Hyper-spectral retrieval of soil nutrient content of various land-cover types in Ebinur Lake Basin

The soil nutrient affects soil quality, vegetation type, crops growth and yield. To rapidly and accurately determine soil nutrient contents, an indoor spectral data (measured by ASD FieldSpec3) of 75 soil samples of Jinghe County of Ebinur Lake Basin were analyzed. Then the collected data were processed at resampling interval of 10 nm to suppress noise. Soil nutrient hyper-spectral forecast models were used to forecasts soil nutrient contents in three transformation conditions. The performance of the models was evaluated based on stepwise multiple linear regression (SMLR) analysis, partial least squares regression (PLSR) analysis and artificial neural network (ANN) analysis and the optimal model determined by comparison. The results showed that the transformation of first-order and second-order differential dramatically enhanced correlation between spectroscopy data and soil nutrient content. Specifically, the first-order differential of soil spectroscopy had a good correction with soil nutrient content. The correlation coefficients for organic matter and total nitrogen were 0.87 and 0.91, respectively. In conclusion, spectral transformation technique increased the sensitivity of high spectral data to soil nutrient change, and it produced far better forecasting results. Although all the three models had good predictive ability, ANN model had the best predictive effect, followed by PLSR model. The ANN model estimation test based on the second-order differential of spectroscopy data with independent datasets from different soil samples respectively produced R² and RMSE values of 0.885 and 0.984 for organic matter and 2.614 and 0.147 for total nitrogen. The prediction effect of total nitrogen was obviously better than that of organic matter. This indicated that the sensitivity of soil hyper-spectral reflectance to the soil total nitrogen content was much better. Overall, the ANN model based on the second-order differential of spectroscopy data rapidly and precisely predicted soil nutrients contents. It was beneficial for monitoring spatial distributions and dynamic changes of soil nutrients in Ebinur Lake Basin.