Abstract
The sensitivity of the Indian monsoon to the full spectrum of climatic
conditions experienced during the Pleistocene is estimated using the
climate model HadCM3. The methodology follows a global sensitivity
analysis based on the emulator approach of Oakley and O'Hagan (2004)
implemented following a three-step strategy: (1) development of an
experiment plan, designed to efficiently sample a five-dimensional
input space spanning Pleistocene astronomical configurations (three
parameters), CO2 concentration and a Northern Hemisphere
glaciation index; (2) development, calibration and validation of an emulator of
HadCM3 in order to estimate the response of the Indian monsoon over
the full input space spanned by the experiment design; and (3)
estimation and interpreting of sensitivity diagnostics, including
sensitivity measures, in order to synthesise the relative importance
of input factors on monsoon dynamics, estimate the phase of the
monsoon intensity response with respect to that of insolation, and
detect potential non-linear phenomena.
By focusing on surface temperature, precipitation, mixed-layer depth and
sea-surface temperature over the monsoon region during the summer season
(June-July-August-September), we show that precession controls the response
of four variables: continental temperature in phase with June to July
insolation, high glaciation favouring a late-phase response,
sea-surface temperature in phase with May insolation, continental
precipitation in phase with July insolation, and
mixed-layer depth in antiphase with the latter. CO2
variations control temperature variance with an amplitude similar
to that of precession. The effect of glaciation is dominated by the
albedo forcing, and its effect on precipitation competes with that
of precession. Obliquity is a secondary effect, negligible on most
variables except sea-surface temperature. It is also shown that
orography forcing reduces the glacial cooling, and even has
a positive effect on precipitation.
As regards the general methodology, it is shown that the emulator
provides a powerful approach, not only to express model sensitivity
but also to estimate internal variability and detect anomalous simulations.