Article Downscaling Regional Crop Yields to Local Scale Using Remote Sensing
Paresh B. Shirsath 1,* , Vinay Kumar Sehgal 2 and Pramod K. Aggarwal 1
1 CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS), Borlaug Institute for South Asia (BISA), International Maize and Wheat Improvement Centre (CIMMYT), New Delhi 110012, India; [email protected] 2 Division of Agricultural Physics, Indian Agricultural Research Institute (IARI), ICAR, New Delhi 110012, India; [email protected] * Correspondence: [email protected]; Tel.: +91-(11)-25842940
Received: 26 January 2020; Accepted: 10 February 2020; Published: 2 March 2020
Abstract: Local-scale crop yield datasets are not readily available in most of the developing world. Local-scale crop yield datasets are of great use for risk transfer and risk management in agriculture. In this article, we present a simple method for disaggregation of district-level production statistics over crop pixels by using a remote sensing approach. We also quantified the error in the disaggregated statistics to ascertain its usefulness for crop insurance purposes. The methodology development was attempted in Parbhani district of Maharashtra state with wheat and sorghum crops in the winter season. The methodology uses the ratio of Enhanced Vegetation Index (EVI) of pixel to total EVI of the crop pixels in that district corresponding to the growth phase of the crop. It resulted in the generation of crop yield maps at the 500 m resolution pixel (grid) level. The methodology was repeated to generate time-series maps of crop yield. In general, there was a good correspondence between disaggregated crop yield and sub-district level crop yields with a correlation coefficient of 0.9.
Keywords: crop yield; downscaling; remote sensing; disaggregation
1. Introduction Crop production statistics are vital to agricultural planning and policy design and play an important role in the planning and allocation of inputs for the development of the agricultural sector. In an agrarian country, like India, where agriculture is still the main source of livelihood, crop production statistics are even more important. In India, crop area statistics and crop yield statistics are reported using different methods based on the state institutional mechanism. The crop production statistics are being collected at different scales. These statistics are often available at the district scale and to some extent at the sub-district scale. The district/sub-district scale is probably good for the planning of agricultural inputs where the majority of these transactions happen, and also for setting up the minimum support price and for national/international selling/procurement decisions. Recently, the crop insurance schemes were launched in India at the village scale and it required village level time-series of yields [1]. The major challenge in implementation is the absence of historical yield series at the village level. In order to undertake effective implementation of a crop insurance scheme with creditability, there is a need to produce a consistent long time-series of historical crop yield data at the local scale (such as village), which may be useful for fixing key insurance metrics like threshold yield and indemnity level. There have been only some piecemeal studies in India on the use of technologies for yield estimation and insurance contract design. The evidence base of these technologies, under diverse agro-ecological regions of the country, is also limited to support implementation in crop insurance
Agriculture 2020, 10, 58; doi:10.3390/agriculture10030058 www.mdpi.com/journal/agriculture Agriculture 2020, 10, 58 2 of 14 programs [1]. Besides insurance, several new information and communications technologies are emerging that deliver targeted and actionable advisories to farmers [2], and the absence of village level yield series is one of the key limitations. Estimating or measuring yields at the field level is of great use to farmers, service dealers, commodity traders, and policy-makers. Several methods have been tried to estimate or forecast crop yields at different resolutions. These methods can be classified into various categories or types based on how they model yield. The process can be date driven or process based. In a data driven approach where empirical models are built based on observed data, such models analyse data and fit equation(s) to the data. These models are often regarded as black-box (e.g., artificial neural network-based models) since they do not provide explanations on the mechanism that drives the yield [3]. The process-based models, unlike empirical models, provide detailed information on crop growth and penology on a daily basis. The yield modelling tools are dominated by crop forecasting tools and early warning systems [4,5] and operational crop yield forecasting systems [5–7]. There are several spatial crop-modelling systems available, like Monitoring Agricultural ResourceS (MARS), CCAFS Regional Agricultural Forecasting Toolbox (CRAFT), Geospatial Simulation (GeoSIM). These spatial crop modelling systems generate high-resolution yield surfaces (1 1 km: MARS, 10 10 km: CRAFT), which then can be used as a × × substitute for local scale yields. These models need detailed databases on weather, soil, crop cultivars, and management practices. Often getting these datasets at the required granularity is a challenge. Among the models successfully used to simulate crop yield in India are InfoCrop [8–11], DSSAT [12,13], APSIM [14], CERES [15–17], CropSyst [17], ORYZA [18], and many others. Most of these model applications focused on understanding the climate change impacts on crop productivity [10,11,13,15]. In addition to crop modelling [8,9,16], several other methods, like weather based models and satellite data driven yield models [19,20], are commonly used for monitoring crop growth and yield. Crop yield estimation using remote sensing remained a widely researched topic and several detailed reviews are also available on it [21–24]. Most of the remote sensing approaches used Normalized Difference Vegetation Index (NDVI) as a key parameter in crop yield estimation [19]. Many of the remote sensing based methods often follow an integrated approach, which essentially combines crop growth simulation models, weather data, remote sensing data, and scalable models [25,26]. Crop masking is one of the important input parameters to demarcate a crop boundary for simulation, which can be achieved by employing different classification techniques using remote sensing data [26,27]. All inputs for crop growth simulation models (masked with crop growing areas) and simulated yields are used to develop statistical models for different combinations of possible image acquisition dates, and these are then applied to high spatial resolution and gridded weather data [28–30]. These methods, however, require huge input datasets and are often challenging to implement. Downscaling yields to the desired level using these methods will need high resolution climatological, remote sensing, and biophysical datasets, which are not readily available. These methods also fail to incorporate or validate existing productions statistics, which are often available and being used. Considering the acceptance of the district-level crop production statistics and the fact that there is an institutional mechanism in place for collection of this data, it is required to fuse the methodologies that use these reported statistics for the purpose of attaining downscaled statistics at the required administrative level. In this study, a simple downscaling approach is presented for generating a time-series yield of multiple crops at the local scale by disaggregating the district scale yield/production statistics using remote sensing data. We also quantified the error in the disaggregated statistics to ascertain its usefulness for crop insurance purposes. The pixel level yields generated here can add vital information to the implementation of insurance schemes and contingency planning at a disaggregated level. The methodology development part is attempted in the Parbhani district of Maharashtra State with two crops in the winter (rabi) crop season. We understand that the study lacks the validation with the ground truth measurements of crop yield and area, and it is likely that the performance of this approach might not be similar for other crops, especially in the rainy season where the availability of quality remote sensing data is low. Agriculture 2020, 10, x FOR PEER REVIEW 3 of 14 Agriculture 2020, 10, 58 3 of 14 2. Materials and Methods 2. Materials and Methods 2.1. Study District and Crops 2.1.The Study study District was andundertaken Crops for the Parbhani district (Figure 1) of Maharashtra State of India. The ParbhaniThe district study wasis spread