Conference PaperPDF Available

Ensemble of Cubist models for soy yield prediction using soil features and remote sensing variables

Authors:

Abstract and Figures

The goal of this work is to develop a predictive model for selecting elite soy variants for commercial production. Current breeding practices for new soy variants require rigorous evaluation over three stages of field tests, corresponding to three successive growing seasons. We propose to leverage machine learning methods for identifying high yielding variants using remote sensing and soil features. To support this proposition, we trained an ensemble of fifteen decision tree models, one for each relative maturity band. Collectively, our models identified fifteen elite varieties from 21 predictive variables to forecast soybean yields in 2015 at 58 test locations. This method can boost commercial soy yields by about 5% and shorten the time for commercial variant development.
Content may be subject to copyright.
Ensemble of Cubist models for soy yield prediction using
soil features and remote sensing variables
Tzvi Aviv
AvivInnovation
Toronto, ON, Canada
tzvi@avivinnovation.com
Vanessa Lundsgaard-Nielsen
University of Toronto
Toronto, ON, Canada
vanessa.nielsen@mail.utoronto.ca
ABSTRACT
The goal of this work is to develop a predictive model for
selecting elite soy variants for commercial production. Current
breeding practices for new soy variants require rigorous
evaluation over three stages of field tests, corresponding to three
successive growing seasons. We propose to leverage machine
learning methods for identifying high yielding variants using
remote sensing and soil features. To support this proposition, we
trained an ensemble of fifteen decision tree models, one for each
relative maturity band. Collectively, our models identified fifteen
elite varieties from 21 predictive variables to forecast soybean
yields in 2015 at 58 test locations. This method can boost
commercial soy yields by about 5% and shorten the time for
commercial variant development.
CCS Concepts
CCS Applied computing Computers in other
domains Agriculture
Keywords
yield prediction; decision trees; cubist; remote sensing; NDVI;
Land Surface Temperature; MODIS; Random Forest .
1. INTRODUCTION
Crop yields are highly variable across fields as a result of complex
interactions among factors such as environmental conditions, soil
properties, management practices, and disease and pest attack [3,
10, 4]. In particular, environmental conditions play a vital role in
yield variability, and crop productivity is limited by water
availability, light, temperature, and nutrients [15,21]. Here, we
augmented the soil and weather data supplied by Syngenta with
publicly available remote sensing data and selected 21 variables
that were the most predictive of soybean yield. Our solution uses
vegetation indices calculated from satellite images to predict crop
yields on a site-by-site basis. Vegetation indices, such as the
normalized difference vegetation index (NDVI) have been widely
used for agricultural mapping and monitoring and are calculated
using the red and near-infrared wavelengths [5]. In the last
decade, remote sensing-focused yield forecasting has shifted to
the National Aeronautics and Space Administration’s (NASA)
Moderate Resolution Imaging Spectroradiometer (MODIS),
which provide spatial data at a fine resolution [8,24].
Soybean variants are classified according to relative
maturity (RM), which reflects the time it takes a variety to reach
physiological maturity [2]. Depending on the day length and
flowering time, RM is designated with a numerical system
ranging from 0 to 10 from north to south [2, 12]. Effectively, RM
is an indicator of the fit of genetically coded flowering time to the
environmental conditions in geographic locations. The dataset
provided in this challenge includes yield data for variants mostly
in RM bands ranging from 2.0 to 3.5 covering the soy regions of
the US Midwest (Indiana, Nebraska, Illinois, Iowa and
Minnesota). The differences in yields, weather and soil conditions
across RM bands prompted us to avoid a global yield model for
all RM bands, and we opted instead to generate models for each
RM band, creating an ensemble of predictive models, one for each
RM band. We applied decision tree methods to explore the
predictive properties of these variables. Other reported work
explore correlation, multiple linear regression, decision trees and
neural networks to explore potential predictive models for crop
yields with a varying degree of success [20, 9, 1]. Our approach
differs from the existing body of published work by predicting
soybean yields for specific variants of interest at the farm site
level.
2. CRITERIA FOR ELITE VARIETIES
An ‘elite’ variety is stably producing high soy yields across years
and locations. Our primary goal was the development of
quantitative models for predicting soy yields of each variant and
the identification of elite soy variants. We used the median yield
of commercial (‘CHECK’) varieties in each RM band as a
benchmark for comparison (Table 1) and selected 15 elite
varieties exhibiting yields 2.5 bu/ac higher than corresponding
commercial varieties. These elite lines are expected to boost soy
yields by about 4.5% relative to commercial varieties, in each
RM. Time series analysis of these elite variants indicates stable
productivity over the course of the study and reinforces their elite
designation.
3. ESTIMATES OF TYPE I ERRORS
The current process of seed selection is exposed to potential Type
I errors in which variants are designated as highly productive elite
lines, yet fail to produce high yields in subsequent years. To
evaluate the potential of our yield analytic procedure to reduce
Type I errors we examined the potential utility of our Cubist
ensemble (described in details in the next section) to reduce Type
I error by applying our model to the class of 2013 to predict yields
in 2014 using soil and remote sensing variables (RMSE=1.16,
correlation =0.84, Table 2). When a similar 2.5 bu/ac gain over
RM matched check lines is used as a cutoff to select elite lines,
our algorithm eliminated six soy lines from the elite list due to
insufficient yield consistency. We conclude that our predictive
analytics method can successfully ‘weed out’ non-elite lines and
enhance precision in soy seed selection by Syngenta.
