- Metric Name:  California Black Oak Stands

- Data Vintage:  2020

- Unit Of Measure: Value, 0 to 1000

California black oak serves as important wildlife habitat and as a traditional food source for indigenous Californians. The map is intended to be used to inform – and potentially prioritize – management of California black oak stands (e.g., fuels treatments to protect the resource) and to assist those seeking stands for acorn collection (i.e., for reforestation or food).

A satellite-derived map of California black oak (Quercus kelloggii; QUKE) stand distribution from a model trained to Landsat imagery.- Creation Method:  Statistical models were fit to seasonal median Landsat 8 spectral bands 1 – 7 for the period encompassing 2016 – 2020. Training occurrence data spanned the Sierra Nevada RRK project boundary and consisted of 325 30m radius plots assessed via aerial imagery to have ≥ 90% California black oak (QUKE) canopy cover and filtered to exclude plots that experienced greater than 10% loss of absolute tree canopy cover after the date of the image used to assess QUKE canopy cover (Wang et al. 2022). Training occurrence data were combined with 98,506 pseudo-absence locations. From a candidate set that included multiple model-fitting approaches (e.g., Maxent, Random Forests, LDA) Maxent (default settings, version 3.4.3) was selected for its consistently high out-of-sample predictive performance. Seasonal periods of Landsat imagery were defined as follows: Winter (Jan 1 – March 1), Spring (March 31 – May 20), Summer (June 1 – Aug 18), Fall (Oct 17 – Nov 26). Spatial predictions form the statistical model were masked to exclude agricultural urban areas (FVEG), riparian areas (Abood et al. 2022), meadows (UC Davis & USDA Forest Service 2017), and areas with canopy height less than 5 m (Salo Sciences, Spring 2020). Spatial predictions were multiplied by 1000 and rounded to the nearest integer to reduce file size.

Resulting out-of-sample predictive performance was high for delineating areas of ≥ 90% QUKE canopy cover from the broader landscape (AUC = 0.997; mean QUKE cover in sample = 95%). Though the model was trained on plots with ≥ 90% QUKE canopy cover, out-of-sample performance remained relatively high for areas of 50 – 90% QUKE canopy cover (AUC = 0.981; mean QUKE cover in sample = 80%) and areas of 10 – 50% QUKE canopy cover (AUC = 0.959; mean QUKE cover in sample = 34%). The model appears to have moderate skill in predicting continuous QUKE cover – in our sample (biased toward higher QUKE canopy cover plots with mean QUKE cover of 82%) the Spearman's rank correlation coefficient between the model output QUKE score and QUKE canopy cover was 0.54. Notable areas of commission error include certain other deciduous vegetation types, such as aspen.

QUKE Score

Interpretation

0

Very low likelihood of overstory QUKE dominance or very low QUKE overstory cover.

1 – 50

Low likelihood of overstory QUKE dominance or low QUKE overstory cover.

51 – 500

Moderate likelihood of overstory QUKE dominance or moderate QUKE overstory cover.

501 – 1000

High likelihood of overstory QUKE dominance or high QUKE overstory cover.

- Credits:  --	Center for Watershed Sciences, UC Davis – see Meadows

--	California Forest Observatory (Salo Sciences), 2020