Estimation of rainfall distribution and its relation to rice production in Laos

Yukiyo YAMAMOTO

Jun FURUYA

Kenji SUZUKI

Shiro OCHI

Rainfall in Laos is brought by the southwest monsoon from May to October; annual rainfall ranges from 1300–1800 mm. Rice is the primary crop in Laos and approximately 80% of rice production depends on rainfed agriculture during the rainy season; however, drought and flood owing to erratic rainfall can cause unreliable production. To clarify the regional rainfall characteristics, monthly rainfall distribution for 1991 to 2002 was estimated by Kriging interpolation using a semi-variogram representing spatial correlation among 72 rainfall stations. The resolution of estimation was 1 km. Estimation error to the observed dataset was within 30 mm, equivalent to less than 5% of each monthly rainfall figure. In addition, the correlation coefficient of estimation for 100 samples to which Kriging interpolation was not applied was 0.7902.

The harvested area and yield of lowland rice for 1991 to 2002 were estimated by multi regression analysis with rainfall variables. The adjusted R-square values were 0.9470 and 0.5632 respectively. In the regression to estimate harvested area, rainfall variables in May, June, and July showed positive coefficients, and August and September showed negative coefficients, implying that rainfall in the late rainy season reduces harvest area. The equation was finally applied to map calculations in GIS to illustrate the expected harvestable area per person by rainfall pattern in each province.

1. Introduction
Laos occupies the largest area of 6 countries located in the Mekong River basin. More than 50% of GDP depends on agriculture and forestry. By the census data in 1998/1999, 668,000 households, corresponding to 83.7% in all of households in Laos, are agricultural holdings; however, most are primarily subsistence farms that produce for their own consumption (Agricultural Census Office 2000). Rice is the staple diet and primary crop in Laos. Various farming types such as rainfed paddies, lowland irrigation and slash-and-burn uplands have been extensively exploited for rice production. The areas of rainfed, irrigated, and upland rice fields in 2004 were 575,520 ha, 76,840 ha, and 117,960 ha respectively. More than 80% of rice is produced by rainfed agriculture (Dept. of Planning, 2004).

Rainfall in Laos is brought by the southwest monsoon from May to October and annual rainfall ranges from 1300–1800 mm; however, fluctuation by year and region is observed. Droughts and floods are significant factors in unreliable unstable production. The aim of this study was to reveal the relation between rainfall and rice production, both spatially and quantitatively (Yamamoto et.al. xxxx).

2. Data and methodology

2.1 Estimation of rainfall distribution

Rainfall distribution can be estimated by spatial interpolation of rainfall data at points of observation. Meteorological monitoring in Laos has been conducted by the Department of Meteorology and Hydrology at the Ministry of Agriculture and Forestry; they publish an annual data book of 104 rainfall gauges (Dept. of Meteorology and Hydrology). We selected 72 points for the study which had a time series dataset after 1991 and latitude/longitude information (Figure 1). Monthly rainfall data in 72 stations in the rainy season from May to October for 1991-2002, i.e., a 72-month dataset consisting of 6 months in 12 years, was converted into point data format with an attribute table in GIS.

Data was interpolated using Kriging interpolation, taking into account spatial correlation quantified by a ‘semi-variogram’. The experimental semi-variogram ?(h) is defined by the following equation (1):

Υ(h) : Semi-variogram
N(h): Number of pairs of observed points separated by distance h
Zi, Zj : Observed rainfall value at points i, j

The semi-variogram chart with Υ(h) versus Distance h shows that Υ(h) will increase with h over a certain distance, called ‘range’. Υ(h) will level out beyond the range and the spatial correlation will be lost over the range. The value of ?(h) when flat is called the ‘sill’ and the value of Υ(h) at h = 0 is the ‘nugget’. The nugget effectively prevents erratic swinging over very short distances. Accordingly, the semi-variogram expressing spatial correlation is represented by determination of range, sill, nugget, and function fitting the 2-d scattering plot with Υ(h) and h (Robertson 1987, Jayawardene et al. 2005).

Those semi-variogram parameters, determined for each 72 months respectively, were applied as a weighting factor in Kriging interpolation to generate a rainfall distribution map. Using the above process, 72 maps with a grid resolution of 1 km were produced.

2.2 Estimation of rice production by rainfall factor
Multi regression analysis was applied to reveal the relation between rainfall pattern and rice production. The harvested area and yield of lowland rice in the provinces were assigned as response variables. The monthly mean rainfall summarized by provinces and dummy variables corresponding provinces were assigned as predictor variables. In addition, the ratio of increased rice production over the previous year was added to the predictor variables to estimate yield, taking time trends into account. Non-significant variables were eliminated by t-value and p-value, and the two regressions to estimate harvested area and yield were finally obtained. The regression that performed accurate estimation was applied to map calculation in GIS to generate a spatial estimation map.

