The signature event of the Fukushima nuclear meltdowns was the large mushroom-cloud explosion of Unit 3 on March 14th. In contrast, the explosion of Unit 1 lacked any notable vertical projection. Yet Tokyo Electric Power Company assumes each was a hydrogen explosion in the upper-deck above the reactor. However, because dramatically different effects suggest different causes, let us consider an evidence-based model wherein the Unit-3 explosion was a steam explosion that vaporized tons of injected seawater into a mushroom cloud and that triggered secondary hydrogen explosions.
Figure 1: Unit 1 lacked the vertical magnitude and mushroom cloud of Unit 3. So it seems something extra happened at Unit 3. The mushroom cloud is composed of tons of mass consistent with tons of vaporized water.
The risk of a steam explosion during a meltdown in the containment vessel housing a reactor has been a matter of considerable concern and research, as noted in Moriyama et al.:
The steam explosion caused by the contact of molten core and coolant [water] is recognized as one of the potential threats to the integrity of the containment vessel during a severe accident of light water reactors and one of the important sources of uncertainty in the evaluation of frequencies of large early fission product releases. 
Because seawater was injected into Unit-3 reactor in an effort to cool it during its meltdown, the necessary ingredients for a steam explosion were in the containment before the explosion. So given that a steam explosion is a recognized risk under such circumstances, the possibility of a steam explosion requires investigation, which we shall embark upon forthwith.
Distinct steam plumes from the containment
As soon as the clouds of the explosion cleared, two distinct steam plumes were seen rising from the demolished upper deck of Unit 3. Figure 2(a) shows Unit 3 three minutes after it exploded, and there we see two distinct steam plumes. Those two plumes were seen throughout the early Spring when Unit 3 steamed, as in Figure 2(b,c,d).
Figure 2 (a through d): distinct steam plumes billow from Unit 3 seen throughout the early Spring after it exploded. (e) The persistent steam-plume pattern maps onto steam coming from the containment vessel.
Figure 2(e) maps the steam plumes to the Unit-3 blueprint. Not surprisingly the large volume of billowing steam correlates with a large container of boiling water.  The only other body of water on site is the spent-fuel pool on the south side of Unit 3 (see the spent-fuel pool in Figures 2(e) and 3). However, the steam plumes emanate from points around the center of Unit 3, and billow out with some gusto just like steam from holes in a container of boiling water. Clearly, these distinct steam plumes are not coming from the fuel pool.
Well-cap hotspots match steam plumes
Figure 3 maps hots spots on infra-red heat-detecting photographs to the floor plan of Unit 3, finding that key hot spots line up with the rim of the reactor-well cap. These hots spots in turn line up with the steam plumes in Figure 2 and with explosives forces to be seen in Figure 4.
Figure 3: Animation: hotspots correspond with well cap, steam and blasts in Figure 4. Note that the fuel pool on left is off-center; it is also hot from stored spent fuel.
Explosion plumes match steam plumes
Figure 5 runs the initial video frames of the Unit-3 explosion. Notice that there are distinct explosive plumes, most obviously the fiery plume shaped like a fist that punches out through the top of the south sunlit wall. Note too that the initial explosive plumes did not project straight up like the mushroom cloud that followed them, but instead they blasted out along roughly 45° angles. Their angled vectors converge on the rim of the reactor-well cap where the steam plumes also come from. Therefore, in this steam-explosion model these explosive plumes are a phase of the steam plumes that immediately followed seen in Figure 2.
Figure 4 Animation: model of the initial phase of explosion based on data in Figures 2 and 3. The water in our model is darkened by contamination from molten off-gassing fuel.
Figure 5 demonstrates the explosion-triggering mechanism, an ex-vessel steam explosion as described in Moriyama et al. wherein water has pooled at the bottom of the containment vessel below the reactor. Then molten fuel falling through a melt-through hole in the reactor’s bottom triggers a steam explosion as it strikes the water below the reactor.  So in our model for Fukushima, seawater injected into the Unit-3 reactor flowed out of the reactor and pooled in the containment vessel. Molten fuel dropping from the reactor then triggers a steam explosion that then triggers secondary hydrogen explosions. [1,3]
Figure 5 Animation: ex-vessel steam explosion triggered by molten fuel falling in water.
In Figure 6 all our observations come together to form a consistent and coherent ex-vessel steam-explosion model that maps perfectly onto the explosion of Unit 3. We run this model here further than the clip in Figure 4 to the point of ‘mushroom blossoming’, which thereafter follows as expected, a large ball of fuel-dirtied steam rolling upwards into the sky. We presume that the force of the explosion in the containment momentarily lifted the reactor-well cap, allowing a significant portion of the seawater to escape before falling shut again. But blast damage to the cap’s seal allowed steam to billow out for weeks as seen in Figure 2.
Figure 6 Animation: ex-vessel steam-explosion model mapped onto the explosion of Unit 3.
Data from instrumentation shows that the Unit-3 explosion was associated with a significant rate of pressure change (a pressure drop) in the containment vessel (aka the drywell, or D/W) just as would expected with a sudden explosive ejection therefrom. 