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. To copy
otherwise, or republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee.
KDD’17, Month 8, 2017, Hallifax, N.S., Canada.
Copyright 2010 ACM 1-58113-000-0/00/0010 …$15.00.
DOI: http://dx.doi.org/10.1145/12345.67890
Table 1. Predicted yields of elite soy variants (2014 class)
RM
VARIETY_ID
Predicted
Check
Yield
Yield
Yield
(2015)
Gain
2
V140364
60.9
55.1
5.8
2.1
V140393
56.9
54.2
2.8
2.2
V111237
60.5
55.9
4.6
2.5
V114553
60.7
58
2.7
2.5
V114655
61.3
58
3.3
2.6
V114569
62.9
60.2
2.6
2.7
V114530
62.3
59.8
2.6
2.8
V114564
63.8
60.9
2.9
2.8
V114585
63.6
60.9
2.7
2.8
V152312
63.5
60.9
2.6
3
V114565
62.1
59.3
2.9
3.1
V114589
62.8
59.5
3.3
3.1
V152320
64.1
59.5
4.6
3.5
V152324
62.9
59.7
3.1
3.5
V152415
62.4
59.7
2.7
Yields are reported in bu/ac
4. METHODOLOGY
We supplemented soil features provided by Syngenta with
publicly available remote sensing data. Reflectance of vegetation
in multiple light spectrums (vegetation indices) is a good
estimator of crop yield, fruit ripening, biomass, and plant
senescence [11,18, 25].Vegetation indices can be calculated from
space and are conveniently supplied by NASA in Moderate
Resolution Imaging Spectroradiometer (MODIS) and Normalized
Difference Vegetation Index (NDVI) data products. Here we use a
smoothed and gap-filled NDVI product generated for the
conterminous US for the period 2012-Jan-01 through 2015-Dec-
31. The data was obtained from the NASA Stennis Time Series
Product Tool (TSPT) generating smoothed NDVI data from both
the Terra satellite (MODIS MOD13Q1 product) and Aqua
satellite (MODIS MYD13Q1 product) instruments. The data is
provided at a spatial resolution of 250 m and a temporal resolution
of 8 calendar days. NDVI was chosen over other vegetation
indices due to its ability to cancel out a proportion of variability
caused by changing sun angles, topography, clouds, and various
atmospheric conditions [17]. We merged data from locations
identical in their longitude, latitude, soil, and climatic variables
(six sites). We then focused on sites that were present only in
stage three of the class of 2014, thus NDVI data was collected for
58 locations. For each growing season per each location, we
determined maximal NDVI value (NDVIMAX, in a 0 to 1 scale),
the timing in the year of the NDVI peak (MAXTIME, and the
sum of all NDVI values (SUMNDVI).
Surface temperature and soil moisture can be estimated
successfully from space and provide valuable variables for soy
yield prediction [13]. We obtained daytime and nighttime land
surface temperatures (LST) from MODIS (product MOD11A1).
This data was downloaded at a spatial resolution of 1 km and a
temporal resolution of 8 calendar days for the 58 locations of
interest. Plotting the NDVI and LST time series data reveals the
expected seasonal periodicity in the remote sensing NDVI and
LST data and the potential for yield prediction. The potential of
remote sensing to explore correlations of surface temperatures and
biomass production are illustrated in Figure 1 using remote
sensing data from one location (location 4370) as an example. In
this location low soy yields in 2012 are correlated with high day
time temperatures and low NDVI scores (Figure 1).
Figure 1: Yields (top), LST, and NDVI (bottom) in location
4370.
We extracted several features from the NDVI and LST time-series
and performed correlation analyses to select features that correlate
with soy yields (Figure 2). We found a positive correlation of
NDVIMAX with soy yield and a negative correlation of
JULYSUM and AUGSUM with soy yield. In addition strong
negative correlation of NDVIMAX with JULYSUM and
AUGSUM are noticeable. JULYSUM and AUGSUM are also
strongly correlated with each other. The sensitivity of soy to high
temperature is well known, and we surveyed cumulative monthly
daytime temperatures and minimum nightly temperatures as
potential predictors of soy yield [6,7].We identified cumulative
daytime temperature in July and August (JULYSUM, AUGSUM),
and monthly minimum nighttime temperatures in May and July
(MAYMIN, JULYMIN), as most inverse correlated with yields
and selected these variables as predictive variables in our models
(Figure 2).
Figure 2: Correlation plots of remote sensing variables and
soy yields.
Figure 3: Correlation plots of soil variables and soy yield.
We also surveyed the correlations of soil variables with each other
and with soy yields (Figure 3). The strongest correlation was
observed with the irrigation variable. We decided to retain all soil
variables in our models, despite strong correlations of some soil
variables with each other.
