todo

article 2015
    CVG = free sunlight, moonlight, straylight, HEP
    CF_CVF = as above + cloud free
    RCFC = raw cloud free. removes above. includes lights, biomass, aurora, gas flares, volcanoes, noise
    ORCFC = removes outliers for biomass and aurora
    VNL = removes all except lights and persistent gas flares

NOA/NASA /VIIRS + GEE
    VCMCFG = cloud free
    VCMSLCFG = stray light corrected

datasets
    random forest to predict census 775 areas adjusted for UNPD estimates.
    built settlement extents 2013 (already in above)
    landscan


worldpop methodology
    census counts
        any census free methods
        dead - geohive.com

    mapping populations land cover (now replaced by random forest?)
        majority mapped in settlements. remainder using methods elsewhere
            dead - http://www.springerlink.com/content/p12p45n763v67888/?p=76f5da312100404e8c6673bb713ef389&pi=1
        settlements refine land cover data
        high res data used too determine regional/land cover densities - source of high res? countries?
    random forest
        Stevens et al mentions use of settlement locations, land cover etc.
        random forest predicts census data and provides weights
        weights applied to local areas at 100m resolution (though not all data at that resolution)
    VTS bottom up mapping
        where census outdate or unreliable
        oak ridge
        satellite + household surveys
    intra urban
        urban areas only


    worldpop borders
        same as GADM? in geopackage?
        std codes or map names of areas

###########################################################################################
geopandas
rasterio
rasterstats
geopy

gee

plotly
google earth pro
google earth engine
qgis

jupyter notebook
xonsh
###############################################################################################