Dr Helen Greatrex
(International Research Institute for Climate and Society Columbia University, New York)
Geostatistical approaches to the quantification of satellite rainfall uncertainty
| What | ADAPT HOMEPAGE EventsMenuItem |
|---|---|
| When |
Oct 09, 2018 04:00 PM
Oct 09, 2018 05:30 PM
Oct 09, 2018 from 04:00 pm to 05:30 pm |
| Where | Rm 529 Walker Building, Penn State University |
| Add event to calendar |
vCal
iCal
|
Satellite based rainfall estimates (SRFEs) are increasingly being used for climate risk management, especially regions with high climate vulnerability and sparse data. For example, in 2018, over a million farmers in sub-Saharan Africa were covered by weather insurance that was directly triggered by remotely sensed rainfall. It is therefore also increasingly important to be able to quantify the uncertainty on SRFEs and to assess how this might impact downstream products. The process of characterizing the uncertainty on a SRFE is non-trivial and must take into account features including
- Rainfall's mixed non-Gaussian distribution and high spatial intermittency
- Multiple sources of heteroskedastic uncertainty, including bias in the calibration algorithm; error in the interpolation method of the gauge data; measurement errors in the gauge data and the location error of the satellite pixel.
- The inhomogeneity and often low density of observations, and the temporal and spatial variability of spatial correlation patterns.
- The often non-linear relationship between rainfall values and end-user decisions (such as modelled crop yields or insurance pay-outs)
This talk explores how the Monte Carlo, geostatistical technique of stacked sequential simulation can be used to quantify satellite rainfall uncertainty within these constraints, the downstream impact of this uncertainty on agricultural insurance design and how the work is leading to novel satellite-gauge merging techniques.