Gross primary productivity (GPP) is an important metric of ecosystem function and is the primary way carbon is transferred from the atmosphere to the land surface. Remote sensing techniques are commonly used to estimate regional and global GPP for carbon budgets. However, urban areas are typically excluded from such estimates due to a lack of parameters specific to urban vegetation and the modeling challenges that arise in mapping GPP across heterogeneous urban land cover. In this study, we estimated typical midsummer GPP within and among vegetation and land use types in the Minneapolis-Saint Paul, Minnesota metropolitan region by deriving light use efficiency parameters specific to urban vegetation types using in situ flux observations and WorldView-2 high spatial resolution satellite imagery. We produced a land cover classification using the satellite imagery, canopy height data from airborne lidar, and leaf-off color-infrared aerial orthophotos, and used regional GIS layers to mask certain land cover/land use types. The classification for built-up and vegetated urban land cover classes distinguished deciduous trees, evergreen trees, turf grass, and golf grass from impervious and soil surfaces, with an overall classification accuracy of 80% (kappa = 0.73). The full study area had 52.1% vegetation cover. The light use efficiency for each vegetation class, with the exception of golf grass, tended to be low compared to natural vegetation light use efficiencies in the literature. The mapped GPP estimates were within 11% of estimates from independent tall tower eddy covariance measurements. The order of the mapped vegetation classes for the full study area in terms of mean GPP from lowest to highest was: deciduous trees (2.52 gC m-2 d-1), evergreen trees (5.81 gC m-2 d-1), turf grass (6.05 gC m-2 d-1), and golf grass (11.77 gC m-2 d-1). Turf grass GPP had a larger coefficient of variation (0.18) than the other vegetation classes (~0.10). Mean land use GPP for the full study area varied as a function of percent vegetation cover. Urban GPP in general, both including and excluding non-vegetated areas, tended to be low relative to natural forests and grasslands. Our results demonstrate that, at the scale of neighborhoods and city blocks within heterogeneous urban landscapes, high spatial resolution GPP estimates are valuable to develop comparisons such as within and among vegetation cover classes and land use types.

Plants can respond to drought events in a variety of ways, including adjustments in physiological processes and changes in canopy structure. Quantifying these changes over large spatial domains and through time can be challenging, especially with variable vegetation cover types. During 2012-2016, California experienced one of the most severe droughts in its modern history, with limited precipitation and exceptionally high temperatures over an extended time period. Urban vegetation, such as trees and turfgrass lawns, provides many ecosystem services for people living in cities, such as cooling through shading and evapotranspiration, but these benefits may be difficult to maintain through extreme drought, especially in water-limited cities. Therefore, it is critical to understand how drought response in vegetation may vary across urban landscapes. In this dissertation, I used remote sensing time series to quantify how urban vegetation responded to drought in Santa Barbara and Los Angeles, California. In Chapter 1, I examined drought response in turfgrass and across nineteen urban tree species in the city of Santa Barbara using data from repeat flights of the Airborne Visible Infrared Imaging Spectrometer (AVIRIS) and AVIRIS-Next Generation (AVIRIS-NG). I compared many spectral indicators that may be expected to change within plant canopies during drought. Compared with data from before the drought in 2011, all tree species and turfgrass had four or more spectral indicators with significantly lower mean values (p < 0.05) during the drought in 2014, and evidence of recovery was observed for some species in 2017, after the drought had ended. Based on the spectral indicators, turfgrass cover senesced in the middle of the drought but recovered soon after the drought ended. Nearly all tree species showed significant canopy changes in the middle of the drought, but in comparison to turfgrass, most tree species did not fully recover after the drought ended. In Chapter 2, I evaluated how drought manifests seasonally and interannually during 2010-2019 across dominant types of trees and grass in the Santa Barbara area using Landsat and AVIRIS imagery. I compared the condition of dominant types of trees and grasses as they changed throughout the year using the Normalized Difference Vegetation Index (NDVI), difference in vegetation land surface temperature from impervious surfaces (∆LST), and equivalent water thickness (EWT). NDVI was lower and ΔLST was closer to zero during drought years but they were seasonally correlated for only some vegetation types. Changes in EWT revealed seasonal adjustments by vegetation that were not readily apparent in the NDVI time series. I