Upscaling water productivity in irrigated agriculture using remote sensing and GIS technologies

W.G.M. Bastiaanssen*/**, Mobin-ud-Din Ahmad**/*** and Zubair Tahir***

* International Water Management Institute (IWMI), Colombo, Sri Lanka,

** International Institute for Aerospace Survey and Earth Sciences (ITC), Enschede, The Netherlands,

*** International Water Management Institute, Lahore, Pakistan,

Abstract

When fresh water resources are getting scarcer, such as in the irrigated Indus Basin, it is necessary to describe the depletion of the water resource in relation to the agricultural production. Despite that such approach is important, data required to monitor the productivity of the land and water resources over vast areas, are usually not available or not accessible. Satellite measurements from the NOAA weather satellite are in this study merged with ancillary in situ data into a GIS system. Remote sensing measurements are converted to crop yield, actual evapotranspiration and indirectly to net groundwater use. The ancillary data consists of canal water deliveries and rainfall records. For each of the canal commands, the productivity of water and the gross value of production is calculated. A large variability among the canal commands in the river Indus is witnessed. It is concluded that water productivity is controlled more by crop yields than by water factors. The spatial variability of water productivity per unit diverted is more than per unit depleted. This can be ascribed to wide variations in the relationship between canal water supply vs. actual evapotranspiration. This is an issue classically covered by irrigation efficiencies. Upscaling of water productivity was established by merging together the various canal command areas. The results show that the productivity of water is getting constant at a spatial scale of 6 million ha and higher. The definition of water productivity is then not longer relevant because diversion and depletion are balancing each other. It is concluded that the Indus Basin is an example of substantial groundwater recycling and that this needs to be taken into account when working on local improvements of water productivity.

  1. Introduction

When fresh water resources are getting scarcer, such as in the irrigated Indus Basin, it is necessary to describe the depletion of the water resource in relation to the agricultural production. This is known as the “Productivity of Water” (PW). Frameworks for the formulation and assessment of water productivity have been developed by Molden et al (1998) and subsequently used in various water management studies (e.g. Droogers and Kite, 1999). Different definitions of the water productivity exist, and they differ in produced physical crop yield (kg), the economic value of production (US$), diverted irrigation supply (m3) and water consumed by crop evapotranspiration (m3).

Physical productivity of water is often related to crop yield in relation to water depletion by crop evapotranspiration and canal water supply or diversion. This per definition implies that productivity is related to the classical irrigation efficiency which relates supply to consumption.

Water management techniques are focussing often on water savings, but in water scarce conditions, water is just diverted from one place or use to another. It is therefore of extreme importance to get a perspective of the efficiencies and productivities at larger scales. Traditional field surveys and field scale water balance measurements cannot give a comprehensive description of the water flows at the regional scale. The flow path of water contains also processes such as recharge, capillary rise and groundwater extractions, which are difficult to measure or estimate for sub-systems present in a river basin. These terms are mentioned in particular, as they reflect the processes of water recycling. But also information on crop acreage, yields and canal water deliveries are difficult to obtain, as practices in canal operation may deviate substantially from the expected operation.

Data availability required to monitor the productivity of the land and water resources, especially over vast irrigation schemes and river basins can hamper the application of the water productivity framework for understanding and action in scarce resources management. The aim of this paper is to demonstrate how remote sensing and GIS as a tool can help in assessing water productivity and how productivity varies with scale.

  1. Hydrological approach

The soil water balance and crop production figures form the prime basis for water productivity analysis. The soil water balance relates total supply to total consumption and has a storage term for cases when they are not balanced (see also Fig. 1).

S = (P + Icw + Itw + q) – (ETa + q)(mm)(1)

where S is the storage change, P is the precipitation, Icw is the canal water supply, Itw is the groundwaterwater supply through tubewells, q is the capillary rise, ETa is the actual evapotranspiration and q is the recharge. Since several terms of Eq. (1) are difficult to quantify, the three terms related to groundwater are taken together:

NGW = Itw + q – q(mm)(2)

where NGW is the Net GroundWater use, i.e. the extractions of groundwater minus the recharge. The recharge comprises the return flow from tubewell irrigation Itw but can also arise from precipitation P and canal water irrigation Icw. NGW represents the net withdrawal of groundwater. After insertion of Eq. (2) into Eq. (1), a simplified water balance equation emerges:

S= P + Icw + NGW – ETa (mm)(3)

For the current case study in the Indus Basin, P is taken from rain gauges, Icw from flow records and ETa from remote sensing. The two unknowns are then S and NGW. If the storage changes are further ignored, which is not true for all places in the canal command areas, NGW can be computed as the residual term of the water balance. The storage changes depend on the groundwater table fluctuation, which in some cases can be significant with declanations of 1 meter per year. At other places, however, groundwater table measurements reveal minor variations. There wer not a sufficient amount of piezometric readings available to estimate S in a systematic manner across the entire Indus Basin. Because of this limitation, we were forced to ignore S and apply the usual concept of disregarding storage changes for longer time periods in hydrological studies. We will show that later that S is a minor part of NGW.