Decision tree based methods are versatile and powerful machine
learning techniques that are well suited for this challenge. Our
preliminary analysis compared random forests [14] and Cubist
[22] models for feature discovery and model tuning in the R
software package [23]. We used the training data set that we
developed above to compare the predictive potential of the
randomForest and Cubist packages in R. With minimal tuning,
randomForest models achieved lower error rates relative to Cubist
models. However, when random Forest models where cross
validated using data outside of the training set (or when applied to
real test data on the CodaLab portal) we noticed poorer predictive
power relative to Cubist models (Random Forest models achieved
lower scores on the CodaLab platform). We postulate that
Random Forest could be over fitting the models to the data and we
therefore present here only our Cubist models.
5. QUANTITATIVE RESULTS
We used the Cubist package in R to generate regression decision
trees for yields in each RM band using 21 predictor variables
(VARIETY_ID, LATITUDE, LONGITUDE, FIPS, AREA,
IRRIGATION, CEC, PH, ORGANIC.MATTER, CLAY,
SILT_TOP, SAND_TOP, AWC_100CM, NDVIMAX,
MAXTIME, SUMNDVI, JULYSUM, AUGSUM, MAYMIN,
JULYMIN, FAMILY). The correlations of these models with the
training data varied from 0.3 to 0.85 (Table 3). Out of fifteen RM
bands with sufficient number of cases for modeling, we generated
three high quality models (correlation coefficient ~0.85), eight
medium quality models (correlation coefficient~0.7), and four
models with poor predictive power (<0.4). Low quality models
were typically generated when smaller datasets were used for
training (average of 114 cases) and are thus less reliable for yield
prediction. Generation of more experimental data for variants in
RM bands with low data concentration is recommended to
enhance RM coverage by elite soy variants (RMs: 2.3, 2.9, 3.1,
3.3 and 3.6). We validated this modeling procedure by applying
this methodology to ‘predict’ the yields of the class of 2013 in
2014 (see section 3, Table 2). We demonstrated the utility of the
‘divide-and-conquer’ procedure to (a) predict yields from soil and
remote data and (b) select consistently high yielding soy variants.
We than used these models to predict the yields of 28 soy variants
grown in 58 locations in 2015 (Table 1, using soil and remote
sensing variables) and obtained high prediction accuracy on the
Coda-Lab portal (0.45 FMEASURE, 0.99 ACCURACY, 0.47
MATHEWSCC) reflecting the predictive value of our approach.
One advantage of using modern decision tree tools is
the interpretable ranking of variables according to their
contribution to the model. Here we list top four variables in each
RM model (Table 3). Notably, the high variability in ranking
orders likely reflects the high heterogeneity among different RM
bands and supports our decision to segment the data according to
RM. Grouping RM into proximity zones (RM 2.0-2.5, RM2.6-2.9,
RM 3.0-RM3.5) drove down internal RMSE values to about 4.
However the Codalab score generated from RM zones models
were below scores obtained with the separate RM models. Future
work will focus on (a) adding additional remote sensing variables
to our models to enhance precision and (b) utilizing time-series
prediction methods to extrapolate future NDVI and LST values
using data trends in each location to allow true predictions of
future yield values.
6. ACKNOWLEDGMENTS
We thank Syngenta for providing data and funding for this
project, IdeaConnection and AI For Good for organizing this
competition and the R community for developing and maintaining
the tools used in this analysis.
7. REFERENCES
[1] Ainong, L., Shunlin, L., Angsheng, W., Jun, Q. (2007)
Estimating Crop Yield from Multi-temporal Satellite Data
Using Multivariate Regression and Neural Network
Techniques. Photogrammetric Engineering & Remote
Sensing, 10, 1149-1157
[2] Alliprandini, L. F., C. Abatti, P. F. Bertagnolli, J. E.
Cavassim, H. L. Gabe, A. Kurek, M. N. Matsumoto, M. A.
R. de Oliveira, C. Pitol, L. C. Prado, and C. Steckling. 2009.
Understanding Soybean Maturity Groups in Brazil:
Environment, Cultivar Classification, and Stability Crop Sci.
49: 801-808.
[3] Al-Kaisi, M.M., Elmore, R.W., Guzman, J.G., Hanna, H.M,
Hart, C.E., Helmers, M.J., Hodgson, E.W., Lenssen, A.W.,
Mallarino, A.P., Robertson, A.E., and Sawyer, J.E. 2012
Drought impact on crop production and the soil environment:
2012 experiences from Iowa. Journal of Soil and Water
Conservation. 68, 19 24
[4] Altieri, M.A., Nicholls, C.I., Henao, A. et al. Agroecology
and the design of climate change-resilient farming systems
(2015) Agron. Sustain. Dev. 35, 869 890.
[5] Das, A., Qi, J. T., Oneto, M., Shere, S. & Ma, Z. Soybean
Yield Prediction: Using Satellite Imagery to Predict
Commodity Yields. (2016).
[6] Djanaguiraman, M., P. V V Prasad, D. L. Boyle, and W. T.
Schapaugh. 2013. “Soybean Pollen Anatomy, Viability and
Pod Set under High Temperature Stress.” Journal of
Agronomy and Crop Science 199: 17177.
