ECOSTRESS georectification errors

[jondho]

I have just started using ECOSTRESS data - downloaded from the Earthsearch portal and have noticed some large geolocation errors. Attached is a very small (due to file upload limitations) animated gif showing this issue. In one scene (ECOv002_L2T_LSTE_42186_015_19KDR_20251214T063845_0713_01), the data lines up perfectly with a road, but in a second (ECOv002_L2T_LSTE_42064_013_19KDQ_20251206T094808_0713_01), the data appears around 2km to the south and 600m to the west of the true position. Do you have any advice on how to identify these errors? I can't see any information in the metadata which refers to the position accuracy.

The point of interest is at -22.631383,-69.145513

The data downloaded was:
- Good: https://data.lpdaac.earthdatacloud.nasa.gov/lp-prod-protected/ECO_L2T_LSTE.002/ECOv002_L2T_LSTE_42064_013_19KDQ_20251206T094808_0713_01/ECOv002_L2T_LSTE_42064_013_19KDQ_20251206T094808_0713_01_LST.tif
- Bad: https://data.lpdaac.earthdatacloud.nasa.gov/lp-prod-protected/ECO_L2T_LSTE.002/ECOv002_L2T_LSTE_42186_015_19KDR_20251214T063845_0713_01/ECOv002_L2T_LSTE_42186_015_19KDR_20251214T063845_0713_01_LST.tif

[lucas89898]

What you’re seeing is a known issue that can occur in some ECOSTRESS L2T scenes, particularly those affected by attitude or ephemeris uncertainties during acquisition. The L2T products are terrain-corrected, but residual geolocation offsets on the order of hundreds of meters to a few kilometers have been reported for certain orbits.

A good first step is to compare the GEO or L1B_GEO products for the same granule to check scan geometry and look for along-track or cross-track shifts. Many users also validate alignment against a stable reference (e.g., Landsat or Sentinel-2) and apply a manual correction when needed. Unfortunately, explicit per-scene geolocation accuracy metrics are not currently exposed in the metadata, so reporting the affected granules to LP DAAC or the ECOSTRESS science team is the best way to flag potential upstream issues.

[LP DAAC - dgolon]

Hello @jondho We are looking into this. I will report back when we have additional details. Thanks -- Danielle

[LP DAAC - dgolon]

Hi @jondho We spoke via email but also posting here in case any other users are interested in the answer.

The first image (ECOv002_L2T_LSTE_42186_015_19KDR_20251214T063845_0713_01) has a geolocation flag of "best" - the user can expect the geolocation to be 50 meters or less with what is mentioned in the user guide: When we successfully match a scene in an orbit, the geolocation accuracy is better than 50 meters. Beginning with Build 7, if the co-registration is poor or fails, up to four re-tries are attempted (using modified parameters) to match the scene. However, we do encounter orbits where no scenes successfully match (Figure 4.2.1). This might occur because scenes are over water, or the scenes are cloudy, or we just have scenes without a lot of features to match. In these cases, the geolocation error is significantly larger. As described in the ATBD, the worst-case geolocation error for uncorrected data is 7 km. Typically we see smaller errors in the 2-4 km range when only the ISS attitude information is available for geolocation.

The second scene that the user reported that is around 2km to the south and 600m to the west is has been flagged as "suspect" - so the user can expect to see errors in the 2-4km range. For future reference the user can reference either the corresponding ECO_L1B_GEO file or consult the summary text file that is available on the dataset landing page.

Thanks! -- Danielle