search

terraref / computing-pipeline

Pipeline to Extract Plant Phenotypes from Reference Data
BSD 3-Clause "New" or "Revised" License
23 stars 13 forks source link

Examine outliers from canopy cover time series #590

Open max-zilla opened 5 years ago

max-zilla commented 5 years ago

image.png

You can select Season 6 Canopy Cover on traitvis and hover over the spikes to see which days have the spikes (you've probably done this already): https://traitvis.workbench.terraref.org/

My approach for these was to use files on Globus, e.g. for May 6: /ua-mac/Level_2/rgb_fullfield/2018-05-06/rgb_fullfield_L2_ua-mac_2018-05-06_stereovis_ir_sensors_fullfield_sorghum6_shade_may2018_mask_canopycover_bety.csv This contains a row for each plot with the CC value, easier to find which plots are above 5%.

/ua-mac/Level_2/rgb_fullfield/2018-05-06/rgb_fullfield_L2_ua-mac_2018-05-06_stereovis_ir_sensors_fullfield_sorghum6_shade_may2018_mask.tif The _mask.tif files were actually used to calculate the CC, but they can be large.

/ua-mac/Level_2/rgb_fullfield/2018-05-06/rgb_fullfield_L2_ua-mac_2018-05-06_stereovis_ir_sensors_fullfield_sorghum6_shade_may2018_rgb_10pct.tif The RGB 10% resolution files don't have the soil mask, but they are smaller in size and you can still see targets or other things that might affect the CC estimation.

I've attached a JSON plotmap file you can drag and drop into QGIS to overlay the plots onto the GeoTIFF images. The plot names are included as fields. This makes it fast to see where in the image the higher values come from: https://app.zenhub.com/files/41696258/9b78b746-deec-4df6-a0b8-1ba3f1d89334/download

ZongyangLi commented 5 years ago

To coordinate with May 7 error data, I ran a full day canopy cover ration process. 52 image will return a value over 5% canopy cover, in totally 9096 images.

Cases show below: 0 190524407343 0 118628149841 0 0671433894571 0 0937474225893

Some caused by calibration target in the field, some caused by some unexpected object, and also there are some area the algorithm can not work well. But a 0.5% error data seems not a huge mistake.