[7] Djanaguiraman, M, P.V. Vara Prasad, and W. T. Schapaugh.
2013. “High Day- or Nighttime Temperature Alters Leaf
Assimilation, Reproductive Success, and Phosphatidic Acid
of Pollen Grain in Soybean [Glycine Max (L.) Merr.].” Crop
Science 53(4): 15941604.
[8] Doraiswamy, P.C., Hatfield, J.L., Jackson, T.J., Akhmedov,
B., Prueger, J., Stern, A., (2004) Crop condition and yield
simulations using Landsat and MODIS. Remote Sens.
Environ, 92, 548559
[9] Elizondo, D.A., McClendon, R. W., Hoogenboom, G (1994)
Neural Network Models for Predicting Flowering and
Physiological Maturity of Soybean. American Society of
Agricultural and Biological Engineers 37(3): 981-988.
[10] Hartman, G.L. Chang, H.X., Leandro, L.F. (2015) Research
advances and management of soybean sudden death
syndrome. Crop Protection. 73, 60 66.
[11] Hatfield, J.L., Prueger, J.H., 2010. Value of using different
vegetative indices to quantify agricultural crop characteristics
at different growth stages under varying management
practices. Remote Sens. 2, 562578.
[12] Henderson Communications LLC “Syngenta Seeds Develops
First Ever Soybean Relative maturity Map for Canada (2009)
AgriMarketing Global Hub for Agribusiness
[13] Holzman, M. E., Rivas, R., & Piccolo, M. C. (2014).
Estimating soil moisture and the relationship with crop yield
using surface temperature and vegetation index. International
Journal of Applied Earth Observation and Geoinformation,
28, 181-192.
[14] Liaw A. and Wiener M.(2002). Classification and Regression
by randomForest. R News 2(3), 18--22.
[15] Lobell, David B, and Claudia Tebaldi. 2014. “Getting Caught
with Our Plants down: The Risks of a Global Crop Yield
Slowdown from Climate Trends in the next Two Decades.”
Environmental Research Letters 9(7): 74003.
[16] Maccherone, B. & Frazier, S. MODIS Vegetation Index
Products (NDVI and EVI). Available at:
https://modis.gsfc.nasa.gov/data/dataprod/mod13.php.
[17] Matsushita, B., Yang, W., Chen, J., Onda, Y. & Qiu, G.
Sensitivity of the Enhanced Vegetation Index (EVI) and
Normalized Difference Vegetation Index (NDVI) to
Topographic Effects: A Case Study in High-density Cypress
Forest. Sensors 7, 26362651 (2007).
[18] Merzlyak,M.N., Gitelson, A.A., Chivkunova, O.B., Rakitin,
V.Y., 1999. Non-destructive opti- cal detection of pigment
changes during leaf senescence and fruit ripening. Physiol.
Plant. 106, 135141.
[19] Myers, S. S. et al. Climate Change and Global Food Systems:
Potential Impacts on Food Security and Undernutrition.
Annu. Rev. Public Health 38, 259277 (2017).
[20] Norouzi, M., Ayoubi, S., Jalalian, A., Khademi, H.,
Dehghani, A.A. Predicting rainfed wheat quality and quantity
by artificial neural network using terrain and soil
characteristics (2010) Acta Agriculturae Scandinavica 60,
341 - 352
[21] Powell, Nicola et al. 2012. “Yield Stability for Cereals in a
Changing Climate.” Functional Plant Biology 39(7): 539–52.
[22] Quinlan, J. R. (1992). Learning with continuous classes. In
5th Australian joint conference on artificial intelligence (Vol.
92, pp. 343-348).
[23] R Core Team (2013). R: A language and environment for
statistical computing. R Foundation for Statistical
Computing, Vienna, Austria.URL: http://www.R-
project.org/.
[24] Ren, J., Chen, Z., Zhou, Q., Tang, H. (2008) Regional yield
estimation for winter wheat with MODIS-NDVI data in
Shandong China. Int. J. Appl. Earth Obs. Geoinf, 10, 403
413
[25] Yu, N., Li, L., Schmitz, N., Tian, L.F., Greenberg, J.A.,
Diers, B.W., Development of methods to improve soybean
yield estimation and predict plant maturity with an unmanned
aerial vehicle based platform (2016) Remote Sensing
Environment 187, 97 - 10.
Table 2. Error estimates in the Class of 2013
RM
VARIETY_ID
FAMILY
Prediction
(2014)
Check
Yield
Gain
ELITE
CALL
Yields
(2014)
2.1
V156516
FAM14408
58.2
54.2
4.0
TRUE
60.2
2.1
V156786
FAM11169
58.2
54.2
4.0
TRUE
60.6
2.4
V156783
FAM11183
58.8
56.9
2.0
FALSE
58.8
2.5
V156565
FAM14238
63.1
58.0
5.1
TRUE
62.8
2.6
V156806
FAM11189
63.2
60.2
3.0
TRUE
64.1
2.7
V152079
FAM11179
62.3
59.8
2.5
TRUE
62.8
2.7
V156574
FAM14486
62.6
59.8
2.8
TRUE
62.1
2.7
V156807
FAM11189
62.6
59.8
2.8
TRUE
61.4
2.9
V156553
FAM14238
62.4
60.7
1.7
FALSE
61.7
3
V152053
FAM14333
60.7
59.3
1.5
FALSE
60.0
3
V156642
FAM14133
58.8
59.3
-0.4
FALSE
56.5
3.2
V156763
FAM06492
61.5
60.3
1.3
FALSE
61.4
3.2
V156797
FAM06502
61.5
60.3
1.3
FALSE
62.0
3.4
V152061
FAM14581
63.6
61.0
2.6
TRUE
63.2
3.5
V156774
FAM14774
65.1
59.7
5.4
TRUE
65.7
Yields are reported in bu/ac
Table 3: Evaluation of Soy Yield Models
RM
N
RMSE
Relative Error
Cor. Coeff.
