Monitoring and Forecasting Forest Drought Stress to a Changing Climate

Abstract

Globally, trees are increasingly dying from extreme droughts and heatwaves, a trend that is expected to increase with climate change. Loss of trees has significant ecological, biophysical, and biogeochemical consequences. In this thesis, I explored how forests will respond to increases in droughts and heatwaves projected under climate change, by studying the impacts of the 2011 drought on the forests and woodlands of Texas. I first developed new methods using remote sensing that improved monitoring of forest disturbances from droughts and heatwaves at regional scales. I then explored multiple modeling approaches, to improve forecasts of forest vulnerability to future droughts and heatwaves. Drought-induced tree mortality is unique because it often is limited to only subtle and diffuse changes in forest cover. Therefore, I first developed a new approach towards quantifying drought-driven canopy loss in open canopy woodland systems using remotely sensed imagery, across a Landsat scene in central Texas (>30,000 km2). I used classifications of 1-m orthophotos to calibrate and validate 30-m Landsat imagery. In scaling up to create regional canopy loss maps, I assembled a Landsat time-series and separated mortality pixels experiencing persistent canopy loss from pixels with only background noise by applying the Landtrendr algorithm. I then estimated percent tree canopy loss within each of these mortality pixels by comparing two models capable of handling zero-inflated continuous proportions: random forest and a zero-or-one inflated beta (ZOIB) regression model. I found that the ZOIB regression model had the highest accuracy in predicting canopy loss (mean absolute error = 5.16%, root mean square error = 8.01%). Again using remotely sensed imagery, I developed a second method to quantify canopy loss due to the 2011 drought across the many diverse systems of Texas, from the eastern pine/hardwood forests to the western shrublands. I then used these maps to better understand which ecological systems were most impacted and climatic trends that could explain spatial patterns of canopy loss. Canopy loss observations in ~200 multi-temporal fine-scale orthophotos (1-m) were used to train coarser Landsat imagery (30-m) to create 30-m binary statewide canopy loss maps. I found that canopy loss occurred across all major ecoregions of Texas, with an average loss of 9.5%. The drought had the highest impact in post-oak woodlands, pinyon-juniper shrublands, and Ashe juniper woodlands. Focusing on a 100-km by ~1000-km transect spanning the State’s 5-fold east-west precipitation gradient (~1500 to ~300 mm), I compared spatially explicit 2011 climatic anomalies to our canopy loss maps. Much of the canopy loss occurred in areas that passed specific climatic thresholds: warm season anomalies in mean temperature (+1.6 °C) and vapor pressure deficit (VPD, +0.66 kPa), annual percent deviation in precipitation (-38%), and 2011 difference between precipitation and potential evapotranspiration (-1206 mm). Although similarly low precipitation occurred during the landmark 1950s drought, the VPD and temperature anomalies observed in 2011 were even greater. Furthermore, future climate data under the representative concentration pathway 8.5 trajectory project that average values will surpass the 2011 VPD anomaly during the 2070-2099 period and the temperature anomaly during the 2040-2099 period. Identifying vulnerable ecological systems to drought stress and climate thresholds associated with canopy loss will aid in predicting how forests will respond to a changing climate and how ecological landscapes will change in the near term.As climate change continues, forest vulnerability to droughts and heatwaves is increasing, but vulnerability varies regionally and locally through landscape position. However, most models used in forecasting forest responses to heat and drought do not incorporate relevant spatial processes. Therefore, thirdly, to improve spatial predictions of tree vulnerability, I employed a non-linear stochastic model of soil moisture dynamics accounting for landscape differences in aspect, topography, and soils. Across a watershed in central Texas I modeled dynamic water stress for a dominant tree species, Juniperus ashei and projected future dynamic water stress through the 21st century. Modeled dynamic water stress tracked spatial patterns of drought-impacted area derived using remote sensing. Accuracy in predicting drought-impacted stands increased from 60%, accounting for only soil variability, to 72% when also including lateral redistribution of water and radiation/temperature effects attributable to aspect. Our analysis also suggests dynamic water stress will increase through the 21st century, with minimal buffering from the landscape. Favorable microsites/refugia may exist across a landscape where trees can persist; however, if future droughts are too severe, the buffering capacity of a heterogenous landscape could be overwhelmed. Incorporating spatial data will improve projections of future tree water stress and identification of potential resilient refugia.Lastly, droughts impact tree species unequally and therefore have the potential to alter the species composition of forests. As droughts intensify under climate change, models that can account for both changing environmental conditions as well as joint species-specific responses are needed to forecast shifts in species ranges, composition, abundance, and mortality. Therefore, lastly, I used a Generalized Joint Attribute Model, GJAM, to simultaneously model live and dead basal area by species, across Texas, using Forest Inventory and Analysis plots from 2001-2015 and covariates related to climate, topography and soils. I then projected shifts in species abundance into the 21st century under multiple climate warming scenarios. Model projections show that many of the eastern hardwood species will likely shift eastward throughout the 21st century. Many of the western woodland species are also projected to shift eastward and become increasingly dominant throughout Texas. By modeling species abundance and mortality simultaneously and by using covariates related to climate variability, we can improve projections of forest responses to continuing climate change.