Spatial Delineation of Rice Yield and Production by the Combination of Rice Crop Modelling and Remote Sensing: A Case Study in Soc Trang Province, Vietnam

A bstract — Vietnam's economy mainly depends on the growth of agricultural production. The forecasting yields to consolidate information to assure food security, monitoring, and mapping rice accurately are essential. The study is aimed to (i) delineate the rice distribution status; (ii) simulate the rice yield by using the AQUACROP model; (iii) integrate maps and rice yield to delineate the rice production extent in Soc Trang province. The method involves combining MODIS remote sensing data (MOD13Q1, MOD13A1 at 250m resolution, eight-day intervals) with the AQUACROP rice yield prediction results at 9 study sites within the area. The study delineated three primary rice cropping seasons (Winter-Spring, Spring-Autumn, Autumn-Summer), separated into eight specific rice cropping seasons, depending on the date or rice sowing/transplanting and harvesting. Besides, the difference in rice sowing time of the three main rice cropping seasons was identified and delineated. The simulated rice yield always higher observed rice yields at all seasons, from -8.65 to 22.20% in the Winter-Spring season, 5 to 28.6% in Spring-Summer season, and -1 to 49% in Summer-Autumn. The results were validated by comparing with the government statistical results, which was a very close correlation. The results suggested that we can use the MODIS satellite image for delineating rice cropping status. The rice yield and production can be simulated and delineated by the combination with the rice crop model.


I. INTRODUCTION
Among crops, rice is the most important for approximately 50% population of the World [1], [2]. However, the world's rice production growth rate has slowed down during recent years [3]. Vietnam is one of the largest rice exporters, in which, Mekong Delta River (MRD) and Red Delta River (RDR) occupy 95% of the annual rice exports [4]. Therefore, the rice output forecast is useful in ensuring domestic demand and serving export demand. Especially in the current situation, the rice area has been decreasing. Soc Trang is one of the provinces in the MRD that is affected by climate change [5]. The ability to estimate rice production before the harvesting date is significant for national and provincial planning and decision-making regarding food security and economic planning [6]. Today, the use of remote sensing techniques combined with Geographic Information System (GIS) and statistical methods, spatial interpolation, has been applied to study the spatial distribution characteristic in many countries around the World [7].
The AquaCrop model was developed in 2009, and it has been used worldwide in different agro-ecological conditions. AquaCrop is a crop water productivity model developed by FAO to improve water productivity in rainfed and irrigated fields. It simulates yield response to water of herbaceous crops and is particularly suited to address conditions where water is a key limiting factor in crop production [8]. The AquaCrop model is a good tool to simulate various rice genotypes' behavior, especially improved ones, for management practices established upon seedling age variability [9]. Therefore, the study mainly uses the MODIS satellite images and GIS to delineate the time of sowing/transplanting and then combined with the AquaCrop model to predict and delineate the rice yield in Soc Trang province of Vietnam.
B. Data Preparation 1) Climate, soil, and crop data: AquaCrop requires daily, 10-daily, or monthly temperature (Tmax), minimal temperature (Tmin), rainfall (P), reference evapotranspiration (ETo), and CO2. ETo is estimated from the parameter of Tmin, Tmax, relative humidity, sunshine hour, and wind speed by the ETo calculator program [11] based on the FAO Penman-Monteith method. Physical soil characteristics, such as Texture, Soil moisture content (Permanent Wilting Point -PWP, Field Capacity -FC and saturation -Sat, and saturated hydraulic conductivity (Ksat) obtained to run the model. The crop data of rice cultivar collected at the Cuu Long Rice Research Institute in Can Tho city.
2) Remotely sensed data: The images collected from NASA, code MOD09Q1, the spatial resolution of 250m, and temporal resolution with the eight days from 2012 to 2013. The calculation of NDVI images for rice showing stages and cropping calendar determination. The Red and Near-Infrared bands used for calculation NDVI 3) Field observation data: Field observation to obtain rice sowing, crop calendar, and rice yield was incorporated to simulate the rice yield model. There are 50 field observation sites to monitor rice sowing, crop calendar, and nine field measurement sites to collect actual yield for accuracy assessment.

C. Running AquaCrop Model for Rice Yield Simulate
AquaCrop-FAO [11] crop-water productivity model to simulate yield response to water. AquaCrop used several parameters and input data to simulate most of the major fields' yield response to water. Before running a simulation, the user specifies the Main menu, the sowing date, and the appropriate environmental, off-season conditions. The collected data from 9 sites used to feed the AquaCrop model to simulate rice yields.

