search

ssj-delta-cu / ssj-overview

Delta Service Area Boundaries and Associated data
Other
2 stars 1 forks source link

Boundaries for land IQ #41

Open josue-medellin opened 7 years ago

josue-medellin commented 7 years ago

@andybell Greetings from POA. Would you please put a three maps with boundaries on the Delta: Land IQ current coverage, Legal Delta and Delta Service Area? I want to make sure Land IQ is inclusive of both the Legal Delta and DSA. I will send this to Land IQ principal so we are all in the same page for both 2015 and 2016 surveys. is there something else you would see beneficial to change?

Kind regards,

Josue

andybell commented 7 years ago

@josue-medellin Land IQ is fully inclusive for both the Legal Delta and the DSA.

landiq_wbounds

josue-medellin commented 7 years ago

@andybell you sent me this last week. are the missing acres in the legal delta because of ommission of boundaries and farmstead in fields? is about 1% of the area. Land IQ claim (a good one I think) is that they report real acreages is land. Thank you, josue

LandIQ_level2_TabulatedArea_EE.xlsx

andybell commented 7 years ago

@josue-medellin I finally think I've tracked down the issue. There are a couple of things going on here:

  1. Reducing by wrong cell size - the script in earth engine to tabulate the areas for each of the level 2 crop categories uses a reducer with a cell size of 25m while the landcover raster asset has a cell size of 30m. This adds a tiny bit of error when estimating the categories since EE resamples the raster.
  2. Land IQ asset in Earth Engine is clipped to bounding box of DSA prior to upload - the land IQ shapefile and geotiff from https://github.com/ssj-delta-cu/ssj-landuse are fully inclusive of the legal delta and delta service area (shown in the map above - https://github.com/ssj-delta-cu/ssj-overview/issues/41#issuecomment-251154296). The asset that is in Earth Engine is clipped in two areas to the DSA bounding box (see images for locations https://github.com/ssj-delta-cu/ssj-overview/issues/36#issuecomment-249671427) . These two areas add up to about 7,000 acres.
nickrsan commented 7 years ago

Nice find - to add maybe something tiny that you both already know, Quinn had a reason for using 25m cell sizes - I'm struggling to remember what it is, but maybe @jrmerz knows (and Andy, maybe you do too). It was something about getting perfect alignment between some other datasets. Sounds like maybe it introduced the errors still, but just wanted to make sure you both knew that.

andybell commented 7 years ago

The amount of error with the reducer resolution is minuscule but it helps explain some of the discrepancies when calculating the totals outside of EE. Plus, the ET summary by Land IQ category (ssj-overview/et_comparisons.js) used a 200 m reducer, so I don't think it is relevant to the end product.

The alignment issue could be significant though - I tried uploading landuse_2015.tif as a new asset to earth engine. The raster did not align with the landuse asset that is already there. I initially thought it might be a projection error or something. Are the steps used to create the landuse asset in earth engine documented somewhere?

Here is a screenshot of the difference between the two landuse assets - solid grey is no difference. Note the jagged edges that show the issues with alignment:

image

josue-medellin commented 7 years ago

Thanks again Andy. @qjhart Would you please explain us the EE transformations undertaking to upload Land IQ data? Thanks, josue

jrmerz commented 7 years ago

@nickrsan sorry, I do not know this issue. Will have to defer to quinn.

qjhart commented 7 years ago

@andybell, The Landuse data was uploaded to EE in epsg:3310. Originally, the results files were supposed to be in EPSG:3310, but since everyone used UTM, we just gave up and made the final product match the landsat imagery. So it would probably be worthwhile to recalculate the raster in that coordinate system, and then upload. 25m was chosen as it's a integer nmber of pixels to fit with other statewide EPSG:3310 grids, BTW. The reason for the jags at a high resolution is an artifact of the way that EE make higher resolution images on the fly, and we use nearest neighbor for sub sampling.

qjhart commented 7 years ago

https://github.com/ssj-delta-cu/ssj-landuse/blob/master/src/raster.sql shows the raster conversion.

josue-medellin commented 7 years ago

@andybell please follow up on this at your earliest convenience. I noticed you uploaded some land use info but not sure how it relates to this.