|Shaw, David - MSU|
|Reddy, Kambham - MSU|
|Bruce, Lori - MSU|
|Tamhankar, Hrishikesh - MSU|
Submitted to: Weed Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: March 4, 2004
Publication Date: September 14, 2004
Citation: Henry, W.B., Shaw, D.R., Reddy, K.R., Bruce, L.M., Tamhankar, H.D. 2004. Spectral reflectance curves to distinguish soybean from common cocklebur (xanthium strumarium) and sicklepod (cassia obtusifolia)grown with varying soil moisture. Weed Science. 52:5 p.788-796. Interpretive Summary: Remote sensing data may be used to distinguish between weeds and crops. In order to determine if these data can be useful in a field with varying soil moisture, weeds and crops were grown at different moisture levels and were compared spectrally. Weeds and soybean were grown at three moisture levels and various parameters were measured. Moisture stress did not make it harder to distinguish between weeds and crop. Several analysis techniques were used to discriminate between weeds and crop. Although most of these techniques were successfully used to identify crop versus weed (between species), they were for the most part unable to identify the various moisture levels (within species). The models developed from this research are ready to be tested on field scale data.
Technical Abstract: Experiments were conducted to examine the utility of hyperspectral remote sensing for discriminating between weed species and soybean across moisture levels. Weed species and soybean were grown at three moisture levels and hyperspectral reflectance data, and leaf water potential (LWP), were collected every other day following the imposition of moisture stress at eight weeks after planting. Moisture stress did not reduce the ability to discriminate between species. Regardless of analysis technique, the trend was that as moisture stress increased, so too did the ability to distinguish between species. Signature amplitudes (SA) of the top 5 bands, discrete wavelet transforms (DWT), and multiple indices were promising analysis techniques. Discriminant models created from one year's data set and validated on additional data sets provided, on average, approximately 80% accurate classification among weeds and crop. This suggests that these models are relatively robust and could potentially be used across environmental conditions in field scenarios. Distinguishing between leaves grown at high-moisture stress and no-stress was met with limited success, primarily because there was substantial variation among samples within the treatments. Leaf water potential (LWP) was measured, and these were classified into three categories using indices. Classification accuracies were as high as 68%; however, there was a tendency to overestimate leaf moisture stress or underestimate LWP. Hyperspectral vegetation response recorded reflectance intensity in 2151 individual bands. In an effort to determine which of the 2151 spectral bands contributed most significantly to LWP, correlations were generated between LWP and the 2151 spectral bands for sicklepod and for common cocklebur. The 10 bands most highly correlated to LWP were selected; however, there were no obvious trends or patterns in these top 10 bands with respect to time, species or moisture level, suggesting that LWP is an elusive parameter to quantify spectrally.