Prediction of land suitability class of alfalfa, potato and irrigated wheat using random forest learning machine and auxiliary data

Improper use of land resources due to increased human food needs has led to the destruction and reduction of arable land. One way to increase production per unit area is to land suitability assessment. Land suitability assessment is the fitness of a type of land for defined use. Assessing spatial variability of land suitability class is necessary to increase production and prevent land degradation. Determining the land suitability class requires measuring soil, topography, moisture and climate properties, which are costly and time consuming. One solution to this problem is to use learning machines and auxiliary data. Learning machines are used to relate various properties with auxiliary variables to assess their spatial and temporal variability. Random forest learning machine is one of the most common and widely used learning machines. The aim of this study is to assess land suitability based on FAO land suitability framework and parametric method for three important irrigated crops of the region, including alfalfa, potato and irrigated wheat, and to predict their land suitability classes using random forest learning machine and auxiliary data.

122 soil profiles were dug, described and sampled in the Ghorveh area of Kurdistan Province (covers 6500 ha). Soil texture, acidity, organic carbon, CaCO3, gypsum, ESP, electrical conductivity and gravel were measured in all soil samples. Moreover, topography and climate data were also recorded. Environmental variables in this research were terrain attributes, land unit components map, and data of ETM+ image. To make a relationship between land suitability class and auxiliary data, random forest (RF) learning machine were applied and using cross validation method and statistic indices including overall accuracy and kappa index was validated.

The results showed that suitability class of the study area has 37, 41 and 57% N2 class, 21, 34 and 27% N1 class and 48, 19 and 16% S3 class for irrigated wheat, alfalfa and potato, respectively. The major limitations of the study area to plant the crops are included high slope, shallow soil depth, high pH and gravel.To predict land suitability class of alfalfa, potato and irrigated wheat, auxiliary variables including MRRTF index, MRVBF index, wetness index, LS factor, elevation and land unit components map were the most important. The results of this study showed that the random forest learning machine for prediction of land suitability class of irrigated wheat with 0.78, and 0.71, alfalfa with 0.75 and 0.70 and potato with 0.79 and 0.72 for overall accuracy and kappa index, respectively, had a suitable accuracy.

Topography is the most important soil forming factor and is effective in distribution of land suitability class. The study area, because of limitation of soil and topography has low to non-suitable suitability to plant these crops and it is suggested proper land improvement operations to increase production and land sustainability management. Random forest learning machine had suitable accuracy for predicting land suitability class. Therefore, it is suggested to map land suitability class learning machine techniques (such as randomized forest) and auxiliary data such as terrain attributes, land unit components map and satellite images were applied.