Mapping Crop Residue with Landsat Data using ENVI for ArcGIS®

Customer Challenge

Traditionally, farmers have “worked” or tilled their land to control weeds and kill volunteer crops. These tillage methods remove crop residues, disturb the natural soil structure, and expose the soil to erosion. Conservation tillage, on the other hand, uses either a no-till or minimal tilling method that leaves varying amounts of the previous year’s crop residue on fields before and after planting the next crop.

Tillage information is an important component in agricultural and environmental analysis and policy because of the impact these practices have on soil erosion, moisture retention, and crop yields. An effective technique was needed to extract tillage information from Landsat data, even though previous studies suggested this wouldn’t be feasible.

Solution Achieved

Because the timing of tillage implementation and planting varies, the solution was to extract minimum NDTI (Normalized Difference Tillage Index) values of a time-series (minNDTI) using ENVI^® and IDL^® routines to get a clearer picture about when and where tillage was occurring.

Viewing Fields from Afar

Remote sensing with image analysis software like ENVI allows for rapid acquisition of tillage extent information, which is an important input for agricultural greenhouse gas estimates, soil erosion, water quality models, and can be used to verify compliance with soil conservation and carbon credit trading programs. While hyperspectral and ASTER shortwave infrared (SWIR) data are ideal for this purpose, the required data stream for large scale analysis is difficult to obtain.

During his post-doctoral work, Dr. Guy Serbin was using hyperspectral data to study crop residue cover and tillage practices at the USDA/ ARS (Agricultural Research Service) Hydrology and Remote Sensing Laboratory in Beltsville, MD with Dr. Craig Daughtry and Dr. E. Raymond Hunt Jr. Serbin’s research showed that for single scenes, Landsat data produced poor results in comparison with hyperspectral and ASTER-based indices. Hyperspectral and ASTER-based indices are preferred for this type of analysis because the sensors target specific absorption features that are present in crop residues (also referred to as plant litter, senescent vegetation, or non-photosynthetic vegetation) but not in soil minerals.

Following a meeting at the 2009 ASPRS Conference in Baltimore, MD, Serbin and Baojuan Zheng, a Virginia Tech Ph.D. graduate student advised by Dr. James Campbell, began to collaborate on remote sensing of tillage using Landsat. Dr. Zheng successfully tried a new approach by using a time series of Landsat-derived NDTI, and tested her results against ground truth from Serbin’s postdoctoral research. Her new methodology provided results that were similar in quality to those from hyperspectral and ASTER-based images. Following this, Serbin, who currently works as an imagery processing analyst for Inuteq LLC on a contract with the USDA, decided to take Zheng’s work and develop an operational process for mapping tillage residue with Landsat data using ArcGIS and ENVI+IDL.

“Tilling happens at different times, as does planting,” says Serbin. “For this reason, looking at a single scene wasn’t helpful. However, comparing a time-series provided a much clearer picture of what was happening.” Zheng used NDTI (Normalized Difference Tillage Index) time-series, but because NDTI is more sensitive to green vegetation than crop residues or soils. The best way to isolate tillage practices was to extract minimum NDTI values of a time-series (minNDTI).

Zheng discovered that a temporal sequence was required to cover the whole planting season to capture the existence and timing of tilling practices. By using scenes over the growing season, minNDTI provided an accurate measure of tillage where individual NDTI scenes failed. Long-term global coverage of Landsat and Sentinel-2 imagery provides opportunities to map tillage practices at site-specific detail, at broad scales, over time. While this work is not currently being used for regional modeling, Serbin ventured that it would be fairly easy to implement on a wide scale.

While Zheng’s approach was successful, it utilized surface reflectance measures which can entail intensive processing steps that prove impractical for use on large datasets. Correcting Landsat digital numbers to Top of Atmosphere (TO A) reflectance measures can provide a basis for standardized comparison of data in a single scene or between images acquired on different dates like surface reflectance, but less process intensively. Serbin found that the TO A-corrected results yielded acceptable accuracy and allowed him to eliminate the surface reflectance correction step and improve processing time.

