Project Advisor(s) (Students Only)

Dr. Reuben Heine, Dr. Thomas Albright

Presentation Type (All Applicants)

Poster Presentation

Disciplines (All Applicants)

Climate

Description, Abstract, or Artist's Statement

The use of gridded temperature products is becoming increasingly prevalent in ecological research due to their accessibility, low cost, and spatial and temporal coverage. While a few studies have compared gridded products against each other and against weather station data, little research exists that attempts to verify the accuracy of these gridded products on finer spatial scales in field settings. In this study, we use two networks of temperature sensors to evaluate the effectiveness of these widely used gridded products in modeling ambient temperatures and compare tradeoffs between spatial and temporal resolution of gridded products.

We deployed 65 temperature sensors in radiation shields (Holden 2013) at the Kofa wildlife refuge in Southwestern Arizona and 80 sensors on the Snake Range of Eastern Nevada. From 2014 to 2015, the sensors recorded hourly temperatures. We then compared the sensor-collected temperatures against three widely used gridded temperature products that have varying spatial and temporal resolutions: NLDAS 10 km at hourly intervals, PRISM 4 km at daily intervals and Daymet 1 km at daily intervals. In order to compare the daily products, it was necessary to interpolate hourly values from daily minima and maxima. To do this, two methods of hourly interpolation (a cosine fit with variable sunrise and the Chillr package in R) were compared against sensor readings. We find that gridded products provide strong overall fits with sampled datasets but have a tendency to underestimate maxima and overestimate minima. Studies involving processes that are sensitive extremes and threshold based indices may be negatively affected by these biases. Of the gridded products used, Daymet was the most accurate at capturing Tmax and hourly temperatures (averageR2 > 0.90), while NLDAS was the least accurate (R2 = 0.70). While this suggests that the benefits of finer spatial resolution may outweigh the benefits of finer temporal resolution, other factors unrelated to resolution like topographic homogeneity across pixels may have contributed to the differences among products.

Comments

This project was generously supported by funding from NSF Value of Snow REU, National Science Foundation ESPCoR EPS–0814372,and National Aeronautics and Space Administration NNX13AB65G

This project was conducted as an REU and then used for a Senior Inquiry project under the guidance and direction of Dr. Heine

Included in

Climate Commons

Share

COinS
 
May 3rd, 12:00 AM May 3rd, 12:00 AM

An Evaluation of Gridded Temperature Products and their Effectiveness in Modeling Small Scale Ambient Temperature

The use of gridded temperature products is becoming increasingly prevalent in ecological research due to their accessibility, low cost, and spatial and temporal coverage. While a few studies have compared gridded products against each other and against weather station data, little research exists that attempts to verify the accuracy of these gridded products on finer spatial scales in field settings. In this study, we use two networks of temperature sensors to evaluate the effectiveness of these widely used gridded products in modeling ambient temperatures and compare tradeoffs between spatial and temporal resolution of gridded products.

We deployed 65 temperature sensors in radiation shields (Holden 2013) at the Kofa wildlife refuge in Southwestern Arizona and 80 sensors on the Snake Range of Eastern Nevada. From 2014 to 2015, the sensors recorded hourly temperatures. We then compared the sensor-collected temperatures against three widely used gridded temperature products that have varying spatial and temporal resolutions: NLDAS 10 km at hourly intervals, PRISM 4 km at daily intervals and Daymet 1 km at daily intervals. In order to compare the daily products, it was necessary to interpolate hourly values from daily minima and maxima. To do this, two methods of hourly interpolation (a cosine fit with variable sunrise and the Chillr package in R) were compared against sensor readings. We find that gridded products provide strong overall fits with sampled datasets but have a tendency to underestimate maxima and overestimate minima. Studies involving processes that are sensitive extremes and threshold based indices may be negatively affected by these biases. Of the gridded products used, Daymet was the most accurate at capturing Tmax and hourly temperatures (averageR2 > 0.90), while NLDAS was the least accurate (R2 = 0.70). While this suggests that the benefits of finer spatial resolution may outweigh the benefits of finer temporal resolution, other factors unrelated to resolution like topographic homogeneity across pixels may have contributed to the differences among products.