UC Berkeley

Mountain pine beetle (Dendroctonus ponderosae; MPB) population has existed at endemic levels in the pine forests of western North America for centuries, but in recent decades it grew to epidemic levels and outbroke over extensive areas from British Columba in Canada to New Mexico in the United States. The current MPB outbreaks have impacted large expanses of lodgepole and ponderosa pine forests, reduced their ability to act as carbon sinks, altered wildfire hazards, affected wildlife populations, changed regional climate, modified local surface energy balance and water quality. Those effects are predicted to increase as a consequence of the direct and indirect effects of climate changes. Despite severe impacts of MPB, substantial unknowns and uncertainties still exist about its historical and current spatial-temporal patterns, future potential distributions, disturbance regime characteristics, ways of interaction with other major disturbance events, and impacts on forest resilience mechanisms.

In this dissertation, I first explored the potential of medium resolution satellite imagery in mapping the chronic insect disturbance in the Southern Rocky Mountains Ecoregion. A forest-growth trend analysis method that integrates temporal trajectories in Landsat images and decision tree techniques was introduced to derive annual forest disturbance maps over a period of one decade. This workflow is able to capture the disturbance events as represented by spectral-temporal segments after the removal of observational noises from temporal trajectories in Landsat images, and efficiently recognizes and attributes events based on the characteristics of the segments. Higher overall accuracy (OA) was achieved when compared with the traditional single-date classifications, and a smaller number of training sample units is required compared with maximum likelihood and random forest classifiers.

To test the feasibility of the trajectory-based approach at broader scales, I advanced this method by replacing the decision tree based semi-automatic event labeling procedure with an automatic attribution step via random forest, which was run on a set of segment features containing information on spatial-temporal neighborhoods. Meanwhile, I developed a new sampling strategy that intensively selects sample units in overlapping areas among images acquired from adjacent rows, and automatically adds spectrally dissimilarity sample units from non-overlapping areas, to improve the efficiency of representative sample selection at the ecoregion scale. The mean OA for all scenes was 82%. The satellite derived multi-temporal landscape quantification results revealed that MPB accounted for 70% of the total area of disturbance. I found that whether fire and MPB are linked disturbances depended on their occurring sequences. Fire severity was largely unrelated to pre-fire MPB outbreak severity, whereas post-fire beetle severity was shown to decrease with fire severity. The recovery rate varied among different disturbance types. Half of the clearcut and fire areas were at various stages of recovery, but the regeneration rate was much slower at MPB disturbed sites. Beetle outbreaks and fire created a positive compound effect on the seedling reestablishment, which suggests that beetle-killed serotinous lodgepole pines might have a new forest resilience mechanism to subsequent wildfire.

Following the depiction of the disturbance pattern in landscapes, I further assessed the effects of a variety of biotic and abiotic factors on the outbreak dynamics in Grand County, Colorado. Thirty-four variables were included to develop a number of general linear models (GLM). Case and control samples were extracted from maps derived from satellite image. I first removed non-significant predictors based on the Bayesian Information Criterion in a multiple backward stepwise selection, and then built the model using the retained variables. A correction factor was added into the traditional GLM to account for model bias introduced by different ratios of case and control observations in the sample and in the population. Finally, I evaluated the model performance with an independent validation dataset, and generated predictive maps of MPB mortality. The final model had an average area under the curve value of 0.72 in predicting the annual area of new mortality. The results showed that neighborhood mortality, winter mean temperature anomaly, and residential housing density were positively associated with MPB mortality, whereas summer precipitation was negatively related. The extent of MPB mortality will expand under both RCP 4.5 and 8.5 climate-change scenarios, which implies that the impacts of MPB outbreaks on vegetation composition and structure, and ecosystem functioning are likely to increase in the future.

Disturbance is the main driver for the heterogeneous landscape mosaic, and the understanding about its pattern, regime characteristics, impacts on forest resilience system and future trend is of great importance to many fields of research, such as carbon cycling, biological conservation, and environmental protection. The overall working approach in this dissertation provides feasible algorithms that can be applied to other regions, and can aid in generating consistent and high temporal frequency data on insect mortality and other disturbances impacting a variety of ecosystem services.