  1. Hydrological results

Data on precipitation and canal water supply were taken from Habib et al. (1999) and from Tahir and Habib (2000). The Indus Basin comprises 44 canal command areas and some areas had incomplete data and were not further considered in this study. The data shows a variation between 50 to 800 mm canal water supply per unit culturable command area during rabi season (Fig. 2). This suggests a highly non-uniform distribution of irrigation water across the Indus Basin. Similar heterogeneities in canal water supply were found for kharif. Partly this large deviations may result from measurement and interpretation erros in the main canals because the flows through these huge irrigation canals are not straightforward to measure. Another issue is that the actual area receiving irrigation water may deviate from the area expected to receive irrigation water. The numbers used in this study are presented in appendix 1.

The actual evapotranspiration data are withdrawn from Bastiaanssen et al. (2001) and are based on remotely sensed data. Raw data from the NOAA-AVHRR (National Oceanic and Atmospheric Administration – Advanced Very High Resolution Radiometer) satellite has been used. The Surface Energy Balance Algorithm for Land (SEBAL) has been used to convert the raw data into broadband surface albedo, vegetation index and surface temperature. The major concept of SEBAL is to explore the range of surface albedo for describing net radiation, the range of vegetation index to assess soil heat flux variability and the range of surface temperature for estimating sensible heat flux. The concept of the energy balance equation is used to compute actual evapotranspiration from the energy left to latent heat flux:

LE24 = Rn24- H24(W m-2)(4)

where LE24 is the 24 hour latent heat flux, Rn24 is the 24 hour net radiation and H24 is the 24 hour sensible heat flux. The soil heat flux on a 24 hour basis is usually small and can be ignored. LE24 can be converted into actual evapotranspiration (mm d-1) from the energy required to vaporize 1 kg of water at a given temperature. Equation (4) has been used to compute the actual evapotranspiration for NOAA images acquired during 20 different days. Individual day results were temporally integrated by preserving the evaporative fraction between two successive satellite acquisition days. The evaporative fraction for a daily time basis reads as LE24/Rn24. This energy partitioning was fixed until the next available AVHRR image. Since net radiation changes considerably due to cloud cover that may arise during satellite fly-over days, day-to-day variations of net radiation have been taken into account to compute actual evapotranspiration. The Indus Basin was for this purpose divided into 5 climatic zones, and daily global radiation (shortwave radiation reaching the land surface) was computed for every climatic zone.

Fig. 3 shows the map of the annual actual evapotranspiration. Validation in the Indus Basin was realized through the application of the well calibrated field scale transient moisture flow model SWAP (Sarwar et al., 2000), in situ Bowen ratio measurements and water balance residual analyses for an area of 3 million ha. The accuracy of assessing time integrated consumptive use was found to vary from 0.3 % at field scale to 4.5 % at the regional scale of 3 million ha.

The actual evapotranspiration during the dry winter season (rabi) was on average 350 mm and the wet summer season (kharif) had a total consumption of 620 mm. The area shown in Fig. 3 covers all pixels encompassed by the gross command areas. The variation is again – like for canal water supply – very high; the values of the annual evapotranspiration ranges between 450 to 1270 mm yr-1.

Surface supply to cropped area can be from three difference sources i.e. canal irrigation Icw, groundwater irrigation Itw and net precipitation Pn.(gross precipitation P minus interception losses Pi and surface runoff). For the sustainable of irrigation systems, it is important to estimate the reliance on groundwater resources. Groundwater resource ratio can be estimated by the ratio of groundwater irrigation Itw to the total inflow from all sources:

(-)

A map of the groundwater resource ratio  is shown in Fig. 4. The results reveals that there is little contribution of groundwater during kharif and a significant amount of groundwater resource use during the dry rabi season (the rainfall in many areas varies between 25 to 50 mm). Some areas rely during rabi for 80% of their water resources on groundwater, which basically implies that canal water is reaching these areas only to a limited extent. Although not shown in Fig. 4, several canal command areas have a negative NGW, which implies that deep percolaton occurs and that the groundwater system is recharged.