also assessed the correlations of NDVI and LST with the Standardized Precipitation Evapotranspiration Index (SPEI) to test the effects of drought length and severity on vegetation. NDVI and ΔLST were most strongly correlated with SPEI during summer for most vegetation types, except for annual grass NDVI (winter). Annual grass was correlated with SPEI at spans ≤12 months, whereas trees and turfgrass were correlated with SPEI at spans >12 months in addition to seasonal time spans. In Chapter 3, I assessed annual changes in fractional cover of trees, turfgrass, non-photosynthetic vegetation (NPV; e.g., senesced grass, plant litter), and non-vegetated urban surfaces across the Los Angeles metropolitan area during 2013-2018 using AVIRIS imagery. During the drought time series from 2013 to 2018, mean turfgrass cover decreased and NPV cover increased, but tree cover did not show a strong trend with drought until 2018. The interior valleys of the study area (San Gabriel and San Fernando) consistently lost more turfgrass than coastal areas, and the San Gabriel Valley had strong losses of total vegetation cover (tree + turfgrass + NPV) overall. I also used datasets of median household income from census tracts and of typical non-drought outdoor water use from postal carrier routes to compare the magnitude and timing of different vegetation cover type changes at different income and water use levels. There were larger absolute changes in vegetation cover in higher income and higher water use areas, likely due to the higher baseline of mean vegetation cover in these areas. Once normalized for their mean values, the magnitude of changes often became more similar across different income and water use levels, but not always, with lower income and water use areas showing greater relative changes for trees. Overall, this dissertation quantifies drought responses across different urban vegetation types during a severe, long-term drought event at an array of spatial and temporal scales, providing implications for vegetation sustainability planning in cities with frequent droughts.

In cities, microclimates are created by local mixtures of vegetation, constructed materials, vertical structure, and moisture, with significant consequences for human health, air quality, and resource use. Vegetation can moderate microclimates through evapotranspiration, however this function is dependent on local conditions so its effect may vary over space and time. This dissertation used hyperspectral and thermal remote sensing imagery to derive key observations of urban physical and biophysical properties and model urban microclimates across the megacity of Los Angeles. In Chapter 1, I used Multiple Endmember Spectral Mixture Analysis (MESMA) to map sub-pixel fractions of different vegetation types, as well as other types of urban cover, at 4 m and 18 m resolution over Santa Barbara, California (Wetherley et al., 2017). Fractional estimates correlated with validation fractions at both scales (mean R2 = 0.84 at 4 m and R2 = 0.76 at 18 m), with accuracy affected by image spatial resolution, endmember spatial resolution, and class spectral (dis)similarity. Accuracy was improved by using endmembers measured at multiple spatial resolutions, likely because they incorporated additional spectral variability that occurred across spatial scales. In Chapter 2, I applied this methodology to derive sub-pixel cover for the greater Los Angeles metropolitan area (4,466 km2) (Wetherley et al., 2018). Further improvement in quantifying sub-pixel vegetation types was achieved by modifying the MESMA shade parameter. Land surface temperature (LST), derived from thermal imagery, was used to model temperature change along vegetation fractional gradients, with slopes of LST change showing significant differences between trees and turfgrass (p < 0.001). Expected per-pixel LST was derived from these gradients based on sub-pixel composition, and when compared to measured LST was found to deviate with a standard deviation of 3.5°C across the scene. These deviations were negatively related to irrigation and income, while building density was observed to affect tree LST more than it affected turfgrass LST. In Chapter 3, I used the map of Los Angeles landcover, along with data from LiDAR, GIS, and WRF climate variables, to parameterize an urban climate model (Surface Urban Energy and Water Balance Scheme: SUEWS) for 2,123 neighborhoods (each 1 km2) across Los Angeles. Modeled latent fluxes were correlated with remote sensing LST (R2 = 0.39) collected over a period of 5 hours, with an overall diurnal pattern modified by irrigation timing. Spatial variability across the study area was related to local landcover, with albedo and vegetation fraction strongly influencing latent and sensible fluxes. A strong regional climatic gradient was observed to affect latent fluxes based on coastal proximity. Overall, this dissertation quantifies the key drivers of urban vegetation function in a large city, and further demonstrates the potential of hyperspectral and thermal imagery for observing city scale surface and microclimate variability.