Empirically validated drought vulnerability mapping in the mixed conifer forests of the Sierra Nevada

Abstract

AbstractSevere droughts are predicted to become more frequent in the future, and the consequences of such droughts on forests can be dramatic, resulting in massive tree mortality, rapid change in forest structure and composition, and substantially increased risk of catastrophic fire. Forest managers have tools at their disposal to try to mitigate these effects but are often faced with limited resources, forcing them to make choices about which parts of the landscape to target for treatment. Such planning can greatly benefit from landscape vulnerability assessments, but many existing vulnerability analyses are unvalidated and not grounded in robust empirical datasets. We combined robust sets of ground‐based plot and remote sensing data, collected during the 2012–2016 California drought, to develop rigorously validated tools for assessing forest vulnerability to drought‐related canopy tree mortality for the mixed conifer forests of the Sequoia and Kings Canyon national parks and potentially for mixed conifer forests in the Sierra Nevada as a whole. Validation was carried out using a large external dataset. The best models included normalized difference vegetation index (NDVI), elevation, and species identity. Models indicated that tree survival probability decreased with greenness (as measured by NDVI) and elevation, particularly if trees were growing slowly. Overall, models showed good calibration and validation, especially for Abies concolor, which comprise a large majority of the trees in many mixed conifer forests in the Sierra Nevada. Our models tended to overestimate mortality risk for Calocedrus decurrens and underestimate risk for pine species, in the latter case probably due to pine bark beetle outbreak dynamics. Validation results indicated dangers of overfitting, as well as showing that the inclusion of trees already under attack by bark beetles at the time of sampling can give false confidence in model strength, while also biasing predictions. These vulnerability tools should be useful to forest managers trying to assess which parts of their landscape were vulnerable during the 2012–2016 drought, and, with additional validation, may prove useful for ongoing assessments and predictions of future forest vulnerability.