3. Results and discussion

Table 1. Semi-variogram parameters for Kriging interpolation.

3.1 Rainfall patterns in Laos
By means of Kriging interpolation with the semi-variogram for each month (Table 1), monthly rainfall in 1-km grids over 72 months was estimated (Figure 2). The estimation error in observed rainfall value at the 72 points ranged from 1–30 mm, equivalent to less than 5% of the monthly average (Table 2). In addition, to verify the adaptability with respect to points which were not applicable to semi-variogram modeling, the observed value of 100 data items recorded in 2001 and 2002 by a data logger in 10 locations (Shown in Figure 1) was compared to the estimated value at these points. It tended to underestimate in low-rainfall parts and to overestimate in heavy rainfall parts; however, the correlation of all pairs was 0.7902 (Figure 3). These verification results suggest that rainfall was reasonably well estimated.

Estimated rainfall was summarized by province. Annual mean rainfall by province was 1200 – 2300 mm; however, Huaphanh, Attapeu, and Champasak showed large yearly fluctuations (Table 3).

Table 2. Average of estimation error in ovserved rainfall at the 72 points used for modeling (mm) .

Figure 3. Verification of rainfall estimation respect to the points which were not applied for semi-variogram modeling.
( Data set : Recorded by data logger in 2001 & 2002).

3.2 Estimation of rice production by rainfall factor
Two kinds of multi regressions to estimate harvested area and yield of lowland rice were obtained (Table 4). Contribution of dummy variables was significant in both regressions, and monthly rainfall variables from May to October were also selected to estimate harvested area. Monthly rainfall in July and time trend variables were selected to estimate yield. The squared multiple correlation coefficient adjusted for degrees of freedom (Adjusted R-square) was 0.9470 for the harvested area and 0.5632 for yield. In the regression to estimate the harvested area, rainfall variables for May, June, and July showed a positive coefficient but rainfall in August and September showed a negative coefficient. This implies that rainfall in the late rainy season risks cause damage to rice production by decreasing the harvested area.

The regression for estimation of harvested area, which gave a better correlation, was applied to map calculation by GIS using the rainfall distribution maps produced in 3.1, to illustrate harvestable area per person (Figure 4). The expected harvestable area in the northern region was consistently less than 0.1 ha per person

Figure 2. Estimated rainfall maps by Kriging interpolation with semi-variogram.

Table 3. Mean rainfall in province estimated by Kriging with semi-variogram (mm).

through 12 years, but the central and southern regions varied according to the rainfall conditions each year. In Champasak Province, for instance, the expected harvestable area was comparatively large until 1997, but decreased in 1998, slightly increased in 1999 and 2000, then decreased in 2001 and 2002 again. In Saravane Province, a large harvestable area was expected through 12 years, but the western portion was smaller than the eastern portion in 1996 and 2000. Thus, spatial association between rainfall and rice production was shown by the grid image.

4. Conclusion
This paper investigated the relation between rainfall and rice production in Laos using spatial analysis. Domestic rice production in Laos increased from 660,000 ton in 1976 to 2,530,000 ton in 2004 (Dep. of Planning, 2004). Although irrigation developed after the 1990s, rice production by irrigation accounts for only 16% of total production. On the other hand, rice production by rainfed agriculture is steady at 77% of overall production. In addition, upland rice is a primary crop in mountainous regions known to be high-poverty areas. Accordingly, rainfed paddies and upland fields are still crucial to both self-sufficiency and livelihood in Laos. In regions where socio-economic infrastructure and distribution networks have not been sufficiently developed, it is important not only to increase total countrywide rice production but also to stabilize regional production in each region. Rainfall cannot be controlled; however, it should be possible to devise countermeasures to fluctuations by employing farming methods suited to regional rainfall conditions. A better understanding of annual / regional fluctuations and the spatial correlation between rainfall and rice production is likely to provide useful information.

Figure 4. Potential harvestable area expected by rainfall condition in each year (per person).

References

• Agricultural Censes Office (2000) Lao Agricultural Census, 1998/99 Highlights. Vientiane
• Dept. of Meteorology and Hydrology,Dairy Recorded Rainfall of Lao P.D.R. 1991-1996; 1999; 2002; 2003..
• Dept. of Planning in Ministry of Agriculture and Forestry,2004,Agricultural Statistics Year Book 2004.
• Jayawardene,H.K.W.I., Sonnadara, D.U., Jayewardene, D.R., 2005, Spatial interpolation of weekly rainfall depth in the dry zone of Sri Lanka. Climate Research 29, pp.223-231.
• Robertson, G.P., 1987, Geostatistics in ecology: Interpolating with know varianve. Ecology 68(3), pp.744-748.
• Yamamoto, Y., Furuya J., Suzuki K., Ochi S., xxxx, Characteristics of Rainfall and Rice Production in Laos Described by Spatial Analysis, Jl. Of the Japanese Agricultural Systems Society [In Japanese] (contributing).