How GLOBE Data Supports Science

Improving Earth Observation Capabilities

GLOBE’s surface-based observations have been used to assess the accuracy of satellite derived data products and to enhance dynamic models. Featured research projects demonstrate how GLOBE data can synergize with models and other datasets in advancing the fundamental scientific understanding of the Earth. Furthermore they demonstrate the complementary nature of citizen science and remote sensing in addressing complex scientific questions.

GLOBE data have served as surface-based validation for snow products derived from satellites, and to assess the accuracy and feasibility of contrail observations. Researchers have also explored the use of GLOBE data in monitoring aerosols in the Netherlands, and the length of the growing season in Alaska.  

Projects

Authors: Abdelkader et al., 2024

DOI: https://doi.org/10.3390/rs16081368

Summary

In this study researchers focused on developing a reliable method for detecting river ice formation in rural Alaska. Using Google Earth Engine (GEE) to combine data from multiple satellites (moderate and high-resolution, along with Synthetic Aperture Radar) with citizen science observations, this research devised a new automated system to analyze for lake and river ice conditions in rural Alaska. Land cover data observations collected using the GLOBE Observer include users taking photos in all four cardinal directions, up and down, are integrated into the ‘Fresh Eyes on Ice’ project. These photos are shared in near-real time with the National Weather Station in the Alaska-Pacific River Forecast Center. These photographs serve as a vital source of information relating to winter travel safety warnings.

This research focused on detecting ice and snow conditions in a specific ice jam in 2023. Google Earth Engine provides ready access to constantly updated satellite images from Landsat, Sentinel and other NASA assets. Abdelkader et al., developed automatic processes to analyze data, applying algorithms to detect and map river ice content, snow cover and other features. This study introduces an automated framework that supports FAIR (Findable, Accessible, Interoperable, and Reusable) science principles. 

Six text boxes with varying shades of color detailing method used in this research

 

 

Figure outlining the methodology framework employed by the researchers - reproduced from Abdelkader, M., Bravo Mendez, J. H., Temimi, M., Brown, D. R. N., Spellman, K. V., Arp, C. D., Bondurant, A., & Kohl, H. (2024). A Google Earth Engine Platform to Integrate Multi-Satellite and Citizen Science Data for the Monitoring of River Ice Dynamics. Remote Sensing, 16(8), 1368. https://doi.org/10.3390/rs16081368. 
 

 

 

 

Inputs

Data from multiple satellites which include Landsat (8 & 7), Sentinel (1,2,3) and NOAA-20 were integrated into the study to compensate for spatial, temporal challenges, cloud cover, revisit times etc. GEE provides rapid processing capability, extensive data repository, and computational power. The system leverages high-resolution images from the Visible Infrared Imaging Radiometer (VIIRS) sensor on the NOAA-20 satellite.

Left image showing a satellite above the earth with clouds, land and sea visible three images showing different satellites above the earth with clouds, land and sea visible

 

 

 

 

 

 


Landsat 8 satellite (left) NASA. Sentinel 1-3 satellites -(right) European Space Agency. 

Using ground truth photos and reports of ice conditions from citizen scientists which were collected through the ‘Fresh Eyes on Ice’ project platform, together with fixed camera data positioned along rivers allowed for integration of observations by citizen scientists. These photographic contributions flow in near-real time to the National Weather Service Alaska-Pacific River Forecast Center (APRFC), serving as a crucial resource for spring flood forecasting and disseminating winter travel safety warnings.

Outputs

The system generated near-real time maps of detailed delineated classes such as ice, river, land, cloud, snow, vegetation and land. It demonstrates reliability in detecting ice and snow conditions in analyzing a specific ice jam in 2023. The results were quantitatively confirmed using existing ice extent maps from northeastern US and New Brunswick, Canada.

Conclusion

The study concludes this methodology is a significant advancement in river ice monitoring and demonstrates that citizen science data can be converted into valuable quantitative information.  After this research, the authors created a user-friendly River Ice Monitoring System for River Ice conditions that is an open-source system facilitated by CUASHI (Consortium of Universities for the Advancement of Hydrological Science, Inc.)  and available for residents of Alaska and other areas to be better informed on river ice conditions. 

