Distinguishing irrigated from non-irrigated locations constitutes a Level IV refinement of Level III crop type land use/land cover (LULC) mapping. While seasonal maximum NDVI from general cropland has been found to have some utility for identifying irrigated locations, the diverse agricultural landscape and pronounced climatic gradients across Kansas undermine the general effectiveness of approaches based solely on this information. Consequently, we developed an alternative procedure for creating a binary irrigated lands map for Kansas that is generally applicable to the 2003-2012 time period and which utilizes multiple datasets and methods facilitative of this task. Five crop types (alfalfa, corn, sorghum, soybeans, and winter wheat) dominate the Kansas agricultural landscape, and these crop types were considered for this exercise. The mapping effort is summarized in the following steps.

(1) “Place of Use” (POU) spatial information provided by the Division of Water Resources – Kansas Department of Agriculture, which indicates where high-volume irrigation is permissible in Kansas, was used to restrict the mapping exercise to locations contained therein. All irrigated locations in the final dataset fall within POU boundaries.

(2) Based on agricultural management tendencies, climate gradients, and irrigation use extensiveness, we split the state into two regions, west-central (WC) and east (E), following Agricultural Statistics District (ASD) boundaries. Region-specific boosted decision tree models developed using MODIS NDVI time series were used to provide initial annual estimates for irrigated locations. Models were trained using ground reference data consisting of USDA Farm Service Agency (FSA) annual cropping records from 2003-2007 that were spatially linkable to FSA Common Land Unit (CLU) polygons (c.2007). Models for 2003-2005 were constructed to simultaneously map both crop type and irrigation status, whereas models for 2006-2012 were constructed to map crop-specific irrigation status. For the latter period, USDA National Agricultural Statistics Service (NASS) Cropland Data Layer (CDL) spatial information was used to map individual crop types prior to irrigation status model development. Classified raster data were generalized to field boundaries. Field boundaries were created by combining multi-year CLU data and manual delineations using heads-up digitizing and NAIP imagery. We refer to the final field boundary dataset as the “sub-CLU” data.

(3) Initializing the irrigated locations map using model output from 2007, LULC trajectories were examined for 2007 non-irrigated cropland locations within the POU. A rule set was developed based on mapped irrigation frequency during 2003-2012 that was used to “flip” irrigation status from non-irrigated to irrigated for fields in the 2007 layer that met particular criteria.

(4) Statewide irrigated area, by crop, was computed for the 2007 map and compared to USDA NASS irrigated cropland area estimates from 2007. Mapped irrigated area for alfalfa and corn were found to exceed the USDA numbers (by 14% and 3%, respectively), so no further irrigation status changes were applied to locations mapped to those crop types in 2007. It was determined that 2007 total mapped soybean area within the POU fell 15% short of the 2007 USDA estimate for irrigated soybean area in Kansas, so all soybeans within the POU were assigned a status of irrigated, and no further irrigation status changes were applied to fields mapped to this crop in 2007. Mapped irrigated area for 2007 sorghum and winter wheat were found to fall short of 2007 USDA irrigated area estimates (by 44% and 26%, respectively), and non-irrigated areas for these crops were sufficiently prevalent within the POU to consider further irrigation status changes.

(5) Non-irrigated 2007 sorghum and winter wheat parcels were ranked according to 10 size classes reflective of common field sizes for center pivot irrigation. The same fields were then ranked according to 2007 maximum MODIS NDVI, and then again by mapped irrigation frequency during 2003-2012. A crop-specific weighted sum of these three rankings was calculated to provide each parcel with a score reasoned to reflect its likelihood of being irrigated. For each crop, county-level NASS irrigated area estimates from 2007 were compared with respective county specific, mapped irrigated area totals from 2007. For each county whereby the mapped irrigated area exceeded the NASS irrigated area, no action was taken. For each county whereby the mapped total area within the POU fell short of the NASS irrigated area, all relevant parcels within the POU were assigned a status of irrigated. For the final case, whereby mapped irrigated area fell short of NASS irrigated area but total mapped area within the POU exceeded NASS irrigated area, one by one, the highest ranked parcels were reassigned to irrigated status until NASS totals were met or exceeded. The irrigated location map obtained at the end of this step is the final map (irrigated = 1, non-irrigated = 0).

