Closed ghost closed 6 years ago
Just as a spot (gage or later any watershed "outlet" lat/lon) provides a cumulative exposure of sunlight from its latitude (solar at top of atmosphere [the data being discussed in this thread]) + net loss for cloud cover + net loss because of humidity, a generalized dryness beyond temperature is expressed. All things equal between a Alabama and Texas watershed, the Texas one should go dry more frequently because of the generalization of more direct solar radiation in much of Texas relative to Alabama.
I should have added this figure too. It shows our gages. Blue are sites with no zero-flow days. Red are those with the the MOST dry shading towards green as the number of days decreases. The + are small (<10 square mile watersheds) [part of some experiments of mine]. You might notice a belt of sublinear RED sites just above the south of Texas. These are aligned downstream of the Edwards aquifer outcrop in which all base flow coming from upstream can be lost into the subsurface. One can see real geological context in this plot.
This figure shows the direct solar irradiance from the citation:
We can look at the loo-cv errors for the logistic regression with and without solar radiation and see if we are justified in pursuing a routed version for NHDPlus V2 catchements.
National Renewable Energy Laboratory, 2018a, Geospatial Data Science---Monthly and annual average solar resource potential for 48 contiguous United States, lower 48 DNI 10-km Resolution 1998--2005: Alliance for Sustainable Energy, LLC for the U.S. Department of Energy, accessed on January 2, 2018 at https://www.nrel.gov/gis/data-solar.html and https://www.nrel.gov/gis/assets/data/us9805_dni.shp.xml
National Renewable Energy Laboratory, 2018b, Geospatial Data Science---Monthly and annual average direct normal irradiance for Hawaii and the contiguous United States, lower 48 and Hawaii DNI 10-kilometer resolution, 1998--2009: Alliance for Sustainable Energy, LLC for the U.S. Department of Energy, accessed on January 2, 2018 at https://www.nrel.gov/gis/data-solar.html and https://www.nrel.gov/gis/assets/data/us_dni.shp.xml
During week after Christmas, I have been testing the 1998--2009 version of solar radiation as a conceptual experiment in estimating percentage of zero flow as a logistic regression problem using a mgcv::gam(). Some promising results for 2000 decade shown below. This is really promising without yet deep study of local issues (karst recharge zone in Texas or other sites dominated by springflow, or urban locations, etc.) Perhaps 50% of the variation is easily explained right off the cuff.
Family: quasibinomial; Link function: logit Formula: pplo ~ s(CDA) + s(ANN_DNI) + s(MAY, DEC, k = 10) + s(east, north, k = 15) Parametric coefficients: Estimate Std. Error t value Pr(>|t|)
(Intercept) -5.26258 0.03648 -144.2 <2e-16 ***
Approximate significance of smooth terms: edf Ref.df F p-value
s(CDA) 5.149 6.251 2281.2 <2e-16 s(ANN_DNI) 7.964 8.661 868.1 <2e-16 s(MAY,DEC) 8.259 8.836 638.6 <2e-16 s(east,north) 13.285 13.862 847.6 <2e-16 R-sq.(adj) = 0.544 Deviance explained = 63.9% GCV = 325.23 Scale est. = 2.9851 n = 568
CDA figure shows the smooth contribution (y-axis) of the log10 of the drainage area. Message: percent decadal no flow goes up as drainage area goes down. INVERSE RELATION
ANN_DNI (annual direct normal total solar input) smooth shows that as the irradiance goes up (think moving west into and through Texas) that percent decadal no flow goes up. DIRECTION RELATION.
Bivariate smooth of December and May direct normal irradiance showed maximum input of more information based on analyses of all month permutations. Interesting that they are nearly 1/2 year apart from each other.
The east,north bivariate smooth is the albers equal area coordinates. You can now see the distribution of the gages in the general study area.