The following link provides access to the user-friend River Ice Monitoring System for River Ice Conditions.  Mapping interface display for the latest river and lake ice conditions

Abdelkader, M., Bravo Mendez, J. H., Temimi, M., Brown, D. R. N., Spellman, K. V., Arp, C. D., Bondurant, A., & Kohl, H. (2024). A Google Earth Engine Platform to Integrate Multi-Satellite and Citizen Science Data for the Monitoring of River Ice Dynamics. Remote Sensing, 16(8), 1368. https://doi.org/10.3390/rs16081368

Authors: Jessica Robin et al., 2005

DOI: https://doi.org/10.1016/j.ecolmodel.2004.11.022 

Introduction

Soil-vegetation-atmosphere (SVAT) models are an important tool in understanding water and energy dynamics associated with plant growth. Evapotranspiration, soil moisture, and temperature changes all affect the length of the growing season. Detailed measurements over a wide spatial scale at regular temporal intervals are important to create accurate and reliable model outputs. 

Student environmental data from the GLOBE Program and data derived from satellites were used to test simulated water and energy fluxes with the General-Purpose Atmosphere Plant Soil Simulator (GAPS). GAPS is a computer model that is used to simulate the interactions between the soil, plants, and atmospheric processes. The input parameters and degree of detail required to run a simulation, depend on the model algorithms chosen. General to more complex simulations can be performed based on the available data for a given site, which makes this an ideal model to run with GLOBE Program data.

Model inputs:
Between 1995 and 2003 students at Reynolds Junior/Senior High School in Greenville, Pennsylvania, collected GLOBE measurements. Following GLOBE protocols students measure a wide range of variables including:

  • Soil horizon depth,
  • Soil texture,
  • Bulk density,
  • Soil water content,
  • Minimum and maximum daily temperature,
  • Precipitation,
  • Snow water equivalent, and
  • Soil temperature. 

Soil samples were also collected by the Cornell University Lab of Soil and Water to undertake additional soil moisture measurements. Data collected by the National Weather Service Cooperative Weather Station (NWSC) was also used as a comparison nearest to the student's sampling locations. 

Normalized Difference Vegetation Index (NDVI) is an index that measures the amount of green vegetation over a given area and can be used to measure vegetation health. NDVI data from the European Space Agency SPOT4 satellite, which captures data in a 1km resolution, was also used as an input to the GAPS model. Spectral resolution in the red (0.61-0.68 µm) and near-infrared (0.79-0.89 µm) bands are used to determine the NDVI values.

Land cover of the NDVI pixels were determined from the National Land Cover Data (NLCD) with the majority of pixels classified as pasture, with remaining pixels registering as residential (33%) or deciduous/mixed forests (10%). Field measurements verified that the site vegetation was composed of orchard grass.

Results

The researchers used the GAPS model to simulate the total daily water within the root zone, daily drainage out of the root zone, and daily transpiration. 

graph with dates on x-axis and millimeters of water on y-axis. different types of lines dotter and solid show trends in soil mositure across the timeframe

Figure 1 shows the simulated and measured soil moisture content in the root zone at the Reynolds site from 1998-2001. Reproduced from: Utilizing satellite imagery and GLOBE student data to model soil dynamics, Ecological Modelling, Volume 185, Issue 1, 2005,
Pages 133-145, ISSN 0304-3800, https://doi.org/10.1016/j.ecolmodel.2004.11.022.


The model simulations of the daily water in the root zone were compared with the soil water measured by the students at the Reynolds school site and from the NWCS. Simulations that used the GLOBE climate data outperformed the data from the NWCS for the growing seasons in 1999, 2000, and 2001.

Both datasets (GLOBE and NWCS) were statistically examined to determine how much agreement was presented when compared with the outputs of the model simulations. Three different statistical analyses were conducted.

  1. d-index, where a value of 1 indicates complete agreement, and a value of 0 indicating complete disagreement.

  2. Root Mean Square Error (RMSE) values which is a common metric used to evaluate the accuracy of models. A lower RMSE indicates a better fit of the model to the data.

