LA CEDAW COMO HITO EN LA CIUDADANÍA DE LAS MUJERES ARGENTINAS

3.4.2.1 Crop yield and irrigated areas

Time series of groundnut crop yields and irrigated area for the period 1966–1993 were obtained from a previous GLAM study (Challinor et al., 2004). Using the district-level harvested area and total production, yearly crop yields were calculated. These crop yields were first linearly de-trended to remove any monotonic trend (due to improvement of technologies, higher fertilizer use, and new varieties) to the technology levels of 1966, and then scaled onto the IMD grid by assuming that the crop is evenly distributed within each district. Irrigated area data at the district level were also scaled onto the IMD grid. Whenever a grid cell was composed by fractions of various districts, the detrended yield or irrigated fraction of the grid cell was calculated as the weighted-area average of all districts. Districts and years with missing data were not used in calculating grid cell values for those specific years, but they were used for those years for which data were available. The resulting spatially observed gridded yield data are shown in Figure 3.4.

(a) ¯Y (b) σY

Figure 3.4: _{Observed mean ( ¯Y ) and standard deviation (σY) of groundnut yields. Grid} cells marked with an “X” indicate locations where area harvested is less than 0.2 % of

the total grid cell area.

The highest mean yield areas are located in northern Gujarat, and along the east coast of India (states of Andhra Pradesh, Tamil Nadu and Orissa). The largest yield variability was found in northern Gujarat as well as in central India. Marginal yields (i.e. green areas of Figure 3.4) were generally observed in areas where irrigation rates were also very low

(Figure 3.5) (e.g. Rajasthan, Uttar Pradesh, and southern Gujarat). In these areas, there is generally a low use of inputs, thus leading to large yield gaps (Bhatia et al.,2009). A more complete description of these data has been done byChallinor et al. (2003,2004).

Figure 3.5: _{Observed mean irrigated fraction scaled to the IMD grid. Grid cells with} a black “X” indicate places where mean irrigated fraction equals zero, white “X” are in places where irrigated area is between 0 and 1 %, and white dots indicate places where

irrigated area is between 1 and 3 %.

3.4.2.2 Crop presence and absence data

Two different presence-absence datasets were used. The first one was used for calibrating EcoCrop, while the second was used for evaluating the model.

For the first dataset, occurrences (i.e. presence observations) of groundnut were gathered from the Global Biodiversity Information Facility (GBIF, http://data.gbif.org), and the study of (Bhatia et al.,2006). The data consisted of geographic coordinates of 1,716 locations of groundnut (Arachis hypogaea L.) representing areas where the crop is grown within India. The data were carefully verified for the consistency of its geographic coor- dinates (latitude, longitude) and corrected or removed as needed. Only unique locations in a 30 arc-second spatial resolution grid were used for all further steps (1,464 locations, “EcoCrop calibration dataset” hereafter, Figure 3.6). Crop locations were used since the alternative approach of using crop distribution gridded data (Monfreda et al., 2008; You

Chapter 3. Data and models 75

Figure 3.6: _{Crop locations used for EcoCrop’s calibration overlaid with total annual} rainfall. Total annual rainfall as in the CL WCL-QA dataset.

et al., 2009) can lead to inaccuracies due to the known spatial scale differences in those datasets (Licker et al., 2010). The crop presence points dataset was used to calibrate EcoCrop.

For the second dataset, the district-level harvested area described above (Sect. 3.4.2.1) were used to produce a presence-absence evaluation dataset, whereby any districts where there was no harvested area reported between 1966–1992 were considered absences. This dataset was gridded at two resolutions: 2.5 arc-minutes (roughly 5 km), and 1x1 degree (roughly 100x100 km) (Figure3.7) in order to evaluate the baseline suitability predictions at those two resolutions (see EcoCrop’s evaluation procedure in Sect. 7.3.2).

3.4.2.3 Crop calendar data

Two crop calendar datasets were used. The first dataset was used in GLAM simulations. This source consisted in the planting windows from the global study ofSacks et al.(2010).

Sacks et al. (2010) assembled a global dataset of planting and harvest dates for 19 major crops using six different sources: FAOs Global Information and Early Warning System (GIEWS) (FAO, 2007), USDA (2006), USDA-FAS (2008), USDA-NASS (1997), USDA- FAS (2003), and IMD-AGRIMET (2008). The dataset of Sacks et al. (2010) is the first

(a) 2.5 arc-minute (b) 1x1 degree

Figure 3.7: _{Gridded district-level crop presence evaluation data at 2.5 arc-minute and} 1x1 degree spatial resolutions. Absence areas are those where no harvested area was

reported between 1966 and 1993.

global dataset with georeferenced crop planting and harvesting information. The filled dataset for groundnut was used because (1) it is trustable for India, and (2) it provides spatially continuous values of planting windows for the region of analysis. The data were aggregated onto the TS IDM-GM precipitation grid using area-weighted averages and carefully checked for inconsistencies. Sacks et al. (2010) crop calendar data are hereafter referred to as GLAM-SPD dataset

The second source was the growing season data ofEricksen et al.(2011), who determined the start and length of the growing season across the global tropics and sub-tropics using a simple water balance module (fully described byJones 1987) and spatially-explicit time series of weather generated using the MarkSim weather generator (see Jones and Thorn- ton 2000) at a resolution of 5 arc-minutes. In this dataset (further referred to as E-LGP dataset), the growing season starts after 5 consecutive growing days and ends after 12 consecutive non-growing days, with a growing day defined as that with ratio of actual to potential evapotranspiration (Ea/_Ep) greater than 0.35 and minimum daily temperature greater than 6 ◦_{C (Ericksen et al., 2011; Jones et al., 2009). These estimations of grow-}

ing season duration were used instead of the growing season data of Sacks et al. (2010) (described above) since these were available at a sufficiently high resolution for EcoCrop

Chapter 3. Data and models 77

calibration and captured the spatial variation in the rainfall-driven growing season start and duration (Mehrotra,2011). The E-LGP dataset was used directly to drive the EcoCrop model (see Chapter7). In turn, these estimates were not used for GLAM simulations since it was preferred that GLAM planted the crop according to its own water balance model using an automatic planting routine (see Chapter 5, Sect. 5.3.2).