1. Introduction
Maize is an important cereal crop in Uganda, cultivated on over 23.4% of the land dedicated to food crops [
1]. Currently, maize is a major staple food for low income earners in rural and urban areas, providing a varied diet to households and institutions (schools, prisons, factories, etc.) in the form of roasted green cobs, steamed green cobs, and maize flour prepared as
posho (maize porridge). Maize is a source of income for smallholder famers, providing a living to over 3.8 million Ugandans and a source of government revenue through foreign exchange, contributing to over US
$ 100 million in forex earnings [
1].
Over the years, Uganda’s agriculture has mostly depended on rainfall which is distributed in a bimodal pattern, with short rains coming between March to June and the long rains from September to December. For the period 1940 to 2009, annual rainfall amounts ranged from 660 to 1600 mm [
2]. Because of this rainfall distribution, many parts of Uganda—with the exception of the northern part of the country—produce maize during the two seasons. The Northern region, which is a relatively drier area, produces maize in only one season (October to January).
However, the required rainfall for agricultural production is increasingly becoming unreliable due to climate change affecting rainfall distribution [
3,
4], and subsequently affecting food production. Data indicates that maize output is only 2.663 × 10
6 tons from 1.149 × 10
6 ha [
5]. This output can’t respond to the needs of the growing population, i.e., 3.2% per annum [
6], and hence, threatens food security. Thus, Uganda is food insecure, with 6% of Ugandans surviving on one meal a day, 14% of children stunted, and 48% of Ugandans being food energy deficient [
7].
With the increasing threat of food insecurity on the population, there is a need for efficient and effective food production systems if the demands of the high population are to be met. However, these production systems require more water use, yet agricultural water use faces competition from other sectors like municipal, industrial, and ecological [
8] activities. Although smart irrigation water saving technologies (rain sensor, soil water sensor and evapotranspiration controller) [
9,
10,
11] have been developed, Uganda’s agriculture is dominated by small and medium scale farmers with national average land holding sizes of 1.1 ha [
12]. These farmers cannot afford the investment and maintenance costs of systems such as smart irrigation equipment. Therefore, this study sought to find an alternative method to accurately predicate the yields (biomass and grain harvest) of Uganda’s staple crop, maize, under rainfed agriculture. This was in an attempt to save farmers labour and input costs in situations where the grain harvest would be zero due to crop failure. Rockström and Barron [
3] showed that it is possible to at least double rainfed staple food production by producing more ‘crop per drop’ of rainwater. Crop yield simulation models like APSIM [
13], DSSAT [
14], and FAO AquaCrop [
15] have been widely used as decision support tools in the agricultural sector.
However, these models are often applicable only to the fields for which they are calibrated, and require a number of parameters for their application, hence limiting their application in developing countries like Uganda, where there are challenges of data collection due to inadequate equipment and funds to conduct research.
The FAO AquaCrop model, originating from the “yield response to water” [
16], developed in 2009 into a normalized crop water productivity concept [
17,
18]. In comparison to other models, the AquaCrop is relatively easy to run, as it requires only a set of easy to acquire input parameters [
18,
19]. The AquaCrop model is a robust model capable of simulating crop performance in multiple scenarios.
The AquaCrop model has been tested on grain crops in different environments, for example maize in California, USA [
19], maize in Zaragoza, Spain [
20], barley in Ethiopia [
21], and maize in Kenya [
22]. However, no such study has been reported for Uganda. Therefore, the purpose of this study was to assess the capability of the AquaCrop model to simulate the growth and yields of maize in Uganda.
2. Materials and Methods
2.1. Description of the Study Area
The study was conducted at the Makerere University Agricultural Research Institute Kabanyolo (MUARIK), Wakiso district, Uganda (Latitude 0°28′00.38″ N, Longitude 32°36′46.01″ E and 1161 m above sea level). The weather station at MUARIK is located at the following coordinates: Latitude 0°27′49.99″ N, Longitude 32°36′28.68″ E.
MUARIK is characterized by a typical tropical climate with average maximum temperatures of 28.5 °C and minimum temperatures of 14 °C. The mean annual rainfall approximates 1200 mm in a bimodal distribution, with the short rains coming between March to June, and the long rains from September to December [
23]. The soil type is clay-loamy.
Longe5, popularly known as “Nalongo”, a local cultivar commonly grown by farmers, was used in the experiment, and the local practices of farm management (ploughing, weeding, pest management, no fertilizer amendments) were followed. Longe5 was released by the National Agricultural Research Organization in 2000 in response to farmers’ concerns about the low productivity of maize crops in Uganda. Farmers have shown a preference for this variety because of its quality protein, easy access to seed, and good adaptability [
24]. Plot area was 64 m
2 in all the growing seasons, replicated three times in a randomized complete block design. Some of the relevant information required for the model is presented in the
Table 1. The September to December 2014 season received high precipitation, averaging 5.12 mm/day, and fairly distributed throughout the whole season. In contrast, the March to July 2015 season had relatively low precipitation, averaging 2.82 mm/day, with low downfall at the start of the season and gradually decreasing rainfall towards the end. The September to December 2015 season had high rainfall, averaging 5.31 mm/day and fairly distributed throughout the entire season. The mean seasonal temperature was 20 °C. The mean annual air humidity was 78%, and the average wind speed at a height of 2 m was 3.2 m/s.