undertaken over ~650,000 for the Parbhaniha with 9 district sub-districts (Figure (tehsils/t1) of Maharashtraaluks) and State 848 villages of India. (https://parbhani.gov.in/village_list/The Parbhani district is spread over accessed ~650,000 on ha 25 with January 9 sub-districts 2020). The (tehsils Parbhani/taluks) district and 848is slightly villages larger(https: in// termsparbhani.gov.in of area compared/village_list to the/ accessed average onIndian 25 January district 2020).area of The ~450,000 Parbhani ha. This district district is slightly is in a largerhot semi-arid in terms ecological of area compared region with to the an average average Indian annual district rainfall area of 830 of ~450,000 mm (http://maharain.gov.in/ ha. This district is in a accessedhot semi-arid on 25 January ecological 2020). region In withParbhani, an average cotton, annual sorghum, rainfall soybean, of 830 and mm green (http: gram//maharain.gov.in are the key / cropsaccessed in the on rainy 25 January season 2020). (kharif) In Parbhani, and wheat cotton, and sorghum,sorghum soybean,in the winter and green season. gram In arethis the study, key cropswe attemptedin the rainy the season disaggregation (kharif) and of winter wheat and(rabi) sorghum crops of in Jowar the winter (sorghum) season. and In wheat. this study, The district we attempted level cropthe disaggregationstatistics for the of Parbhani winter (rabi) district crops for of 1999–2000 Jowar (sorghum) to 2010–2011 and wheat. period The were district downloaded level crop from statistics the websitefor the of Parbhani Department district of for Agriculture 1999–2000 and to 2010–2011 Cooperation, period Ministry were downloaded of Agriculture from and the websiteFarmers’ of Welfare,Department Govt. of of Agriculture India (https://aps.dac.gov.in/AP and Cooperation, MinistryY/Index.htm of Agriculture accessed and on Farmers’25 January Welfare, 2020). Owing Govt. of toIndia an overlap (https: //ofaps.dac.gov.in the agricultural/APY calendar/Index.htm (June accessed to May onin next 25 January year) and 2020). the Owingcrop area, to an production overlap of statisticsthe agricultural are often calendar referred (Juneto with to the May current in next an year)d next and calendar the crop year area, (e.g., production 2001–2002 statistics for agriculture are often yearreferred starting to within 2001). the current The sub-district and next calendarlevel (tehsil/taluk) year (e.g., level 2001–2002 crop area for agriculture and yield statistics year starting were in collected2001). The from sub-district the Office level of Commissionerate (tehsil/taluk) level of crop Agriculture, area and yield Maharashtra statistics wereState, collectedPune. The from sub- the districtOffice level of Commissionerate statistics were used of Agriculture, for cross validati Maharashtrang the accuracy/error State, Pune. The of disaggregation sub-district level of statisticsdistrict levelwere crop used statistics. for cross validating the accuracy/error of disaggregation of district level crop statistics.
Figure 1. Map of sub-districts (tehsils/taluks) of Parbhani, Maharashtra (background image depicts the Figure 1. Map of sub-districts (tehsils/taluks) of Parbhani, Maharashtra (background image depicts Enhanced Vegetation Index (EVI)). the Enhanced Vegetation Index (EVI)). 2.2. Satellite Datasets 2.2. Satellite Datasets In order to meet the requirement of generating a long time-series of yield statistics, we used the Terra-MODISIn order to (Moderatemeet the requirement Resolution of Imaging generating Spectrometer) a long time-series derived of 500 yield m Vegetation statistics, we Indices used (VIs)the Terra-MODISimage product (Moderate for the 2000–2001Resolution to Imaging 2013–2014 Spectr periodometer) (14 derived years) at 500 a frequencym Vegetation of 16 Indices days in(VIs) this imagestudy. product A total for of the 322 2000–2001 VI image to products 2013–2014 pertaining period (14 to theyears) study at a areafrequency (MODIS of 16 Granule days in h25v7)this study. were Adownloaded total of 322 through VI image the USGSproducts Global pertaining Visualization to the Viewer study (http: area//glovis.usgs.gov (MODIS Granule/). MODIS h25v7) VI were Image downloadedProduct: Global through MOD13A1 the USGS (version Global 5) image Visualization products Viewer are generated (http://glovis.usgs.gov/). and provided every MODIS 16 days VI at a Image500-m Product: spatial resolutionGlobal MOD13A1 as a gridded (version level-3 5) image product products in the Sinusoidal are generated projection. and provided Each file every contains 16 daysfour at reflectance a 500-m spatial band imagesresolution (blue, as a red, gridded near infrared,level-3 product and mid in infrared)the Sinusoidal and two projection. vegetation Each indices file containsimages four (NDVI reflectance and EVI). band Global images MODIS (blue, vegetation red, near indices infrared, are and designed mid infrared) to provide and consistent two vegetation spatial indicesand temporal images (NDVI comparisons and EVI). of vegetation Global MODIS conditions. vegetation Blue, indices red, and are near-infrared designed to provide reflectance, consistent centred
AgricultureAgriculture2020 2020, 10, ,10 58, x FOR PEER REVIEW 4 of4 14 of 14
spatial and temporal comparisons of vegetation conditions. Blue, red, and near-infrared reflectance, atcentred 469, 645, at and469, 858-nanometers, 645, and 858-nanometers, respectively, respec aretively, used toare determine used to determine the MODIS the dailyMODIS vegetation daily indices.vegetation The dailyindices. vegetation The daily index vegetation images index are composited images are for composited 16 days to for produce 16 days a maximumto produce value a compositemaximum (MVC) value VIcomposite image for (MVC) the 16-day VI image period. for the This 16-d studyay period. also usedThis study the Enhanced also used Vegetation the Enhanced Index (EVI)Vegetation times-series Index images(EVI) times-series for classification images for of cropclassi typesfication and of crop disaggregation types and disaggregation of crop yields. of crop EVI is anyields. ‘optimised’ EVI is index an ‘optimised’ designed toindex enhance designed the vegetationto enhance signal the vegetation with improved signal sensitivitywith improved in high sensitivity in high biomass regions. EVI is more responsive to canopy structural variations, including biomass regions. EVI is more responsive to canopy structural variations, including leaf area index leaf area index (LAI), canopy type, plant physiognomy, and canopy architecture [32]. It thus supports (LAI), canopy type, plant physiognomy, and canopy architecture [31]. It thus supports improved improved vegetation monitoring through a de-coupling of the canopy background signal and a vegetation monitoring through a de-coupling of the canopy background signal and a reduction in reduction in atmosphere influences. atmosphere influences. 2.3. Methodology 2.3. Methodology The overall methodology followed in this study is depicted as a flow chart in Figure 2. The steps The overall methodology followed in this study is depicted as a flow chart in Figure2. The steps shown in the first column are those related to the acquisition and pre-processing of satellite data. The shownsteps inshown the first in the column second are column those are related those to related the acquisition to the identification and pre-processing of pixels of ofthe satellite crop of interest data. The stepsin a shown season in in the thesecond study districts. column areThe those steps relatedshown toin the third identification column are of pixelsthose related of the crop to finding of interest a inweighting a season in criterion the study using districts. satellite The derived steps index shown to distribute in the third district column crop arestatistics those over related the identified to finding a weightingcrop pixels. criterion Then, using the satellitedistributed derived crop indexstatistics to distribute are re-aggregated district crop at statisticsthe sub-district, over the i.e., identified the croptehsil/taluk pixels. Then, level, the and distributed compared crop with statistics the reported are re-aggregated values for an aterror the sub-district,analysis of the i.e., disaggregated the tehsil/taluk level,values. and Based compared on this with meth theodology, reported a time-series values for anof disaggregated error analysis yield of the statistics disaggregated were generated values. Basedfor onthe this 2000–2001 methodology, to 2013–2014 a time-series period. of disaggregated The methodology yield statistics is described were generatedfor these three for the columns 2000–2001 in to 2013–2014subsequent period. sections. The methodology is described for these three columns in subsequent sections.