Distinguishing irrigated from non-irrigated locations constitutes a Level IV refinement of Level III crop type land use/land cover (LULC) mapping. While seasonal maximum NDVI from general cropland has been found to have some utility for identifying irrigated locations, the diverse agricultural landscape and pronounced climatic gradients across Kansas undermine the general effectiveness of approaches based solely on this information. Consequently, we developed an alternative procedure for creating a binary irrigated lands map for Kansas that is generally applicable to the 2003-2012 time period and which utilizes multiple datasets and methods facilitative of this task. Five crop types (alfalfa, corn, sorghum, soybeans, and winter wheat) dominate the Kansas agricultural landscape, and these crop types were considered for this exercise. The mapping effort is summarized in the following steps.

(1) “Place of Use” (POU) spatial information provided by the Division of Water Resources – Kansas Department of Agriculture, which indicates where high-volume irrigation is permissible in Kansas, was used to restrict the mapping exercise to locations contained therein. All irrigated locations in the final dataset fall within POU boundaries.

(2) Based on agricultural management tendencies, climate gradients, and irrigation use extensiveness, we split the state into two regions, west-central (WC) and east (E), following Agricultural Statistics District (ASD) boundaries. Region-specific boosted decision tree models developed using MODIS NDVI time series were used to provide initial annual estimates for irrigated locations. Models were trained using ground reference data consisting of USDA Farm Service Agency (FSA) annual cropping records from 2003-2007 that were spatially linkable to FSA Common Land Unit (CLU) polygons (c.2007). Models for 2003-2005 were constructed to simultaneously map both crop type and irrigation status, whereas models for 2006-2012 were constructed to map crop-specific irrigation status. For the latter period, USDA National Agricultural Statistics Service (NASS) Cropland Data Layer (CDL) spatial information was used to map individual crop types prior to irrigation status model development. Classified raster data were generalized to field boundaries. Field boundaries were created by combining multi-year CLU data and manual delineations using heads-up digitizing and NAIP imagery. We refer to the final field boundary dataset as the “sub-CLU” data.

(3) Initializing the irrigated locations map using model output from 2007, LULC trajectories were examined for 2007 non-irrigated cropland locations within the POU. A rule set was developed based on mapped irrigation frequency during 2003-2012 that was used to “flip” irrigation status from non-irrigated to irrigated for fields in the 2007 layer that met particular criteria.

(4) Statewide irrigated area, by crop, was computed for the 2007 map and compared to USDA NASS irrigated cropland area estimates from 2007. Mapped irrigated area for alfalfa and corn were found to exceed the USDA numbers (by 14% and 3%, respectively), so no further irrigation status changes were applied to locations mapped to those crop types in 2007. It was determined that 2007 total mapped soybean area within the POU fell 15% short of the 2007 USDA estimate for irrigated soybean area in Kansas, so all soybeans within the POU were assigned a status of irrigated, and no further irrigation status changes were applied to fields mapped to this crop in 2007. Mapped irrigated area for 2007 sorghum and winter wheat were found to fall short of 2007 USDA irrigated area estimates (by 44% and 26%, respectively), and non-irrigated areas for these crops were sufficiently prevalent within the POU to consider further irrigation status changes.

(5) Non-irrigated 2007 sorghum and winter wheat parcels were ranked according to 10 size classes reflective of common field sizes for center pivot irrigation. The same fields were then ranked according to 2007 maximum MODIS NDVI, and then again by mapped irrigation frequency during 2003-2012. A crop-specific weighted sum of these three rankings was calculated to provide each parcel with a score reasoned to reflect its likelihood of being irrigated. For each crop, county-level NASS irrigated area estimates from 2007 were compared with respective county specific, mapped irrigated area totals from 2007. For each county whereby the mapped irrigated area exceeded the NASS irrigated area, no action was taken. For each county whereby the mapped total area within the POU fell short of the NASS irrigated area, all relevant parcels within the POU were assigned a status of irrigated. For the final case, whereby mapped irrigated area fell short of NASS irrigated area but total mapped area within the POU exceeded NASS irrigated area, one by one, the highest ranked parcels were reassigned to irrigated status until NASS totals were met or exceeded. The irrigated location map obtained at the end of this step is the final map (irrigated = 1, non-irrigated = 0).