Super interesting, thanks! Any clue why accuracy declined for the tropics over 2004-2006 in figure 4? Just an anomaly?
And out of curiousity, do any authors or experts have a sense for how likely it is that machine learning will significantly change the situation? (I would expect rich countries to drive this innovation so I doubt it’s of philanthropic interest)
Hey Ben—thanks for your question. Unfortunately we didn’t come across anything in our research that would quickly explain the decrease for 2004-2006 (or the increase in 2012-2016). It could have been differences in the atmospheric conditions at that time that the model was less able to handle, or changes to the ECMWF forecasting methods.
Re. machine learning, it does seem likely that there’s scope for that here. While we didn’t look into this in detail, we mention two ideas briefly in this spreadsheet of potential interventions.
Use machine learning to downscale (i.e. get better resolution) of existing forecasts: Tim Palmer at Oxford is a proponent of this.
Use machine learning to improve forecasts: this seems plausible and interesting, and I don’t know that we should expect that rich countries will drive improvements that close the gap in forecast quality between the tropics and NH. It also seems likely that the quality/ quantity of data from the tropics will put some limits on how successful ML can be for some locations.
Would be interested to hear if you have any further thoughts on either topic!
Thanks! That all makes sense. I think I was imagining ML-based improvements to drive accuracy in absolute terms—so it wouldn’t close the NH-tropic gap, but could raise tropic accuracy overall. But provided there are incentives for improved accuracy in the NH, I’d expect private investment to pursue it.
I agree—the data quality/quantity seems like larger bottlenecks to improving tropic accuracy. It seems possible that ML-approaches that work better with poor quality/quantity data may be sufficiently different to NH problems such that the expected private investment wouldn’t translate into improvements for the tropics, maybe opening up potential for philanthropy to have an impact… but that’s a long chain and I don’t anywhere near enough about ML/weather forecasting to make a good guess.
Super interesting, thanks! Any clue why accuracy declined for the tropics over 2004-2006 in figure 4? Just an anomaly?
And out of curiousity, do any authors or experts have a sense for how likely it is that machine learning will significantly change the situation? (I would expect rich countries to drive this innovation so I doubt it’s of philanthropic interest)
Hey Ben—thanks for your question. Unfortunately we didn’t come across anything in our research that would quickly explain the decrease for 2004-2006 (or the increase in 2012-2016). It could have been differences in the atmospheric conditions at that time that the model was less able to handle, or changes to the ECMWF forecasting methods.
Re. machine learning, it does seem likely that there’s scope for that here. While we didn’t look into this in detail, we mention two ideas briefly in this spreadsheet of potential interventions.
Use machine learning to downscale (i.e. get better resolution) of existing forecasts: Tim Palmer at Oxford is a proponent of this.
Use machine learning to improve forecasts: this seems plausible and interesting, and I don’t know that we should expect that rich countries will drive improvements that close the gap in forecast quality between the tropics and NH. It also seems likely that the quality/ quantity of data from the tropics will put some limits on how successful ML can be for some locations.
Would be interested to hear if you have any further thoughts on either topic!
Thanks! That all makes sense. I think I was imagining ML-based improvements to drive accuracy in absolute terms—so it wouldn’t close the NH-tropic gap, but could raise tropic accuracy overall. But provided there are incentives for improved accuracy in the NH, I’d expect private investment to pursue it.
I agree—the data quality/quantity seems like larger bottlenecks to improving tropic accuracy. It seems possible that ML-approaches that work better with poor quality/quantity data may be sufficiently different to NH problems such that the expected private investment wouldn’t translate into improvements for the tropics, maybe opening up potential for philanthropy to have an impact… but that’s a long chain and I don’t anywhere near enough about ML/weather forecasting to make a good guess.