Application of hyperspectral canopy reflectance measurement and partial least square regression to predict within-field spatial variation in crop growth and nitrogen status before heading stage of rice
In: Precision Agriculture ‘05Search for other papers by Hung. T. Nguyen in
Current site
Google Scholar
PubMed
Search for other papers by Jun Han Kim in
Current site
Google Scholar
PubMed
Search for other papers by Anh T. Nguyen in
Current site
Google Scholar
PubMed
Search for other papers by Jin Chul Shin in
Current site
Google Scholar
PubMed
leebw@snu.ac.kr
Search for other papers by Byun-Woo Lee in
Current site
Google Scholar
PubMed
Purchase instant access (PDF download and unlimited online access):
For site-specific prescription of fertilizer topdressing in rice cultivation, nondestructive diagnosis of the rice growth and nutrition status is indispensable. Three experiments were carried to develop and test a model using canopy reflectance for nondestructive diagnosis of the rice growth and N status. Two experiments for model development were conducted, one in 2000 and one in 2003 at the Experimental Farm (37o16’ N, 126o59’ E) of Seoul National University, Suwon, Korea. The experiment included two rice varieties and four nitrogen (N) rates in year 2000 and four rice varieties and 10 nitrogen treatments in year 2003. Hyperspectral canopy reflectance (300 to 1100 nm) data recorded at various growth stages before heading were used in developing a partial least square regression (PLS) model to predict eight crop variables such as plant biomass and N nutrition status. 342 observations were split for model calibration (75%) and validation (25%). The PLS model was then tested to predict within-field spatial variation of three crop variables using hyperspectral canopy reflectance data measured for 10 × 10m grids from a paddy field of 6500 m2 in 2004. Coefficient of determination (R2), root mean square of error in prediction (RMSEP) and relative error of prediction (REP) were calculated for the model quality evaluation. The results revealed that PLS using hyperspectral canopy reflectance data to predict eight plant variables in year 2000 and 2003 produced an acceptable model precision and accuracy. The model R2 and REP ranged from 0.81 to 0.88 and 10.0 to 23.8% for calibration and 0.76 to 0.85 and 11.1 to 24.6% for validation, respectively. The model R2 was reduced in the test data of year 2004 but the error in prediction (RMSEP and REP) was smaller. The PLS model using canopy reflectance data could be a promising method to predict within-field spatial variation of rice crop growth and N status if error sources occurring during canopy reflectance measurement were successfully controlled.