Top Contributing Variables
Model Quality
2
115
8.9
1.01
0.48
organic.matter, augsum, area, julysum
Medium
2.1
257
4.14
0.53
0.85
sumndvi, julysum,ndvimax,maymin
High
2.2
211
7.28
0.82
0.64
ndvismax,julysum,maymin,augsum
Medium
2.3
57
2.4
198
9.47
1.06
0.43
sand_top, silt_top, julysum,ndvismax
Low
2.5
358
5.18
0.66
0.76
julysum,clay,ndvimax,longitude
Medium
2.6
190
7.05
0.89
0.56
maymin,julysum,sumndvi,longitude
Medium
2.7
426
4.94
0.61
0.76
clay,pH,sumndvi, irrigation
Medium
2.8
402
5.11
0.65
0.76
clay,pH, irrigation, cec
Medium
2.9
143
8.79
1.09
0.29
irrigation,cec,maymin,julymin
Low
3
395
6.39
0.71
0.71
maymin,julysum,irrigation,cec
Medium
3.1
105
11.08
1.12
0.23
latitude, sumndvi,fips,julymin
Low
3.2
310
4.9
0.52
0.84
irrigation, sand_top,maxtime,augsum
High
3.3
0
3.4
442
4.64
0.51
0.85
julysum, cec, sand_top, organic.matter
High
3.5
256
6.23
0.67
0.75
cec, julysum, ndvimax, maymin
Medium
3.6
161
9.45
1.13
0.3
julysum,awc_100cm,latitude, julymin
Low
... In cropping situations in New Zealand grid sampling is rarely undertaken at points less than one value per hectare. The Cubist regression model is primarily used in remote sensing studies for handling large datasets [14,15]. Cubist is a rule-based decision tree algorithm, modified from Ross Quinlan's M5 model tree [16,17] and introduced to the R statistical software by Max Kuhn in the R development group [18]. ...
... A specific set of predictor variables will then choose an actual prediction model based on the rule that best fits the predictors [18,19]. Promising results have been reported when predicting continuous variables [14,15]. ...
... Two rain-fed, non-irrigated sites in Waikato were chosen for this study, see Figure 2 because of their consistent within-site management histories, where the grain harvester was equipped with a yield monitor. based on the rule that best fits the predictors [18,19]. Promising results have been reported when predicting continuous variables [14,15]. This paper aims to examine the use of statistical modelling techniques on precision farming data for estimating within-paddock maize-grain yield potential. ...
... In cropping situations in New Zealand grid sampling is rarely undertaken at points less than one value per hectare. The Cubist regression model is primarily used in remote sensing studies for handling large datasets [14,15]. Cubist is a rule-based decision tree algorithm, modified from Ross Quinlan's M5 model tree [16,17] and introduced to the R statistical software by Max Kuhn in the R development group [18]. ...
... A specific set of predictor variables will then choose an actual prediction model based on the rule that best fits the predictors [18,19]. Promising results have been reported when predicting continuous variables [14,15]. ...
Article
Full-text available
Spatial variability in soil, crop, and topographic features, combined with temporal variability between seasons can result in variable annual yield patterns within a paddock. The complexity of interactions between yield-limiting factors such as soil nutrients and soil water require specialist statistical processing to be able to quantify variability, and thus inform crop management practices. This study uses multiple linear regression models, Cubist regression and feed-forward neural networks to predict spatial maize-grain (Zea mays) yield at two sites in the Waikato Region, New Zealand. The variables considered were: crop reflectance data from satellite imagery, soil electrical conductivity, soil organic matter, elevation, rainfall, temperature, solar radiation, and seeding density. This exercise explores methods which may be useful in predicting yield from proximal and remote sensed data with higher resolution than traditional low spatial resolution point sampling using soil testing and yield response curves.
... This method can boost commercial soy yields by about 5% and shorten the time for commercial variant development. (Aviv 2018) So, the technology helps maximize yield for specific soil conditions, and speeds crop optimization, which serves SDGs 1 and 2 (Poverty, Hunger), as well as 15, which seeks to protect, restore and promote sustainable use of terrestrial ecosystems. Similar points could be made about the company's Farm360AI platform which predicts corn and soybean yields from satellite imagery and weather data. ...
