The question you should ask is why did the models predict an increased OLR even though the CO2 level had risen. The answer is that the region observed is right in the center of the ENSO pattern of variation. If you look at the pattern of ENSO variation, you will see that while there were slightly cool ENSO conditions in that zone in 1997, they were very much cooler in 1970:
Remember, warmer temperatures increase OLR, and the 1997 temperatures were distinctly warmer, and warmer beyond the mere expectation from global warming due to the ENSO pattern. That additional warmth above the AGW trend increased OLR beyond the additional reduction due to the slight increase in CO2 over that period. Indeed, it was only because of the additional warmth due to ENSO that the OLR increased. Had the increase in warmth been only that of the trend, the net OLR would have declined slightly.
Harries et al did not leave it there. They used a model to correct for the temperature difference, thereby showing the impact of greenhouse gases apart from the changes in temperature:
As expected, the change in GHG concentration reduces OLR.
I know that pseudoskeptics attempt to dismiss this data because a model was used to generate it. It was not, however, climate model. It was a radiation model (specifically Modtran3). This is the sort of accuracy you can get with radiation models:
Because the adjustment was done with radiation code, denying the validity of the adjustment is tantamount to denying radiative physics altogether. It puts those who do it into flat earth society territory as regards to the level of their pseudo-science.