Data Driven Model to Estimate Furrow Irrigation Sediment Loss in Western U.S.

Highlights Kohonen self-organizing maps (KSOM) combined with linear regressions were used to estimate furrow sediment loss. A transfer learning approach was used based on KSOM to cluster furrow soil conditions and hydraulic parameters. Mean sediment loss of KSOM clusters was adjusted for furrow flow rate and furrow length to predict sediment loss. Predicted sediment loss was on average 17% less than measured with an R2 of 0.83. The furrow sediment loss prediction model was implemented in a spreadsheet for ease of use without special software. ABSTRACT. The area irrigated by furrow irrigation in the U.S. has been steadily decreasing but still represents about 20% of the total irrigated area. Furrow irrigation sediment loss is a major water quality issue in the western U.S. and a method for estimating sediment loss is needed to quantify the environmental impacts and estimate the effectiveness and economic value of conservation practices. The objective of the study was to investigate the use of the unsupervised machine learning technique Kohonen self-organizing maps (KSOM) to predict furrow sediment loss. Historical published and unpublished data sets containing measurements of furrow irrigation sediment loss in the western U.S. were assembled into a furrow sediment loss data set comprising over 2000 measurements. The KSOM model underpredicted furrow sediment loss by 57% on average with a linear regression coefficient of determination of 0.51. The KSOM model was placing little weight on measured sediment loss in the input data set, indicating that it was clustering the data based on input parameters defining the hydraulic and soil conditions and placing little weight on measured sediment loss, likely due to inherent variability in sediment loss data. This outcome led to developing a transfer learning approach for predicting furrow sediment loss. The transfer learning approach used a KSOM to cluster data records of similar furrow hydraulic and soil conditions in the data set. Mean measured sediment loss and furrow flow rate of each cluster was determined based on data set vectors assigned to each cluster by the KSOM. The KSOM was then used to determine the best matching cluster based on a new input vector. Then, the mean measured sediment loss of that cluster was adjusted using linear regression to account for differences between the input vector's furrow flow rate and length and the cluster's mean values. This adjustment provided the predicted furrow sediment loss for the KSOM input vector. On average, the predicted furrow sediment loss was 17% less than the measured sediment loss, with a coefficient of determination of 0.83. When the data set was randomly split into 90% model development and 10% validation data sets for k-fold validation, the prediction results were consistent. The furrow sediment loss prediction model was implemented in a spreadsheet for easy prediction of furrow sediment loss without the need for special software. Keywords: Artificial intelligence, Furrow irrigation, Model, Sediment loss, Self-organizing maps, Surface irrigation, Transfer learning.