TY - JOUR
T1 - Synthesis of proximal sensing, terrain analysis, and parent material information for available micronutrient prediction in tropical soils
AU - Pelegrino, M. H.P.
AU - Weindorf, D. C.
AU - Silva, S. H.G.
AU - de Menezes, M. D.
AU - Poggere, G. C.
AU - Guilherme, L. R.G.
AU - Curi, N.
N1 - Publisher Copyright:
© 2018, Springer Science+Business Media, LLC, part of Springer Nature.
PY - 2019/8/15
Y1 - 2019/8/15
N2 - In developing countries, the use of proximal and remotely sensed data is of critical importance as a less expensive means of obtaining soils information. While proximal sensor approaches such as portable X-ray fluorescence (pXRF) spectrometry are becoming increasingly used to predict soil properties worldwide, remotely sensed data has also been used for terrain analysis in recent decades with the aid of powerful interpretive algorithms. The aims of this work were to apply a random forest algorithm to model and predict the available contents of Fe, Cu, Mn, and Zn from pXRF data in addition to terrain attributes (TAs) with 5 and 10 m spatial resolution and parent material information. The data were used separately and together in an area with high variability of parent materials. Soil samples (n = 153) were collected, analyzed by pXRF, and subjected to laboratory analyses to determine the available contents of Fe, Cu, Mn, and Zn. Twelve TAs were generated from digital elevation models (DEM). These data were divided into five datasets (or random forest inputs): pXRF data; TA 5 m data; TA 10 m; pXRF + TA 5 m; and pXRF + TA 10 m. Predictions were performed to assess the importance of such variables. Models were validated with an independent set of samples. Finally, the best models were spatially rendered to cover the entire study area and maps were also validated. The combination of pXRF data and TA covariates in addition to parent material information allowed accurate predictions of available Fe, Mn, Cu, and Zn through the random forest algorithm. Parent material information improved the predictions. Pixel size of 10 m resolution promoted better results than 5 m resolution. Available Fe contents were better predicted using only TA data. For the spatial prediction of available micronutrients, validation of maps resulted in R2 of 0.88, RMSE of 59.97 mg kg−1 and ME of 24.00 mg kg−1 for Fe; 0.85, 29.65 mg kg−1, 9.70 mg kg−1 for Mn, 0.64, 3.11 mg kg−1, 0.71 mg kg−1 for Zn and 0.82, 1.17 mg kg−1, 0.43 mg kg−1 for Cu, respectively. Available micronutrient contents can be accurately predicted using pXRF data in association with terrain and parent material information.
AB - In developing countries, the use of proximal and remotely sensed data is of critical importance as a less expensive means of obtaining soils information. While proximal sensor approaches such as portable X-ray fluorescence (pXRF) spectrometry are becoming increasingly used to predict soil properties worldwide, remotely sensed data has also been used for terrain analysis in recent decades with the aid of powerful interpretive algorithms. The aims of this work were to apply a random forest algorithm to model and predict the available contents of Fe, Cu, Mn, and Zn from pXRF data in addition to terrain attributes (TAs) with 5 and 10 m spatial resolution and parent material information. The data were used separately and together in an area with high variability of parent materials. Soil samples (n = 153) were collected, analyzed by pXRF, and subjected to laboratory analyses to determine the available contents of Fe, Cu, Mn, and Zn. Twelve TAs were generated from digital elevation models (DEM). These data were divided into five datasets (or random forest inputs): pXRF data; TA 5 m data; TA 10 m; pXRF + TA 5 m; and pXRF + TA 10 m. Predictions were performed to assess the importance of such variables. Models were validated with an independent set of samples. Finally, the best models were spatially rendered to cover the entire study area and maps were also validated. The combination of pXRF data and TA covariates in addition to parent material information allowed accurate predictions of available Fe, Mn, Cu, and Zn through the random forest algorithm. Parent material information improved the predictions. Pixel size of 10 m resolution promoted better results than 5 m resolution. Available Fe contents were better predicted using only TA data. For the spatial prediction of available micronutrients, validation of maps resulted in R2 of 0.88, RMSE of 59.97 mg kg−1 and ME of 24.00 mg kg−1 for Fe; 0.85, 29.65 mg kg−1, 9.70 mg kg−1 for Mn, 0.64, 3.11 mg kg−1, 0.71 mg kg−1 for Zn and 0.82, 1.17 mg kg−1, 0.43 mg kg−1 for Cu, respectively. Available micronutrient contents can be accurately predicted using pXRF data in association with terrain and parent material information.
KW - Brazil
KW - Portable X-ray fluorescence spectrometer
KW - Random forest
KW - Soil property prediction
KW - Tropical soils
UR - http://www.scopus.com/inward/record.url?scp=85053673041&partnerID=8YFLogxK
U2 - 10.1007/s11119-018-9608-z
DO - 10.1007/s11119-018-9608-z
M3 - Article
AN - SCOPUS:85053673041
SN - 1385-2256
VL - 20
SP - 746
EP - 766
JO - Precision Agriculture
JF - Precision Agriculture
IS - 4
ER -