FigureFigure 2. 2.The The overall overall methodology methodology flowchartflowchart showing the the steps steps followed followed in in this this study. study. The The steps steps on on thethe left left describe describe data data pre-processing, pre-processing, thethe middlemiddle column describes describes crop crop mapping, mapping, and and the the column column on on thethe right right describes describes yield yield downscaling. downscaling.
2.3.1.2.3.1. Pre-Processing Pre-Processing of of Satellite Satellite Data Data TheThe pre-processing pre-processing of of time-series time-series ofof MODISMODIS image granules granules was was carried carried out out in inENVI™ ENVI ™+ IDL™+ IDL ™ V4.8V4.8 Image Image Processing Processing software. software. Each Each MODISMODIS GranuleGranule in in HDF-EOS HDF-EOS format format was was imported imported into into native native binary format of Image Processing Software, and among the scientific datasets present in the granule, only the EVI image dataset was extracted. All the 322 extracted EVI images in original sinusoidal Agriculture 2020, 10, 58 5 of 14 projection were stacked together in a time-series starting from Julian day 129 of 2000 (9-May-2000) to Julian day 113 of 2014 (23-April-2014). The Julian day of an EVI image refers to the starting day of the 16-day period. The stacked images were reprojected from Sinusoidal projection to Lambert Conformal Conic (LCC) projection with WGS-84 datum and a pixel size of 500 m using nearest neighbour resampling method. The stacked EVI images were subsetted, corresponding to the geographical extent of the study district. The original image stores scaled EVI values in integer. By applying the scale factor for recalibration, the images were converted into true EVI values in float.
EVI Time-Series Filtering Although the 16-day maximum value compositing of EVI is implemented to minimise the noise due to cloud contamination, it has been reported that the VI signal is still contaminated by significant residual noise due to atmospheric variability and bi-directional effects [32,33]. Such residual noise is more prominent under Indian conditions, especially during the rainy (kharif) season, when cloudy conditions persist for months together. These noises greatly undermine the basic assumption that the changes in VI are related to changes in vegetation conditions, thus affecting the monitoring of land cover and terrestrial ecosystems [34,35]. For this reason, a number of methods for reducing noise and constructing high-quality VI time-series data sets for further analysis have been formulated, applied, and evaluated. Among the different methods proposed in literature, we adopted an algorithm-based Savitzky–Golay filter, to efficiently reduce the contamination in the VI time-series that is caused primarily by cloud contamination and atmospheric variability. The algorithm is mainly based on the approach proposed by Chen et al., 2004. This method is based on two assumptions: (1) that the VI data from a satellite sensor is primarily related to vegetation changes and follows annual cycle of growth and decline; and (2) that clouds and poor atmospheric conditions mostly reduce VI values, requiring that sudden drops in VI, which are not compatible with the gradual process of vegetation change, be regarded as noise and removed. This algorithm makes the data approach the upper VI envelope and to best fit the VI variations over the season through an iterative process. At the heart of the algorithm is the Savitzky–Golay filter, which is a weighted moving average filter with weighting given as a polynomial of a certain degree (3rd degree in this case). The weight coefficients, when applied to a signal, perform a polynomial least squares fit within the filter window. This polynomial is designed to preserve higher moments within the data and to reduce the bias introduced by the filter. We cross-checked the filter (Savitzky–Golay) output for inconsistencies on a sample basis. Please see Figure3, where we highlight the raw and filtered EVI profile over 14 years. The blue coloured profile shows noises in terms of sudden dips in data, while the filtered red colour profile fits the outer EVI envelope, is smoothened, and preserves the periods of growth and decline well. Agriculture 2020, 10, x FOR PEER REVIEW 6 of 14 decision rule classification above, the winter (rabi) season crops were classified for each crop season of this study.
