Soil moisture estimation using novel bio-inspired soft computing approaches

Abstract

Soil moisture (SM) is of paramount importance in irrigation scheduling infiltration runoff and agricultural drought monitoring. This work aimed at evaluating the performance of the classical ANFIS (Adaptive Neuro-Fuzzy Inference System) model as well as ANFIS coupled with three bio-inspired metaheuristic optimization methods including whale optimization algorithm (ANFIS-WOA) krill herd algorithm (ANFIS-KHA) and firefly algorithm (ANFIS-FA) in estimating SM. Daily air temperature relative humidity wind speed and sunshine hours data at Istanbul Bolge station in Turkey and soil temperature values measured over 2008-2009 were fed into the models under six different scenarios. ANFIS-WOA (RMSE = 1.68 MAPE = 0.04) and ANFIS (RMSE = 2.55 MAPE = 0.07) exhibited the best and worst performance in SM estimation respectively. All three hybrid models (ANFIS-WOA ANFIS-KHA and ANFIS-FA) improved SM estimates reducing RMSE by 34 28 and 27% relative to the base ANFIS model respectively. A more detailed analysis of model performances in estimating moisture content over three intervals including [15-25) [25-35) and >= 35% revealed that ANFIS-WOA has had the lowest errors with RMSEs of 1.69 1.89 and 1.55 in the three SM intervals respectively. From the perspective of under- or over-estimation of moisture values ANFIS-WOA (RMSE = 1.44 MAPE = 0.03) in under-estimation set and ANFIS-KHA (RMSE = 1.94 MAPE = 0.05) in over-estimation set showed the highest accuracies. Overall all three hybrid models performed better in the underestimation set compared to overestimation set.