In full transparency, I did a somewhat deep dive using the Esri Support AI Chatbot for this question as it falls outside of my expertise. Testing of suggested workflows below are strongly recommended:
<CLIP>
Let me know if this works (or does not work) and I'll examine support cases for you.
Large min/max EVI values in ArcGIS Pro with Landsat 8/9 **Collection 2 Level-2 SR** are usually caused by one (or more) of these:
1) **Computing EVI on unscaled integer DN values (or not forcing float math)**
L2 SR bands are stored as integers; EVI should be computed on **scaled reflectance as floating point**:
**SR = DN * 0.0000275 + (-0.2)**
2) **Including invalid pixels (clouds/shadows/water-fill) or SR values outside a valid range (often negatives)**
Those pixels can blow up the EVI denominator and produce extreme outputs. Esri recommends masking out unwanted pixels using the QA band and excluding out-of-range SR values because they can cause band indices to be wrong.
Reference workflow: [Clean up your Landsat imagery: removing cloud and cloud shadow](https://www.esri.com/arcgis-blog/products/arcgis-pro/imagery/clean-up-your-landsat-imagery-removing-...)
3) **Wrong band mapping/order**
EVI expects **NIR, Red, Blue**. For Landsat 8/9 that is **B5 (NIR), B4 (Red), B2 (Blue)**.
EVI formula reference: [Band Arithmetic function—ArcGIS Pro](https://pro.arcgis.com/en/pro-app/3.4/help/analysis/raster-functions/band-arithmetic-function.htm)
## How to obtain valid EVI
Try this approach:
- Mask clouds/shadows (and other unwanted categories) using the **QA** product (set unwanted to NoData).
- Optionally mask SR DN outside a reasonable range (commonly **0–10000** in DN space) before scaling (per Esri’s guidance in the blog above).
- Scale B2/B4/B5 to float reflectance (DN * 0.0000275 - 0.2).
- Compute EVI: **2.5 * (NIR - Red) / (NIR + 6*Red - 7.5*Blue + 1)**
<END CLIP>