  3. Coefficient of determination (denoted as r2 and pronounced R-squared) measures the proportion of variance that can be explained by the independent variable(s) in a regression model. It shows how well the data fits the model.  r2 values range from 0-1, with higher values indicating that the model explains more of the variation in the data.

There was strong agreement between the field measurements and those from the model.

Table 1: Statistical analysis modeled to the measured root zone soil water content for each growing season.
Dataset 1998 1999 2000 2001
  d RMSE r2 d RMSE r2 d RMSE r2 d RMSE r2
GLOBE 0.80 49 0.86 0.95 26 0.94 0.84 29 0.58 0.87 29 0.62
NWCS 0.84 40 0.82 0.88 42 0.70 0.81 26 0.50 0.83 33 0.62

 

 

 

 

 


The GLOBE dataset had higher d-indexes (0.80 or higher) and r2 values and lower RMSE than the NWCS dataset.

The NDVI results show similar seasonal patterns for all four years as the soil and air temperatures collected by the GLOBE students. The students four datasets (average air temperature, maximum air temperature, soil temperature at 5cm and soil temperature at 10cm) had r2 values of 0.80, 0.77, 0.70, 0.70 with NDVI.

Conclusion

This research highlights the potential for GLOBE data to serve as an important input and source of data for models such as the GAPS. The authors noted that the GLOBE data is an underutilized resource that could provide valuable ground truth data for models in various environments across the world.

Jessica Robin, Elissa Levine, Susan Riha, Utilizing satellite imagery and GLOBE student data to model soil dynamics, Ecological Modelling, Volume 185, Issue 1,2005, Pages 133-145, ISSN 0304-3800, https://doi.org/10.1016/j.ecolmodel.2004.11.022. (https://www.sciencedirect.com/science/article/pii/S0304380004006003)

Authors: Ault et al., 2006
DOI: https://doi.org/10.1016/j.rse.2006.07.004

Introduction

Snow cover is a major study in climate change. Snowpacks act as water storage which can greatly impact aquifers and reservoirs. It's important to measure the regional and global snow cover and understand the accuracy of the measurements. Low snow cover in the Great Lakes area between 1999 and 2002 resulted in the lowest water levels recorded in 35 years for Lakes Superior, Michigan-Huron, and Erie. The low water levels had negative impacts on commercial freight transport, hydropower production, and recreational boating.

view of space and clouds above earths surface noted as blue and white. Also showing a digital impression of satellite passing over earth

Figure 1: Moderate Resolution Imaging Spectroradiometer (MODIS) on board the Terra satellite (Image courtesy of NASA) 

In 2006, Ault et al., published a paper in ScienceDirect where GLOBE student data was a central component of the research along with data from the Student and Teachers Evaluating Local Landscapes to Interpret the Earth from Space (SATELLITES) Program, and the National Weather Service (NWS).

This research examines the accuracy of the MODIS swath snow product (MOD10), MODIS cloud mask (MOD35), and the Liberal Cloud Mask (LCM). Data collected by the GLOBE Program and SATELLITES program were used to compare against the National Weather Service's cooperative extensions snow observations (point measurements). 

The MOD10 and MOD35 are two satellite-derived products generated by NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS). MOD10 provides snow cover information, while MOD35 is a cloud mask that helps to identify pixels where clouds are present. Ground-truth data was used to explore the accuracy of satellite products including MODIS snow products (MOD 10_L2 swath snow product) and cloud classifications (Cloud mask MOD35, and the Liberal Cloud Mask LCM) throughout the Great Lakes area. The Great Lakes area in North America is challenging for these products due to low amounts of snow and high amounts of cloud cover. Snow and clouds are difficult to differentiate in the visible spectrum. 

Figure 2: Reproduced from Ault et al., 2006. SATELLITES, GLOBE and NWS cooperation extension observation sites used in this study (Remote Sensing of Environment 105 (2006) 341–353).

What GLOBE data was collected?

1447 cloud and snow observations from 77 GLOBE schools were submitted along with 1025 cloud and snow observations from 52 schools associated with the SATELLITES Program. Data from the first 2 weeks of December 2001, and the last and first weeks of Jan and Feb 2002 were submitted. Teachers selected their own sites, free from disturbance and away from tall structures. Snowfall and snow water equivalent measurements followed protocols by Colorado Climate Center's and the NWS.