Article
Full-text available
Does AI conform to humans, or will we conform to AI? An ethical evaluation of AI-intensive companies will allow investors to knowledgeably participate in the decision. The evaluation is built from nine performance indicators that can be analyzed and scored to reflect a technology’s human-centering. The result is objective investment guidance, as well as investors empowered to act in accordance with their own values. Incorporating ethics into financial decisions is a strategy that will be recognized by participants in environmental, social, and governance investing, however, this paper argues that conventional ESG frameworks are inadequate to companies that function with AI at their core. Fully accounting for contemporary big data, predictive analytics, and machine learning requires specialized metrics customized from established AI ethics principles. With these metrics established, the larger goal is a model for humanist investing in AI-intensive companies that is intellectually robust, manageable for analysts, useful for portfolio managers, and credible for investors.
... The Cubist regression model is a technique primarily used in remote sensing studies for handling large datasets and has, in the past, reported promising results when predicting continuous variables. It has a faster training speed than other computationally intensive machine learning methods such as random forest and neural networks (Aviv and Lundsgaard-Nielsen, 2017;Noi et al., 2017). For MLR, the dependent variable (yield) needs to be normally distributed. ...
Conference Paper
Full-text available
Spatial variability in soil, crop, and topographic features, combined with temporal variability in weather can result in variable annual yield patterns within a paddock. The complexity of interactions between these yield-limiting factors requires specialist statistical processing to be able to quantify spatial and temporal variability, and thus inform crop management practices. This paper evaluates the role of multivariate linear regression and a Cubist regression model to predict spatial variability of maize-grain yield at two sites in the Waikato Region, New Zealand. The variables considered were: crop reflectance data from satellite imagery (Sentinel 2 and Landsat 8), soil electrical conductivity (EC), soil organic matter (OM), elevation, rainfall, temperature, solar radiation, and seeding density. The datasets were split into training and validation sets, proportionally 75% and 25% respectively. Both models learn using 10-fold cross-validation. Statistical performance was evaluated by leaving out one year of yield data as the validation set for each iteration, with all remaining years included in the training set for building the prediction models. In the multiple-year analysis, the Cubist model (RMSE=1.47 and R 2 =0.82 for site 1; RMSE=2.13 and R 2 =0.72 for site 2) produced a better statistical prediction than the MLR model (RMSE=2.41 and R 2 =0.51 for site 1; RMSE=3.37 and R 2 =0.30 for site 2) for the prediction of the validation set. However, for the leave-one-year-out analyses, the MLR model provided better statistical predictions (RMSE=1.57 to 4.93; R 2 = 0.15 to 0.31) than the Cubist model (RMSE = 2.62 to 5.9; R 2 = 0.05 to 0.14) for Site 1. For Site 2, both models produced poor results. Yield data for additional years and inclusion of more independent variables (e.g. soil fertility and texture) may improve the models. This analysis demonstrates that there is potential to use statistical modelling of spatial and temporal data to assist farm management decisions (e.g. variable rate application, precision land levelling, irrigation, and drainage). Once the functional relationship between within-paddock yield potential and complementary variables is established, it should be possible to provide an accurate management prescription, enabling variable rates of an input (e.g. plant density, fertiliser) to be applied automatically across the paddock based on the "yield-input" response curve.
... The Cubist regression model is a technique primarily used in remote sensing studies for handling large datasets and has, in the past, reported promising results when predicting continuous variables. It has a faster training speed than other computationally intensive machine learning methods such as random forest and neural networks (Aviv and Lundsgaard-Nielsen, 2017;Noi et al., 2017). For MLR, the dependent variable (yield) needs to be normally distributed. ...
Conference Paper
Spatial variability in soil, crop, and topographic features, combined with temporal variability in weather can result in variable annual yield patterns within a paddock. The complexity of interactions between these yield-limiting factors requires specialist statistical processing to be able to quantify spatial and temporal variability, and thus inform crop management practices. This paper evaluates the role of multivariate linear regression and a Cubist regression model to predict spatial variability of maize-grain yield at two sites in the Waikato Region, New Zealand. The variables considered were: crop reflectance data from satellite imagery (Sentinel 2 and Landsat 8), soil electrical conductivity (EC), soil organic matter (OM), elevation, rainfall, temperature, solar radiation, and seeding density. The datasets were split into training and validation sets, proportionally 75% and 25% respectively. Both models learn using 10-fold cross-validation. Statistical performance was evaluated by leaving out one year of yield data as the validation set for each iteration, with all remaining years included in the training set for building the prediction models. In the multiple-year analysis, the Cubist model (RMSE=1.47 and R 2 =0.82 for site 1; RMSE=2.13 and R 2 =0.72 for site 2) produced a better statistical prediction than the MLR model (RMSE=2.41 and R 2 =0.51 for site 1; RMSE=3.37 and R 2 =0.30 for site 2) for the prediction of the validation set. However, for the leave-one-year-out analyses, the MLR model provided better statistical predictions (RMSE=1.57 to 4.93; R 2 = 0.15 to 0.31) than the Cubist model (RMSE = 2.62 to 5.9; R 2 = 0.05 to 0.14) for Site 1. For Site 2, both models produced poor results. Yield data for additional years and inclusion of more independent variables (e.g. soil fertility and texture) may improve the models. This analysis demonstrates that there is potential to use statistical modelling of spatial and temporal data to assist farm management decisions (e.g. variable rate application, precision land levelling, irrigation, and drainage). Once the functional relationship between within-paddock yield potential and complementary variables is established, it should be possible to provide an accurate management prescription, enabling variable rates of an input (e.g. plant density, fertiliser) to be applied automatically across the paddock based on the "yield-input" response curve.