Figure 7: the explosion coincided with a sudden containment-pressure drop.
TEPCO’s theory that the Unit-3 explosion only involved an explosion of hydrogen gas in the upper-deck space above the containment is challenged by the simultaneous loss of pressure from the containment vessel, clearly indicating its involvement with the explosion.
There is also an indication that seawater injected into the reactor was leaking out, which would thereby flood the containment vessel as depicted in Figure 5. Twenty hours before Unit 3 exploded, TEPCO also reported in a press release (underscore added):
Taking account of the situation that the water level within the pressure vessel did not rise for a long time and the radiation dose is increasing, we cannot exclude the possibility that the same situation occurred at Unit 1 on Mar 12 will occur. 
That the water did not rise for a long time is consistent with the water flowing out of the reactor. And that it eventually did rise is consistent with the level in the containment eventually rising high enough to allow the level in the reactor to finally rise. However, bear in mind that these are inferences from one statement about a complex situation and that even those on site at the time could not be certain about the meanings of water-level data.
Given that an ex-vessel steam explosion during a meltdown is a recognized by the nuclear industry and scientists as a serious risk, it is surprising that the only mention of it with respect to the Fukushima meltdowns found via Google is in a report by Greenpeace Germany.  Also surprising is that there has to date been no explanation or even acknowledgement of the dramatic differences between the Fukushima explosions from industry, government or academic sources. And yet understanding exactly how nuclear plants have exploded would obviously help safeguard the public from future nuclear catastrophes.
In the Japanese Government’s report, the Unit-3 explosion is explained as: "An explosion, which was likely a hydrogen explosion, occurred at the upper part of the reactor building at 11:01 on March 14."  That’s it! For an explanation universally accepted without question to be asserted in passing as merely likely is surprising. Moreover, it is likely relative to what? If I say "Rain is likely," we know that means it is likely relative to not raining, and we know what not-raining is. Yet there is no mention of any other possible cause relative to which this likelihood is favored. The term steam explosion does not even appear in the report. So it seems either only Greenpeace is familiar with the nuclear literature, or the government and TEPCO have opted to keep quiet about other possible causes.
Considering that leakage of coolant in the containment is a precondition for a much-feared ex-vessel steam explosion, it is curious that TEPCO stated in almost every press release before Unit 3 exploded: "Currently, we do not believe there is any reactor coolant leakage inside the reactor containment vessel."  Prefaced on what is believed, that is primarily a statement about belief that serves as a way of saying: We know nothing about any leakage. Such a denial of knowing that a critical ex-vessel steam-explosion precondition may exist smacks of pre-litigation maneuvering intended to reduce TEPCO’s potential liability.
A possible reason for failure disclose an explosion in the primary containment is that if the public knew that the contents of nuclear containments could be blasted straight out and onto them, that could put the health of the government-created nuclear industry at risk versus believing it was a hydrogen explosion in a clean room above the containment. Unlike TEPCO’s theory, a containment explosion necessitates massive nuclear fallout, and if this was recognized as a real risk that did happen, this could allow the public to defend itself and thereby jeopardize the survival of the nuclear-welfare industry.
In closing, the evidence in this report points consistently to an explosion within the containment vessel and thus most likely to an ex-vessel steam explosion within that large container of boiling water.  This type of steam explosion is the most likely type because research indicates that an in-vessel steam explosion (aka an alpha-mode containment failure) occurring inside the rector itself is considered very unlikely to breach the containment vessel, and thus would be unlikely to produce the dramatic explosion of Unit 3. 
The multimodal empirical evidence reviewed above demonstrates that (a) plumes of steam, (b) thermal hotspots, (c) explosive forces and (d) a steam-like mushroom cloud all correspond with vectors whose origins converge around the lid of a large container of boiling water known as the containment vessel. Additionally, instrumental measurements show that pressure within the containment vessel dropped suddenly with the explosion (consistent with an explosion from the containment) and that the day before the explosion, water levels failed to rise in the reactor for a long time despite in-flowing water (consistent with water flowing out of the reactor and pooling in the containment vessel). Finally, given the presence of fire in two of the explosive plumes (Figures 4 and 6), the explosion in the containment probably initiated secondary explosions of hydrogen gas accumulated in both containment-vessel space and the upper-deck space above the containment vessel.
 We need not hypothesize that the container of water was boiling because its boiling is a fact accepted by any knowledgable observer because (1) water around melting and molten nuclear fuel is necessarily boiling and has to be replaced constantly to quench the rapid rate of boil off, and (2) the steam plumes seen in Figure 2 clearly demonstrate that water within Unit 3′s leaking containment vessel was boiling. Recognizing that the containment vessel was a large container of boiling water, like a large pressure cooker before its seals failed, the theory that it suffered a steam explosion is recognizable as the default theory.
 Prime Minister of Japan and His Cabinet. (2011). Report of Japanese Government to the IAEA Ministerial Conference on Nuclear Safety – The Accident at TEPCO’s Fukushima Nuclear Power Stations, Chapter 4.