Land Cover Image The global MODIS/Terra + Aqua Land Cover Type (MCD12Q1 product for h25v7 granule) for the year 2012 period, available from http://reverb.echo.nasa.gov/ at a resolution of 500 m in sinusoidal projection and HDF-EOS format, was downloaded. The first layer in the IGBP Land Cover Type Classification with sixteen classes was extracted and reprojected. The global land use layer was subsetted, corresponding to the bounds of the study area (Figure 4a). The grids with land use class of Cropland (class = 12) was retained and other land-use classes were masked out. The croplands mask was overlaid with a study area boundary to generate the cropland mask of the study area. The croplandsAgriculture 2020 land, 10 use, 58 mask for the study area is shown in Figure 4b. 6 of 14
Figure 3. The raw and filtered EVI profile of pixels over 2000–2001 kharif and rabi crops. The inset Figureshows 3. zoomed The raw up and raw filtered and filtered EVI profile EVI profiles of pixels for theover 2004–2005 2000–2001 crop kharif season. and rabi crops. The inset shows zoomed up raw and filtered EVI profiles for the 2004–2005 crop season. 2.3.2. Crop Area Identification Crop area estimation methodology and its efficacy depends on various operational factors, such as land configuration, field shape, crop type, cropping pattern, available skills, and resources [36]. Here, we used the Hierarchical Decision Rule method for crop area identification based on the growth profile of EVI. The scientific rationale for separability of different landuse/crop classes using hierarchical decision rules is based on using the a-priori knowledge of the temporal spectral growth profiles of the various land covers including crops [37]. MODIS Vegetation Image Product was used in the study, and the EVI values ranged from 2000 to +10,000 with a scale factor of 0.0001. In the first step, − the pixels, which were masked out during filtering, were labelled as “Non-crop”. In the next step, the current fallow pixels with an EVI below 0.2 during November, as well as Early February periods, were identified. The current fallow represents the agricultural land where sowing has not been done in the current season. Then, the pixels with wheat crop were identified based on the decision that the EVI profile shows a peak anywhere between January and February. Using the decision rule classification above, the winter (rabi) season crops were classified for each crop season of this study.
Land Cover Image The global MODIS/Terra + Aqua Land Cover Type (MCD12Q1 product for h25v7 granule) for the year 2012 period, available from http://reverb.echo.nasa.gov/ at a resolution of 500 m in sinusoidal projection and HDF-EOS format, was downloaded. The first layer in the IGBP Land Cover Type Classification with sixteen classes was extracted and reprojected. The global land use layer was subsetted, corresponding to the bounds of the study area (Figure4a). The grids with land use class of Cropland (class = 12) was retained and other land-use classes were masked out. The croplands Agriculture 2020, 10, 58 7 of 14
mask was overlaid with a study area boundary to generate the cropland mask of the study area. The croplandsAgriculture land 2020, use 10, x mask FOR PEER for theREVIEW study area is shown in Figure4b. 7 of 14
(a) (b)
FigureFigure 4. (a)4. (a)The The MODIS MODIS land land cover cover map map of of the the study study region region with with land land cover cover classes classes and and(b) (b)cropland cropland maskmask of theof the study study region. region.
2.3.3.2.3.3. Disaggregation Disaggregation of of District District Yield Yield Statistics Statistics WeighingWeighing criterion: criterion: The The basic basic premise premise we we adopted adopted in in this this study study to to disaggregate disaggregate district district statistics statistics overover the the crop crop pixels pixels of of interest interest was was that that the the integrated integrated EVI EVI for for the the growth growth period period of of the the crop crop is directlyis directly relatedrelated to to the the crop crop yield. yield. The The higher higher the the value value of of integratedintegrated EVI,EVI, thethe higher the crop yield yield will will be. be. In Inorder to scale thisthis relation,relation, aa weightingweighting factor factor (F) (F) was was calculated calculated as as the the ratio ratio of of “Crop “Crop pixel pixel integrated integrated EVI”EVI” to to the the “Total “Total integrated integrated EVI EVI over over all all the the Crop Crop pixels pixels in thein the district”. district”.
EVIܧܸܫ௫ ܨൌ pixel (1) F = ܧܸܫ (1) EVIdistrictௗ௦௧௧
where, EVIpixel is the integrated EVI value of the pixel of the crop of interest, EVIdistrict is the sum of where, EVIpixel is the integrated EVI value of the pixel of the crop of interest, EVIdistrict is the sum of EVIpixel over all the crop pixels in that district, i.e., EVIpixel over all the crop pixels in that district, i.e., n ܧܸܫௗ௦௧௧Xൌܧܸܫ௫ (2) EVIdistrict = ୀEVIpixel (2) i=0 where, the total number of pixels of the crop of interest in the district is n. So, an F value map for crop where,pixel the was total calculated. number of pixels of the crop of interest in the district is n. So, an F value map for crop pixel wasAs calculated.the EVIdistrict was an integrated value, we correspondingly disaggregated district wise crop productionAs the EVI statisticsdistrict was over an the integrated crop pixels, value, instead we correspondingly of district yield disaggregated statistics. For district a crop wisepixel, crop the F productionvalue was statistics multiplied over by the the crop district pixels, production instead of statis districttics yield and statistics.divided by For the a pixel crop pixel,area (i.e., the F25 value ha), to wasarrive multiplied at the yield by the value district of the production crop at the statistics pixel level and dividedin kg/ha. by the pixel area (i.e., 25 ha), to arrive at the yield value of the crop at the pixel level in kg/ha. ܨൈܲௗ௦௧௧ ܻ݈݅݁݀ ൌ (3) ௫ ܣ F Pdistrict Yieldpixel = × (3) where, Pdistrict is the district crop production in (kg), andA A is the area of one pixel in (ha). P where,3. Resultsdistrict is the district crop production in (kg), and A is the area of one pixel in (ha).
3.1. Crop Mapping The decision tree approach was allied to generate crop maps for each year. Figure 5 shows the classified map of the Parbhani district for the rabi season year 2010–2011 as an example; this map was generated by inputting time-series of 13 filtered EVI bands between 14-September-2010 and 22- March-2011.
Agriculture 2020, 10, 58 8 of 14
3. Results
3.1. Crop Mapping The decision tree approach was allied to generate crop maps for each year. Figure5 shows the classified map of the Parbhani district for the rabi season year 2010–2011 as an example; this map was generated by inputting time-series of 13 filtered EVI bands between 14-September-2010 andAgriculture 22-March-2011. 2020, 10, x FOR PEER REVIEW 8 of 14
Figure 5. The crop area map of the Parbhani district using hierarchical decision rule-based classification Figure 5. The crop area map of the Parbhani district using hierarchical decision rule-based of time-series of an EVI profile. classification of time-series of an EVI profile. In order to verify the classification accuracy, the wheat and rabi sorghum crop area was aggregated In order to verify the classification accuracy, the wheat and rabi sorghum crop area was at the sub-district (tehsil/taluk) level and compared with the statistics reported by the State Department aggregated at the sub-district (tehsil/taluk) level and compared with the statistics reported by the of Agriculture. The comparison for the year 2010–2011 is shown in Figure6. The error in the remote State Department of Agriculture. The comparison for the year 2010–2011 is shown in Figure 6. The sensing (RS) estimated area of wheat as compared to values reported by the State Department of error in the remote sensing (RS) estimated area of wheat as compared to values reported by the State Agriculture (SDA) varied between a 17.6% underestimation for the Parbhani sub-district (tehsil/taluk) Department of Agriculture (SDA) varied− between a −17.6% underestimation for the Parbhani sub- to a 12.0% overestimation in the Selu sub-district (tehsil/taluk). Overall, at the district level, the error district (tehsil/taluk) to a 12.0% overestimation in the Selu sub-district (tehsil/taluk). Overall, at the was a 3.6% underestimation. The error in remote sensing estimated that the rabi sorghum area at the district level, the error was a 3.6% underestimation. The error in remote sensing estimated that the sub-district (tehsil/taluk) level varied between 11.0% in the Manwat sub-district (tehsil/taluk) to +20% rabi sorghum area at the sub-district (tehsil/taluk)− level varied between −11.0% in the Manwat sub- in the Pathri sub-district (tehsil/taluk), with an overall 1.0% overestimation error at the district level. district (tehsil/taluk) to +20% in the Pathri sub-district (tehsil/taluk), with an overall 1.0% overestimation error at the district level.
