Using FAO NewClim to Build Climatologies Primary attempt – May a. verdin 08/20/2010
FAO - LTM Precipitation > 7300 locations within US/S.Canada/N.Mexico Predictors used for interpolation: – LST (land surface temperature) – IRM (infrared mean) – IRS (infrared std. dev.) – DEM (digital elevation model) – CMORPH (satellite precip estimate) dmax(weighted centers radius; 1degree points): 550 IDW(inverse distance weighting) max integer: 12
FAO Precipitation – May grid
FAO Precip – Station Distribution
FAO Precip – Station Representation Mean Absolute Error = mm R2 = RASTERVALU = predicted precip
FAO Precip – Cross-Validation stn mean val: mm pred. mean val: mm 10th percentile for stn: 17 mm 50th percentile for stn: 88 mm 90th percentile for stn: 126 mm 10th percentiles for pred: mm 50th percentiles for pred: mm 90th percentiles for pred: mm R2:
FAO - LTM Minimum Temperature > 5100 locations within US/S.Canada/N.Mexico Predictors used for interpolation: – LST (land surface temperature) – IRM (infrared mean) – IRS (infrared std. dev.) – DEM (digital elevation model) dmax(weighted centers radius; 1degree points): 550 IDW(inverse distance weighting) max integer: 12
FAO Min Temp – May grid
FAO Min Temp – Station Distribution
FAO Min Temp – Station Representation Mean Absolute Error = deg. C R2 = RASTERVALU = predicted min temp
FAO Min Temp – Cross-Validation stn mean val: degrees C pred. mean val: degrees C 10th percentile for stn: 2.7 degrees C 50th percentile for stn: 8.1 degrees C 90th percentile for stn: 15.3 degrees C 10th percentiles for pred: degrees C 50th percentiles for pred: degrees C 90th percentiles for pred: degrees C R2:
FAO - LTM Maximum Temperature > 5100 locations within US/S.Canada/N.Mexico Predictors used for interpolation: – LST (land surface temperature) – IRM (infrared mean) – IRS (infrared std. dev.) – DEM (digital elevation model) dmax(weighted centers radius; 1degree points): 550 IDW(inverse distance weighting) max integer: 12
FAO Max Temp – May grid
FAO Max Temp – Station Distribution
FAO Max Temp – Station Representation Mean Absolute Error = deg. Celsius R2 = RASTERVALU = predicted max. temp
FAO Max Temp – Cross-Validation stn mean val: degrees pred. mean val: degrees 10th percentile for stn: 17.2 degrees 50th percentile for stn: 22.5 degrees 90th percentile for stn: 29.2 degrees 10th percentiles for pred: degrees 50th percentiles for pred: degrees 90th percentiles for pred: degrees R2:
FAO - LTM Mean Temperature > 5300 locations within US/S.Canada/N.Mexico Predictors used for interpolation: – LST (land surface temperature) – IRM (infrared mean) – IRS (infrared std. dev.) – DEM (digital elevation model) dmax(weighted centers radius; 1degree points): 550 IDW(inverse distance weighting) max integer: 12
FAO Mean Temp – May grid
FAO Mean Temp – Station Distribution
FAO Mean Temp – Station Representation Mean Absolute Error = degrees Celsius R2 = RASTERVALU = predicted mean temp
FAO Mean Temp – Cross-Validation stn mean val: degrees pred. mean val: degrees 10th percentile for stn: 10.1 degrees 50th percentile for stn: 15.3 degrees 90th percentile for stn: 22.6 degrees 10th percentiles for pred: degrees 50th percentiles for pred: degrees 90th percentiles for pred: degrees R2:
FAO PET (potential evapotranspiration) < 200 locations within US/S.Canada/N.Mexico Predictors used for interpolation: – LST (land surface temperature) – IRM (infrared mean) – IRS (infrared std. dev.) – DEM (digital elevation model) – CMORPH (satellite precip estimate) FIRST ATTEMPT… dmax(weighted centers radius; 1degree points): 550 IDW(inverse distance weighting) max integer: 12 SECOND ATTEMPT… dmax(weighted centers radius; 1degree points): 750 IDW(inverse distance weighting) max integer: 20
FAO PET – May grid (FIRST ATTEMPT…) Even just visually analyzing this interpolation causes worry… Statistics agree – REDO!
FAO PET – May grid (SECOND ATTEMPT…) Visually speaking, this looks much better. Let us compare the statistics, just to be certain!
FAO PET – Station Distribution The sparseness of information led to an increase in dmax & IDW values in the second attempt… …Statistics to come… NOW!
FAO PET – Station Representation FIRST ATTEMPT… As expected, the lack of information leads to a poor station representation.. Mean Absolute Error = R2 = RASTERVALU = predicted PET values
FAO PET – Station Representation SECOND ATTEMPT… The increase in dmax & IDW values improve our station values… STILL CHECK CV! Mean Absolute Error = R2 = RASTERVALU = predicted PET values
FAO PET – Cross-Validation FIRST ATTEMPT… stn mean val: pred. mean val: th percentile for stn: th percentile for stn: th percentile for stn: th percentiles for pred: th percentiles for pred: th percentiles for pred: R2: ***Judging solely on the cross-validation summary, we may be deceived into believing the first attempt is acceptable. We know the station representation fails… So let’s take a look at the SECOND ATTEMPT….
