Calibration of the disturbance magnitude threshold

To determine an appropriate minimum magnitude threshold for the MODTrendr outputs, we applied a range of different thresholds to the training sample data (0-0.2 dNBR), and estimated omission and commission errors from the resulting MODTrendr predictions. The “best” threshold minimizes both commission and omission errors across test sites.

Commission and omission error rates in relationship to absolute change magnitude (dNBR) at test sites. The vertical gray line represents the chosen minimum disturbance magnitude threshold (at dNBR=0.02).

In the Pacific Northwest (USA), a magnitude threshold of 0.08 dNBR best minimized commission and omission errors. For the MODTrendr mapping and assessment of forest disturbances across Northern Eurasia, we selected a threshold of 0.02 dNBR.

Credit:

• D. Pflugmacher, D. Sulla-Menashe, O. Krankina. “ASSESSMENT OF THE MODTRENDR ALGORITHM FOR MAPPING FOREST DISTURBANCES IN NORTHERN EURASIA” NELDA-II REPORT (UNPUBLISHED)