1) Creating the Normalized Different Vegetation Index (NDVI):
The agronomic parameter Normalized Difference Vegetation Index (NDVI) was proposed by [2], which is a numeric indicator of the difference between the wavelengths of the red and near-infrared bands divided by their sum. The NDVI shows an excellent linear correlation with biomass compared to other vegetation indices and expected to be less influenced by changes in atmospheric conditions. Moreover, it is a good indicator of vegetation biomass when the terrain exhibits good vegetation cover [12]. The intermediate evaluation differentiated the characteristics of the materials such as biomass, leaf area index, and photosynthetic capacity seasonally. Those characteristics are relevant and highly dependent on the type of plant cover and weather, physiological characteristics, biochemical, and pests. NDVI averaged in time series data as the essential tool to monitor vegetation status changes, helping understand the impact of weather and climate on the biosphere. NDVI calculated as follows: 2) The relationship between NDVI values and rice growing stages: Each crop with an NDVI value range fluctuates within a certain limit (because on each soil with different characteristics, on fertile lands, the crop that grows well, the NDVI value will reach high and vice versa), in general, the laws of their fluctuations are the same. Based on the results of research by BR Parida et al. [13] in The Gujarat of India proposed NDVI oscillatory intervals for some of the following subjects (table 2): Because the NDVI value of crops changing in a certain situation (each soil type has different properties). The NDVI value changes in the same direction. Based on the research results of [13] in India's Gujarat, the proposed NDVI values of different land use, as shown in Table 2: From the results of processing, combining the actual survey, the NDVI index over time of the whole Mekong Delta, compared with the current rice crop's current status, finds the relationship between the NDVI and the current status of rice teas. Based on the theory that the NDVI number of the rice region in the Mekong Delta is low value at the beginning of the crop, gradually increases and reaches the highest at the time of the rice crop at the post-brunch stage, and decreases when the rice plant begins to ripen and the lowest at the end of the crop (Figure 1) 3) Monitoring the progress of rice sowing and delineating the rice cropping calendar: The time series NDVI values of the MRD calculated and delineated, which combined with rice-growing states to find out the relationship between NDVI indexes for rice growing stages [14]. On the basis, low NDVI value at the early stage increased gradually and peaked at the tillering stage and decreased when rice began mature and harvested [15] (Fig 2).  Table 3. There were four soil groups: group 1 had the highest clay content, including D1, D5, D6, D9 sites; group 4 had the lowest clay content, including D2, D3, D4, D8 sites; group 2 (D7 sites); group 3 (D9 sites).
2) Climatic Conditions: Sets of climatic data were collected from meteorological institutes in Soc Trang province from 2012 to 2013.

B. Rice Yield Simulation
Before running a simulation, all input data must be created, including climate, crop, management, and soil, presented above and in Table 4.
The result in Table 5 showing that the simulated and observed rice yields at different sites, but rice yield at all seasons always higher than the observed rice yield from -8.65 to 22.20% in the Winter-Spring season, 5 to 28.6% in Spring-Summer season, and -1 to 49% in Summer-Autumn. Since the differences changing site to site due to the differences in soil, pest, cultivation techniques.   These results can assist the government in developing the major rice cropping calendars homogeneously for all of the areas to avoid the effect of pests and diseases. Otherwise, the field management for using water, fertilizers, herbicide, and pesticides is easy and effective [16], [17]. The distribution of rice cropping season and date of starting/ending of rice sowing/transplanting of each rice crop are showed in Fig 5, 6 , 7, 8. The time of the rice sowing date was changing at a different district. Depending on the flooding season, the rice field receives the water from the rivers to start the new season and leach out the toxicity from saline intrusion and acidic soils.
The situation was the same explanation for those fields within the district.
By estimating the starting/ending date of rice sowing for each rice cropping season, the number of rice cropping seasons in different districts identified. It used to predict the district's total rice cultivation area, combining with the expected yield to calculate the rice production.      Table 9. The difference comparison between statistics and interpretation data shows a close correlation between the two results data. The predicted area always higher than the statistic data. However, during the Winter-Spring season, the difference was higher than the Autumn-Winter season; this means the expected yield was higher than the statistical result. Due to this season, several rice fields were harvested, but the rice leaflets of some alive rice remained in the field. Otherwise, at the end of the rainy and the start of dry seasons with saline water intrusion, several crops such as corn, grass, sugarcane, etc., impaired the satellite image interpretation results.

D. Delineation of Rice Yield and Production
The map of time of rice sowing combined with the result of rice yield simulated to create a rice yield map. Fig 9, 10,  and 11 showing that the different colors mean different rice yields. The concentration of higher yield on the alluvial soil, which receives enough water for irrigation. While the lower rice yield was mostly on acid sulfate soils and saline water affected from the sea. The white color was non-rice, such as urban, fruit crops, mangrove, sugarcane, pineapple, etc. The rice production was calculated in ton/ha by multiplying the rice yield in ton/ha with rice cropping season in ha. The results show in Table 10. The difference between the estimated rice yield and observed is inevitable. Many causes lead to this discrepancy, such as acid sulfate soils with high toxicity, low pH, and saline or soil fertility levels, which limited the rice-growing and then rice yield. The simulated rice yield, affected by pests and diseases, was not included in the model. Table 11 showing the difference between estimated or predicted from the model with statistical rice production from government varies by season and area.

IV. CONCLUSION
Monitoring the date of starting rice sowing based on the temporal variation of MODIS -NDVI satellite images are very effective and useful to determine the current rice evolution states, rice sowing stages, rice cropping calendar. This assists the government in understanding the near realtime and its distribution on the field to develop the strategies for rice sowing/transplanting homogeneously for better rice management and rice yield. However, to increase the accuracy of predicted data at the district level, it should combine high spatial resolution satellite images to simulate the rice yields. It has been successful when combining MODIS satellite images as rice sowing stages maps to delineate and predict the rice yield and production.
As a result, simulated yields are usually higher than actual rice yield, and accuracy varies by season and area. The model does not consider the limitation of fertility, weeds, and pests for rice yield and production during the running process. These factors can affect the rice yield, which causes the deviation of simulated rice yield.

ACKNOWLEDGMENT
This study is funded in part by the Can Tho University Improvement Project VN14-P6 (supported by a Japanese ODA loan), and Ministry of Education support for the annual study, and VLIR (CTU-Belgium) projects, and Department of Science and Technology of the province.