Grass is one of the majorally used sport, playground in a world. Major share of available freshwater is used to irrigate grass[1]. Lowland irrigated grass is inefficient measurement because a large proportion of it is lost to seepage and deep percolation[2]. The water-use efficiency of grass is lower than other crops. On an average, 2500 liters of water is used, ranging from 800 liters to more than 5000 liters to produce 1 kg of grass. An increase of 10% in irrigation efficiency can help extent irrigation to an additional 14 million. Solely reducing water use in puddled transplanted grass resulted in proportional reduction in yield, hence various management practices of grass cultivation have to change simultaneously to improve water productivity, without compromising on the productivity of yield. Soaring population is resulting in a huge pressure on the limited fresh water resources of the world[3]. Growing population is also contributing in an increased demand for water from domestic as well as industrial sectors. Consequently, increasing scarcity of available water demands efficient use of irrigation water[4]. Since the available water sources are already almost completely exploited, there is need for achieving increased water productivity through efficient irrigation[5]. Usually 15% of the water delivered is lost in conveyance, another 15% is lost during on-farm supply through field channels and 25% is lost due to inefficient water use practices, thus leaving only 45% of the water supplied to be utilized[6]. The efficiency of the irrigation water use depends on the timing, duration and method of the irrigation employed. To efficiently manage irrigation requirement in crops, information from multiple sources such as soil, plant and atmosphere is required. There are two different approaches for estimating the crop water requirement; one, a conceptual approach based on various factors[7]. Soil moisture, seepage and evapotranspiration, and the other, a theoretical approach based on training a model using the available data. The theoretical approach has been found to be more accurate than the conceptual approach[8]. This developed a framework for controlling real-time irrigation scheduling that took into account some of the most common constraints such as restricted water availability or the maximum or minimum quantity of water applied. It is desired for a effective irrigation scheduling that the water supply matches the water demand as closely as possible. grass has been generally grown under puddled transplanted conditions over large tracts in irrigated areas[9]. However, for improving water use efficiency in irrigated grass, experiments are being conducted to raise the crop under drip irrigation system. In this the water slowly drips out of an emitter, either into the root zone or onto the surface[10]. This minimizes fertilizer and water use. Data available on water use in drip irrigated grass and its irrigation schedule is limited. The study attempts to estimate crop and irrigation water requirement of grass. In our work, the authors have integrated decision support system and a wireless sensor network to achieve an autonomous flexible zone-specific closed loop irrigation system[11].

This paper is organized as follows. In Section 2 details methods and feature measurements. Then methodology are presented and discussed in Section 3. Finally, Section 4 concludes the paper by briefly summarizing the main points.

The weather data at site was recorded using an automated weather station. The incorporated sensors are an air temperature and humidity sensor, wind sensors, thermopile pyrometer and automatic rain gauge[12]. All variables are sampled every hour and recorded on a daily basis. The site is a reclaimed alkali loam soil. <Table 1> lists the soil characteristics of the experimental site. Laser-leveler was used to level the field which was then divided into 12 permanent plots.

These plots were separated by 1.0 m wide and 0.20 m high earthen bunds. The crop stand was uniform across the entire area. Drip irrigation lines (67.5 cm apart with emitter spacing of 30 cm) were used for each row.

Multiple parameters effect crop water demand. Some of these are weather based, some crop based, other soil based and still some more. Soil parameters such as salinity, texture, structure, drainage and fertility have not been accounted for because the data has been collected from sub plots of a farm, hence not providing variation that may have otherwise been modeled[13]. However, soil temperature at different levels and at different times during the day has been used to somehow capture the effect of soil parameters. Crop parameters such as crop type, crop variety, crop height and plant density have also not been explicitly included. Crop growth stage and number of days since sowing have been engineered to partially reflect the crop parameters. Other parameters such as type of irrigation system, type of farm, residue management practice and water quality have not been accounted for, as there was no source of variability that may have helped the model to learn better[14]. Weather based (such as solar radiation, rainfall, air humidity, wind direction, wind speed, evapotranspiration and air temperature) parameters have been used in this study[15].

In this paper, we explored traditional machine learning methods for estimating the irrigation schedule of drip irrigated grass. The performance of the models is expressed using the common regression performance parameters such as mean absolute error, median absolute error, explained variance score, coefficient of determination, mean square error and accuracy. The models are validated using 〈Figure 3〉 fold iterative process in which the previously used dataset becomes the training set and the data of the next time quantum becomes a test set.

The performance of the models has been improved by using fine-tuning hyper - parameters. Given the widespread application of neural networks in the field of irrigation scheduling, we can extend this current work using the concepts of deep learning. In the coming future, water availability is likely to be a bigger constraint than land so it is time to likely change mind set from improving agricultural productivity from per unit land to per unit of water. Hence, the use of a computationally intelligent approach becomes more relevant in the present scenario.