Article
Full-text available
Great progress has been made in addressing global undernutrition over the past several decades, in part because of large increases in food production from agricultural expansion and intensification. Food systems, however, face continued increases in demand and growing environmental pressures. Most prominently, human-caused climate change will influence the quality and quantity of food we produce and our ability to distribute it equitably. Our capacity to ensure food security and nutritional adequacy in the face of rapidly changing biophysical conditions will be a major determinant of the next century's global burden of disease. In this article, we review the main pathways by which climate change may affect our food production systems-agriculture, fisheries, and livestock-as well as the socioeconomic forces that may influence equitable distribution. Expected final online publication date for the Annual Review of Public Health Volume 38 is March 20, 2017. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.
Article
Full-text available
Diverse, severe, and location-specific impacts on agricultural production are anticipated with climate change. The last IPCC report indicates that the rise of CO 2 and associated " greenhouse " gases could lead to a 1.4 to 5.8 °C increase in global surface temperatures, with subsequent consequences on precipitation frequency and amounts. Temperature and water availability remain key factors in determining crop growth and productivity; predicted changes in these factors will lead to reduced crop yields. Climate-induced changes in insect pest, pathogen and weed population dynamics and in-vasiveness could compound such effects. Undoubtedly, climate-and weather-induced instability will affect levels of and access to food supply, altering social and economic stability and regional competiveness. Adaptation is considered a key factor that will shape the future severity of climate change impacts on food production. Changes that will not radically modify the monoculture nature of dominant agroecosystems may moderate negative impacts temporarily. The biggest and most durable benefits will likely result from more radical ag-roecological measures that will strengthen the resilience of farmers and rural communities, such as diversification of agroecosytems in the form of polycultures, agroforestry systems , and crop-livestock mixed systems accompanied by organic soil management, water conservation and harvesting, and general enhancement of agrobiodiversity. Traditional farming systems are repositories of a wealth of principles and measures that can help modern agricultural systems become more resilient to climatic extremes. Many of these ag-roecological strategies that reduce vulnerabilities to climate variability include crop diversification, maintaining local genetic diversity, animal integration, soil organic management, water conservation and harvesting, etc. Understanding the ag-roecological features that underlie the resilience of traditional agroecosystems is an urgent matter, as they can serve as the foundation for the design of adapted agricultural systems. Observations of agricultural performance after extreme climatic events (hurricanes and droughts) in the last two decades have revealed that resiliency to climate disasters is closely linked to farms with increased levels of biodiversity. Field surveys and results reported in the literature suggest that agroecosystems are more resilient when inserted in a complex landscape matrix, featuring adapted local germplasm deployed in diversified cropping systems managed with organic matter rich soils and water conservation-harvesting techniques. The identification of systems that have withstood climatic events recently or in the past and understanding the agroecological features of such systems that allowed them to resist and/or recover from extreme events is of increased urgency , as the derived resiliency principles and practices that underlie successful farms can be disseminated to thousands of farmers via Campesino a Campesino networks to scale up agroecological practices that enhance the resiliency of agroecosystems. The effective diffusion of agroecological technologies will largely determine how well and how fast farmers adapt to climate change.
Article
Full-text available
In many discussions of climate change impacts in agriculture, the large magnitudes of expected impacts toward the end of the century are used to emphasize that most of the risks are to future generations. However, this perspective misses the important fact that demand growth for food is expected to be much slower after 2050 than before it, and that the next two decades represent the bulk of growth before 2050. Thus, impacts of smaller magnitude in the near-term can be as or more consequential for food prices or food security as larger magnitude impacts in the future. Here we estimate the risks that climate trends over the next 10 or 20 years could have large impacts on global yields of wheat and maize, with a focus on scenarios that would cut the expected rates of yield gains in half. We find that because of global warming, the chance of climate trends over a 20 year period causing a 10% yield loss has increased from a less than 1 in 200 chance arising from internal climate variability alone, to a 1 in 10 chance for maize and 1 in 20 chance for wheat. Estimated risks for maize are higher because of a greater geographic concentration than wheat, as well as a slightly more negative aggregate temperature sensitivity. Global warming has also greatly increased the chance of climate trends large enough to halve yield trends over a 10 year period, with a roughly 1 in 4 chance for maize and 1 in 6 chance for wheat. Estimated risks are slightly larger when using climate projections from a large ensemble of a single climate model that more fully explores internal climate variability, than a multi-model ensemble that more fully explores model uncertainty. Although scenarios of climate impacts large enough to halve yield growth rates are still fairly unlikely, they may warrant consideration by institutions potentially affected by associated changes in international food prices.
