Citation: Kansas Applied Remote Sensing Program. 2014. BACC FLUD Kansas Land Cover Database (2003-2012). Available at http://kars.ku.edu/geonetwork/srv/en/main.home(accessed {DATE}; verified {DATE}. Kansas Applied Remote Sensing Program, Kansas Biological Survey, University of Kansas, Lawrence, Kansas.

Acronyms:

FSA = USDA Farm Service Agency

NASS = USDA National Agricultural Statistics Service

LULC = land use/land cover

ASD = Agricultural Statistics District (nine of these in KS; labeled 10, 20, …, 90)

W = western ASDs (10,20,30)

C = central ASDs (40,50,60)

E = eastern ASDs (70,80,90)

WC = western + central ASDs (10-60)

CE = central + eastern ASDs (40-90)

I = irrigated

NI = non-irrigated

DT = boosted decision tree model

Overview:LULC maps for 2003-2005 were produced using MODIS NDVI-based DT models. Mapped classes included alfalfa (I & NI), corn (I & NI), sorghum (I & NI), soybeans (I & NI), winter wheat (I & NI), double crop (I & NI), and fallow (NI). The ground reference dataset used for model training was developed using FSA 578 Compliance data tables from 2003-2007 and the FSA CLU spatial coverage from 2007 (more on this below). Non-cropland classes (grass/pasture, urban, water) were burned in from the 2005 Kansas Land Cover Patterns dataset previously developed at KARS for a different project. General LULC for 2006-2012 was based on USDA NASS CDL layers from these years, which do not include irrigation status. CDL crops not included in the above list were recoded to “other crop”. MODIS NDVI-based DT models subsequently were developed to map irrigation status (I or NI) for the above crops in the 2006-2012 LULC map set.

Ground reference data:All monocropped CLUs with crop types from the above list were identified in the 2003-2007 FSA datasets and formed the basis for the ground reference dataset. The vast majority of irrigated records were associated with subCLUs and thus did not filter into the ground reference dataset. To supplement the population of irrigated ground reference samples, a rule set was created to identify subCLUs that were highly likely to correspond with mono-cropped, quarter-section sized center pivots (the most common irrigated footprint). Fields identified during this process were appended to the ground reference dataset, which approximately doubled the number of irrigated ground reference data samples.

Models:A number of DT models were developed to complete the 2003-2013 LULC map set. Due to regional similarities and differences in crop phenologies, management practices, and climatic tendencies, the state was divided into two modelling regions, WC and E. For the WC region, ground reference data from the WC ASDs were used for model development. For the E region, ground reference data from the CE ASDs were used. For the E region, ground reference data from the C region was included because there were not enough samples using E region data alone for many classes.

2003-2005, WC, inside POU:A boosted decision tree was developed using WC ground reference data to map the following 13 classes: Alfalfa (I & NI), Corn (I & NI), Sorghum (I & NI), Soybeans (I & NI), Winter Wheat (I & NI), Double Crop (I & NI), and Fallow (NI).

2003-2005, WC, outside POU:A boosted decision tree model was developed using non-irrigated WC ground reference data to map the following 7 classes (all non-irrigated): Alfalfa, Corn, Sorghum, Soybeans, Winter Wheat, Double Crop, and Fallow.

2003-2005, E, inside POU:A boosted decision tree model was developed using CE ground reference data to map the following 8 classes: Alfalfa (NI), Corn (I & NI), Sorghum (NI), Soybeans (I & NI), Winter Wheat (NI), and Double Crop (NI).

2003-2005, E, outside POU:A boosted decision tree model was developed using non-irrigated CE ground reference data to map the following 6 classes (all non-irrigated): Alfalfa, Corn, Sorghum, Soybeans, Winter Wheat, and Double Crop.

2006-2012, W, inside POU:A boosted decision tree model (one model per crop) was developed using WC ground reference data to map I & NI status for each of Alfalfa, Corn, Sorghum, Soybeans, Winter Wheat, and Double Crop.

2006-2012, E, inside POU:A boosted decision tree model (one model per crop) was developed using CE ground reference data to map I & NI status for each of Corn and Soybeans.

Irrigation Mask Development:Based on the consistency of irrigation water use in Kansas during 2002-2013, as well as management and economic difficulties that preclude substantial annual spatial variations in the set of irrigated fields, it was assumed that the set of irrigated fields in Kansas was largely stable during the 10-year study period. Furthermore, past experience of the mapping team strongly suggested that the variability of mapping irrigation status on a year-by-year basis likely would result in more mapping error than creating a single, constant irrigation mask that would be applied to all 10 years of LULC maps. Consequently, a procedure was developed to utilize all of the “inside POU” map results and produce a single irrigation mask. Upon completion, the irrigation status layer for all 10 years of maps was modified to coincide with this final, static irrigation mask.

The final procedure relied on the creation and implementation of a hierarchical logical rule set that utilized LULC trajectories, USDA NASS irrigated acreage statistics, and maximum MODI NDVI values from individual fields. All available county-level USDA NASS irrigated acreage data were obtained to facilitate this process, ultimately steering crop-specific irrigated acreage totals toward the reported values.