Total snow depth were determined using a measuring stick. 24-hr snowfall was measured (daily) on a standard 40cmX40cm white snowboard. Cloud observations were categorized as follows:

  • clear (0-10%),
  • isolated (11-25%),
  • scattered (26-50%),
  • broken (%1-90%),
  • overcast (91-100%) and,
  • obscured (fog, smoke or heavy precipitation).

Observations taken 3 or more hours before or after the MOD10_L2 image were discarded. Ideally the NWS and school observations would reach the same conclusions as each other when compared with the MOD10_L2 snow product.

Findings

The researchers found that the MODIS cloud mask can sometimes misclassify snow as cloud, especially in areas where there is patchy snow or thin snow cover. This misclassification was a major source of error in the MODIS snow product, with accuracy of 41% in snow detection when the snow was less than 10mm deep. When snow was not present the MODIS snow product was determined to be 96% accurate. When snow was 10mm or more, the MODIS snow product was likely to correctly classify the snow cover.

The researchers found that in areas with complex terrain with variable snow conditions, that there was a need to improve the cloud masking algorithms of the MODIS snow product. The results of this research contributed to a better understanding of the limitations of the MODIS data snow product and provided valuable insights into necessary improvements.

 Figure 3: Comparing the MODIS snow product with data collected by students from the GLOBE and SATELLITES programs and National Weather Service Coop. Percentages indicate the degree of correct snow detection by the MODIS product with varying depths of snow.  

During the research it was determined that cloud cover was the most significant factor impacting the accuracy of snow-cover detection. Improvements in cloud detection algorithms was suggested as a means to improve snow cover accuracy.

An unexpected finding of this research was the misidentification of pixels (as inland water and low altitude cloud) by the Liberal Mask Cloud. Cloud observations from the school programs allowed the researchers to explore this new finding and determine that when low altitude clouds occurred, the Liberal Mask Cloud was prone to high error (59% of the time).

The observations provided by students involved in the GLOBE Program and the SATELLITES Program allowed for a comparative determination between two datasets which highlighted the erroneous identification of low-altitude clouds. An additional benefit was students involved in the data collection process gained valuable knowledge in taking scientific measurements.

Timothy W. Ault, Kevin P. Czajkowski, Teresa Benko, James Coss, Janet Struble, Alison Spongberg, Mark Templin, Christopher Gross, Validation of the MODIS snow product and cloud mask using student and NWS cooperative station observations in the Lower Great Lakes Region, Remote Sensing of Environment, Volume 105, Issue 4, 2006, Pages 341-353, ISSN 0034-4257.

Authors: Robin et al., 2008
DOI: https://doi.org/10.1029/2007JG000407

Introduction

Normalized Difference Vegetation Index (NDVI) is an index that measures the amount of green vegetation over a given area and can be used to measure vegetation health. NDVI has been used to document phenological changes but much of this research in northern latitudes does not incorporate field validation. In high northern latitudes such as those found in Alaska many phenological changes have been observed, but it is important to establish whether NDVI is the proper tool to apply in these regions. NDVI is a proxy measurement of plant photosynthetic activity. 

NDVI values derived from the Advanced Very High-Resolution Radiometer (AVHRR) have shown an increase in NDVI in northern latitudes indicating a longer growing season but it's difficult to validate this without field observations. Previous studies of NDVI have indicated that northern latitude especially boreal forests exhibit strong changes, but these changes are affected by forest fires (significant source of aerosols which impact NDVI values up to a month later). Observed changes in earlier spring phenological events in both plants and animals has been attributed to anthropogenic climate change. NDVI values differ by vegetation types and in landscapes such as those found in Alaska are difficult to implement because of excessive snow and cloud coverage during winter months. 

Researchers in this study used six quadratic regression models with NDVI as a function of the accumulated growing degree days (AGDD) with various inputs. These included NDVI products from AVHRR 14-day 1 km resolution and MODIS, as well as climate products sourced from the National Oceanic and Atmospheric Administration and the Bonanza Creek Long-term Ecological Research program at the University of Fairbanks, Alaska. 