(a) (b)
Figure 6. Comparison remote sensing estimated (RS) wheat crop (a) and sorghum (b) area with those reported by the State Department of Agriculture (SDA) for the sub-districts (tehsils/taluks) of Parbhani.
Agriculture 2020, 10, x FOR PEER REVIEW 8 of 15
Figure 5. The crop area map of the Parbhani district using hierarchical decision rule-based classification of time-series of an EVI profile.
In order to verify the classification accuracy, the wheat and rabi sorghum crop area was aggregated at the sub-district (tehsil/taluk) level and compared with the statistics reported by the State Department of Agriculture. The comparison for the year 2010–2011 is shown in Figure 6. The error in the remote sensing (RS) estimated area of wheat as compared to values reported by the State Department of Agriculture (SDA) varied between a −17.6% underestimation for the Parbhani sub- district (tehsil/taluk) to a 12.0% overestimation in the Selu sub-district (tehsil/taluk). Overall, at the district level, the error was a 3.6% underestimation. The error in remote sensing estimated that the
Agriculturerabi sorghum2020, 10 area, 58 at the sub-district (tehsil/taluk) level varied between −11.0% in the Manwat9 sub of 14- district (tehsil/taluk) to +20% in the Pathri sub-district (tehsil/taluk), with an overall 1.0% overestimation error at the district level.
(a) (b)
AgricultureFigure 2020 6., Comparison10, x FOR PEER remote REVIEW sensing estimated (RS) wheat crop ( a) and sorghum (b) area with those9 of 14 reported by by the the State State Department Department of Agriculture of Agriculture (SDA) (SDA) for the for sub-districts the sub-districts (tehsils/taluks) (tehsils/taluks) of Parbhani. of 3.2. DisaggregatedParbhani. Crop Yield Map 3.2. Disaggregated Crop Yield Map Using the disaggregation criterion, the rabi sorghum and wheat yield maps were generated for
the cropUsing pixels the disaggregationfor each crop year criterion, of the the stud rabiy. sorghumFor example, and wheatFigure yield7 shows maps the were rabi generated sorghum fordisaggregated the crop pixels yield for map. each For crop the yeardisaggregation, of the study. the For district example, production Figure statistics7 shows for the 2010–2011 rabi sorghum were disaggregatedused, and the yieldEVI was map. integrated For the disaggregation, at pixels for the the period district between production 3-January-2011 statistics for and 2010–2011 30-January- were used,2011. The and thepixel EVI level was yield integrated generated at pixels varied for betw the periodeen 694 between and 1447 3-January-2011 kg/ha for the andyear 30-January-2011. 2010–2011, and Thethe EVI pixel was level integrated yield generated at pixels varied for betweenthe period 694 between and 1447 1-January-2011 kg/ha for the year and 2010–2011, 2-February-2011. and the EVIThe waspixel integrated level wheat at yield pixels generated for the period by disaggregation between 1-January-2011 varied between and 2-February-2011. 900 and 2400 kg/ha The for pixel the levelyear wheat2010–2011. yield generatedTo validate by disaggregationthe disaggregation varied methodology between 900 andand 2400 analyse kg/ha forthe theextent year 2010–2011.of error, the To validatedisaggregated the disaggregation pixel level yield methodology values were and aggregated analyse the to extentthe sub-district of error, the (tehsil/taluk) disaggregated level pixel and levelthen yieldcompared values with were the aggregated values reported to the sub-districtby the State (tehsil Department/taluk) level of Agriculture and then compared (SDA) at withthe sub-district the values reported(tehsil/taluk) by the level. State The Department comparison of of Agriculture estimated (SDA)and reported at the sub-district rabi sorghum (tehsil and/taluk) wheat level. yield The is comparisonshown in Figure of estimated 8. In general, and reported there was rabi good sorghum correspondence and wheat yield between is shown the two in Figure with 8a. Incorrelation general, therecoefficient was goodof 0.9 correspondencefor both crops. The between methodology the two underestimated with a correlation the coe sorghumfficient ofyield 0.9 in for sub-districts both crops. The(tehsils/taluks) methodology having underestimated higher yield the sorghumvalues, while yield init sub-districtsoverestimated (tehsils the /yieldtaluks) in having sub-districts higher yield(tehsils/taluks) values, while with it lower overestimated yield values. the yield in sub-districts (tehsils/taluks) with lower yield values.
(a) (b)
Figure 7. The disaggregated wheat (a) and sorghum (b) yield map for the year 2010–2011.
At the sub-district (tehsil/taluk) level, the error between the estimated and reported sorghum yield varied between −22.0% in the Selu sub-district (tehsil/taluk) to +15.8% in the Pathri sub-district (tehsil/taluk) (Figure 8). Overall, the average root mean square error was 167.87 kg/ha, which is about 15.4% of the average district yield. For wheat, at the sub-district (tehsil/taluk) level, the error between the estimated and reported yield varied between −16.0% in the Selu sub-district (tehsil/taluk) to +9.0% in the Manwat sub-district (tehsil/taluk). Overall, the average root mean square error was 178.9 kg/ha, which is about 10.5% of the average district yield.