Citation: Kansas Applied Remote Sensing Program. 2014. BACC FLUD Kansas Land Cover Database (2003-2012). Available at http://kars.ku.edu/geonetwork/srv/en/main.home(accessed {DATE}; verified {DATE}. Kansas Applied Remote Sensing Program, Kansas Biological Survey, University of Kansas, Lawrence, Kansas.

Acronyms:

FSA = USDA Farm Service Agency

NASS = USDA National Agricultural Statistics Service

LULC = land use/land cover

ASD = Agricultural Statistics District (nine of these in KS; labeled 10, 20, …, 90)

W = western ASDs (10,20,30)

C = central ASDs (40,50,60)

E = eastern ASDs (70,80,90)

WC = western + central ASDs (10-60)

CE = central + eastern ASDs (40-90)

I = irrigated

NI = non-irrigated

DT = boosted decision tree model

Overview:LULC maps for 2003-2005 were produced using MODIS NDVI-based DT models. Mapped classes included alfalfa (I & NI), corn (I & NI), sorghum (I & NI), soybeans (I & NI), winter wheat (I & NI), double crop (I & NI), and fallow (NI). The ground reference dataset used for model training was developed using FSA 578 Compliance data tables from 2003-2007 and the FSA CLU spatial coverage from 2007 (more on this below). Non-cropland classes (grass/pasture, urban, water) were burned in from the 2005 Kansas Land Cover Patterns dataset previously developed at KARS for a different project. General LULC for 2006-2012 was based on USDA NASS CDL layers from these years, which do not include irrigation status. CDL crops not included in the above list were recoded to “other crop”. MODIS NDVI-based DT models subsequently were developed to map irrigation status (I or NI) for the above crops in the 2006-2012 LULC map set.

Ground reference data:All monocropped CLUs with crop types from the above list were identified in the 2003-2007 FSA datasets and formed the basis for the ground reference dataset. The vast majority of irrigated records were associated with subCLUs and thus did not filter into the ground reference dataset. To supplement the population of irrigated ground reference samples, a rule set was created to identify subCLUs that were highly likely to correspond with mono-cropped, quarter-section sized center pivots (the most common irrigated footprint). Fields identified during this process were appended to the ground reference dataset, which approximately doubled the number of irrigated ground reference data samples.

Models:A number of DT models were developed to complete the 2003-2013 LULC map set. Due to regional similarities and differences in crop phenologies, management practices, and climatic tendencies, the state was divided into two modelling regions, WC and E. For the WC region, ground reference data from the WC ASDs were used for model development. For the E region, ground reference data from the CE ASDs were used. For the E region, ground reference data from the C region was included because there were not enough samples using E region data alone for many classes.

2003-2005, WC, inside POU:A boosted decision tree was developed using WC ground reference data to map the following 13 classes: Alfalfa (I & NI), Corn (I & NI), Sorghum (I & NI), Soybeans (I & NI), Winter Wheat (I & NI), Double Crop (I & NI), and Fallow (NI).

2003-2005, WC, outside POU:A boosted decision tree model was developed using non-irrigated WC ground reference data to map the following 7 classes (all non-irrigated): Alfalfa, Corn, Sorghum, Soybeans, Winter Wheat, Double Crop, and Fallow.

2003-2005, E, inside POU:A boosted decision tree model was developed using CE ground reference data to map the following 8 classes: Alfalfa (NI), Corn (I & NI), Sorghum (NI), Soybeans (I & NI), Winter Wheat (NI), and Double Crop (NI).

2003-2005, E, outside POU:A boosted decision tree model was developed using non-irrigated CE ground reference data to map the following 6 classes (all non-irrigated): Alfalfa, Corn, Sorghum, Soybeans, Winter Wheat, and Double Crop.

2006-2012, W, inside POU:A boosted decision tree model (one model per crop) was developed using WC ground reference data to map I & NI status for each of Alfalfa, Corn, Sorghum, Soybeans, Winter Wheat, and Double Crop.

2006-2012, E, inside POU:A boosted decision tree model (one model per crop) was developed using CE ground reference data to map I & NI status for each of Corn and Soybeans.

Irrigation Mask Development:Based on the consistency of irrigation water use in Kansas during 2002-2013, as well as management and economic difficulties that preclude substantial annual spatial variations in the set of irrigated fields, it was assumed that the set of irrigated fields in Kansas was largely stable during the 10-year study period. Furthermore, past experience of the mapping team strongly suggested that the variability of mapping irrigation status on a year-by-year basis likely would result in more mapping error than creating a single, constant irrigation mask that would be applied to all 10 years of LULC maps. Consequently, a procedure was developed to utilize all of the “inside POU” map results and produce a single irrigation mask. Upon completion, the irrigation status layer for all 10 years of maps was modified to coincide with this final, static irrigation mask.

The final procedure relied on the creation and implementation of a hierarchical logical rule set that utilized LULC trajectories, USDA NASS irrigated acreage statistics, and maximum MODI NDVI values from individual fields. All available county-level USDA NASS irrigated acreage data were obtained to facilitate this process, ultimately steering crop-specific irrigated acreage totals toward the reported values.