Watershed managers are challenged by the need for predictive temperature models with sufficient accuracy and geographic breadth for practical use. We described thermal regimes of New England rivers and streams based on a reduced set of metrics for the May–September growing season (July or August median temperature, diurnal rate of change, and magnitude and timing of growing season maximum) chosen through principal component analysis of 78 candidate metrics. We then developed and assessed spatial statistical models for each of these metrics, incorporating spatial autocorrelation based on both distance along the flow network and Euclidean distance between points. Calculation of spatial autocorrelation based on travel or retention time in place of network distance yielded tighter‐fitting Torgegrams with less scatter but did not improve overall model prediction accuracy. We predicted monthly median July or August stream temperatures as a function of median air temperature, estimated urban heat island effect, shaded solar radiation, main channel slope, watershed storage (percent lake and wetland area), percent coarse‐grained surficial deposits, and presence or maximum depth of a lake immediately upstream, with an overall root‐mean‐square prediction error of 1.4 and 1.5°C, respectively. Growing season maximum water temperature varied as a function of air temperature, local channel slope, shaded August solar radiation, imperviousness, and watershed storage. Predictive models for July or August daily range, maximum daily rate of change, and timing of growing season maximum were statistically significant but explained a much lower proportion of variance than the above models (5–14% of total).

Uncertainty in the impacts of climate change and development on freshwater resources pose significant challenges for water resources management. Integrated and adaptive approaches to water resources management are a promising means of addressing uncertainty that afford flexibility in balancing multiple stakeholder objectives. However, guidance on designing such plans is lacking. In this study, we use multiobjective optimization to strategically incorporate green infrastructure (GI) into water resources management plans that maximize reductions in nutrient loads, minimize stormwater runoff, and minimize costs. Robust decision-making methods are applied to the resulting plan options to evaluate how optimized GI implementation varies under different possible future climates and to determine which solutions would be robust under a range of plausible future conditions. We demonstrate these coupled methods using a case study in southern Massachusetts to address water quality issues related to point and nonpoint source nutrients in a rapidly developing watershed.