AgricultureAgriculture 20202020,, 1010,, 58x FOR PEER REVIEW 10 of 14
Figure 8. Comparison of disaggregated yield (RS) and reported yield (SDA) at the sub-district Figure 8. Comparison of disaggregated yield (RS) and reported yield (SDA) at the sub-district (tehsil/taluk) level, and the per cent error for the winter (rabi) season of 2010–2011. (tehsil/taluk) level, and the per cent error for the winter (rabi) season of 2010–2011. At the sub-district (tehsil/taluk) level, the error between the estimated and reported sorghum 4. Discussion yield varied between 22.0% in the Selu sub-district (tehsil/taluk) to +15.8% in the Pathri sub-district − (tehsilUse/taluk) of remote (Figure 8sensing). Overall, data the for average disaggregating root mean exis squareting errorcrop wasproduction 167.87 kg statistics/ha, which to the is about local 15.4%scale ofcan the be average highly district useful yield. to agricultural For wheat, atpl theanning sub-district and implementation (tehsil/taluk) level, of risk the errormanagement between thestrategies. estimated The and remote reported sensing-based yield varied methods between provid16.0%e quick in the crop Selu yield sub-district estimates (tehsil (downscaled/taluk) to + or9.0% in- − inseason) the Manwat covering sub-district a vast geographical (tehsil/taluk). area. Overall, The the in-season average rootestimates mean squarecan be error achieved was 178.9 by minor kg/ha, whichmodifications is about in 10.5% the ofmethodolog the averagey districtdescribed yield. here. A few minor modifications would make this method a predictor model; possible modifications would be the use of vegetation indices at a crop 4.growth Discussion stage where yield-vegetation and crop identification are highly correlated. UseHigh of resolution remote sensing yield surface data for can disaggregating add substant existingial value crop to productionrisk management statistics and to planning the local scaleprocess can in be agriculture. highly useful These to agricultural yield surfaces planning ca andn also implementation give insight ofto risk commodity management markets strategies. and Theagricultural remote sensing-basedinput planning. methods provide quick crop yield estimates (downscaled or in-season) coveringWith a the vast space geographical data revolution area. The and in-season availability estimates of more can and be more achieved datasets by minor with increased modifications spatial in theand methodologytemporal coverages, described the here.accuracy A few has minor been su modificationsbstantially increased would make while this costs method have come a predictor down model;significantly. possible Still, modifications there are challenges would be in the obtaining use of vegetation estimates indicesin the rainy at a crop season growth because stage of wherecloud yield-vegetationcover. and crop identification are highly correlated. HighUse of resolution disaggregated yield surfacecrop yields can addin insurance substantial schemes: value to Globally, risk management crop insurance and planning is the preferred process inoption agriculture. for risk transfer. These yield The demand surfaces for can improved also give insurance insight to products commodity is increasing. markets Several and agricultural countries inputnow operate planning. insurance schemes at the village level; however; unavailability of crop yield and area statisticsWith at the this space level data remains revolution the key and bottleneck availability in implementing of more and moreor designing datasets such with schemes. increased In spatial India, andgovernments temporal coverages,provide a huge the accuracy subsidy has on beencrop substantiallyinsurance premiums, increased and while th costse new have crop come insurance down significantly.scheme in India Still, is there being are implemented challenges in obtainingwith the estimatesvillage as in the the insurance rainy season un becauseit [1,39]. of The cloud scheme cover. parameters,Use of disaggregated like threshold cropyield yields and loss in insuranceassessment, schemes: are calculated Globally, on cropan area–yield-based insurance is the approach. preferred optionUnder forthe riskprevious transfer. yield The index demand insurance for improved schemes, insurance the insurance products unit was is increasing. typically at Several the sub-district countries nowlevel. operate In the new insurance scheme, schemes the insurance at the village product level; desi however;gn and price unavailability is based on ofhistorical crop yield yield and values area statisticsat the village at this level. level A remainsmajor challenge the key bottleneckin implementi in implementingng a new scheme or designing at the village such level schemes. is the Inabsence India, of historical yield series. So, in order to undertake effective implementation of crop insurance schemes with creditability, there is a need to produce a consistent long time-series of historical crop
Agriculture 2020, 10, 58 11 of 14
governments provide a huge subsidy on crop insurance premiums, and the new crop insurance scheme in India is being implemented with the village as the insurance unit [1,38]. The scheme parameters, like threshold yield and loss assessment, are calculated on an area–yield-based approach. Under the previous yield index insurance schemes, the insurance unit was typically at the sub-district level. In the new scheme, the insurance product design and price is based on historical yield values at the village level. A major challenge in implementing a new scheme at the village level is the absence of Agriculturehistorical 2020 yield, 10, x FOR series. PEER So, REVIEW in order to undertake effective implementation of crop insurance11 schemes of 14 with creditability, there is a need to produce a consistent long time-series of historical crop yield data yieldat thedata local at scalethe local (such scale as village), (such as which village), may which be useful may for be fixing useful indemnity for fixing limits, indemnity threshold limits, yields, thresholdpremium yields, rates premium at the village rates panchayat, at the village or pancha other equivalentyat, or other units. equivalent This methodology units. This methodology is the first step is thetowards first step getting towards a historical getting time-seriesa historical time-ser of yieldies at of the yield village at the level village using level district-level using district-level production productiondata and data remote and sensing remote techniques. sensing techniques. Figure9 showsFigure coe9 showsfficient coefficient of variation of variation of sorghum of sorghum yield from yield1999–2000 from 1999–2000 to 2013–2014. to 2013–2014.
FigureFigure 9. Coefficient 9. Coefficient of variation of variation of sorghum of sorghum yield yield from from 1999–2000 1999–2000 to 2013–2014. to 2013–2014.
LimitationsLimitations of the of thestudy: study: This Thisstudy study presented presented a simple a simple method method for downscaling for downscaling the district the level district statisticslevel statistics to the sub-district to the sub-district or village or villageor pixel or level pixel using level a using remote a remote sensing sensing approach. approach. Overall, Overall, the averagethe average root mean root square mean error square was error 167.87 was kg/ha 167.87 and kg 178.9/ha and kg/ha, 178.9 which kg/ ha,is about which 15.4% is about and 10.5% 15.4% of and the10.5% average of thedistrict average yield district for sorghum yield for and sorghum wheat, re andspectively. wheat, respectively. The error can The be attributed error can be to attributedseveral sources,to several such sources, as the accuracy such as theof the accuracy generated of the crop generated mask/decision crop mask tree/decision approach tree and approach to the data and to resolution issues considering prevalent small holding farming in the region. Although, the study also validates the approach with the sub-district level using reported statistics. We understand that the study was done over a small geographic area and lacks the validation with the ground truth measurements of crop yield and area. It is likely that the performance of this approach might not be similar for other crops, especially in the rainy season where the availability of quality remote sensing data is low.
5. Conclusions
Agriculture 2020, 10, 58 12 of 14 the data resolution issues considering prevalent small holding farming in the region. Although, the study also validates the approach with the sub-district level using reported statistics. We understand that the study was done over a small geographic area and lacks the validation with the ground truth measurements of crop yield and area. It is likely that the performance of this approach might not be similar for other crops, especially in the rainy season where the availability of quality remote sensing data is low.