2.2. Model Input Data
Weather data (Rainfall, mm; Sunshine hours, h; Wind speed, m/s; Temperature, °C and Relative humidity, %), were obtained from the MUARIK weather station located 0.60 km southwest of the experimental field. The Mean annual atmospheric carbon dioxide concentration for recent years, measured at Mauna Loa Observatory in Hawaii, is provided in the AquaCrop and regularly updated [
20]. The daily reference evapotranspiration (ETo) for each growing season was calculated based on the FAO Penman-Monteith method [
25] using the ETo calculator [
26].
Three Soil samples were collected from the study area for each of the four laboratory analyses. The soil properties laboratory analysis results are presented in
Table 2. However, permanent wilting point moisture content (Ɵ
wp), field capacity moisture content (Ɵ
fc), and hydraulic conductivity at saturation (
ksat) were the only properties recommended as model inputs as Heng et al. [
20]. The saturated hydraulic conductivity was obtained by a constant head method; the soil moisture content at field capacity and wilting capacity were obtained by the pressure plate apparatus method [
27]. Given the soil type and the plant patterns, the Curve Number (CN) for the field was 67 [
28]. The Readily Evaporative water value used in the model was 5 mm.
The AquaCrop model uses three categories of crop parameters. Conservative parameters (
Table 3) do not necessarily change with time, location, cultivar or management practices [
19,
20], and they are provided with the model. Management and cultivar specific parameters are also required by the model; these vary with cultivar and management. Some of these parameters are presented in
Table 1.
2.3. Description of AquaCrop Model
The model was proposed by FAO in 2009, with the details outlined in Steduto et al. [
18], and Raes et al. [
15]. The model relates its components (the soil, the crop, and the atmosphere) through soil and water balance, the atmosphere (precipitation, temperature, evapotranspiration, and carbon dioxide concentration), and crop conditions (phenology, crop cover, root depth, biomass production, and harvestable yield) and field management (irrigation, fertility and field agronomic practices) components [
17,
18]. It computes the daily water balance and separates the evapotranspiration into evaporation and transpiration. Transpiration is associated with canopy cover which is proportional to the scope of soil cover, whereas evaporation is related to the area of soil not covered. As the crop matures, it responds to water changes through four stress coefficients (leaf expansion, stomata closure, canopy senescence, and change in harvest index). The AquaCrop uses a normalized crop water productivity (WP*) to calculate the daily aboveground biomass production [
18,
19], which is considered constant for a given climate and crop. The yield is obtained by multiplying the biomass with the harvest index (HI). The HI for maize was set between 48% and 52%.
2.4. Field Data Collection
The canopy cover values during the first season were obtained through aerial images of three randomly selected representative plants from each of the four experimental plots. The images were analyzed in Imagej software [
29] to obtain the canopy cover.
Final biomass and grain yield was obtained following maturity using samples obtained from the experimental plot. The final aboveground biomass samples were collected by cutting the crop at the ground level; they were then oven dried for two days, and finally, weighed on a digital scale. The final yield samples were also harvested, dried to 12.5% moisture content in a drier, and weighed on the digital scale.
2.5. Model Evaluation Criterion
The model was calibrated using measured canopy cover accumulation values for the September to December 2014 growing season. The model was run after entering all the input data sets, and an iterative process was conducted by adjusting the model parameters until the best match between the simulated and measured data was obtained. The validation was done using the measured final biomass and grain yield against simulated values for all the three seasons.
The performance of the model was evaluated using statistical parameters which include Root Mean Square Error (RMSE) and Nash-Sutcliffe (E). E assesses the predictive power of the model [
30] (Equation (1)), while RMSE (Equation (2)) indicates the error in model prediction; these statistical indices were used to compare the measured and simulated values.
where
and
are the simulated and observed data,
is the mean value of
, and N is the number of observations.
RMSE measures the overall deviation between observed and simulated values, thereby estimating the model uncertainty. It takes the same units of the variable being simulated, and therefore, the closer the value is to zero, the better the model simulation performance. However, E compared to RMSE evaluates the model performance over the entire simulation period. RMSE does not account for the large deviations occurring in some parts of the season, or the small deviations in other part of the season; E accounts for the different deviations, as they depart from () along the season and expresses an efficiency of the model performance, that is, the smaller the departure from (), the higher the performing efficiency of the model. E is unitless and may assume values ranging from −∞ to +1, with better model simulation efficiency when values are closer to +1.