Serbin set out to automate the process for identifying when tillage or planting had occurred. Since ArcGIS served as a familiar platform for tool operation and data visualization, Serbin chose to implement a python script callable from the ArcGIS interface that entailed the following procedures:

• Selecting a list of input files, ordered by Landsat WRS-2 Path, Row, and acquisition date in binary form, which included the year and the Julian day divided by the number of days in the year.
• Calling ENVI processing capabilities from ArcMap using the envipy.RunTool to convert data to Top of Atmosphere (TO A) reflectance; calculate NDVI, NDTI, and a cloud excessive NDVI (>0.3); and create a zero-value pixel mask for each scene.
• Using ENVI and IDL routines to stack masked NDTI values for each Path/Row combination in order of time, calculate minimum NDTI values and occurrence dates for the time series, and use the results to render separate images.

The results from the minNDTI process were comparable to those from hyperspectral and ASTER-based indice and also agreed with field-level data—validating that the results were indeed accurate and not the result of noise in the time-series. The outcome of the study demonstrates the viability of the widely available Landsat data source for this type of work.

These tillage maps show coherent spatial patterns of tillage applications that are related to soil, cropping patterns, and other landscape characteristics. The patterns may also be related to the indirect effects of government conservation programs that encourage farmers to adopt conservation and no-till practices.

Conservation Tilling — On the Ground

There are both environmental and practical considerations for conservation tilling. Soil erosion can be reduced by as much as 60 to 90 percent since crop residue like corn stalks or wheat stubble shield soil particles from rain and wind until new plant growth can create a protective canopy over the soil. As crop residue decomposes, it adds organic matter to the soil that improves soil quality, letting water absorb more easily and further reducing runoff. This tillage method decreases water evaporation at the soil surface and reduces potential air pollution from dust and diesel emissions.

Jeremy Patterson currently farms near Copeland, Kansas. His family has been farming there since 1924, but the record drought these last few years prompted him to try no-till and conservation tilling practices. “The major benefit is saving moisture, which is critical in this drought. We currently don’t do any full tillage,” says Patterson. “This is our third year of record drought conditions. Since January, we have only had 4.22 inches of moisture. 2011 was the driest year on record here, 2012 was only slightly better," according to Patterson.

Patterson uses strip till on irrigated corn acres. “That basically means we put all our fertilizer on in the fall with a strip every 30 inches that we work. We leave the rest of the soil and residue undisturbed and then plant directly in the strip in the spring,” says Patterson. “On our dryland acres, we use conservation tillage, working the ground only one time in a two-year period. We do this mainly to kill any resistant weeds.”

Other practical applications of mapping senescent vegetation with remote sensing are for rangeland health monitoring and fire fuel detection. Senescent vegetation cover is an important indicator of rangeland soil health and grazing patterns. Also, dry plant material easily catches fire, and wildfires cause loss to human life and billions of US dollars in damage annually. The work that Serbin and his group performed will help to map these features.

Conclusion

Greater access to imagery is creating opportunities to improve productivity and process efficiencies across a variety of environmental applications. Whether calculating agricultural yields or producing fire mitigation plans, ENVI image analysis techniques yield information and data layers that enable a GIS to address complex spatial problems. By leveraging ENVI and IDL image analysis capabilities to implement a Landsat-based tillage mapping application within an ArcGIS framework, Drs. Serbin and Zheng demonstrated the power and flexibility of integrating remote sensing and GIS methods to achieve scientifically-accurate agricultural analysis results.

ENVI for ArcGIS has over twenty prebuilt image analysis tools available via the ArcToolbox that allow ArcGIS users to extract information from virtually any type of geospatial imagery, including hyperspectral, multispectral, panchromatic, LiDAR and Synthetic Aperture Radar (SAR). And, since ENVI is the only image analysis software that is highly integrated with ArcGIS and fully extensible via a high-level development language (IDL), customized desktop or web-based image processing workflows can enable desktop, server, or Online ArcGIS users to integrate information from imagery with their GIS, regardless of device type and location.