5. Conclusions The study was undertaken to develop and validate a methodology of disaggregation of district level crop yield statistics at the pixel level (i.e., local scale) using satellite remote sensing images. The study procured and pre-processed 500 m MODIS EVI times-series images for 2000–2001 to 2013–2014 crop years for the Parbhani district of Maharashtra State in India. The Savitzky–Golay based filtering technique was adopted in this study for the removal of residual noise in EVI time-series images. A hierarchical decision rule-based image classification technique was adopted for identification and mapping of crop pixels of interest. In the absence of specific ground-truth, the crop growth profile derived phenology events were the basis for fixing the rules during classification. A simple method was developed and applied for disaggregation of crop district production statistics over crop pixels by using the ratio of EVI of pixel to total EVI of the crop pixels in that district corresponding to the growth phase of crops. It resulted in the generation of crop yield maps at the 500 m resolution pixel (grid) level. The methodology was repeated for all the 14 years to generate time-series of crop yield map. In general, there was good correspondence between disaggregated crop yield and tehsil crop yields with a correlation coefficient of 0.9 for rabi crops in Parbhani (wheat and sorghum). At the sub-district (tehsil/taluk) level, the error between estimated and reported sorghum yield varied between 22.0% in the Selu sub-district (tehsil/taluk) to +15.8% in the Pathri sub-district − (tehsil/taluk), and for wheat, 16.0% in the Selu sub-district (tehsil/taluk) to +9.0% in the Manwat − sub-district (tehsil/taluk). Overall, the average root mean square error was 178.9 kg/ha, which is about 10.5% of the average district wheat yield. The local scale yield estimates are of immense use in agricultural planning and implementation of risk management strategies. Future research should be targeted at improving these estimates in the rainy season where cloudy conditions pose challenges.
Author Contributions: P.B.S.; V.K.S. and P.K.A. designed the research, V.K.S. and P.B.S. developed methodology and software and analysed the results; V.K.S. and P.B.S. wrote the first draft of the article, which was improved with substantial input from P.K.A. All authors have read and agreed to the published version of the manuscript. Funding: This research was funded by CGIAR Fund Donors and through bilateral funding agreements. Acknowledgments: This work was implemented as part of the CGIAR Research Program on Climate Change, Agriculture, and Food Security (CCAFS), which is carried out with support from the CGIAR Trust Fund and through bilateral funding agreements. For details, please visit https://ccafs.cgiar.org/donors. The views expressed in this document cannot be taken to reflect the official opinions of these organizations. Conflicts of Interest: The authors declare no conflict of interest.
References
1. Aggarwal, P.K.; Chand, R.; Bhutani, A.; Kumar, V.; Goel, S.K.; Rao, K.N.; Poddar, M.K.; Sud, U.C.; Krishna Murthy, Y.V.N.; Ray, S.S.; et al. Report of the Task Force on Enhancing Technology Use in Agriculture Insurance; CMFRI: Kochi, India, 2016. 2. Gangopadhyay, P.K.; Khatri-Chhetri, A.; Shirsath, P.B.; Aggarwal, P.K. Spatial targeting of ICT-based weather and agro-advisory services for climate risk management in agriculture. Clim. Chang. 2019, 154, 242–256. [CrossRef] 3. Yeh, H.; Lee, C.; Hsu, K.; Chang, P. GIS for the assesment of the groundwater recharge potential zone. Environ. Geol. 2009, 58, 185–195. [CrossRef] Agriculture 2020, 10, 58 13 of 14
4. Shelia, V.; Hansen, J.; Sharda, V.; Porter, C.; Aggarwal, P.; Wilkerson, C.J.; Hoogenboom, G. A multi-scale and multi-model gridded framework for forecasting crop production, risk analysis, and climate change impact studies. Environ. Model. Softw. 2019, 115, 144–154. [CrossRef] 5. Gyawali, D.R.; Shirsath, P.B.; Kanel, D.; Burja, K.; Arun, K.C.; Aggarwal, P.K.; Hansen, J.W.; Rose, A. Inseason Crop Yield Forecasting for Early Warning Planning of Food Security Using Ccafs Regional Agricultural Forecasting Toolbox (Craft) in Nepal; CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS): Wageningen, The Netherlands, 2018. 6. Cantelaube, P.; Terres, J.M. Seasonal weather forecasts for crop yield modelling in Europe. Tellus A Dyn. Meteorol. Oceanogr. 2005, 57, 476–487. [CrossRef] 7. Motha, R.P. Agricultural Drought: USDA Perspectives; Drought Mitigation Center Faculty Publications: Reston, Virginia, USA, 2011. 8. Aggarwal, P.K.; Banerjee, B.; Daryaei, M.G.; Bhatia, A.; Bala, A.; Rani, S.; Chander, S.; Pathak, H.; Kalra, N. InfoCrop: A dynamic simulation model for the assessment of crop yields, losses due to pests, and environmental impact of agro-ecosystems in tropical environments. II. Performance of the model. Agric. Syst. 2006, 89, 47–67. [CrossRef] 9. Aggarwal, P.K.; Kalra, N.; Chander, S.; Pathak, H. InfoCrop: A dynamic simulation model for the assessment of crop yields, losses due to pests, and environmental impact of agro-ecosystems in tropical environments. I. Model description. Agric. Syst. 2006, 89, 1–25. [CrossRef] 10. Byjesh, K.; Kumar, S.N.; Aggarwal, P.K. Simulating impacts, potential adaptation and vulnerability of maize to climate change in India. Mitig. Adapt. Strateg. Glob. Chang. 2010, 15, 413–431. [CrossRef] 11. Kumar, S.N.; Aggarwal, P.K.; Saxena, R.; Rani, S.; Jain, S.; Chauhan, N. An Assessment of Regional Vulnerability of Rice to Climate Change in India. Climatic Chang. 2013, 118, 683–699. 