Article
Full-text available
The United Nations Food and Agriculture Organisation (FAO) forecasts a 34% increase in the world population by 2050. As a consequence, the productivity of important staple crops such as cereals needs to be boosted by an estimated 43%. This growth in cereal productivity will need to occur in a world with a changing climate, where more frequent weather extremes will impact on grain productivity. Improving cereal productivity will, therefore, not only be a matter of increasing yield potential of current germplasm, but also of improving yield stability through enhanced tolerance to abiotic stresses. Successful reproductive development in cereals is essential for grain productivity and environmental constraints (drought, cold, frost, heat and waterlogging) that are associated with climate change are likely to have severe effects on yield stability of cereal crops. Currently, genetic gains conferring improved abiotic stress tolerance in cereals is hampered by the lack of reliable screening methods, availability of suitable germplasm and poor knowledge about the physiological and molecular underpinnings of abiotic stress tolerance traits.
Article
Full-text available
Soybean [Glycine max (L.) Merr.] is often exposed to high daytime and nighttime temperatures during critical growth stages. Threshold mean daily temperature for photosynthesis, respiration, and reproductive process in soybean is >= 26 degrees C. In future, the magnitude of increase in nighttime temperatures will be greater than in daytime temperatures. The objectives were to determine effects of high daytime or nighttime temperatures on (i) leaf photosynthetic and respiration rates; (ii) pollen germination, pod-set, and seed weight; and (iii) pollen phospholipids profile. Soybean plants were exposed to high daytime temperature (39/20 degrees C), high nighttime temperatures (30/23 degrees C, 30/26 degrees C, and 30/29 degrees C), or optimum temperature (30/20 degrees C) for 10 d at flowering stage. High daytime temperature (39/20 degrees C) or nighttime temperatures (30/29 degrees C) increased leaf respiration rates and decreased leaf chlorophyll content, photosynthetic rate, photochemical quenching, and electron transport rate compared to optimum temperature. Likewise, high temperature decreased pollen viability and germination. Lower pollen germination at high temperature may be due to decreased levels of saturated phospholipids and phosphatidic acid in pollen grains compared with optimum temperature. Pod-set and seed weight were decreased by high daytime or nighttime temperature. In conclusion, high daytime (39/20 degrees C) or nighttime (30/29 degrees C) temperature decreased leaf photosynthetic rate and pollen germination, leading to lower pod-set and seed weight.
Article
It is important for farmers to know when various plant development stages occur for making appropriate and timely crop management decisions. Although computer simulation models have been developed to simulate plant growth and development, these models have not always been very accurate in predicting plant development for a wide range of environmental conditions. The objective of this study was to develop a neural network model to predict flowering and physiological maturity for soybean (Glycine max L. Merr.). An artificial neural network is a computer software system consisting of various simple and highly interconnected processing elements similar to the neuron structure found in the human brain. A neural network model was used because it has the capabilities to identify relationships between variables of rather large and complex data bases. For this study, field-observed flowering dates for the cultivar 'Bragg' from experimental studies conducted in Gainesville and Quincy, Florida, and Clayton, North Carolina, were used. Inputs considered for the neural network model were daily maximum and minimum air temperature, photoperiod, and days after planting or days after flowering. The data sets were split into training sets to develop the models and independent data sets to test the models. The average relative error of the test data sets for date of flowering prediction was + 0.143 days (n = 21, R2 = 0.987) and for date of physiological maturity prediction was + 2.19 days (n = 21, R2 = 0.950). It can be concluded from this study that the use of neural network models to predict flowering and physiological maturity dates is promising and needs to be explored further.
Article
Fusarium virguliforme causes soybean sudden death syndrome (SDS) in the United States. The disease was first observed in Arkansas in 1971, and since has been reported in most soybean-producing states, with a general movement from the southern to the northern states. In addition to F. virguliforme, three other species, Fusarium brasiliense, Fusarium crassistipitatum, and Fusarium tucumaniae, have been reported to cause SDS in South America. Yield losses caused by F. virguliforme range from slight to 100%. Severely infected plants often have increased flower and pod abortion, reduced seed size, increased defoliation, and prematurely senescence. Foliar symptoms observed in the field are most noticeable from mid to late reproductive growth stages. To manage SDS, research on crop rotations, soil types, tillage practices, seed treatments, and the development and utilization of host resistance has been investigated. This review focuses on what is known about F. virguliforme, the management of SDS in the United States, and how genetic engineering along with other traditional management options may be needed as integrated approaches to manage SDS.
Article
Accurate, objective, reliable, and timely predictions of crop yield over large areas are critical to helping ensure the adequacy of a nation’s food supply and aiding policy makers on import/export plans and prices. Development of objective mathematical models of crop yield prediction using remote sensing is highly desirable. In this study, we develop a new methodology using an artificial neural network (ANN) to estimate and predict corn and soybean yields on a county-by-county basis, in the “corn belt” area in the Midwestern and Great Plains regions of the United States. The historical yield data and long time-series NDVI derived from AVHRR and MODIS are used to develop the models. A new procedure is developed to train the ANN model using the SCE-UA optimization algorithm. The performance of ANN models is compared with multivariate linear regression (MLR) models and validation is made on the model’s stability and forecasting ability. The new algorithms can effectively train ANN models, and the prediction accuracy can be as high as 85 percent.