Image Processing

Advanced Analysis

Surface Reflectance Calibration

GDA has developed advanced algorithms for imagery calibration to the Surface Reflectances (SR).  The SR calibration improves radiometric properties of the imagery and allows for superior land cover and crop mapping, time series and change analysis, as well as image compositing and comparison.

GDA’s SR products are a per-pixel, per-band estimates of the surface spectral reflectance as it would have been measured at ground level as if there were no atmospheric or topographic contributions and deviations from satellite nadir view or solar zenith position.  Our SR calibration methodology relies on physical models that employ (i) sensor specific orbital and imaging information, (ii) sensor and band specific atmospheric radiative transfer calculations, (iii) sensor, location, and time dependent estimations of surface radiative budget, and (iv) modeling of surface topographic and land cover anisotropy. Reliance on physics based models ensures the predictability, transparency, and accuracy of the results, allows for the simulation of data acquisition, and enhances the comparison and analysis of data from various sensors and/or acquisitions.

GDA SR calibration approach is comparable to NASA’s Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS), USGS’ Landsat Surface Reflectance Code (LaSRC), and the ESA SEN2COR calibration.  However, our SR models are also applicable to a range of other sensors, account for per pixel variation in solar and view angles within the scene, use MODIS based coincident atmospheric properties and BDRF coefficients, and cross-calibrate imagery to the MODIS ‘gold standard’ SR products from NASA. Cross-calibration with MODIS further improves calibration of the imagery and ensures cross-compatibility of data from various sensors.

We offer SR products to our clients and use them in our analytics and to generate higher level products.  Our SR deliverables are 16-bit raster images, linearly scaled between 0 (representing 0% reflectance) to 10,000 (representing 100% reflectance). 


GDA’s Surface Reflectance calibrated Landsat (left) draped over USGS’ Surface Reflectance calibrated MODIS c6 (right)

Back to top



GDA has developed industry-leading methodologies and automated algorithms for the sensor-agnostic image sharpening.  Our sharpening algorithms are based on the cutting-edge, state-of-art super-resolution approach.  The goal of the super-resolution is to enhance spatial sharpness of a specific scene or even specific features within a given scene.  It uses pre-built libraries to understand the underlying structure of the target scene and uses this structure to enhance the resolution and interpretability of the imagery.  This approach has only recently started finding its way in to the satellite remote sensing applications.  The result of the sharpening is an analysis ready imagery with all bands at desired spatial resolution and the original spectral / radiometric properties preserved.  The application does not expect co-incident / co-located higher resolution imagery for each sharpened scene; as pan-sharpening, substitution, PCA and other applications would require.  An example of GDA image sharpening can be seen below.


Landsat scene at original 30 meter resolution (left) vs GDA Landsat product at 10 meter resolution (right)

Back to top

Crop ID Maps

GDA is releasing a new and unique Crop ID product for the 2020 US growing season -- In-season Crop ID maps at 10m resolution. GDA’s 2020 Crop ID maps will provide an excellent foundation for crop-specific services, products, or analysis that need to be generated on a timely basis during the growing season. Based on GDA’s years of experience with satellite imagery analysis and proprietary sharpened imagery at a 10m resolution, the crop ID product has key attributes valuable for in-season use:

  • Large coverage: All major and some minor US production states – minimum of 32 states
  • Useful release dates: April – October monthly map releases on the 5th of the month, plus December 5th, based on imagery through the 1st of the month (Apr-Jun winter wheat only)
  • Sharp resolution: 10m resolution provides significantly better granularity than products based on unprocessed 30m resolution imagery
  • Historical maps: 2018 and 2019 maps are available based on the same imagery dates for comparison
  • Delivery format: View and analyze through GDA’s webGIS platform GeoSynergy™, or receive as a GIS file for integration with your own system

GDA Crop ID products rely on our 10m/10day GeoChronicles™ imagery time series.

To ensure model accuracy, the performance of GDA in-season crop ID mapping models is tested against historical end-of-season USDA Cropland Data Layer (CDL) maps and USDA final crop statistics.

GDA crop ID products already contribute to the decision making of various clients worldwide including the USDA Foreign Agricultural Service which uses our crop ID products for the last 15 years.