Since 1999, 200 schools around the world have made 120,000 phenology measurements. Nearly half of these are measurements have been collected in Alaska providing a largely untapped method to validate satellite-derived phenology for boreal regions.

Soil temperature data were obtained from Bonanza Creek Long-Term Ecological Research (LTER) program. Two sites were selected offering different types, LTER1 is an upland site, while LTER2 is a floodplain site. Sites where GLOBE data was collected were characterized as upland and floodplain groups based on their elevation and soil classification. 

Start of growing season (SOS) observations were made by students in or near Fairbanks, Alaska, from 2001 through 2004. Students made SOS observation on birch, poplar, and willow trees, with all three species being native to Alaska. SOS is defined when 50% of the buds for all sites of the same genus at one school had leafed out or burst. Sites with similar locations, cloud cover classification and temporal NDVI were classified into 3 groups: forest, mixed and urban. 

colored blocks of text with connecting lines outlining method framework Model Inputs
1. Climate products

  • Meteorological data from NOAA weather stations
  • Soil temperature data from Bonanza Creek Long-term Ecological Research program. 
  • Annual accumulated growing degree (AGDD) days required for SOS were calculated from daily temperatures greater than 0°C from March 1 through time of observed SOS.

2. NDVI Products

  • Maximum value, 14-day -1 km AVHRR NDVI composites – data from NOAA-16 and NOAA-17 satellites atmospherically corrected for ozone, water vapor and Rayleigh scattering. High cloud cover prevents use of 7-day 1 km AVHRR composites. 
  • Maximum value biweekly 8km AVHRR NDVI composites from 1982-2004 from GIMMS group (Global Inventory Monitoring and Modeling Systems) - NOAA-7 through NOAA-17 satellites
  • MODIS/Terra Vegetation Indices 16-Day L3 1 km grid composites from 2001-2004
  • 8-day product was not used because too few unobstructed observations were available of the sites.
  • Only composites from the observation season were used to ensure snow-free NDVI values.

3. AVHRR and MODIS models

  • 1 km AVHRR and MODIS NDVI datasets from 2001-2004 for Fairbanks.
  • A 3X3 mean filter around each GLOBE site was applied to the biweekly NDVI composites.
  • Mean AGDD computed to correspond to the 14-day AVHRR NDVI composites or 16-day MODIS composite. 
  • NDVI for start of growing season were determined from the model equations and field observation of SOS (GLOBE)

4. SOS TEST

  • Biweekly NDVI composites from 1982-2003 extracted for 8 km AVHRR for each GLOBE site. 
  • Annual SOS 1982-2003 determined from the University of Alaska Fairbanks (UAF1) observations – converted to Annual accumulated growing degree of the same biweekly 8 km AVHRR. 
  • Applied to the AVHRR and MODIS models to compute NDVI for each year. 

Results

Field observations made by students showed similar annual SOS dates by species. Overall, observations by GLOBE students were earlier by four days or less, than observations made by University of Alaska Fairbanks researchers. However, most GLOBE observations were made in the valley, while University of Alaska observations were made on a hillside close to the UAF campus. This could explain the difference given that green-up was reported by Thoman and Fathauer (1998) to begin on lower elevations of south facing slopes moving quickly down valleys, and more slowly up higher elevation hillsides. 

Results showed significant limitations on continuity between AVHRR and MODIS NDVI datasets. 

AVHRR NDVI has a significantly longer time series but with wider spectral near-infrared bands, it is more sensitive to water vapor interference which dampens NDVI values. MODIS NDVI has higher processing and greater spectral characteristics making it able to detect small changes in SOS.

The AVHRR and MODIS regression models showed significant differences in NDVI curves, and subsequently different start, peak and length of growing season.  Both models produced significantly different NDVI values for the Start of Season observations, which makes long-term monitoring with both models difficult. A more sensitive predictor than NDVI is needed to monitor changes in the start of the growing season. 

GLOBE data provided on-site ground truth data as a comparison with observations made by University of Alaska Fairbanks green-up observations. Researchers highlighted that the research could not have been completed without the data measurements collected by the students and teachers at the GLOBE schools in Alaska - Barnette Elementary, Joy Elementary, Moosewood Farm Home School, Monroe Catholic High School, North Pole Elementary, Ticasuk Brown Elementary, and West Valley High School.

