This project was essentially an observational study of the summertime 2-meter air temperatures across the Detroit Metropolian area. The network used was an integrated network of three different observing networks. Our focus was on the spatial variability of the daily high and low temperatures. Characterization of the magnitude of that spatial variability and investigation of useful statistical models for predicting it was undertaken. The relationships between summer average values for both daily high and low temperatures (at each station) with land cover and position variables were then explored. Lastly the existence of spatial variability during dangerously hot weather was investigated.

Here are links to the working introduction and methods sections:

introduction

methods

The observational network used for this study was actually three observing networks. The first network consisted of five NWS ASOS/AWOS stations at the local airports. The second network was a network of stations operated by the Michigan Department of Environmental Quality (MDEQ). The third network was our own personal fleet of monitors, called HOBOS. These HOBO monitors allowed calibration between the three networks (e.g. co-locate monitors and quantify bias and uncertainity between observations) and subsequently all three networks were combined.

The uncertainity associated with each networks observations was generally different for the daily highs and lows. For this study we only recognize structure beyond the uncertainy. For example, for the airport observations (considered the base of the network) only the instrument uncertainity is assumed for those observations. For the HOBO monitors, uncertainities such as "temporal sampling algorthim" differences and the variability within the HOBO monitors themselves are considered. All the considered uncertainities were summed for each network, seperately for daily highs and lows. The daily high and low observations are converted into a spatial anomalies (i.e. anomaly w.r.t. all monitors at that time) and then the magnitude of that value is reduced by the amount of uncertainity. Here are the uncertianities that we used for each network:

The HOBOs were placed primarily in back yards of participants and on the properties of buisnesses. The instrument siting was choosen specifically to minimize any influences from the microclimate (i.e. immediate surroundings). In the Allen Park MDEQ photo, you can see the HOBO monitor that was co-located with this station. Here are some pictures to give you a feeling:

Due to our networks lack of rural representation, this study characterizes the intra-urban/suburban spatial variability in temperature. We focus on the daily high and low and use the spatial anomalies discussed above. The metric we use derived and use here is called the "range of simultaneously observed spatial anomalies" (range of SOSAs), which is simply the range over the network for any given daily extreme and date. The histograms above show the daily highs are skewed toward the small values. The 110-day mean range of SOSAs was 1.4 °C in the daily highs and 2.8 °C in the lows. For both daily extremes, a Student’s t-test rejected the null hypothesis that either of the true mean range of SOSAs could be zero. Essentially, the daily highs generally have a small amount of spatial variability but the daily lows sometimes have large amounts of spatial variability.

This spatial variability can be linked to the overlying weather conditions. This study linked the averaged afternoon and early morning cloud cover percentages and wind speeds to the range in SOSAs. Using an objective automated method equations were created to predict the spatial variability for the daily lows and highs. Common metrics were used to judge the goodness, and then cross validation tests were used to ensure functionality. Lastly the relationships were explored with similar varialbes derived from the Norther American Regional Reanalysis (NARR) dataset. The daily low equations and performance results are given below:

using averaged 4-8am and 2-6pm airport data inputs

using analogous NARR data inputs:

Shown above are the 110-mean spatial anomalies for each station and both daily extremes. The spatial patterns for the daily low are more coherent and larger in magnitude. The daily low shows highest temperatures downtown and the daily highs are largest on the western, rather than central, parts of Metropolitian Detroit.

This study also explored the ability to predict the 110-mean spatial anomaly at each station from three landcover and position variables. Local percent impervious surface (i.e. % within a radius of the station), distance to city center and distance from large water body were all calculated for each station. Shown above are the scatterplots for each of the variables.

Then using an objective automated method identified which variables were suitable for a predicative model of the mean anomalies. Again, model diagnostics and cross validation were used to confirm the goodness and utility of these models. The equation indicated for the daily low that all three variables were appropriate (% imperviousness was dominant), but for the daily high only distance to city center and distance to water (dominant) were appropriate. Below are the results of the model performances:

The importance of spatial variabilty in daily extreme temperatures across a metropolitian region is that both daily extremes have been shown (when elevated) to be predictors of excess mortality. Essentially making the weather conditions worse for some areas of the city during dangerous weather. So this study also confirmed the existence of spatial variability during these times. However, there was only two afternoons the NWS called for "excessive heat advisories" (June 24 and Aug 9, 2009) and the mean range in SOSAs during those daily highs was 0.7C (a little less than average). There was only one "heat wave" (i.e. 3-day 81st percentile exceedence at airports, both daily extremes), and it was September 21-23 (2009) and the mean range in SOSAs during the daily low was 2.1C. Again, a little less than average.

This study went about other methods to convince decision makers during hot weather. First our range of SOSAs metric as a function a popular method identifying dangerous weather, called the the Spatial Synoptic Classification (SSC) system. This graph is shown above, and it indicates that the dangerous air masses (Moist Tropical +/++, Dry Tropical) have average daily low values a little less than average (2.8C).

Conclusions and Discussion

For brievity, the conclusions here is a link to the current conclusions section and here some discussion points will be mentioned.

Of relevance to the urban climate dicussion, this study related a variable (previous day cloud cover% and SW radiation) not previously related to spatial variability (or "urban heat island (uhi) magnitude", or " uhi intensity") in the literature. Logically a city must be heated for the spatial variability to be robust during the subsequent daily low. Equally interesting, our results suggested local imperiousness was a better predictor of the spatial structure than the very traditional distance to city center variable.

As for the public health discussion it should be interesting that the temperature spatial variability during the daily high does not have a large impact however the warmest region is generally located to the west of the city center and the magnitude of that variability is not a function of how dangerous the weather is w.r.t heat stress. However the daily low spatial pattern of temperatures follows the local (e.g. 0.25km) built up pattern of the city and is higher than average during dry and hot weather but less than average during moist and hot weather.