The example below shows a comparison of the GDA 2017 in-season map to the USDA CDL.

illinois_gdacropid_vs_cdlIllinois: GDA In-Season Crop ID 10m map (left) vs USDA Cropland Data Layer (CDL) 30m map (right)



Missouri: GDA In-Season Crop ID 10m map (left) vs USDA Cropland Data Layer (CDL) 30m map (right)


Contact us for more details or to see a demo of the product.


Crop Type NDVIs

Crop type specific NDVI time series are one-of-a-kind datasets that GDA offers to its clients.   Our crop type NDVIs uniquely represent real crop conditions on the ground and offer our clients with superior understanding of crop conditions and exceptional data for crop monitoring, assessments, and yield analysis.

GDA crop NDVI time series can be requested for individual fields / parcels and for various country and sub-country levels.  Data delivery is triggered by the client’s API data request. 

Our crop type NDVIs are multi-annual datasets which combine almost 20 years of historical data with crop specific NDVI for the most current, on-going season.  GDA crop type NDVI time series are available for major crops in all main agricultural regions of the World. GDA crop type NDVI time series are generated at 250 meters from twice daily MODIS Terra and MODIS Aqua imagery; they rely on time series monitoring of individual fields under a given crop type.  These crop maps provide annual masks for us to sample the field level NDVI data for a given crop, area, and season. 

An example on the figure below shows how 2016 drought impacted Sorghum in New South Wales, Australia.  Notice the difference between a typical sorghum NDVI (grey line) vs an NDVI during the drought year (red line).  Sorghum during the drought experienced conditions dramatically outside of the historically worst conditions.  With our crop NDVI data GDA clients were apprised of the severity and extent of the problem early in the season to adjust their decisions and to take action.


Example for GDA Sorghum NDVI for New South Wales, Australia. Notice the difference between the NDVI for 2016 drought year (red line) vs the typical NDVI (grey line) and historical min / max NDVI range (grey envelope).



GDA sorghum NDVI (left) depicts real crop conditions and provides GDA clients with superior data for crop condition insight and yield anticipation than would have been offered by the commonly used ‘cropland’ NDVIs (right)

Back to top


Crop Yield Assessments

GDA provides its clients with frequent in-season crop yield forecasts at national and sub-national levels for major commodity crops including grains, cereals, oilseeds and cotton.  GDA yield forecasts provide a crop type specific end-of-season yield assessment per administrative unit for the current growing season and its comparison to historical yield statistics.

We have developed our own methodologies and algorithms for providing our clients with timely, accurate, and objective yield forecasts.  In comparison to typical yield assessments based on ground surveys and sampling, GDA's yield estimates predominantly relay on analysis of satellite data that is collected at every point on the globe twice a day and allows for direct observation and analysis of crop status/conditions for 100% of the crop lands under investigation.  

GDA yield models utilize our satellite imagery derived multi-annual crop type NDVIs and measures of crop condition and water content.  They are supported by our proprietary comprehensive global database on historical and most current crop yield, area, production, practices, progress, phenology, calendars along with various crop relevant datasets covering weather, soil moisture, crop type GDD, etc.  The database is constantly being updated with new and revised crop statistics from an unprecedented combination of a multitude of most credible sources across the world.

To produce the yield estimates we train our machine-learning models on vast historical data to predict what yield should be expected this season.  The modeling starts as soon as the crop emerges and becomes visible on the imagery.  The yield estimates are updated a few times a month to accommodate client requirements.  The performance of each model run is tested against historical yield data to ensure the model accuracy.

GDA yield forecasts contribute to the decision making of various clients including the USDA Foreign Agricultural Service which uses our global yield forecasts for the last 10 years.

An example of GDA yield model performance is shown below for US county level estimates for corn. GDA 2005-2017 results for the objective states are compared to USDA stats.  For the vast majority, GDA yield estimates exhibit 0.9 and better R2 correlation with USDA (see the map).  The map demonstrates that GDA achieves high correlations for all counties with typical crop acreage above 25K and with a complete USDA historical yield record.  Not surprisingly, the presumably poorly surveyed counties with small areas under the crop and an incomplete historical record show lower correlation between GDA / USDA.


Example of GDA vs USDA corn yield correlations (R2) by county for objective states for 2005-2017

(Deeper green colors show higher correlation; deeper red colors show lower correlation)

Back to top