Robin, J., Dubayah, R., Sparrow, E., & Levine, E. (2008). Monitoring start of season in Alaska with GLOBE, AVHRR, and MODIS data. Journal of Geophysical Research, 113, https://doi.org/10.1029/2007JG000407

Authors: Boersma and Vroom, 2006

DOI: https://doi.org/10.1029/2006JD007172 

In this research, scientists compared Aerosol Optical Thickness (AOT) observations derived from satellites with observations taken by GLOBE students in the Netherlands.

Introduction

Several satellite instruments measure the atmospheres' aerosol optical thickness (AOT). AOT refers to the amount of light that aerosols (tiny, suspended particles in the air) absorb and scatter in the vertical column of the atmosphere. With instrument degradation, calibration drift and differences in retrieval algorithms, it's important to regularly validate satellite derived AOT. Many of the previous calibration efforts have taken place with few sun photometers often hundreds of kilometers apart. In the Netherlands, a network of schools participated in the GLOBE Aerosol Monitoring Project, which used sun photometers to measure AOT.  

Using the GLOBE Aerosol Monitoring Project, researchers compared the measurements with MODIS (Moderate Resolution Imaging Spectroradiometer) derived AOT. The researchers also address student measurement quality. Boersma and Vroom present error estimates based on the intercomparison of students and professional observations. 

GLOBE Data

Twenty sun photometers were assembled from parts provided by Drexel University and calibrated at KNMI (Royal Netherlands Meteorological Institute). The KNMI and an Environmental Consultant (SME Advies) trained teachers and students on the GLOBE Aerosol protocol. The GLOBE student network consists of 12 schools, with approximately 1700 student observations being reported to GLOBE between 2001 and 2005.

An improved AOT retrieval algorithm was developed for the GLOBE observations to account for the unique characteristics including their broad spectral detectors, and spatiotemporal variations in ozone, which can affect AOT measurements.

The accuracy of the student observations was further validated by comparing them to data collected by professional instruments such as the Sun Photometer UV (SPUV) operated by KMNI. A series of measurements were recorded for each sun photometer for a range of atmospheric conditions and calibrated relative to the SPUV RG2-047 instrument. 

Comparison with professional measurements

Independent calibration of the RG2-047 using the Langley procedure facilitated a suite of other simultaneous observations to be taken at KNML. The SPUV (precision sun photometer measures direct solar spectral irradiance at up to 10 discrete wavelengths) takes measurements in a continuous fashion and the RG2-047 observations were taken right next to the SPUV on the roof of the KNML less than 2m away.

49 coincident measurements were taken between Sept 2002 and March 2003. For both channels, a high correlation coefficient (R2>0.98) was found – root mean squared differences smaller than 0.012AOT. The small bias observed, and RMS provide evidence that the measurements taken by GLOBE Sun photometer compare favorably with observations taken by fully automated instruments. 

Comparison at De Populier (the Hague) with AERONET measurements.

Comparisons were also made between measurements taken by students at the De Populier secondary school and measurements from the CIMEL Sun photometer at TNO-FEL and part of the AERONET – approx. 4km apart. Measurements were taken within 30 minutes of one another with the professional measurements interpolated to the GLOBE Sun photometer. 

The GLOBE student measurements and professional observations are highly correlated (R2>0.92) but the bias and RMS are slightly higher than the measurements taken at KNML. The GLOBE measurements tended to be higher by approx. 0.04 AOT for both channels. This is most likely due to sampling differences. 

All measurements for this study have been taken where the solar zenith angle is less than 70 degrees.

Findings

This research determined that MODIS generally underestimated AOT at low values and overestimated at high values, especially in coastal areas. The research found that MODIS and GLOBE data compare well over land. They found for 13 collocated inland pixels that MODIS overestimated AOT over coastal areas by 0.10 at 470nm and by 0.08 at 660nm. Findings are consistent with Chu et al (2002) who found similar MODIS biases over coastal areas of the US. 

