Remote sensing of burned residue in fields using Landsat sensor imagery

In recent years due to the benefits of conservation tillage and also disadvantages of crop residue burning, extension and education of conservation tillage, has been highlighted on the agenda of agriculture policymakers. In this regard, for farmers who use conservation tillage or crop residue burning on their farms considered subsidies or crimes respectively. Lack of information, cost, and time consuming of information gathering from the farm using conventional methods, led to the poor performance of law enforcement. Therefore, present research was carried out to find an accurate and fast method for monitoring the residue management. In this research, the ability of Landsat-8 satellite imagery for monitoring of burned fields was evaluated using spectral indices and linear spectral unmixing analysis.
The present research was carried out in the Orzooiyeh region of Kerman province. For conducting the experiment, an area with approximate size of 25 square kilometers is considered and 10 farms (include burnt residue of wheat or corn) were selected randomly in this area as 10 replications of experimental plots. Images were downloaded from the landsat-8 website and all features were extracted from images using ENVI software. On the other hand, the data of real burned areas on the farm were collected using handheld GPS device and also the exact date of residue burning was recorded directly on the field. The maps of experimental farms were prepared using ArcGis software. The correlation between data of real burned area on the farms and ENVI extracted data of burned areas were studied and real burned areas were expressed as a function of burned area that extracted from satellite images by a linear regression curve. Finally, the accuracy of regression functions and correlation between real data and satellite data were calculated. For this purpose, spectral indices include; Normalized Difference of Vegetation Index (NDVI), Burned Area Index (BAI), Normalized Burn Ratio (NBR) and Normalized Burn Ratio Thermal (NBRT) was created for experimental lands and four soil surface condition as experimental plots were considered include; no planted field, residue covered field, green vegetation field and burned residue field.
In the present study, because of extracting the pure spectral data of soil and residue, directly from Landsat-8 images, spectral unmixing analysis was not sensitive to the spectral changes that caused by conditions such as moisture content of soil and plant residue (Pacheco and McNairn, 2011). The average value of BAI index obtained 88.39, 9.29, 4.20 and 6.87 for burned residue field, no planted field, residue covered field and green vegetation field respectively. As can be seen, the average value of BAI index for burned residue field is significantly higher than values for other soil surface conditions. This difference is because of the very low percentage of spectral reflectance of ash in the red and near-infrared bands (Alonso, et al., 2007). Therefore, BAI index was selected as an indicator to distinguish between burnt residue and other three surface conditions in the farm. The result showed, there is a significant difference between means in four soil surface conditions of studied indices. Also, the results showed that the BAI index can be used as a good indicator for separation of burned fields. By the BAI index, location and area of trial burned farms were determined with higher accuracy than other indices. The area average of burned fields that had been separated from other fields using BAI index had high correlation (R2=0.95) with ground-truth data. Also, the area of burned fields that had been estimated by linear spectral unmixing analysis had a good correlation (R2=0.89) with obtained data from the ground-based method.
According to the results, BAI index had most accuracy for estimating burned area of farms and BAI index proposed for separation and determining the area of burnt fields. However, there is a slight error in estimating burned area using spectral indicators and linear spectral unmixing analysis, due to pixel nature of satellite images, basically. Since there is only one spectral data for each pixel of satellite images, spectral data of pixels that are more than the threshold value, are considered as the burned pixels while it is possible, only the part of pixels has been burned, it would be overestimating the actual amount of burned area. And for spectral data of pixels that are less than the threshold value, are considered as unburned pixels while that may be part of pixels is burned, it would be estimating the burned area less than the actual amount.