Mapping Potential Agricultural Areas with Remote Sensing

By Kailee Woodbeck on 2015-04-20 19:29

Supervised classification results of orchards and farmland.

The goal of this project was to find potential agricultural areas in Creston Valley that were not currently being used for farming.

The imagery utilized for this project was 30 metre Landsat 8 satellite imagery and 20cm fixed wing captured aerial photography. The satellite imagery was downloaded from this awesome Landsat Image Browser for June to September of 2013 and 2014. The fixed wing aerial photography was obtained from the Regional District of the Central Kootenay (RDCK) for 2012.

Landsat 8 Analysis

To determine pixels with live vegetation, Normalized Difference Vegetation Index (NDVI) was calculated. The NDVI results were reclassed to eliminate all non vegetation pixel values (ie. water, rock, roads, buildings etc). The results were then combined to create one raster for each year that contained the maximum NDVI value for each pixel.

Combined NDVI results of live vegetation for 2014 Landsat Imagery showing light to dark green as low to high NDVI values.

The next step was to determine what was causing low/ high NDVI values across the results. Zonal statistics for each parcel of land were calculated from the NDVI results. From the zonal statistics results, hot spot analysis was performed to determine any spatial relation of the NDVI values.

Hotspot Analysis results showing blue as lower NDVI values and red as higher NDVI values for each parcel.

It was evident from the hotspot analysis results that there was clustering of areas with low/ high NDVI values. Regression analysis was run on the resulting zonal statistics against several variables including type of farm, soil type, elevation and slope to try to explain the NDVI value clustering, A statistically significant relationship for low NDVI values was identified between rockier soil types and orchards.

Using ENVI’s** change detection** tool, the NDVI results from 2013 to 2014 were compared. The results showed a 1-50% reduction in high NDVI values on 55% of the scene.