Improving Soil Available Nutrient Estimation by Integrating Modified WOFOST Model and Time-Series Earth Observations
Cheng, Z., Meng, J., Shang, J., Liu, J., Qiao, Y., Qian, B., Jing, Q., Dong, T. (2019). Improving Soil Available Nutrient Estimation by Integrating Modified WOFOST Model and Time-Series Earth Observations. IEEE Transactions on Geoscience and Remote Sensing, [online] 57(5), 2896-2908. http://dx.doi.org/10.1109/TGRS.2018.2878382
Plain language summary
Information on soil available nutrient at key crop growth stages is critical to generate prescription maps for implementing variable rate fertilization. Integration of time-series remote sensing data with the modified World Food Studies (WOFOST) crop model provides a useful approach to acquiring information on field soil available nutrient; however, the estimation accuracy was low. In this study, three steps were proposed to improve nutrient estimation accuracy further, which include a rapid-nutrient assimilation method, an optimization step for improving estimation of potassium content, and iteration to improve nitrogen, potassium, and phosphorous content. This procedure produced an improved soil nutrient estimation , that could be useful for precision agriculture.
Information on soil available nutrient (SAN) at key crop growth stages is critical to generate prescription maps for implementing variable rate fertilization (VRF). Our previous study showed that integrating time-series remote sensing (RS) data with the modified World Food Studies (WOFOST) crop model provides a useful approach (preliminary RS-WOFOST-based method) to acquiring information on field SAN; however, the estimation accuracy was low for VRF application. In this paper, three steps were proposed to further improve the SAN estimation accuracy. At the first step, the rapid-nutrient assimilation (RNA) method was used to optimize the crop growth simulation process. Compared with the ensemble Kalman filter (EnKF) method, the RNA method showed an improved performance in estimating soil available nitrogen (N), phosphorus (P), and potassium (K) content [EnKF: R2 = 0.48 (N), 0.37 (P), 0.15 (K); RNA: R2 = 0.59 (N), 0.46 (P), and 0.18 (K)]. The improved K estimation at the first step was clearly lower than that of N and P; hence, the K content estimation was further optimized at the second step and the accuracy was improved ( R2 = 0.27 ) by using the estimated N as an input variable during the K estimation. At the third step, an iteration algorithm was implemented based on the first two steps, and the final R2 = 0.71 (N), 0.58 (P), 0.49 (K); root-mean-square error = 14.35 (N), 3.70 (P), and 14.87 (K). In general, the optimized approach can overcome the limitations of the preliminary RS-WOFOST-based method and improve the SAN estimation accuracy.