Four color shaded maps of aerosols over the Netherlands

(a) Winter (DJF) 2004 mean seasonal aerosol optical thickness at 470 nm. Grey areas correspond to areas where the MODIS retrieval has not been performed (i.e., over open sea and ocean), has been subject to persistent cloud cover, or with too scarce sampling (see text). GLOBE locations that contributed to the validation of MODIS AOT data are shown in white. (b) Same as Figure 7a but now for spring (MAM) 2004. (c) Same as Figure 7a but now for summer (JJA) 2004. (d) Same as Figure 7a but now for fall (SON) 2004. Reproduced from Journal of Geophysical Research: Atmospheres, Volume: 111, Issue: D20, First published: 31 October 2006, DOI: (10.1029/2006JD007172)

Significance of GLOBE contributions

Secondary school students equipped with handheld sun photometers, may provide accurate and precise aerosol data at small spatial scales that can be used for validation of satellite-derived AOT. 

From error analysis they estimate that GLOBE AOT precisions are better than 0.02 AOT- dominated by uncertainty in the calibration constant of the GLOBE photometer.

The GLOBE student data collected using handheld sun photometers proved to be a valuable resource for validating MODIS data, especially in areas where professional AERONET stations were absent. Establishing school-based networks can significantly increase the spatial coverage of validation efforts.

K. F. Boersma and J. P. de Vroom , Validation  of MODIS aerosol observations over the Netherlands with GLOBE student measurements, J. Geophys. Res.: Atmos., 2006, 111 , D20311

Authors: D.P. Duda, R. Palikonda, and P. Minnis 2009 

DOI: https://doi.org/10.5194/acp-9-1357-2009  

Contrails, short for condensation trails, are composed of water in the form of ice crystals and form when exhaust from jet engines mix at high altitudes with cold, low-pressure air. Contrails can have impacts on local climate and with growing air traffic volume, they also have the potential for global impacts. The magnitude of contrails remains uncertain as they are crudely parameterized in general circulation models. This study aims to evaluate the potential for using two models to diagnose and predict persistent contrail formation conditions using a variety of datasets.

Image of Earth with blue ocean and green land surrounded by black space. Image of satellite in forefront and center of image

Photo of contrails covering between 25-50% of the sky. 

Contrail formation is dependent on temperature and pressure conditions to allow the mixing of hot, moist exhaust gases with the cold ambient air temperature to form a contrail. Numerical weather analyses often underestimate the upper troposphere relative humidity (UTH), which can in turn impact the detection rate of contrails. In order to improve the realistic simulation of contrails it is necessary to determine how accurately the metrological data provided by the numerical weather analyses and forecasts diagnose contrail formation conditions. The study matched and compared several months of contrail occurrence statistics derived from satellite and surface observations to the Numerical Weather Analysis-derived humidity, vertical velocity, windshear, and atmosphere stability.

Researchers evaluated the potential for using two models, (1) the Rapid Update Cycle (RUC) and (2) the Advanced Regional Prediction Systems (ARPS) to diagnose and predict persistent contrail formation conditions using a variety of datasets. Contrail formation can be observed from both surface observations and satellites. Satellite detection can identify contrails formed above low-level clouds that are missed by ground observers. Conversely, surface observers can see thinner contrail formations that are missed by satellites.

The GLOBE Program started collecting data relating to contrail formation in May 2003. GLOBE observations relating to contrails include reporting the number of contrails, cloud coverage, cloud type and classification of contrails into three categories, short lived, persistent non-spreading, and persistence spreading. Over 18,500 GLOBE contrail observations were submitted from 417 schools between April 2004 and June 2005. Observations from 11 GLOBE schools that had at least 50 observations were chosen for further examination. Collected observations included clouds, contrails and other environmental parameters which were compared with near coincident contrail observations from the Geostationary Operational Environmental Satellite (GOES).  Surface and satellite observations matched 75% of the time. Surface observers typically can detect much narrower and probably optically thinner contrails than GOES. The researchers found that GLOBE observations of clouds and contrails were consistent with the numerical weather outputs.

Geostationary Operational Environmental Satellites - a collaborative network between NOAA and NASA to examine atmospheric conditions and solar activity. 