FAO PET – Cross-Validation SECOND ATTEMPT… stn mean val: pred. mean val: th percentile for stn: th percentile for stn: th percentile for stn: th percentiles for pred: th percentiles for pred: th percentiles for pred: R2:
FAO hPa (water vapor pressure) > 300 locations within US/S.Canada/N.Mexico Predictors used for interpolation: – LST (land surface temperature) – IRM (infrared mean) – IRS (infrared std. dev.) – DEM (digital elevation model) – CMORPH (satellite precip estimate) dmax(weighted centers radius; 1degree points): 550 IDW(inverse distance weighting) max integer: 12
FAO hPa – May grid
FAO hPa – Station Distribution Looks a little sparse as well… Let’s see how the statistics hold up.
FAO hPa – Station Representation Mean Absolute Error = R2 = RASTERVALU = predicted hPa (water vapor pressure) values
FAO hPa – Cross-Validation stn mean val: pred. mean val: th percentile for stn: th percentile for stn: 10 90th percentile for stn: th percentiles for pred: th percentiles for pred: th percentiles for pred: R2: So, the statistics hold up, although our station distribution may not be the best. :)
FAO Sunshine Gradient ~ 200 locations within US/S.Canada/N.Mexico Predictors used for interpolation: – LST (land surface temperature) – IRM (infrared mean) – IRS (infrared std. dev.) – DEM (digital elevation model) FIRST ATTEMPT… dmax(weighted centers radius; 1degree points): 550 IDW(inverse distance weighting) max integer: 12 SECOND ATTEMPT… dmax(weighted centers radius; 1degree points): 750 IDW(inverse distance weighting) max integer: 20
FAO Sunshine Gradient – May grid FIRST ATTEMPT… Even just visually analyzing this interpolation causes worry… Statistics agree – REDO!
FAO Sunshine Gradient – May grid SECOND ATTEMPT… Visually speaking, this looks much better. Let us compare the statistics, just to be certain!
FAO Sunshine Gradient – Station Distribution Hmm… The distribution looks about as sparse as our PET stations… Larger dmax & IDW will fix this as well?
FAO Sunshine Gradient – Station Representation FIRST ATTEMPT… As expected, the lack of information leads to a poor station representation.. Mean Absolute Error = 2.01 R2 = RASTERVALU = predicted Sunshine Gradient values
FAO Sunshine Gradient – Station Representation SECOND ATTEMPT… The increase in dmax & IDW values improve our station values… STILL CHECK CV! Mean Absolute Error = 1.63 R2 = RASTERVALU = predicted Sunshine Gradient values
FAO Sunshine Gradient – Cross-Validation FIRST ATTEMPT… stn mean val: pred. mean val: th percentile for stn: 45 50th percentile for stn: 60 90th percentile for stn: 70 10th percentiles for pred: th percentiles for pred: th percentiles for pred: R2: ***Judging solely on the cross-validation summary, we may be deceived into believing the first attempt is acceptable. We know the station representation fails… So let’s take a look at the SECOND ATTEMPT….
FAO Sunshine Gradient – Cross-Validation SECOND ATTEMPT… stn mean val: pred. mean val: th percentile for stn: 45 50th percentile for stn: 60 90th percentile for stn: 70 10th percentiles for pred: th percentiles for pred: th percentiles for pred: R2: Lookin’ good…
FAO Windspeed > 300 locations within US/S.Canada/N.Mexico Predictors used for interpolation: – LST (land surface temperature) – IRM (infrared mean) – IRS (infrared std. dev.) FIRST ATTEMPT… dmax(weighted centers radius; 1degree points): 550 IDW(inverse distance weighting) max integer: 12 SECOND ATTEMPT… dmax(weighted centers radius; 1degree points): 750 IDW(inverse distance weighting) max integer: 20
FAO Windspeed – May grid FIRST ATTEMPT… Things seem to be just a little … “off”
FAO Windspeed – May grid SECOND ATTEMPT… Overall, a better-looking spatial spread of information….
FAO Windspeed – Station Distribution
FAO Windspeed – Station Representation FIRST ATTEMPT… As expected, the lack of information leads to a poor station representation.. Mean Absolute Error = R2 = RASTERVALU = predicted Windspeed values
FAO Windspeed – Station Representation SECOND ATTEMPT… The increase in dmax & IDW values improve our station values… STILL CHECK CV! Mean Absolute Error = R2 = RASTERVALU = predicted Windspeed values
FAO Windspeed – Cross-Validation FIRST ATTEMPT… stn mean val: pred. mean val: th percentile for stn: th percentile for stn: th percentile for stn: th percentiles for pred: th percentiles for pred: th percentiles for pred: R2:
FAO Windspeed – Cross-Validation SECOND ATTEMPT… stn mean val: pred. mean val: th percentile for stn: th percentile for stn: th percentile for stn: th percentiles for pred: th percentiles for pred: th percentiles for pred: R2:
Conclusions… The large number of observations for the FAO temperature data along with the strong cross-validation and station representation implies a good fit for these interpolations. FAO precip has the most observations for our area of interest, and the statistics imply a good fit Problems with the FAO PET seem to be stemmed from the sparseness of our observations, leading to highly skewed interpolation. A greater dmax value will result in more “neighboring” stations, and thus a finer end grid. The FAO water vapor pressure grid holds up in cross-validation and station representation. With not even 350 observations, this is surprising. The FAO sunshine gradient ALONG WITH the FAO windspeed would be better off with a different set of predictors. Neither LST, IR, CMORPH, nor elevation have a strong relationship with these worldly variables.
End…? a. verdin 08/20/2010