional temperature rise seen in the glass/plastic windowed box. Exactly what mechanism is operating here, is presently uncertain. Since the glass window would block this infared anyway, whilst the plastic one would allow it to enter the box, it suggests that inbound infrared may be heating the glass window itself, which then warms the air inside the box by convection. This would correlate with the fact that an uninsulated glass window cannot produce an increase in temperature.

Implications for climate

Firstly, it has to be remembered that there are significant differences between a Wood's Experiment and the atmosphere's greenhouse effect.  For one, the infrared absorbtion spectra of glass, carbon dioxide and water vapour differ considerably, although all partially block the range of wavelengths typically emitted by objects at room temperature, or just above. 

Nevertheless, the implications of these results would seem to be that in any situation where inbound solar infrared is filtered-out, any further infrared-retaining layer has little effect on surface temperature. Also, it could be hypothesised that elimination of convective losses is essential for the infrared-blocking 'greenhouse effect' effect to be seen, otherwise such convection effects predominate.

The overall effect on climate is really something for the climate science guys to think on, as the implications are by no means straightforward. Though, from our findings it would seem likely that high-altitude filtering of the infrared component of inbound sunlight by CO2 must preclude any such effect nearer to the ground.

It would also seem probable that since the troposphere does experience convection effects, then a near-surface layer of carbon dioxide could not cause any significant surface warming through thermalisation of outbound infrared, since as with the solo glass barrier, the retained heat in the CO2 layer will quickly convect away upwards. Backscattering of outbound infrared could still provide a small amount of surface warming, however. 

Ian Macdonald, June 2015