The research determined that relative humidity in the upper troposphere is the most important factor in determining if contrails are short-lived or persistent. Upper troposphere humidity is typically much higher than surface humidity observed under cloudy skies. The researchers also concluded that the upper troposphere relative humidity index in both models correlates well with satellite observations of cirrus clouds.  Vertical velocity and vertical shear are expected to influence the spreading rate of contrails, with the researchers suggesting that vertical velocity also appears to influence where persistent contrails may form.

Duda, D. P., Palikonda, R., and Minnis, P.: Relating observations of contrail persistence to numerical weather analysis output, Atmos. Chem. Phys., 9, 1357–1364, https://doi.org/10.5194/acp-9-1357-2009, 2009.

Author D. L. Verbyla 
DOI: https://doi.org/10.1080/01431160010030127 

Introduction
Using remote sensing products that depend on the optical spectrum to detect the timing of spring leaf flush (green-up) is limited especially in higher latitudes because frequent cloud cover is common. Clouds scatter and reflect sunlight, and their presence can prevent optical sensors from capturing useful Earth observations.  This research investigated the potential for Synthetic Aperture Radar (SAR) to monitor the timing of green-up as SAR uses microwave radar signals to penetrate through clouds. When new leaves are generated, they alter the forests’ physical structure and moisture content, which affects how the microwave radar signals interact with the canopy. By monitoring these changes over time, this research investigated the feasibility of using SAR to monitor the start of the growing season. When leaves develop, they affect how radar waves scatter and reflect back to the satellite, which is called radar backscatter. 

Two satellite sensors were investigated – 
1.    European Remote Sensing Satellite-2 (ERS-2), and 
2.    Radarsat Standard Beam Synthetic Aperture Radar (SAR)

European Remote Sensing Satellite-2 (ERS-2) - image courtesy of European Space Agency.

The researchers identified broadleaf and conifer forest areas from the Landsat images. The selected study area is a level floodplain near the Bonanza Creek Experimental Forest (BNZ LTER), near Fairbanks, Alaska. The site has a distinct spring growing season and is composed of a relatively homogenous stand of Balsam poplar (dominant broadleaf) and a coniferous stand of white spruce or black spruce. SAR images from 1997, 1998, and 1999 were derived and calibrated and co-registered to a Landsat image. 

The mean decibel (dB) value was calculated for each of the forest types. For SAR, decibel measurement is a unit to represent the backscatter intensity of a radar signal. Since the canopy changes in broadleaf forests is much greater than conifers, the change in mean backscatter intensity was expected to change more than the mean backscatter of conifer stands. 

  • Green-up data reported by GLOBE students and scientists at the BNZ LTER were consistently reported within 5 days. This data served as ground truth data providing a valuable validation set.
  • ERS-2 SAR full resolution scenes were acquired for 1997 and 1998 (C-Band sensor).
  • Radarsat SAR full resolution images for 1998 and 1999 (C-Band sensor).
  • Both were geometrically and radiometrically processed.
  • Ten broadleaf and ten conifer stands were digitized as sample polygons. Different stands were delineated for 1997, 1998 and 1999.
  • The mean SAR dB from each sample stand was computed from images taken prior to and after spring leaf flush of broadleaves.

Results
This study found no consistent trends in either the ERS-2 or Radarsat SAR in detecting spring leaf flush in broadleaf forests. The trend for detecting leaf flush in conifer stands mean dB was also inconsistent. 

Discussion
The ERS SAR data were not useful for detecting leaf flush even under extreme pre-budburst versus full leaf stages. The 1998 Radarsat SAR backscatter increased after leaf flush for all broadleaf stands and most conifer stands. This may have been due to recent rain.

Coniferous stands were chosen as controls as these do not present significant changes in canopy cover (leaf flush). However, the results suggested that transpiration of water could be an important factor influencing SAR backscatter in conifer stands.

Conclusion
There was no consistent change detected using SAR pre-versus post leaf flush backscatter of broadleaf stands relative to conifer stands. Inconsistencies noted during the research suggested that other factors affect the ability of SAR to confidently detect changes such as surface moisture, freeze/thaw events and other environmental conditions.

Verbyla, D. L. (2001). A test of detecting spring leaf flush within the Alaskan boreal forest using ERS-2 and Radarsat SAR data. International Journal of Remote Sensing, 22(6), 1159–1165. https://doi.org/10.1080/01431160010030127