12. Timsina, J.; Godwin, D.; Humphreys, E.; Kukal, S.S.; Smith, D. Evaluation of options for increasing yield and water productivity of wheat in Punjab, India using the DSSAT-CSM-CERES-Wheat model. Agric. Water Manag. 2008, 95, 1099–1110. [CrossRef] 13. Subash, N.; Ram Mohan, H.S. Evaluation of the impact of climatic trends and variability in rice-wheat system productivity using Cropping System Model DSSAT over the Indo-Gangetic Plains of India. Agric. For. Meteorol. 2012, 164, 71–81. [CrossRef] 14. Mohanty, M.; Probert, M.E.; Reddy, K.S.; Dalal, R.C.; Mishra, A.K.; Subba Rao, A.; Singh, M.; Menzies, N.W. Simulating soybean-wheat cropping system: APSIM model parameterization and validation. Agric. Ecosyst. Environ. 2012, 152, 68–78. [CrossRef] 15. Singh, P.K.; Singh, K.K.; Bhan, S.C.; Baxla, A.K.; Singh, S.; Rathore, L.S.; Gupta, A. Impact of projected climate change on rice (Oryza sativa L.) yield using CERES-rice model in different agroclimatic zones of India. Curr. Sci. 2017, 112, 108–115. [CrossRef] 16. Jones, J.; Hoogenboom, G.; Porter, C.; Boote, K.; Batchelor, W.; Hunt, L.; Wilkens, P.; Singh, U.; Gijsman, A.; Ritchie, J. The DSSAT cropping system model. Eur. J. Agron. 2003, 18, 235–265. [CrossRef] 17. Singh, A.K.; Tripathy, R.; Chopra, U.K. Evaluation of CERES-Wheat and CropSyst models for water-nitrogen interactions in wheat crop. Agric. Water Manag. 2008, 95, 776–786. [CrossRef] 18. Sudhir-Yadav, S.; Li, T.; Humphreys, E.; Gill, G.; Kukal, S.S. Evaluation and application of ORYZA2000 for irrigation scheduling of puddled transplanted rice in north west India. Field Crops Res. 2011, 122, 104–117. [CrossRef] 19. Dadhwal, V.K.; Sehgal, V.K.; Singh, R.P.; Rajak, D.R. Wheat yield modelling using satellite remote sensing with weather data: Recent Indian experience. Mausam 2003, 54, 253–262. 20. Becker-Reshef, I.; Justice, C.; Sullivan, M.; Vermote, E.; Tucker, C.; Anyamba, A.; Small, J.; Pak, E.; Masuoka, E.; Schmaltz, J.; et al. Monitoring Global Croplands with Coarse Resolution Earth Observations: The Global Agriculture Monitoring (GLAM) Project. Remote Sens. 2010, 2, 1589–1609. [CrossRef] 21. Lobell, D.B.; Burke, M.B. On the use of statistical models to predict crop yield responses to climate change. Agric. For. Meteorol. 2010, 150, 1143–1152. [CrossRef] 22. Burke, M.; Lobell, D.B. Satellite-based assessment of yield variation and its determinants in smallholder African systems. Proc. Natl. Acad. Sci. USA 2017, 114, 2189–2194. [CrossRef] 23. Jin, X.; Kumar, L.; Li, Z.; Feng, H.; Xu, X.; Yang, G.; Wang, J. A review of data assimilation of remote sensing and crop models. Eur. J. Agron. 2018, 92, 141–152. [CrossRef] Agriculture 2020, 10, 58 14 of 14
24. Weiss, M.; Jacob, F.; Duveiller, G. Remote sensing for agricultural applications: A meta-review. Remote Sens. Environ. 2020, 236, 111402. [CrossRef] 25. Lobell, D.B.; Azzari, G. Satellite detection of rising maize yield heterogeneity in the U.S. Midwest. Environ. Res. Lett. 2017, 12, 014014. [CrossRef] 26. Azzari, G.; Jain, M.; Lobell, D.B. Towards fine resolution global maps of crop yields: Testing multiple methods and satellites in three countries. Remote Sens. Environ. 2017, 202, 129–141. [CrossRef] 27. Lobell, D.B.; Thau, D.; Seifert, C.; Engle, E.; Little, B. A scalable satellite-based crop yield mapper. Remote Sens. Environ. 2015, 164, 324–333. [CrossRef] 28. Chakrabarti, S.; Bongiovanni, T.; Judge, J.; Zotarelli, L.; Bayer, C. Assimilation of SMOS Soil Moisture for Quantifying Drought Impacts on Crop Yield in Agricultural Regions. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2014, 7, 3867–3879. [CrossRef] 29. Ines, A.V.M.; Das, N.N.; Hansen, J.W.; Njoku, E.G. Assimilation of remotely sensed soil moisture and vegetation with a crop simulation model for maize yield prediction. Remote Sens. Environ. 2013, 138, 149–164. [CrossRef] 30. Nearing, G.S.; Crow, W.T.; Thorp, K.R.; Moran, M.S.; Reichle, R.H.; Gupta, H.V. Assimilating remote sensing observations of leaf area index and soil moisture for wheat yield estimates: An observing system simulation experiment. Water Resour. Res. 2012, 48.[CrossRef] 31. Gao, X.; Huete, A.R.; Ni, W.; Miura, T. Optical–Biophysical Relationships of Vegetation Spectra without Background Contamination. Remote Sens. Environ. 2000, 74, 609–620. [CrossRef] 32. Chen, X.; Vierling, L.; Rowell, E.; DeFelice, T. Using lidar and effective LAI data to evaluate IKONOS and Landsat 7 ETM+ vegetation cover estimates in a ponderosa pine forest. Remote Sens. Environ. 2004, 91, 14–26. [CrossRef] 33. Lovell, J.L.; Graetz, R.D. Filtering Pathfinder AVHRR Land NDVI data for Australia. Int. J. Remote Sens. 2001, 22, 2649–2654. [CrossRef] 34. Cihlar, J.; Ly, H.; Li, Z.; Chen, J.; Pokrant, H.; Huang, F. Multitemporal, multichannel AVHRR data sets for land biosphere studies—Artifacts and corrections. Remote Sens. Environ. 1997, 60, 35–57. [CrossRef] 35. Gutman, G.G. Vegetation indices from AVHRR: An update and future prospects. Remote Sens. Environ. 1991, 35, 121–136. [CrossRef] 36. Casley, D.; Kumar, K. The Collection, Analysis, and Use of-Monitoring and Evaluation Data a World Bank Publication; World Bank: Washington, DC, USA, 1988. 37. Poza, M.; Bhagia, N.; Patel, J.H.; Dutta, S.; Dadhwal, V.K. National Wheat Acreage Estimation for 1995-96 Using Multi-Date IRS-1C WiFS Data. J. Indian Soc. Remote Sens. 1996, 24, 243–254. [CrossRef] 38. Gulati, A.; Tewary, P.; Hussain, S. Crop Insurance in India: Key Issues and Way Forward; Working Paper No. 352; Indian Council for Research on International Economic Relations: New Delhi, India, 2018.
© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).