  1. Crop yield

Crop yield is one of the major input factors in water productivity frameworks. Crop yield information is classically collected through field surveys. This is a laboursome activity, especially when one has to deal with vast areas. To aid the ground sampling and to swiftly obtain an overall picture of the crop development, a remote sensing model on crop yield prediction was developed (Bastiaanssen and Ali, 2001). This model is based on Monteith’s equation for biomass production which reads in its simplest form as:

Bio = APAR  (kg m-2 d-1)(6)

where Bio (kg m-2 d-1) is the biomass production, APAR (MJ m-2) is the Absorbed Photosynthetical Active Radiation and  (kg/MJ) is the light use efficiency. APAR is controlled by incoming solar radiation and light interception by the presence of leaves. The solar radiation was computed from the actual hours of sunshine, and the leaf presence from the Normalized Difference Vegetation Index (NDVI), being derived from the NOAA-AVHRR sensor. The light use efficiency depends on c3 or c4 crops, besides on the soil moisture availability to keep the leaf water potential low. Moisture stress reduces the light use efficiency, and this feedback was taken into account by incorporating evaporative fraction (LE24/Rn24) into the light use efficiency. The biomass production rates for single NOAA acquisition days were further integrated in time by keeping the fraction of APAR/PAR and  constant, and making PAR variable according to the day-to-day variation of cloud cover.

Remote sensing estimates of crop yield have been validated against secondary data collected by the Agriculture Department of Pakistan. The validation revealed a root mean square error of 525, 616, 551 and 13484 kg/ha for wheat, rice, cotton and sugarcane yield respectively. The deviation between secondary data and remote sensing data shows that wheat, rice and sugarcane yield can be mapped for approximately 80% of the cases within the 95% confidence levels of the secondary field data. On average, crop yields in Pakistan are at the lower side. The yields are 2276, 1756, 1293 and 47929 kg ha-1 for wheat, rice, cotton and sugarcane respectively. A comparison with the study of Hussain et al. (2000) who collected crop-cutting experimental data in Sindh confirmed the wheat yields to be low in Sindh. The canal commands in Fig. 5 are numbered from the upstream to the downstream end. It is evident that, except for wheat, the yields have no relationship with their position in the basin. The command area containing the highest rice yield has the most appropriate name “Rice Canal”.

  1. Water productivity

One of the first issues in water productvity is the identification of which “crop” and which ‘drop’ is referred to. The differentiation between water productivity per unit depleted and per unit diverted canal water supply has sense as the former one describes how productively water is used that leaves the basin, whereas in the latter case the return from canal management efforts and irrigation sector investments are demonstrated.

In their original paper, Molden et al. (1998) suggest to define the productivity per unit diverted irrigation supply as surface irrigation water diverted to the command area plus net removals from ground water (writers’ italics). There are a few reason why we believe that the expression per unit canal water supply may have some advantages for the Indus Basin:

  • The irrigation manager is responsible for canal water supply and prefers to understand the impact of expensive irrigation infrastructure. Groundwater management is usually hosted at institutions different from Irrigation Departments
  • Adding water flows originating from different sources (P, Icw, Itw) breaks the opportunity to study the impact of individial sources. The role of groundwater on water productivity gets hidden if it is added to the diverted water

We therefore continue with the following set of definitions:

PWETa= Ya/ ETa(kg m-3)(7)

PWIcw= Ya/ Icw(kg m-3)(8)

PW$= GVP/ Icw(US$ m-3)(9)

where Ya (kg ha-1) is the actual crop yield and GVP (US$ kg-1) is the Gross Value of Production. GVP is computed from the crop production of every crop, the market price of every crop and the acreage of every crop. The indicator PW$ is especially suitable to sum up the total production of different crops. Also, it can be used to compare with the water productivity of users other than agriculture, such as fish, ecosystems etc.

Raster data from satellites on Ya and ETa can be easily merged to make crop wise evaluations on PWETa; this is not straightforward for Ya/Icw as crop specific Icw data is seldomly available. Hence, PWETa has the advantage that it can evaluate crop specific evaluations. Table 1 contains an overview of the basin wide crop specific productivities. It shows that sugarcane and cotton are more water consuming than rice because of their longer growing period. Cotton has the lowest PWETa values and sugarcane the highest. Involvement of the world market prices of agricultural products for 1994 shows, however, that cotton is more economical productive than rice and wheat.

Having data on crop yield, actual evapotranspiration and canal water flow available in the GIS database, it became feasible to compute PWETa and PWIcw. The case of wheat is given as an example (Fig. 6). PWETa varies between 0.2 to 0.8 kg m-3 being – grosso modo – a low value. A literature search on PWETa for wheat showed an average figure of approximately 1.0 kg m-3, hence Pakistan is performing low in terms of WPETa. The PWETa line in Fig. 6 shows something important as well: the response of wheat yield to evaporation is not constant. The value for PWETa increases with higher yields (R2=0.83). Since it is apparent that PWETa increases with wheat yield, the conclusion should be drawn that wheat with a higher yield is more efficient in terms of water depletion. This is worth further exploration in future studies.